From f70c3af1a89c9bf796893871de5b87d654fc2248 Mon Sep 17 00:00:00 2001 From: gouzi <530971494@qq.com> Date: Tue, 14 Jul 2026 22:31:35 +0800 Subject: [PATCH 1/5] retire TensorRT tests --- inference/benchmark/python/paddle/run.sh | 60 -- inference/benchmark/python/tensorflow/run.sh | 30 - inference/benchmark/python/torch/run.sh | 42 -- .../test_case/image_preprocess.py | 2 +- .../python_api_test/test_case/infer_test.py | 512 ---------------- .../python_api_test/test_class_model/run.sh | 3 - .../test_class_model/run_ce_win.sh | 5 +- .../test_class_model/run_parallel.sh | 5 - .../test_DarkNet53_trt_fp16.py | 181 ------ .../test_DarkNet53_trt_fp32.py | 178 ------ .../test_DenseNet121_trt_fp16.py | 189 ------ .../test_DenseNet121_trt_fp32.py | 170 ------ .../test_EfficientNetB0_slim.py | 70 --- .../test_GhostNet_x1_0_slim.py | 70 --- .../test_GoogLeNet_trt_fp16.py | 173 ------ .../test_GoogLeNet_trt_fp32.py | 187 ------ .../test_class_model/test_InceptionV3_slim.py | 70 --- .../test_class_model/test_MobileNetV1_slim.py | 70 --- .../test_MobileNetV3_large_x1_0_slim.py | 70 --- .../test_MobileNetV3_large_x1_0_ssld_slim.py | 70 --- .../test_PPHGNet_tiny_slim.py | 70 --- .../test_PPLCNetV2_base_slim.py | 70 --- .../test_PPLCNet_x1_0_slim.py | 76 --- .../test_class_model/test_ResNet50_vd_slim.py | 70 --- .../test_ShuffleNetV2_x1_0_slim.py | 76 --- .../test_SqueezeNet1_0_slim.py | 70 --- .../test_ViT_base_patch16_224_trt_fp16.py | 173 ------ .../test_ViT_base_patch16_224_trt_fp32.py | 169 ------ .../test_ViT_base_patch16_384_trt_fp16.py | 172 ------ .../test_ViT_base_patch16_384_trt_fp32.py | 170 ------ .../test_ViT_base_patch32_384_trt_fp16.py | 172 ------ .../test_ViT_base_patch32_384_trt_fp32.py | 170 ------ .../test_ViT_large_patch16_224_trt_fp16.py | 173 ------ .../test_ViT_large_patch16_224_trt_fp32.py | 170 ------ .../test_ViT_large_patch16_384_trt_fp16.py | 189 ------ .../test_ViT_large_patch16_384_trt_fp32.py | 170 ------ .../test_ViT_large_patch32_384_trt_fp16.py | 172 ------ .../test_ViT_large_patch32_384_trt_fp32.py | 170 ------ .../test_ViT_small_patch16_224_trt_fp16.py | 170 ------ .../test_ViT_small_patch16_224_trt_fp32.py | 167 ----- .../test_pcpvt_base_trt_fp16.py | 170 ------ .../test_pcpvt_base_trt_fp32.py | 170 ------ .../test_class_model/test_resnet50_slim.py | 40 -- .../test_resnet50_trt_fp16.py | 194 ------ .../test_resnet50_trt_fp32.py | 175 ------ .../test_swin_transformer_trt_fp16.py | 206 ------- .../test_swin_transformer_trt_fp32.py | 190 ------ .../test_tnt_small_trt_fp16.py | 214 ------- .../test_tnt_small_trt_fp32.py | 208 ------- .../test_class_model/test_vgg11_trt_fp16.py | 183 ------ .../test_class_model/test_vgg11_trt_fp32.py | 173 ------ .../python_api_test/test_det_model/run.sh | 6 +- .../test_det_model/run_parallel.sh | 4 - .../test_det_model/test_fast_rcnn_trt_fp16.py | 111 ---- .../test_det_model/test_fast_rcnn_trt_fp32.py | 110 ---- .../test_det_model/test_mask_rcnn_trt_fp16.py | 114 ---- .../test_det_model/test_mask_rcnn_trt_fp32.py | 110 ---- .../test_det_model/test_ppyolo_trt_fp16.py | 123 ---- .../test_det_model/test_ppyolo_trt_fp32.py | 127 ---- .../test_det_model/test_ppyolov2_trt_fp16.py | 114 ---- .../test_det_model/test_ppyolov2_trt_fp32.py | 112 ---- .../test_det_model/test_solov2_trt_fp16.py | 121 ---- .../test_det_model/test_solov2_trt_fp32.py | 118 ---- .../test_det_model/test_yolov3_trt_fp16.py | 264 -------- .../test_det_model/test_yolov3_trt_fp32.py | 260 -------- .../python_api_test/test_int8_model/README.md | 24 - .../test_int8_model/backend/__init__.py | 1 - .../test_int8_model/backend/onnxruntime.py | 62 +- .../backend/paddle_inference.py | 48 -- .../test_int8_model/backend/tensorrt.py | 356 ----------- .../test_int8_model/base_nv_trt_fp16.py | 438 ------------- .../test_int8_model/base_nv_trt_int8.py | 438 ------------- .../test_int8_model/base_trt_fp16.py | 258 -------- .../test_int8_model/base_trt_int8.py | 258 -------- .../test_int8_model/convert_onnx.sh | 128 ---- .../test_int8_model/get_benchmark_info.py | 30 +- .../test_int8_model/requirements.txt | 2 - .../test_int8_model/requirements_no_cuda.txt | 1 - .../python_api_test/test_int8_model/run.sh | 4 - .../test_int8_model/run_nv_trt_fp16.sh | 71 --- .../test_int8_model/run_nv_trt_fp32.sh | 35 -- .../test_int8_model/run_nv_trt_int8.sh | 73 --- .../test_int8_model/run_ocr.sh | 15 - .../test_int8_model/run_trt_fp16.sh | 69 --- .../test_int8_model/run_trt_fp32.sh | 67 -- .../test_int8_model/run_trt_int8.sh | 69 --- .../test_int8_model/test_bert_infer.py | 38 +- .../test_image_classification_infer.py | 74 +-- .../test_int8_model/test_nlp_infer.py | 40 +- .../test_int8_model/test_ocr_infer.py | 106 +--- .../test_int8_model/test_ppyoloe_infer.py | 48 +- .../test_segmentation_infer.py | 43 +- .../test_int8_model/test_yolo_series_infer.py | 55 +- .../python_api_test/test_int8_model/xly.sh | 18 +- .../python_api_test/test_nlp_model/run.sh | 4 - .../test_nlp_model/test_AFQMC_PTQ_trt_int8.py | 542 ----------------- .../test_AFQMC_base_trt_fp16.py | 574 ------------------ .../test_AFQMC_base_trt_fp32.py | 574 ------------------ .../test_nlp_model/test_bert_trt_fp16.py | 121 ---- .../test_nlp_model/test_bert_trt_fp32.py | 121 ---- .../test_nlp_model/test_ernie_trt_fp16.py | 121 ---- .../test_nlp_model/test_ernie_trt_fp32.py | 121 ---- .../test_ernie_varlen_trt_fp16.py | 199 ------ .../test_nlp_model/test_lac_trt_fp16.py | 84 --- .../test_nlp_model/test_lac_trt_fp32.py | 89 --- .../python_api_test/test_ocr_model/run.sh | 3 +- .../test_ocr_model/test_ocr_det_mv3_db_gpu.py | 6 +- .../test_ocr_det_mv3_db_trt_fp16.py | 357 ----------- .../test_ocr_det_mv3_db_trt_fp32.py | 359 ----------- .../test_seg_model/test_deeplabv3_trt_fp16.py | 194 ------ .../test_seg_model/test_deeplabv3_trt_fp32.py | 175 ------ .../test_fcn_hrnetw18_trt_fp16.py | 190 ------ .../test_fcn_hrnetw18_trt_fp32.py | 172 ------ .../test_pp_humanseg_lite_trt_fp16.py | 194 ------ .../test_pp_humanseg_lite_trt_fp32.py | 175 ------ .../test_pp_liteseg_stdc1_trt_fp16.py | 190 ------ .../test_pp_liteseg_stdc1_trt_fp32.py | 171 ------ .../test_seg_model/test_unet_trt_fp16.py | 194 ------ .../test_seg_model/test_unet_trt_fp32.py | 175 ------ .../ModelsTestFramework.py | 32 +- models/AutomaticTestSystem/test_3D_acc.py | 27 +- models/AutomaticTestSystem/test_ocr_acc.py | 27 +- .../inference_script/detection_inference.sh | 45 +- .../CE/ppdiffusers/deploy/controlnet/infer.sh | 71 --- .../deploy/ipadapter_sd15/sd15_infer.sh | 78 --- .../deploy/ipadapter_sdxl/sdxl_infer.sh | 77 --- .../CE/ppdiffusers/deploy/script/start.sh | 50 -- .../ppdiffusers/deploy/script/start_hand.sh | 6 +- .../deploy/script/test_paddle_tensorrt.sh | 50 -- .../CE/ppdiffusers/deploy/sd15/infer.sh | 71 --- .../CE/ppdiffusers/deploy/sdxl/infer.sh | 69 --- .../CE/ppdiffusers/deploy/svd/infer.sh | 40 -- .../CE2/ppdiffusers/sdxl/dynamic2static.sh | 16 - .../dynamic2static.sh | 85 --- .../stable_diffusion_deploy/dynamic2static.sh | 84 --- models/PaddleVideo/CI/test_video.sh | 18 - models/paddlecv/test_paddlecv.py | 2 +- .../PaddleClas/base/ImageNet_base.yaml | 25 - ...configs^ImageNet^PPHGNet^PPHGNet_base.yaml | 6 - ...igs^ImageNet^PPLCNetV2^PPLCNetV2_base.yaml | 4 - ...ls^configs^ImageNet^PeleeNet^PeleeNet.yaml | 6 - ...pcls^configs^ImageNet^ResNet^ResNet50.yaml | 4 - ...nfigs^ImageNet^ResNet^ResNet50_amp_O1.yaml | 2 - ...ImageNet^ResNet^ResNet50_amp_O1_ultra.yaml | 3 - ...ImageNet^ResNet^ResNet50_amp_O2_ultra.yaml | 3 - ...nTransformer_base_patch4_window12_384.yaml | 2 - ...Transformer_large_patch4_window12_384.yaml | 2 - ...nTransformer_large_patch4_window7_224.yaml | 2 - ...nTransformer_small_patch4_window7_224.yaml | 2 - ...inTransformer_tiny_patch4_window7_224.yaml | 2 - ...s^configs^ImageNet^Twins^alt_gvt_base.yaml | 2 - ...^configs^ImageNet^Twins^alt_gvt_large.yaml | 2 - ...^configs^ImageNet^Twins^alt_gvt_small.yaml | 2 - ...cls^configs^ImageNet^Twins^pcpvt_base.yaml | 2 - ...ls^configs^ImageNet^Twins^pcpvt_large.yaml | 2 - ...ls^configs^ImageNet^Twins^pcpvt_small.yaml | 2 - .../ppcls^configs^ImageNet^VAN^VAN_B0.yaml | 6 - .../PaddleDetection/base/keypoint_base.yml | 30 - .../PaddleDetection/base/mot_base.yml | 30 - .../PaddleDetection/base/normal_base.yml | 30 - ...ore_config^picodet_lcnet_1_5x_416_coco.yml | 4 - ...icodet^legacy_model^picodet_s_320_coco.yml | 4 - ...nfigs^picodet^picodet_s_320_coco_lcnet.yml | 4 - .../configs^ppyolo^ppyolo_mbv3_large_coco.yml | 4 - ...onfigs^ppyolo^ppyolo_r50vd_dcn_1x_coco.yml | 4 - .../configs^ppyolo^ppyolo_tiny_650e_coco.yml | 4 - ...gs^ppyolo^ppyolov2_r50vd_dcn_365e_coco.yml | 4 - ...onfigs^ppyoloe^ppyoloe_crn_s_300e_coco.yml | 4 - ...gs^ppyoloe^ppyoloe_plus_crn_s_80e_coco.yml | 4 - ...gs^retinanet^retinanet_r50_fpn_1x_coco.yml | 4 - ...alldet^ppyoloe_plus_sod_crn_l_80e_coco.yml | 4 - ...t^ppyoloe_vit_base_csppan_cae_36e_coco.yml | 4 - .../PaddleDetection/cases/test.yml | 4 - ..._configs^ppyolo^ppyolo_mbv3_large_coco.yml | 4 - ...onfigs^ppyolo^ppyolo_r50vd_dcn_1x_coco.yml | 4 - .../diy_build/PaddleDetection_Build.py | 2 - .../cases/model_zoo^ernie-3.0-fastdepoly.yaml | 9 - .../model_zoo^ernie-tiny-fastdeploy.yaml | 17 - .../PaddleOCR/base/ocr_cls_base.yaml | 9 - .../base/ocr_cls_base_pretrained.yaml | 18 - .../PaddleOCR/base/ocr_det_base.yaml | 11 - .../PaddleOCR/base/ocr_det_base_distill.yaml | 11 - .../base/ocr_det_base_pretrained.yaml | 21 - .../base/ocr_e2e_base_pretrained.yaml | 25 - .../base/ocr_kie_base_pretrained.yaml | 31 - .../PaddleOCR/base/ocr_rec_base.yaml | 14 - .../PaddleOCR/base/ocr_rec_base_distill.yaml | 18 - .../base/ocr_rec_base_pretrained.yaml | 27 - .../base/ocr_rec_base_pretrained_distill.yaml | 31 - .../PaddleOCR/base/ocr_sr_base.yaml | 15 - .../base/ocr_sr_base_pretrained.yaml | 25 - .../PaddleOCR/base/ocr_table_base.yaml | 17 - .../base/ocr_table_base_pretrained.yaml | 24 - models_restruct/PaddleOCR/tools/start.py | 14 - .../PaddleSeg/base/normal_base.yml | 30 - ..._resnet101_os8_cityscapes_1024x512_80k.yml | 4 - ...ann_resnet101_os8_voc12aug_512x512_40k.yml | 4 - ...n_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...^ann_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ...attention_unet_cityscapes_1024x512_80k.yml | 4 - ...enet^bisenet_cityscapes_1024x1024_160k.yml | 4 - ..._resnet18_os8_cityscapes_1024x512_160k.yml | 4 - ...t_resnet101_os8_cityscapes_769x769_60k.yml | 4 - ...t_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...anet_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ...net^ddrnet23_cityscapes_1024x1024_120k.yml | 4 - ...t_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...et_resnet50_os8_cityscapes_832x832_80k.yml | 4 - ..._resnet101_os8_cityscapes_1024x512_80k.yml | 4 - ...bv3_resnet101_os8_voc12aug_512x512_40k.yml | 4 - ...3_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...abv3_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ..._resnet101_os8_cityscapes_1024x512_80k.yml | 4 - ...p_resnet101_os8_cityscapes_769x769_80k.yml | 4 - ...v3p_resnet101_os8_voc12aug_512x512_40k.yml | 4 - ...p_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...50_os8_cityscapes_1024x512_80k_rmiloss.yml | 4 - ...bv3p_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ..._resnet101_os8_cityscapes_1024x512_80k.yml | 4 - ..._resnet101_os8_cityscapes_1024x512_80k.yml | 4 - ...net_resnet101_os8_voc12aug_512x512_40k.yml | 4 - ...t_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...lnet_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ..._resnet101_os8_cityscapes_1024x512_80k.yml | 4 - ...net_resnet101_os8_voc12aug_512x512_40k.yml | 4 - ...t_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...anet_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ..._resnet101_os8_cityscapes_1024x512_80k.yml | 4 - ...figs^enet^enet_cityscapes_1024x512_80k.yml | 4 - ...espnet^espnet_cityscapes_1024x512_120k.yml | 4 - ...etv1^espnetv1_cityscapes_1024x512_120k.yml | 4 - ...stfcn_resnet50_os8_ade20k_480x480_120k.yml | 4 - ...cnn^fastscnn_cityscapes_1024x1024_160k.yml | 4 - ...n^fcn_hrnetw18_cityscapes_1024x512_80k.yml | 4 - ...^fcn^fcn_hrnetw18_voc12aug_512x512_40k.yml | 4 - ...n^fcn_hrnetw48_cityscapes_1024x512_80k.yml | 4 - ...^fcn^fcn_hrnetw48_voc12aug_512x512_40k.yml | 4 - ..._resnet101_os8_cityscapes_1024x512_80k.yml | 4 - ...net_resnet101_os8_voc12aug_512x512_40k.yml | 4 - ...t_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...cnet_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ...inet_resnet101_os8_ade20k_520x520_150k.yml | 4 - ..._resnet101_os8_cityscapes_1024x512_80k.yml | 4 - ...net_resnet101_os8_voc12aug_512x512_40k.yml | 4 - ...ginet_resnet50_os8_ade20k_520x520_150k.yml | 4 - ...t_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...inet_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ...e_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...lore_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ...dnet^hardnet_cityscapes_1024x1024_160k.yml | 4 - ...t_W48_contrast_cityscapes_1024x512_60k.yml | 4 - ...t_resnet101_os8_cityscapes_769x769_80k.yml | 4 - ...net_resnet101_os8_voc12aug_512x512_40k.yml | 4 - ...et_resnet50_os8_cityscapes_769x769_80k.yml | 4 - ...anet_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ...pp_mobilenetv3_cityscapes_1024x512_80k.yml | 4 - ...3_cityscapes_1024x512_80k_large_kernel.yml | 4 - ...bilenetv3_cityscapes_1024x512_80k_os32.yml | 4 - ...leseg_ghostnet_cityscapes_1024x512_80k.yml | 4 - ...eg_litehrnet18_cityscapes_1024x512_80k.yml | 4 - ...eg_mobilenetv2_cityscapes_1024x512_80k.yml | 4 - ...eg_mobilenetv3_cityscapes_1024x512_80k.yml | 4 - ...g_shufflenetv2_cityscapes_1024x512_80k.yml | 4 - ...rnet_hrnetw18_cityscapes_1024x512_160k.yml | 4 - ...t^ocrnet_hrnetw18_voc12aug_512x512_40k.yml | 4 - ...rnet_hrnetw48_cityscapes_1024x512_160k.yml | 4 - ...t^ocrnet_hrnetw48_voc12aug_512x512_40k.yml | 4 - ..._resnet101_os8_cityscapes_512x1024_40k.yml | 4 - ...end_resnet101_os8_voc12aug_512x512_40k.yml | 4 - ...d_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...rend_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ...eg^pp_liteseg_stdc1_camvid_960x720_10k.yml | 4 - ...tdc1_cityscapes_1024x512_scale0.5_160k.yml | 4 - ...dc1_cityscapes_1024x512_scale0.75_160k.yml | 4 - ...tdc1_cityscapes_1024x512_scale1.0_160k.yml | 4 - ...eg^pp_liteseg_stdc2_camvid_960x720_10k.yml | 4 - ...tdc2_cityscapes_1024x512_scale0.5_160k.yml | 4 - ...dc2_cityscapes_1024x512_scale0.75_160k.yml | 4 - ...tdc2_cityscapes_1024x512_scale1.0_160k.yml | 4 - ..._resnet101_os8_cityscapes_1024x512_80k.yml | 4 - ...net_resnet101_os8_voc12aug_512x512_40k.yml | 4 - ...t_resnet50_os8_cityscapes_1024x512_80k.yml | 4 - ...pnet_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ...rmer^rtformer_slim_ade20k_512x512_160k.yml | 4 - ...rtformer_slim_cityscapes_1024x512_120k.yml | 4 - ...segformer_b0_cityscapes_1024x1024_160k.yml | 4 - ...segformer_b1_cityscapes_1024x1024_160k.yml | 4 - ...segformer_b2_cityscapes_1024x1024_160k.yml | 4 - ...segformer_b3_cityscapes_1024x1024_160k.yml | 4 - ...segformer_b4_cityscapes_1024x1024_160k.yml | 4 - ...r_vit_small_linear_ade20k_512x512_160k.yml | 4 - ...ter_vit_small_mask_ade20k_512x512_160k.yml | 4 - ..._resnet18_os8_cityscapes_1024x1024_80k.yml | 4 - ..._resnet50_os8_cityscapes_1024x1024_80k.yml | 4 - ...cseg^stdc1_seg_cityscapes_1024x512_80k.yml | 4 - ...stdcseg^stdc1_seg_voc12aug_512x512_40k.yml | 4 - ...cseg^stdc2_seg_cityscapes_1024x512_80k.yml | 4 - ...stdcseg^stdc2_seg_voc12aug_512x512_40k.yml | 4 - ...er^topformer_small_ade20k_512x512_160k.yml | 4 - ...mer^topformer_tiny_ade20k_512x512_160k.yml | 4 - ...w18_small_cityscapes_1024x512_120k_bs3.yml | 4 - ...hrnetw18_small_cityscapes_1024x512_80k.yml | 4 - ...uhrnetw48_cityscapes_1024x512_120k_bs3.yml | 4 - ...^fcn_uhrnetw48_cityscapes_1024x512_80k.yml | 4 - ...igs^unet^unet_cityscapes_1024x512_160k.yml | 4 - ..._resnet101_os8_cityscapes_512x1024_40k.yml | 4 - models_restruct/PaddleSeg/cases/fix_input.yml | 4 - ...segformer_b0_cityscapes_1024x1024_160k.yml | 4 - ...segformer_b0_cityscapes_1024x1024_160k.yml | 4 - models_restruct/PaddleSeg/cases/test.yml | 4 - ...bv3p_resnet50_os8_voc12aug_512x512_40k.yml | 4 - ...n^fcn_hrnetw18_cityscapes_1024x512_80k.yml | 4 - ...rnet_hrnetw18_cityscapes_1024x512_160k.yml | 4 - ...rnet_hrnetw48_cityscapes_1024x512_160k.yml | 4 - ...tdc1_cityscapes_1024x512_scale1.0_160k.yml | 4 - ...dc2_cityscapes_1024x512_scale0.75_160k.yml | 4 - .../tpaddle/test_bert.py | 118 ---- .../tpaddle/test_case/infer_test.py | 410 ------------- .../tpaddle/test_deeplabv3.py | 2 +- .../tpaddle/test_ernie.py | 118 ---- .../tpaddle/test_mobilenetv1.py | 106 +--- .../tpaddle/test_ocr.py | 416 +------------ .../tpaddle/test_resnet50.py | 106 +--- .../tpaddle/test_helper.py | 25 +- .../tpaddle/test_helper.py | 25 +- .../tpaddle/test_case/infer_test.py | 410 ------------- .../tpaddle/test_mobilenet.py | 106 +--- .../tpaddle/test_resnet50.py | 106 +--- 328 files changed, 89 insertions(+), 20378 deletions(-) delete mode 100644 inference/python_api_test/test_class_model/test_DarkNet53_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_DarkNet53_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_DenseNet121_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_DenseNet121_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_GoogLeNet_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_GoogLeNet_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_base_patch16_224_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_base_patch16_224_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_base_patch16_384_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_base_patch16_384_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_base_patch32_384_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_base_patch32_384_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_large_patch16_224_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_large_patch16_224_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_large_patch16_384_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_large_patch16_384_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_large_patch32_384_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_large_patch32_384_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_small_patch16_224_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_ViT_small_patch16_224_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_pcpvt_base_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_pcpvt_base_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_resnet50_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_resnet50_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_swin_transformer_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_swin_transformer_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_tnt_small_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_tnt_small_trt_fp32.py delete mode 100644 inference/python_api_test/test_class_model/test_vgg11_trt_fp16.py delete mode 100644 inference/python_api_test/test_class_model/test_vgg11_trt_fp32.py delete mode 100644 inference/python_api_test/test_det_model/test_fast_rcnn_trt_fp16.py delete mode 100644 inference/python_api_test/test_det_model/test_fast_rcnn_trt_fp32.py delete mode 100644 inference/python_api_test/test_det_model/test_mask_rcnn_trt_fp16.py delete mode 100644 inference/python_api_test/test_det_model/test_mask_rcnn_trt_fp32.py delete mode 100644 inference/python_api_test/test_det_model/test_ppyolo_trt_fp16.py delete mode 100644 inference/python_api_test/test_det_model/test_ppyolo_trt_fp32.py delete mode 100644 inference/python_api_test/test_det_model/test_ppyolov2_trt_fp16.py delete mode 100644 inference/python_api_test/test_det_model/test_ppyolov2_trt_fp32.py delete mode 100644 inference/python_api_test/test_det_model/test_solov2_trt_fp16.py delete mode 100644 inference/python_api_test/test_det_model/test_solov2_trt_fp32.py delete mode 100644 inference/python_api_test/test_det_model/test_yolov3_trt_fp16.py delete mode 100644 inference/python_api_test/test_det_model/test_yolov3_trt_fp32.py delete mode 100644 inference/python_api_test/test_int8_model/backend/tensorrt.py delete mode 100644 inference/python_api_test/test_int8_model/base_nv_trt_fp16.py delete mode 100644 inference/python_api_test/test_int8_model/base_nv_trt_int8.py delete mode 100644 inference/python_api_test/test_int8_model/base_trt_fp16.py delete mode 100644 inference/python_api_test/test_int8_model/base_trt_int8.py delete mode 100644 inference/python_api_test/test_int8_model/convert_onnx.sh delete mode 100644 inference/python_api_test/test_int8_model/run_nv_trt_fp16.sh delete mode 100644 inference/python_api_test/test_int8_model/run_nv_trt_fp32.sh delete mode 100644 inference/python_api_test/test_int8_model/run_nv_trt_int8.sh delete mode 100644 inference/python_api_test/test_int8_model/run_ocr.sh delete mode 100644 inference/python_api_test/test_int8_model/run_trt_fp16.sh delete mode 100644 inference/python_api_test/test_int8_model/run_trt_fp32.sh delete mode 100644 inference/python_api_test/test_int8_model/run_trt_int8.sh delete mode 100644 inference/python_api_test/test_nlp_model/test_AFQMC_PTQ_trt_int8.py delete mode 100644 inference/python_api_test/test_nlp_model/test_AFQMC_base_trt_fp16.py delete mode 100644 inference/python_api_test/test_nlp_model/test_AFQMC_base_trt_fp32.py delete mode 100644 inference/python_api_test/test_nlp_model/test_bert_trt_fp16.py delete mode 100644 inference/python_api_test/test_nlp_model/test_bert_trt_fp32.py delete mode 100644 inference/python_api_test/test_nlp_model/test_ernie_trt_fp16.py delete mode 100644 inference/python_api_test/test_nlp_model/test_ernie_trt_fp32.py delete mode 100644 inference/python_api_test/test_nlp_model/test_ernie_varlen_trt_fp16.py delete mode 100644 inference/python_api_test/test_nlp_model/test_lac_trt_fp16.py delete mode 100644 inference/python_api_test/test_nlp_model/test_lac_trt_fp32.py delete mode 100644 inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_trt_fp16.py delete mode 100644 inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_trt_fp32.py delete mode 100644 inference/python_api_test/test_seg_model/test_deeplabv3_trt_fp16.py delete mode 100644 inference/python_api_test/test_seg_model/test_deeplabv3_trt_fp32.py delete mode 100644 inference/python_api_test/test_seg_model/test_fcn_hrnetw18_trt_fp16.py delete mode 100644 inference/python_api_test/test_seg_model/test_fcn_hrnetw18_trt_fp32.py delete mode 100644 inference/python_api_test/test_seg_model/test_pp_humanseg_lite_trt_fp16.py delete mode 100644 inference/python_api_test/test_seg_model/test_pp_humanseg_lite_trt_fp32.py delete mode 100644 inference/python_api_test/test_seg_model/test_pp_liteseg_stdc1_trt_fp16.py delete mode 100644 inference/python_api_test/test_seg_model/test_pp_liteseg_stdc1_trt_fp32.py delete mode 100644 inference/python_api_test/test_seg_model/test_unet_trt_fp16.py delete mode 100644 inference/python_api_test/test_seg_model/test_unet_trt_fp32.py delete mode 100644 models/PaddleMIX/CE/ppdiffusers/deploy/script/test_paddle_tensorrt.sh diff --git a/inference/benchmark/python/paddle/run.sh b/inference/benchmark/python/paddle/run.sh index eebd94bb3d..f500ff3ddc 100644 --- a/inference/benchmark/python/paddle/run.sh +++ b/inference/benchmark/python/paddle/run.sh @@ -2,78 +2,18 @@ python resnet101.py --device gpu --batch_size 1 python resnet101.py --device gpu --batch_size 4 python resnet101.py --device gpu --batch_size 8 -python resnet101.py --device gpu --batch_size 1 --use_trt True --trt_precision fp32 -python resnet101.py --device gpu --batch_size 4 --use_trt True --trt_precision fp32 -python resnet101.py --device gpu --batch_size 8 --use_trt True --trt_precision fp32 - -python resnet101.py --device gpu --batch_size 1 --use_trt True --trt_precision fp16 -python resnet101.py --device gpu --batch_size 4 --use_trt True --trt_precision fp16 -python resnet101.py --device gpu --batch_size 8 --use_trt True --trt_precision fp16 - -python resnet101.py --device gpu --batch_size 1 --use_trt True --trt_precision int8 -python resnet101.py --device gpu --batch_size 4 --use_trt True --trt_precision int8 -python resnet101.py --device gpu --batch_size 8 --use_trt True --trt_precision int8 - python vgg16.py --device gpu --batch_size 1 python vgg16.py --device gpu --batch_size 4 python vgg16.py --device gpu --batch_size 8 -python vgg16.py --device gpu --batch_size 1 --use_trt True --trt_precision fp32 -python vgg16.py --device gpu --batch_size 4 --use_trt True --trt_precision fp32 -python vgg16.py --device gpu --batch_size 8 --use_trt True --trt_precision fp32 - -python vgg16.py --device gpu --batch_size 1 --use_trt True --trt_precision fp16 -python vgg16.py --device gpu --batch_size 4 --use_trt True --trt_precision fp16 -python vgg16.py --device gpu --batch_size 8 --use_trt True --trt_precision fp16 - -python vgg16.py --device gpu --batch_size 1 --use_trt True --trt_precision int8 -python vgg16.py --device gpu --batch_size 4 --use_trt True --trt_precision int8 -python vgg16.py --device gpu --batch_size 8 --use_trt True --trt_precision int8 - python mobilenetv2.py --device gpu --batch_size 1 python mobilenetv2.py --device gpu --batch_size 4 python mobilenetv2.py --device gpu --batch_size 8 -python mobilenetv2.py --device gpu --batch_size 1 --use_trt True --trt_precision fp32 -python mobilenetv2.py --device gpu --batch_size 4 --use_trt True --trt_precision fp32 -python mobilenetv2.py --device gpu --batch_size 8 --use_trt True --trt_precision fp32 - -python mobilenetv2.py --device gpu --batch_size 1 --use_trt True --trt_precision fp16 -python mobilenetv2.py --device gpu --batch_size 4 --use_trt True --trt_precision fp16 -python mobilenetv2.py --device gpu --batch_size 8 --use_trt True --trt_precision fp16 - -python mobilenetv2.py --device gpu --batch_size 1 --use_trt True --trt_precision int8 -python mobilenetv2.py --device gpu --batch_size 4 --use_trt True --trt_precision int8 -python mobilenetv2.py --device gpu --batch_size 8 --use_trt True --trt_precision int8 - python fast_rcnn.py --device gpu --batch_size 1 python fast_rcnn.py --device gpu --batch_size 4 python fast_rcnn.py --device gpu --batch_size 8 -python fast_rcnn.py --device gpu --batch_size 1 --use_trt True --trt_precision fp32 -python fast_rcnn.py --device gpu --batch_size 4 --use_trt True --trt_precision fp32 -python fast_rcnn.py --device gpu --batch_size 8 --use_trt True --trt_precision fp32 - -python fast_rcnn.py --device gpu --batch_size 1 --use_trt True --trt_precision fp16 -python fast_rcnn.py --device gpu --batch_size 4 --use_trt True --trt_precision fp16 -python fast_rcnn.py --device gpu --batch_size 8 --use_trt True --trt_precision fp16 - -python fast_rcnn.py --device gpu --batch_size 1 --use_trt True --trt_precision int8 -python fast_rcnn.py --device gpu --batch_size 4 --use_trt True --trt_precision int8 -python fast_rcnn.py --device gpu --batch_size 8 --use_trt True --trt_precision int8 - python squeezenet.py --device gpu --batch_size 1 python squeezenet.py --device gpu --batch_size 4 python squeezenet.py --device gpu --batch_size 8 - -python squeezenet.py --device gpu --batch_size 1 --use_trt True --trt_precision fp32 -python squeezenet.py --device gpu --batch_size 4 --use_trt True --trt_precision fp32 -python squeezenet.py --device gpu --batch_size 8 --use_trt True --trt_precision fp32 - -python squeezenet.py --device gpu --batch_size 1 --use_trt True --trt_precision fp16 -python squeezenet.py --device gpu --batch_size 4 --use_trt True --trt_precision fp16 -python squeezenet.py --device gpu --batch_size 8 --use_trt True --trt_precision fp16 - -python squeezenet.py --device gpu --batch_size 1 --use_trt True --trt_precision int8 -python squeezenet.py --device gpu --batch_size 4 --use_trt True --trt_precision int8 -python squeezenet.py --device gpu --batch_size 8 --use_trt True --trt_precision int8 diff --git a/inference/benchmark/python/tensorflow/run.sh b/inference/benchmark/python/tensorflow/run.sh index a28c4ec594..334411351d 100644 --- a/inference/benchmark/python/tensorflow/run.sh +++ b/inference/benchmark/python/tensorflow/run.sh @@ -12,33 +12,3 @@ python clas_keras_benchmark.py --model_name=VGG16 --batch_size=8 --use_gpu python clas_keras_benchmark.py --model_name=MobileNetV2 --batch_size=1 --use_gpu python clas_keras_benchmark.py --model_name=MobileNetV2 --batch_size=4 --use_gpu python clas_keras_benchmark.py --model_name=MobileNetV2 --batch_size=8 --use_gpu - -#ResNet101 tensorrt fp32 -python clas_keras_benchmark.py --model_name=ResNet101 --batch_size=1 --use_gpu --use_trt --trt_precision=fp32 -python clas_keras_benchmark.py --model_name=ResNet101 --batch_size=4 --use_gpu --use_trt --trt_precision=fp32 -python clas_keras_benchmark.py --model_name=ResNet101 --batch_size=8 --use_gpu --use_trt --trt_precision=fp32 - -#VGG16 tensorrt fp32 -python clas_keras_benchmark.py --model_name=VGG16 --batch_size=1 --use_gpu --use_trt --trt_precision=fp32 -python clas_keras_benchmark.py --model_name=VGG16 --batch_size=4 --use_gpu --use_trt --trt_precision=fp32 -python clas_keras_benchmark.py --model_name=VGG16 --batch_size=8 --use_gpu --use_trt --trt_precision=fp32 - -#MobileNetV2 tensorrt fp32 -python clas_keras_benchmark.py --model_name=MobileNetV2 --batch_size=1 --use_gpu --use_trt --trt_precision=fp32 -python clas_keras_benchmark.py --model_name=MobileNetV2 --batch_size=4 --use_gpu --use_trt --trt_precision=fp32 -python clas_keras_benchmark.py --model_name=MobileNetV2 --batch_size=8 --use_gpu --use_trt --trt_precision=fp32 - -#ResNet101 tensorrt fp16 -python clas_keras_benchmark.py --model_name=ResNet101 --batch_size=1 --use_gpu --use_trt --trt_precision=fp16 -python clas_keras_benchmark.py --model_name=ResNet101 --batch_size=4 --use_gpu --use_trt --trt_precision=fp16 -python clas_keras_benchmark.py --model_name=ResNet101 --batch_size=8 --use_gpu --use_trt --trt_precision=fp16 - -#VGG16 tensorrt fp16 -python clas_keras_benchmark.py --model_name=VGG16 --batch_size=1 --use_gpu --use_trt --trt_precision=fp16 -python clas_keras_benchmark.py --model_name=VGG16 --batch_size=4 --use_gpu --use_trt --trt_precision=fp16 -python clas_keras_benchmark.py --model_name=VGG16 --batch_size=8 --use_gpu --use_trt --trt_precision=fp16 - -#MobileNetV2 tensorrt fp16 -python clas_keras_benchmark.py --model_name=MobileNetV2 --batch_size=1 --use_gpu --use_trt --trt_precision=fp16 -python clas_keras_benchmark.py --model_name=MobileNetV2 --batch_size=4 --use_gpu --use_trt --trt_precision=fp16 -python clas_keras_benchmark.py --model_name=MobileNetV2 --batch_size=8 --use_gpu --use_trt --trt_precision=fp16 diff --git a/inference/benchmark/python/torch/run.sh b/inference/benchmark/python/torch/run.sh index 541dacd305..146655a74b 100644 --- a/inference/benchmark/python/torch/run.sh +++ b/inference/benchmark/python/torch/run.sh @@ -18,45 +18,3 @@ python clas_benchmark.py --model_name mobilenet_v2 --device gpu --batch_size 8 python detection_benchmark.py --model_name faster_rcnn --device gpu --batch_size 1 python detection_benchmark.py --model_name faster_rcnn --device gpu --batch_size 4 python detection_benchmark.py --model_name faster_rcnn --device gpu --batch_size 8 - -# FP32 -python clas_benchmark.py --model_name resnet101 --device gpu --batch_size 1 --use_trt --trt_precision fp32 -python clas_benchmark.py --model_name resnet101 --device gpu --batch_size 4 --use_trt --trt_precision fp32 -python clas_benchmark.py --model_name resnet101 --device gpu --batch_size 8 --use_trt --trt_precision fp32 - -python clas_benchmark.py --model_name vgg16 --device gpu --batch_size 1 --use_trt --trt_precision fp32 -python clas_benchmark.py --model_name vgg16 --device gpu --batch_size 4 --use_trt --trt_precision fp32 -python clas_benchmark.py --model_name vgg16 --device gpu --batch_size 8 --use_trt --trt_precision fp32 - -python clas_benchmark.py --model_name squeezenet1_0 --device gpu --batch_size 1 --use_trt --trt_precision fp32 -python clas_benchmark.py --model_name squeezenet1_0 --device gpu --batch_size 4 --use_trt --trt_precision fp32 -python clas_benchmark.py --model_name squeezenet1_0 --device gpu --batch_size 8 --use_trt --trt_precision fp32 - -python clas_benchmark.py --model_name mobilenet_v2 --device gpu --batch_size 1 --use_trt --trt_precision fp32 -python clas_benchmark.py --model_name mobilenet_v2 --device gpu --batch_size 4 --use_trt --trt_precision fp32 -python clas_benchmark.py --model_name mobilenet_v2 --device gpu --batch_size 8 --use_trt --trt_precision fp32 - -python detection_benchmark.py --model_name faster_rcnn --device gpu --batch_size 1 --use_trt --trt_precision fp32 -python detection_benchmark.py --model_name faster_rcnn --device gpu --batch_size 4 --use_trt --trt_precision fp32 -python detection_benchmark.py --model_name faster_rcnn --device gpu --batch_size 8 --use_trt --trt_precision fp32 - -# FP16 -python clas_benchmark.py --model_name resnet101 --device gpu --batch_size 1 --use_trt --trt_precision fp16 -python clas_benchmark.py --model_name resnet101 --device gpu --batch_size 4 --use_trt --trt_precision fp16 -python clas_benchmark.py --model_name resnet101 --device gpu --batch_size 8 --use_trt --trt_precision fp16 - -python clas_benchmark.py --model_name vgg16 --device gpu --batch_size 1 --use_trt --trt_precision fp16 -python clas_benchmark.py --model_name vgg16 --device gpu --batch_size 4 --use_trt --trt_precision fp16 -python clas_benchmark.py --model_name vgg16 --device gpu --batch_size 8 --use_trt --trt_precision fp16 - -python clas_benchmark.py --model_name squeezenet1_0 --device gpu --batch_size 1 --use_trt --trt_precision fp16 -python clas_benchmark.py --model_name squeezenet1_0 --device gpu --batch_size 4 --use_trt --trt_precision fp16 -python clas_benchmark.py --model_name squeezenet1_0 --device gpu --batch_size 8 --use_trt --trt_precision fp16 - -python clas_benchmark.py --model_name mobilenet_v2 --device gpu --batch_size 1 --use_trt --trt_precision fp16 -python clas_benchmark.py --model_name mobilenet_v2 --device gpu --batch_size 4 --use_trt --trt_precision fp16 -python clas_benchmark.py --model_name mobilenet_v2 --device gpu --batch_size 8 --use_trt --trt_precision fp16 - -python detection_benchmark.py --model_name faster_rcnn --device gpu --batch_size 1 --use_trt --trt_precision fp16 -python detection_benchmark.py --model_name faster_rcnn --device gpu --batch_size 4 --use_trt --trt_precision fp16 -python detection_benchmark.py --model_name faster_rcnn --device gpu --batch_size 8 --use_trt --trt_precision fp16 diff --git a/inference/python_api_test/test_case/image_preprocess.py b/inference/python_api_test/test_case/image_preprocess.py index 2a633486a0..d0fa9baffc 100644 --- a/inference/python_api_test/test_case/image_preprocess.py +++ b/inference/python_api_test/test_case/image_preprocess.py @@ -176,7 +176,7 @@ def sig_fig_compare(array1, array2, delta=5, det_top_bbox=False, need_sort=False array1 = array1[:top_count, :] array2 = array2[:top_count, :] elif len(array1.shape) == 1: - # 部分检测模型输出检测框数量，在trt fp16下可能与关闭优化的检测框数量不同，跳过，只关注高置信度检测框 + # 部分fp16检测模型输出检测框数量可能与关闭优化时不同，跳过，只关注高置信度检测框 return if np.any(abs(array2) > 100): normalize_func = np.vectorize(normalize) diff --git a/inference/python_api_test/test_case/infer_test.py b/inference/python_api_test/test_case/infer_test.py index 507f4d0efe..9486abb142 100644 --- a/inference/python_api_test/test_case/infer_test.py +++ b/inference/python_api_test/test_case/infer_test.py @@ -7,7 +7,6 @@ import os import sys import logging -import queue import threading from multiprocessing import Process @@ -42,7 +41,6 @@ def __init__(self): """ __init__ """ - self.errors = queue.Queue() def load_config(self, **kwargs): """ @@ -100,36 +98,6 @@ def get_truth_val(self, input_data_dict: dict, device: str, gpu_mem=1000) -> dic output_data_dict[output_data_name] = output_data return output_data_dict - def collect_shape_info(self, model_path: str, input_data_dict: dict, device: str = "gpu") -> None: - """ - collect_shape_range_info for TRT dynamic shape - Args: - model_path(str): model path - device(str): infer device - input_data_dict(dict): input_data - Returns: - None - """ - if device == "cpu": - self.pd_config.disable_gpu() - elif device == "gpu": - self.pd_config.enable_use_gpu(256, 0) - self.pd_config.enable_memory_optim() - self.pd_config.collect_shape_range_info(f"{model_path}/shape_range.pbtxt") - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for i, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - predictor.run() - - output_names = predictor.get_output_names() - for _, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - def convert_to_mixed_precision_model(self, src_model, src_params, dst_model, dst_params) -> None: """ convert model to mixed precision @@ -437,486 +405,6 @@ def gpu_more_bz_test( print("truth_value_shape:", output_data_truth_val.shape) diff = sig_fig_compare(output_data, output_data_truth_val, delta) - def trt_bz1_slim_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - delta=1e-5, - gpu_mem=1000, - max_batch_size=3, - min_subgraph_size=10, - precision="fp32", - use_static=True, - use_calib_mode=False, - dynamic=False, - shape_range_file="shape_range.pbtxt", - tuned=False, - result_sort=False, - delete_pass_list=None, - with_benchmark=False, - base_latency_ms=np.inf, - benchmark_threshold=np.inf, - ): - """ - test slim model enable_tensorrt_engine() - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - if dynamic: - if tuned: - self.pd_config.collect_shape_range_info("shape_range.pbtxt") - else: - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - self.pd_config.enable_tuned_tensorrt_dynamic_shape() - else: - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - if delete_pass_list: - for ir_pass in delete_pass_list: - self.pd_config.delete_pass(ir_pass) - - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - output_names = predictor.get_output_names() - print("output_names:", output_names) - print("truth_value_names:", list(output_data_dict.keys())) - - predictor.run() - if tuned: # collect_shape_range_info收集动态shape需要predictor后再退出 - return 0 - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data_truth_val = output_data_dict[output_data_name] - print("output_data_shape:", output_data.shape) - print("truth_value_shape:", output_data_truth_val.shape) - diff = sig_fig_compare(output_data, output_data_truth_val, delta) - - # benchmark - if not with_benchmark: - return - # warm up - output_tensor = predictor.get_output_handle(output_names[0]) - for i in range(5): - predictor.run() - batch_output = output_tensor.copy_to_cpu() - predict_time = 0.0 - time_min = float("inf") - time_max = float("-inf") - for i in range(repeat): - start = time.time() - predictor.run() - batch_output = output_tensor.copy_to_cpu() - end = time.time() - timed = end - start - time_min = min(time_min, timed) - time_max = max(time_max, timed) - print("once:", timed) - predict_time += timed - time_avg = (predict_time - time_max) / (repeat - 1) - print( - "[Benchmark] Inference time(ms): min={}, max={}, avg={}".format( - round(time_min * 1000, 2), - round(time_max * 1000, 2), - round(time_avg * 1000, 2), - ) - ) - benchmark_diff = (time_avg * 1000 - base_latency_ms) / base_latency_ms if base_latency_ms != np.inf else np.inf - print(f"benchmark diff: {benchmark_diff}") - assert benchmark_diff <= benchmark_threshold, f"benchmark diff:{benchmark_diff} > {benchmark_threshold}" - - def trt_more_bz_test( - self, - input_data_dict: dict, - output_data_dict: dict, - check_output_list=None, - repeat=1, - delta=1e-5, - gpu_mem=1000, - max_batch_size=3, - min_subgraph_size=10, - precision="fp32", - use_static=True, - use_calib_mode=False, - dynamic=False, - shape_range_file="shape_range.pbtxt", - tuned=False, - auto_tuned=False, - det_top_bbox=False, - need_sort=False, - det_top_bbox_threshold=0.75, - delete_pass_list=None, - delete_op_list=None, - ): - """ - test enable_tensorrt_engine() - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - check_output_list(list): select which outputs to check - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - if dynamic: - if tuned: - self.pd_config.collect_shape_range_info("shape_range.pbtxt") - else: - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - if auto_tuned: - self.pd_config.enable_tuned_tensorrt_dynamic_shape() - else: - self.pd_config.enable_tuned_tensorrt_dynamic_shape(shape_range_file, True) - else: - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - if delete_pass_list: - for ir_pass in delete_pass_list: - self.pd_config.delete_pass(ir_pass) - if delete_op_list: - self.pd_config.exp_disable_tensorrt_ops(delete_op_list) - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - if tuned: # collect_shape_range_info收集动态shape需要predictor后再退出 - return 0 - output_names = check_output_list if (check_output_list) else predictor.get_output_names() - print("output_names:", output_names) - print("truth_value_names:", output_names) - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data_truth_val = output_data_dict[output_data_name] - print("output_data_shape:", output_data.shape) - print("truth_value_shape:", output_data_truth_val.shape) - diff = sig_fig_compare( - output_data, output_data_truth_val, delta, det_top_bbox, need_sort, det_top_bbox_threshold - ) - - def trt_more_bz_dynamic_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - delta=1e-5, - gpu_mem=1000, - max_batch_size=10, - names=None, - min_input_shape=None, - max_input_shape=None, - opt_input_shape=None, - min_subgraph_size=10, - precision="trt_fp32", - use_static=False, - use_calib_mode=False, - ): - """ - test enable_tensorrt_engine() - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - names(list): input names - min_input_shape(list): TensorRT min input shape - max_input_shape(list): TensorRT max input shape - opt_input_shape(list): TensorRT best input shape - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - - self.pd_config.set_trt_dynamic_shape_info( - {names[i]: min_input_shape[i] for i in range(len(names))}, - {names[i]: max_input_shape[i] for i in range(len(names))}, - {names[i]: opt_input_shape[i] for i in range(len(names))}, - ) - - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - - output_names = predictor.get_output_names() - print("output_names:", output_names) - print("truth_value_names:", list(output_data_dict.keys())) - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data_truth_val = output_data_dict[output_data_name] - print("output_data_shape:", output_data.shape) - print("truth_value_shape:", output_data_truth_val.shape) - diff = sig_fig_compare(output_data, output_data_truth_val, delta) - - def trt_bz1_multi_thread_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - thread_num=2, - delta=1e-5, - gpu_mem=1000, - min_subgraph_size=10, - precision="trt_fp32", - use_static=True, - use_calib_mode=False, - delete_pass_list=None, - dynamic=False, - tuned=False, - shape_range_file="shape_range.pbtxt", - ): - """ - test enable_tensorrt_engine() - Multithreading TensorRT predictor - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time - thread_num(int): number of threads - delta(float): difference threshold between inference outputs and thruth value - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - if dynamic: - if tuned: - self.pd_config.collect_shape_range_info("shape_range.pbtxt") - else: - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - self.pd_config.enable_tuned_tensorrt_dynamic_shape() - else: - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - - if delete_pass_list: - for ir_pass in delete_pass_list: - self.pd_config.delete_pass(ir_pass) - - predictors = paddle_infer.PredictorPool(self.pd_config, thread_num) - for i in range(thread_num): - record_thread = threading.Thread( - target=self.run_multi_thread_test_predictor, - args=(predictors.retrieve(i), input_data_dict, output_data_dict, repeat, delta), - ) - record_thread.start() - record_thread.join() - - while not self.errors.empty(): - print("errors queue not empty!!!") - raise self.errors.get() - - def trt_dynamic_multi_thread_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - delta=1e-5, - thread_num=2, - gpu_mem=1000, - max_batch_size=1, - names=None, - min_input_shape=None, - max_input_shape=None, - opt_input_shape=None, - min_subgraph_size=10, - precision="trt_fp32", - use_static=False, - use_calib_mode=False, - ): - """ - test enable_tensorrt_engine() - Multithreading TensorRT predictor - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time - delta(float): difference threshold between inference outputs and thruth value - names(list): input names - min_input_shape(list): TensorRT min input shape - max_input_shape(list): TensorRT max input shape - opt_input_shape(list): TensorRT best input shape - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - self.pd_config.set_trt_dynamic_shape_info( - {names[i]: min_input_shape[i] for i in range(len(names))}, - {names[i]: max_input_shape[i] for i in range(len(names))}, - {names[i]: opt_input_shape[i] for i in range(len(names))}, - ) - predictors = paddle_infer.PredictorPool(self.pd_config, thread_num) - for i in range(thread_num): - record_thread = threading.Thread( - target=self.run_multi_thread_test_predictor, - args=(predictors.retrieve(i), input_data_dict, output_data_dict, repeat, delta), - ) - record_thread.start() - record_thread.join() - - while not self.errors.empty(): - print("errors queue not empty!!!") - raise self.errors.get() - - def run_multi_thread_test_predictor( - self, predictor, input_data_dict: dict, output_data_dict: dict, repeat=1, delta=1e-5 - ): - """ - test paddle predictor in multithreaded task - Args: - predictor: paddle inference predictor - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - Returns: - None - """ - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - output_names = predictor.get_output_names() - print("output_names:", output_names) - print("truth_value_names:", list(output_data_dict.keys())) - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data_truth_val = output_data_dict[output_data_name] - print("output_data_shape:", output_data.shape) - print("truth_value_shape:", output_data_truth_val.shape) - try: - diff = sig_fig_compare(output_data, output_data_truth_val, delta) - except Exception as e: - self.errors.put(e) - def get_gpu_mem(gpu_id=0): """ diff --git a/inference/python_api_test/test_class_model/run.sh b/inference/python_api_test/test_class_model/run.sh index adf0cf6228..3de94cc810 100644 --- a/inference/python_api_test/test_class_model/run.sh +++ b/inference/python_api_test/test_class_model/run.sh @@ -4,10 +4,7 @@ cases="./test_pcpvt_base_gpu.py \ ./test_pcpvt_base_mkldnn.py \ ./test_resnet50_gpu.py \ ./test_resnet50_mkldnn.py \ - ./test_resnet50_trt_fp32.py \ - ./test_resnet50_trt_fp16.py \ ./test_resnet50_slim.py \ - ./test_swin_transformer_trt_fp32.py \ ./test_tnt_small_gpu.py \ " ignore="" diff --git a/inference/python_api_test/test_class_model/run_ce_win.sh b/inference/python_api_test/test_class_model/run_ce_win.sh index d013e494fd..3e722c10f8 100644 --- a/inference/python_api_test/test_class_model/run_ce_win.sh +++ b/inference/python_api_test/test_class_model/run_ce_win.sh @@ -1,9 +1,6 @@ export FLAGS_call_stack_level=2 cases=`find . -name "test*.py" | sort` -ignore="test_swin_transformer_gpu.py \ - test_swin_transformer_trt_fp16.py \ - test_swin_transformer_trt_fp32.py - " +ignore="test_swin_transformer_gpu.py" bug=0 echo "============ failed cases =============" >> result.txt diff --git a/inference/python_api_test/test_class_model/run_parallel.sh b/inference/python_api_test/test_class_model/run_parallel.sh index 62486115a1..0ab0258cb7 100644 --- a/inference/python_api_test/test_class_model/run_parallel.sh +++ b/inference/python_api_test/test_class_model/run_parallel.sh @@ -3,16 +3,11 @@ export FLAGS_call_stack_level=2 # V100 total:690.25s cases="./test_resnet50_gpu.py \ ./test_resnet50_mkldnn.py \ - ./test_resnet50_trt_fp16.py \ - ./test_resnet50_trt_fp32.py \ ./test_resnet50_slim.py \ ./test_swin_transformer_gpu.py \ - ./test_swin_transformer_trt_fp32.py \ ./test_tnt_small_gpu.py \ - ./test_tnt_small_trt_fp32.py \ ./test_vgg11_gpu.py \ ./test_vgg11_mkldnn.py \ - ./test_ViT_base_patch16_224_trt_fp32.py \ " ignore="" bug=0 diff --git a/inference/python_api_test/test_class_model/test_DarkNet53_trt_fp16.py b/inference/python_api_test/test_class_model/test_DarkNet53_trt_fp16.py deleted file mode 100644 index 4b5115c196..0000000000 --- a/inference/python_api_test/test_class_model/test_DarkNet53_trt_fp16.py +++ /dev/null @@ -1,181 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test DarkNet53 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np -import paddle.inference as paddle_infer - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - DarkNet53_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/DarkNet53.tgz" - if not os.path.exists("./DarkNet53/inference.pdiparams"): - wget.download(DarkNet53_url, out="./") - tar = tarfile.open("DarkNet53.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-2 DarkNet53 outputs with true val - """ - check_model_exist() - - file_path = "./DarkNet53" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - # Cause of the diff error, delete the conv2d_fusion when trt_version < 8.0 - ver = paddle_infer.get_trt_compile_version() - if ver[0] * 1000 + ver[1] * 100 + ver[2] * 10 < 8000: - delete_op_list = ["conv2d_fusion"] - else: - delete_op_list = [] - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=1e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - delete_op_list=delete_op_list, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 DarkNet53 outputs with true val - """ - check_model_exist() - - file_path = "./DarkNet53" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=1e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 DarkNet53 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./DarkNet53" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=1e-2, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_DarkNet53_trt_fp32.py b/inference/python_api_test/test_class_model/test_DarkNet53_trt_fp32.py deleted file mode 100644 index 6ba5813964..0000000000 --- a/inference/python_api_test/test_class_model/test_DarkNet53_trt_fp32.py +++ /dev/null @@ -1,178 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test DarkNet53 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np -import paddle.inference as paddle_infer - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - DarkNet53_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/DarkNet53.tgz" - if not os.path.exists("./DarkNet53/inference.pdiparams"): - wget.download(DarkNet53_url, out="./") - tar = tarfile.open("DarkNet53.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-2 DarkNet53 outputs with true val - """ - check_model_exist() - - file_path = "./DarkNet53" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - # Cause of the diff error, delete the conv2d_fusion when trt_version < 8.0 - ver = paddle_infer.get_trt_compile_version() - if ver[0] * 1000 + ver[1] * 100 + ver[2] * 10 < 8000: - delete_op_list = ["conv2d_fusion"] - else: - delete_op_list = [] - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - delete_op_list=delete_op_list, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 DarkNet53 outputs with true val - """ - check_model_exist() - - file_path = "./DarkNet53" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 DarkNet53 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./DarkNet53" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DarkNet53/inference.pdmodel", - params_file="./DarkNet53/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_DenseNet121_trt_fp16.py b/inference/python_api_test/test_class_model/test_DenseNet121_trt_fp16.py deleted file mode 100644 index 043fd76e8e..0000000000 --- a/inference/python_api_test/test_class_model/test_DenseNet121_trt_fp16.py +++ /dev/null @@ -1,189 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test DenseNet121 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - DenseNet121_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/DenseNet121.tgz" - if not os.path.exists("./DenseNet121/inference.pdiparams"): - wget.download(DenseNet121_url, out="./") - tar = tarfile.open("DenseNet121.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-2 DenseNet121 outputs with true val - """ - check_model_exist() - - file_path = "./DenseNet121" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=5e-3, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 DenseNet121 outputs with true val - """ - check_model_exist() - - file_path = "./DenseNet121" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite1 = InferenceTest() - test_suite1.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - test_suite1.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=5e-3, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - tuned=True, - ) - del test_suite1 # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=5e-3, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 DenseNet121 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./DenseNet121" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=5e-3, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_DenseNet121_trt_fp32.py b/inference/python_api_test/test_class_model/test_DenseNet121_trt_fp32.py deleted file mode 100644 index fee6c3d5dc..0000000000 --- a/inference/python_api_test/test_class_model/test_DenseNet121_trt_fp32.py +++ /dev/null @@ -1,170 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test DenseNet121 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - DenseNet121_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/DenseNet121.tgz" - if not os.path.exists("./DenseNet121/inference.pdiparams"): - wget.download(DenseNet121_url, out="./") - tar = tarfile.open("DenseNet121.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-2 DenseNet121 outputs with true val - """ - check_model_exist() - - file_path = "./DenseNet121" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_bz1(): - """ - compared trt fp32 batch_size=1 DenseNet121 outputs with true val - """ - check_model_exist() - - file_path = "./DenseNet121" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 DenseNet121 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./DenseNet121" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./DenseNet121/inference.pdmodel", - params_file="./DenseNet121/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_EfficientNetB0_slim.py b/inference/python_api_test/test_class_model/test_EfficientNetB0_slim.py index a1d7161a18..99eafc7052 100644 --- a/inference/python_api_test/test_class_model/test_EfficientNetB0_slim.py +++ b/inference/python_api_test/test_class_model/test_EfficientNetB0_slim.py @@ -69,77 +69,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 EfficientNetB0_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\EfficientNetB0_act_qat\\inference.pdmodel", - params_file=".\\EfficientNetB0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./EfficientNetB0_act_qat/inference.pdmodel", - params_file="./EfficientNetB0_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./EfficientNetB0_act_qat/inference.pdmodel", - params_file="./EfficientNetB0_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./EfficientNetB0_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\EfficientNetB0_act_qat\\inference.pdmodel", - params_file=".\\EfficientNetB0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./EfficientNetB0_act_qat/inference.pdmodel", - params_file="./EfficientNetB0_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=8e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./EfficientNetB0_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=1.96, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_GhostNet_x1_0_slim.py b/inference/python_api_test/test_class_model/test_GhostNet_x1_0_slim.py index 77127086f9..3411b5bf6c 100644 --- a/inference/python_api_test/test_class_model/test_GhostNet_x1_0_slim.py +++ b/inference/python_api_test/test_class_model/test_GhostNet_x1_0_slim.py @@ -69,77 +69,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 GhostNet_x1_0_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\GhostNet_x1_0_act_qat\\inference.pdmodel", - params_file=".\\GhostNet_x1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./GhostNet_x1_0_act_qat/inference.pdmodel", - params_file="./GhostNet_x1_0_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./GhostNet_x1_0_act_qat/inference.pdmodel", - params_file="./GhostNet_x1_0_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./GhostNet_x1_0_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\GhostNet_x1_0_act_qat\\inference.pdmodel", - params_file=".\\GhostNet_x1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./GhostNet_x1_0_act_qat/inference.pdmodel", - params_file="./GhostNet_x1_0_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=1e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./GhostNet_x1_0_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=1.85, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_GoogLeNet_trt_fp16.py b/inference/python_api_test/test_class_model/test_GoogLeNet_trt_fp16.py deleted file mode 100644 index 510dce3494..0000000000 --- a/inference/python_api_test/test_class_model/test_GoogLeNet_trt_fp16.py +++ /dev/null @@ -1,173 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test GoogLeNet model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - GoogLeNet_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/GoogLeNet.tgz" - if not os.path.exists("./GoogLeNet/inference.pdiparams"): - wget.download(GoogLeNet_url, out="./") - tar = tarfile.open("GoogLeNet.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 GoogLeNet outputs with true val - """ - check_model_exist() - - file_path = "./GoogLeNet" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 2 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 GoogLeNet outputs with true val - """ - check_model_exist() - - file_path = "./GoogLeNet" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 GoogLeNet multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./GoogLeNet" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=2e-2, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_GoogLeNet_trt_fp32.py b/inference/python_api_test/test_class_model/test_GoogLeNet_trt_fp32.py deleted file mode 100644 index d5f17b47a7..0000000000 --- a/inference/python_api_test/test_class_model/test_GoogLeNet_trt_fp32.py +++ /dev/null @@ -1,187 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test GoogLeNet model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np -import paddle - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - GoogLeNet_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/GoogLeNet.tgz" - if not os.path.exists("./GoogLeNet/inference.pdiparams"): - wget.download(GoogLeNet_url, out="./") - tar = tarfile.open("GoogLeNet.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - test_suite.config_test() - - -# note: the diff is 2.294778823852539e-05 in cuda10.2 -# note: the diff is 2.5510787963867188e-05 in win -if paddle.version.cuda() == "10.2" or "win" in sys.platform: - delta = 1e-4 -else: - delta = 1e-5 - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-2 GoogLeNet outputs with true val - """ - check_model_exist() - - file_path = "./GoogLeNet" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - delta=delta, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 GoogLeNet outputs with true val - """ - check_model_exist() - - file_path = "./GoogLeNet" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - delta=1e-4, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 GoogLeNet multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./GoogLeNet" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./GoogLeNet/inference.pdmodel", - params_file="./GoogLeNet/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=delta, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_InceptionV3_slim.py b/inference/python_api_test/test_class_model/test_InceptionV3_slim.py index b8ea0af44e..219e7c3128 100644 --- a/inference/python_api_test/test_class_model/test_InceptionV3_slim.py +++ b/inference/python_api_test/test_class_model/test_InceptionV3_slim.py @@ -67,77 +67,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 InceptionV3_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 299 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\InceptionV3_act_qat\\inference.pdmodel", - params_file=".\\InceptionV3_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./InceptionV3_act_qat/inference.pdmodel", - params_file="./InceptionV3_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./InceptionV3_act_qat/inference.pdmodel", - params_file="./InceptionV3_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./InceptionV3_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\InceptionV3_act_qat\\inference.pdmodel", - params_file=".\\InceptionV3_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./InceptionV3_act_qat/inference.pdmodel", - params_file="./InceptionV3_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=1e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./InceptionV3_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=2.23, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_MobileNetV1_slim.py b/inference/python_api_test/test_class_model/test_MobileNetV1_slim.py index 79a618d8ae..f4d534f245 100644 --- a/inference/python_api_test/test_class_model/test_MobileNetV1_slim.py +++ b/inference/python_api_test/test_class_model/test_MobileNetV1_slim.py @@ -67,77 +67,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 MobileNetV1_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\MobileNetV1_act_qat\\inference.pdmodel", - params_file=".\\MobileNetV1_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./MobileNetV1_act_qat/inference.pdmodel", - params_file="./MobileNetV1_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./MobileNetV1_act_qat/inference.pdmodel", - params_file="./MobileNetV1_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./MobileNetV1_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\MobileNetV1_act_qat\\inference.pdmodel", - params_file=".\\MobileNetV1_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./MobileNetV1_act_qat/inference.pdmodel", - params_file="./MobileNetV1_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=2e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./MobileNetV1_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=0.48, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_MobileNetV3_large_x1_0_slim.py b/inference/python_api_test/test_class_model/test_MobileNetV3_large_x1_0_slim.py index eed5de6d42..cfa066b79e 100644 --- a/inference/python_api_test/test_class_model/test_MobileNetV3_large_x1_0_slim.py +++ b/inference/python_api_test/test_class_model/test_MobileNetV3_large_x1_0_slim.py @@ -69,77 +69,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 MobileNetV3_large_x1_0_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\MobileNetV3_large_x1_0_act_qat\\inference.pdmodel", - params_file=".\\MobileNetV3_large_x1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./MobileNetV3_large_x1_0_act_qat/inference.pdmodel", - params_file="./MobileNetV3_large_x1_0_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./MobileNetV3_large_x1_0_act_qat/inference.pdmodel", - params_file="./MobileNetV3_large_x1_0_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./MobileNetV3_large_x1_0_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\MobileNetV3_large_x1_0_act_qat\\inference.pdmodel", - params_file=".\\MobileNetV3_large_x1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./MobileNetV3_large_x1_0_act_qat/inference.pdmodel", - params_file="./MobileNetV3_large_x1_0_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=5e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./MobileNetV3_large_x1_0_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=1.25, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_MobileNetV3_large_x1_0_ssld_slim.py b/inference/python_api_test/test_class_model/test_MobileNetV3_large_x1_0_ssld_slim.py index 06890fe45b..72e17770b4 100644 --- a/inference/python_api_test/test_class_model/test_MobileNetV3_large_x1_0_ssld_slim.py +++ b/inference/python_api_test/test_class_model/test_MobileNetV3_large_x1_0_ssld_slim.py @@ -69,77 +69,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 MobileNetV3_large_x1_0_ssld_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\MobileNetV3_large_x1_0_ssld_act_qat\\inference.pdmodel", - params_file=".\\MobileNetV3_large_x1_0_ssld_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./MobileNetV3_large_x1_0_ssld_act_qat/inference.pdmodel", - params_file="./MobileNetV3_large_x1_0_ssld_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./MobileNetV3_large_x1_0_ssld_act_qat/inference.pdmodel", - params_file="./MobileNetV3_large_x1_0_ssld_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./MobileNetV3_large_x1_0_ssld_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\MobileNetV3_large_x1_0_ssld_act_qat\\inference.pdmodel", - params_file=".\\MobileNetV3_large_x1_0_ssld_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./MobileNetV3_large_x1_0_ssld_act_qat/inference.pdmodel", - params_file="./MobileNetV3_large_x1_0_ssld_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=3e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./MobileNetV3_large_x1_0_ssld_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=1.24, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_PPHGNet_tiny_slim.py b/inference/python_api_test/test_class_model/test_PPHGNet_tiny_slim.py index f648340bc9..4e61adf600 100644 --- a/inference/python_api_test/test_class_model/test_PPHGNet_tiny_slim.py +++ b/inference/python_api_test/test_class_model/test_PPHGNet_tiny_slim.py @@ -69,77 +69,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 PPHGNet_tiny_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\PPHGNet_tiny_act_qat\\inference.pdmodel", - params_file=".\\PPHGNet_tiny_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./PPHGNet_tiny_act_qat/inference.pdmodel", - params_file="./PPHGNet_tiny_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./PPHGNet_tiny_act_qat/inference.pdmodel", - params_file="./PPHGNet_tiny_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./PPHGNet_tiny_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\PPHGNet_tiny_act_qat\\inference.pdmodel", - params_file=".\\PPHGNet_tiny_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./PPHGNet_tiny_act_qat/inference.pdmodel", - params_file="./PPHGNet_tiny_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=2e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./PPHGNet_tiny_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=1.77, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_PPLCNetV2_base_slim.py b/inference/python_api_test/test_class_model/test_PPLCNetV2_base_slim.py index 4e12d08e72..f38647c1dc 100644 --- a/inference/python_api_test/test_class_model/test_PPLCNetV2_base_slim.py +++ b/inference/python_api_test/test_class_model/test_PPLCNetV2_base_slim.py @@ -69,77 +69,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 PPLCNetV2_base_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\PPLCNetV2_base_act_qat\\inference.pdmodel", - params_file=".\\PPLCNetV2_base_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./PPLCNetV2_base_act_qat/inference.pdmodel", - params_file="./PPLCNetV2_base_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./PPLCNetV2_base_act_qat/inference.pdmodel", - params_file="./PPLCNetV2_base_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./PPLCNetV2_base_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\PPLCNetV2_base_act_qat\\inference.pdmodel", - params_file=".\\PPLCNetV2_base_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./PPLCNetV2_base_act_qat/inference.pdmodel", - params_file="./PPLCNetV2_base_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=3e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./PPLCNetV2_base_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=0.9, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_PPLCNet_x1_0_slim.py b/inference/python_api_test/test_class_model/test_PPLCNet_x1_0_slim.py index c0b3f11bb1..7df3d61f49 100644 --- a/inference/python_api_test/test_class_model/test_PPLCNet_x1_0_slim.py +++ b/inference/python_api_test/test_class_model/test_PPLCNet_x1_0_slim.py @@ -69,83 +69,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 PPLCNet_x1_0_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\PPLCNet_x1_0_act_qat\\inference.pdmodel", - params_file=".\\PPLCNet_x1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./PPLCNet_x1_0_act_qat/inference.pdmodel", - params_file="./PPLCNet_x1_0_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\PPLCNet_x1_0_act_qat\\inference.pdmodel", - params_file=".\\PPLCNet_x1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./PPLCNet_x1_0_act_qat/inference.pdmodel", - params_file="./PPLCNet_x1_0_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./PPLCNet_x1_0_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\PPLCNet_x1_0_act_qat\\inference.pdmodel", - params_file=".\\PPLCNet_x1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./PPLCNet_x1_0_act_qat/inference.pdmodel", - params_file="./PPLCNet_x1_0_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=5e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./PPLCNet_x1_0_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=0.74, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_ResNet50_vd_slim.py b/inference/python_api_test/test_class_model/test_ResNet50_vd_slim.py index dd40a8708e..76d0d5b726 100644 --- a/inference/python_api_test/test_class_model/test_ResNet50_vd_slim.py +++ b/inference/python_api_test/test_class_model/test_ResNet50_vd_slim.py @@ -67,77 +67,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 ResNet50_vd_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\ResNet50_vd_act_qat\\inference.pdmodel", - params_file=".\\ResNet50_vd_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./ResNet50_vd_act_qat/inference.pdmodel", - params_file="./ResNet50_vd_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ResNet50_vd_act_qat/inference.pdmodel", - params_file="./ResNet50_vd_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./ResNet50_vd_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\ResNet50_vd_act_qat\\inference.pdmodel", - params_file=".\\ResNet50_vd_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./ResNet50_vd_act_qat/inference.pdmodel", - params_file="./ResNet50_vd_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=1e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./ResNet50_vd_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=1.65, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_ShuffleNetV2_x1_0_slim.py b/inference/python_api_test/test_class_model/test_ShuffleNetV2_x1_0_slim.py index da92413f7f..d68caf8d05 100644 --- a/inference/python_api_test/test_class_model/test_ShuffleNetV2_x1_0_slim.py +++ b/inference/python_api_test/test_class_model/test_ShuffleNetV2_x1_0_slim.py @@ -69,83 +69,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 ShuffleNetV2_x1_0_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\ShuffleNetV2_x1_0_act_qat\\inference.pdmodel", - params_file=".\\ShuffleNetV2_x1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./ShuffleNetV2_x1_0_act_qat/inference.pdmodel", - params_file="./ShuffleNetV2_x1_0_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\ShuffleNetV2_x1_0_act_qat\\inference.pdmodel", - params_file=".\\ShuffleNetV2_x1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./ShuffleNetV2_x1_0_act_qat/inference.pdmodel", - params_file="./ShuffleNetV2_x1_0_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./ShuffleNetV2_x1_0_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\ShuffleNetV2_x1_0_act_qat\\inference.pdmodel", - params_file=".\\ShuffleNetV2_x1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./ShuffleNetV2_x1_0_act_qat/inference.pdmodel", - params_file="./ShuffleNetV2_x1_0_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=1e-1, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./ShuffleNetV2_x1_0_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=1.13, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_SqueezeNet1_0_slim.py b/inference/python_api_test/test_class_model/test_SqueezeNet1_0_slim.py index 2aceefb758..70cf4b40a5 100644 --- a/inference/python_api_test/test_class_model/test_SqueezeNet1_0_slim.py +++ b/inference/python_api_test/test_class_model/test_SqueezeNet1_0_slim.py @@ -69,77 +69,7 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.win -@pytest.mark.server -@pytest.mark.slim -@pytest.mark.trt_int8 -def test_trt_int8_more_bz(): - """ - compared trt_int8 batch_size=1 SqueezeNet1_0_act_qat outputs with true val - """ - check_model_exist() - file_path = "./case_image_data" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\SqueezeNet1_0_act_qat\\inference.pdmodel", - params_file=".\\SqueezeNet1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./SqueezeNet1_0_act_qat/inference.pdmodel", - params_file="./SqueezeNet1_0_act_qat/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./SqueezeNet1_0_act_qat/inference.pdmodel", - params_file="./SqueezeNet1_0_act_qat/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./SqueezeNet1_0_act_qat/", input_data_dict=input_data_dict, device="gpu" - ) - del test_suite - - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config( - model_file=".\\SqueezeNet1_0_act_qat\\inference.pdmodel", - params_file=".\\SqueezeNet1_0_act_qat\\inference.pdiparams", - ) - else: - test_suite.load_config( - model_file="./SqueezeNet1_0_act_qat/inference.pdmodel", - params_file="./SqueezeNet1_0_act_qat/inference.pdiparams", - ) - test_suite.trt_bz1_slim_test( - input_data_dict, - output_data_dict, - repeat=100, - delta=2e-2, - max_batch_size=max_batch_size, - precision="trt_int8", - min_subgraph_size=30, - dynamic=True, - shape_range_file="./SqueezeNet1_0_act_qat/shape_range.pbtxt", - # use_calib_mode=True, - with_benchmark=True, - base_latency_ms=0.57, - benchmark_threshold=5e-2, - ) - - del test_suite # destroy class to save memory @pytest.mark.win diff --git a/inference/python_api_test/test_class_model/test_ViT_base_patch16_224_trt_fp16.py b/inference/python_api_test/test_class_model/test_ViT_base_patch16_224_trt_fp16.py deleted file mode 100644 index eb85c57995..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_base_patch16_224_trt_fp16.py +++ /dev/null @@ -1,173 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_base_patch16_224 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_base_patch16_224_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_base_patch16_224.tgz" - if not os.path.exists("./ViT_base_patch16_224/inference.pdiparams"): - wget.download(ViT_base_patch16_224_url, out="./") - tar = tarfile.open("ViT_base_patch16_224.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_base_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=5e-4, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16_more_bz_precision -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_base_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=5e-4, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 ViT_base_patch16_224 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_224" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - delta=5e-4, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_base_patch16_224_trt_fp32.py b/inference/python_api_test/test_class_model/test_ViT_base_patch16_224_trt_fp32.py deleted file mode 100644 index 7b5f3adb09..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_base_patch16_224_trt_fp32.py +++ /dev/null @@ -1,169 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_base_patch16_224 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_base_patch16_224_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_base_patch16_224.tgz" - if not os.path.exists("./ViT_base_patch16_224/inference.pdiparams"): - wget.download(ViT_base_patch16_224_url, out="./") - tar = tarfile.open("ViT_base_patch16_224.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_base_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_base_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 ViT_base_patch16_224 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_224" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_224/inference.pdmodel", - params_file="./ViT_base_patch16_224/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_base_patch16_384_trt_fp16.py b/inference/python_api_test/test_class_model/test_ViT_base_patch16_384_trt_fp16.py deleted file mode 100644 index 9c76be7e9b..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_base_patch16_384_trt_fp16.py +++ /dev/null @@ -1,172 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_base_patch16_384 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_base_patch16_384_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_base_patch16_384.tgz" - if not os.path.exists("./ViT_base_patch16_384/inference.pdiparams"): - wget.download(ViT_base_patch16_384_url, out="./") - tar = tarfile.open("ViT_base_patch16_384.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_base_patch16_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - delta=0.005, - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16_more_bz_precision -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_base_patch16_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - delta=0.005, - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 ViT_base_patch16_384 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_384" - images_size = 384 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - delta=0.005, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_base_patch16_384_trt_fp32.py b/inference/python_api_test/test_class_model/test_ViT_base_patch16_384_trt_fp32.py deleted file mode 100644 index 706829a398..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_base_patch16_384_trt_fp32.py +++ /dev/null @@ -1,170 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_base_patch16_384 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_base_patch16_384_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_base_patch16_384.tgz" - if not os.path.exists("./ViT_base_patch16_384/inference.pdiparams"): - wget.download(ViT_base_patch16_384_url, out="./") - tar = tarfile.open("ViT_base_patch16_384.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_base_patch16_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_base_patch16_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 ViT_base_patch16_384 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch16_384" - images_size = 384 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch16_384/inference.pdmodel", - params_file="./ViT_base_patch16_384/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_base_patch32_384_trt_fp16.py b/inference/python_api_test/test_class_model/test_ViT_base_patch32_384_trt_fp16.py deleted file mode 100644 index 8a2de786bf..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_base_patch32_384_trt_fp16.py +++ /dev/null @@ -1,172 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_base_patch32_384 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_base_patch32_384_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_base_patch32_384.tgz" - if not os.path.exists("./ViT_base_patch32_384/inference.pdiparams"): - wget.download(ViT_base_patch32_384_url, out="./") - tar = tarfile.open("ViT_base_patch32_384.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_base_patch32_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch32_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.01, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16_more_bz_precision -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_base_patch32_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch32_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.01, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 ViT_base_patch32_384 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch32_384" - images_size = 384 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - delta=0.01, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_base_patch32_384_trt_fp32.py b/inference/python_api_test/test_class_model/test_ViT_base_patch32_384_trt_fp32.py deleted file mode 100644 index 043189c3c4..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_base_patch32_384_trt_fp32.py +++ /dev/null @@ -1,170 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_base_patch32_384 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_base_patch32_384_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_base_patch32_384.tgz" - if not os.path.exists("./ViT_base_patch32_384/inference.pdiparams"): - wget.download(ViT_base_patch32_384_url, out="./") - tar = tarfile.open("ViT_base_patch32_384.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_base_patch32_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch32_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_base_patch32_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch32_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 ViT_base_patch32_384 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_base_patch32_384" - images_size = 384 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_base_patch32_384/inference.pdmodel", - params_file="./ViT_base_patch32_384/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_large_patch16_224_trt_fp16.py b/inference/python_api_test/test_class_model/test_ViT_large_patch16_224_trt_fp16.py deleted file mode 100644 index d2747718f3..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_large_patch16_224_trt_fp16.py +++ /dev/null @@ -1,173 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_large_patch16_224 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_large_patch16_224_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_large_patch16_224.tgz" - if not os.path.exists("./ViT_large_patch16_224/inference.pdiparams"): - wget.download(ViT_large_patch16_224_url, out="./") - tar = tarfile.open("ViT_large_patch16_224.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_large_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.003, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16_more_bz_precision -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_large_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.003, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 ViT_large_patch16_224 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_224" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=0.003, - min_subgraph_size=1, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_large_patch16_224_trt_fp32.py b/inference/python_api_test/test_class_model/test_ViT_large_patch16_224_trt_fp32.py deleted file mode 100644 index 9ae655feb9..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_large_patch16_224_trt_fp32.py +++ /dev/null @@ -1,170 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_large_patch16_224 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_large_patch16_224_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_large_patch16_224.tgz" - if not os.path.exists("./ViT_large_patch16_224/inference.pdiparams"): - wget.download(ViT_large_patch16_224_url, out="./") - tar = tarfile.open("ViT_large_patch16_224.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_large_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_large_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 ViT_large_patch16_224 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_224" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_224/inference.pdmodel", - params_file="./ViT_large_patch16_224/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_large_patch16_384_trt_fp16.py b/inference/python_api_test/test_class_model/test_ViT_large_patch16_384_trt_fp16.py deleted file mode 100644 index 2fdffb4c60..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_large_patch16_384_trt_fp16.py +++ /dev/null @@ -1,189 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_large_patch16_384 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_large_patch16_384_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_large_patch16_384.tgz" - if not os.path.exists("./ViT_large_patch16_384/inference.pdiparams"): - wget.download(ViT_large_patch16_384_url, out="./") - tar = tarfile.open("ViT_large_patch16_384.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_large_patch16_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.001, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16_more_bz_precision -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_large_patch16_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite1 = InferenceTest() - test_suite1.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - test_suite1.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.001, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - tuned=True, - ) - - del test_suite1 # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.001, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 ViT_large_patch16_384 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_384" - images_size = 384 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=0.001, - min_subgraph_size=1, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_large_patch16_384_trt_fp32.py b/inference/python_api_test/test_class_model/test_ViT_large_patch16_384_trt_fp32.py deleted file mode 100644 index 36733408e5..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_large_patch16_384_trt_fp32.py +++ /dev/null @@ -1,170 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_large_patch16_384 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_large_patch16_384_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_large_patch16_384.tgz" - if not os.path.exists("./ViT_large_patch16_384/inference.pdiparams"): - wget.download(ViT_large_patch16_384_url, out="./") - tar = tarfile.open("ViT_large_patch16_384.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_large_patch16_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_large_patch16_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 ViT_large_patch16_384 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch16_384" - images_size = 384 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch16_384/inference.pdmodel", - params_file="./ViT_large_patch16_384/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_large_patch32_384_trt_fp16.py b/inference/python_api_test/test_class_model/test_ViT_large_patch32_384_trt_fp16.py deleted file mode 100644 index 4fd90e3198..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_large_patch32_384_trt_fp16.py +++ /dev/null @@ -1,172 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_large_patch32_384 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_large_patch32_384_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_large_patch32_384.tgz" - if not os.path.exists("./ViT_large_patch32_384/inference.pdiparams"): - wget.download(ViT_large_patch32_384_url, out="./") - tar = tarfile.open("ViT_large_patch32_384.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_large_patch32_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch32_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.005, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16_more_bz_precision -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_large_patch32_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch32_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.01, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 ViT_large_patch32_384 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch32_384" - images_size = 384 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - delta=0.005, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_large_patch32_384_trt_fp32.py b/inference/python_api_test/test_class_model/test_ViT_large_patch32_384_trt_fp32.py deleted file mode 100644 index 5d04df682c..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_large_patch32_384_trt_fp32.py +++ /dev/null @@ -1,170 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_large_patch32_384 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_large_patch32_384_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_large_patch32_384.tgz" - if not os.path.exists("./ViT_large_patch32_384/inference.pdiparams"): - wget.download(ViT_large_patch32_384_url, out="./") - tar = tarfile.open("ViT_large_patch32_384.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_large_patch32_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch32_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_large_patch32_384 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch32_384" - images_size = 384 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 ViT_large_patch32_384 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_large_patch32_384" - images_size = 384 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_large_patch32_384/inference.pdmodel", - params_file="./ViT_large_patch32_384/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_small_patch16_224_trt_fp16.py b/inference/python_api_test/test_class_model/test_ViT_small_patch16_224_trt_fp16.py deleted file mode 100644 index 59c656d1ef..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_small_patch16_224_trt_fp16.py +++ /dev/null @@ -1,170 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_small_patch16_224 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_small_patch16_224_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_small_patch16_224.tgz" - if not os.path.exists("./ViT_small_patch16_224/inference.pdiparams"): - wget.download(ViT_small_patch16_224_url, out="./") - tar = tarfile.open("ViT_small_patch16_224.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_small_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_small_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.005, - max_batch_size=max_batch_size, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16_more_bz_precision -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ViT_small_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_small_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.007, - max_batch_size=max_batch_size, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 ViT_small_patch16_224 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_small_patch16_224" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=0.005, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_ViT_small_patch16_224_trt_fp32.py b/inference/python_api_test/test_class_model/test_ViT_small_patch16_224_trt_fp32.py deleted file mode 100644 index a4e719c5f3..0000000000 --- a/inference/python_api_test/test_class_model/test_ViT_small_patch16_224_trt_fp32.py +++ /dev/null @@ -1,167 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ViT_small_patch16_224 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ViT_small_patch16_224_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/ViT_small_patch16_224.tgz" - if not os.path.exists("./ViT_small_patch16_224/inference.pdiparams"): - wget.download(ViT_small_patch16_224_url, out="./") - tar = tarfile.open("ViT_small_patch16_224.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_small_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_small_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 ViT_small_patch16_224 outputs with true val - """ - check_model_exist() - - file_path = "./ViT_small_patch16_224" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 ViT_small_patch16_224 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ViT_small_patch16_224" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ViT_small_patch16_224/inference.pdmodel", - params_file="./ViT_small_patch16_224/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_pcpvt_base_trt_fp16.py b/inference/python_api_test/test_class_model/test_pcpvt_base_trt_fp16.py deleted file mode 100644 index 61762e7301..0000000000 --- a/inference/python_api_test/test_class_model/test_pcpvt_base_trt_fp16.py +++ /dev/null @@ -1,170 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test pcpvt_base model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - pcpvt_base_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.1.1/class/pcpvt_base.tgz" - if not os.path.exists("./pcpvt_base/inference.pdiparams"): - wget.download(pcpvt_base_url, out="./") - tar = tarfile.open("pcpvt_base.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 pcpvt_base outputs with true val - """ - check_model_exist() - - file_path = "./pcpvt_base" - images_size = 224 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=1e-1, - max_batch_size=1, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - shape_range_file="./pcpvt_base/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 pcpvt_base outputs with true val - """ - check_model_exist() - - file_path = "./pcpvt_base" - images_size = 224 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=5e-1, - max_batch_size=1, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - shape_range_file="./pcpvt_base/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 pcpvt_base multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./pcpvt_base" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=1e-1, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_pcpvt_base_trt_fp32.py b/inference/python_api_test/test_class_model/test_pcpvt_base_trt_fp32.py deleted file mode 100644 index 609009ff40..0000000000 --- a/inference/python_api_test/test_class_model/test_pcpvt_base_trt_fp32.py +++ /dev/null @@ -1,170 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test pcpvt_base model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - pcpvt_base_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.1.1/class/pcpvt_base.tgz" - if not os.path.exists("./pcpvt_base/inference.pdiparams"): - wget.download(pcpvt_base_url, out="./") - tar = tarfile.open("pcpvt_base.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 pcpvt_base outputs with true val - """ - check_model_exist() - - file_path = "./pcpvt_base" - images_size = 224 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=batch_size, - min_subgraph_size=1, - precision="trt_fp32", - delta=5e-2, - dynamic=True, - auto_tuned=True, - shape_range_file="./pcpvt_base/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 pcpvt_base outputs with true val - """ - check_model_exist() - - file_path = "./pcpvt_base" - images_size = 224 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=batch_size, - min_subgraph_size=1, - precision="trt_fp32", - delta=4e-1, - dynamic=True, - auto_tuned=True, - shape_range_file="./pcpvt_base/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 pcpvt_base multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./pcpvt_base" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pcpvt_base/inference.pdmodel", - params_file="./pcpvt_base/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=1e-2, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_resnet50_slim.py b/inference/python_api_test/test_class_model/test_resnet50_slim.py index 5a7a9c0c07..f02f72b413 100644 --- a/inference/python_api_test/test_class_model/test_resnet50_slim.py +++ b/inference/python_api_test/test_class_model/test_resnet50_slim.py @@ -56,43 +56,3 @@ def test_disable_gpu(): fake_input = np.random.randn(batch_size, 3, 224, 224).astype("float32") input_data_dict = {"image": fake_input} test_suite.disable_gpu_test(input_data_dict) - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_int8 -def test_int8_more_bz(): - """ - compared trt fp32 batch_size=1-10 resnet50 outputs with true val - """ - check_model_exist() - - file_path = "./resnet50_quant" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - if "win" in sys.platform: - test_suite.load_config(model_path=".\\resnet50_quant\\resnet50_quant") - else: - test_suite.load_config(model_path="./resnet50_quant/resnet50_quant") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - fake_output = np.array(npy_list[0:batch_size]).astype("float32") - fake_output = np.squeeze(fake_output, axis=(1, 2)) - input_data_dict = {"image": fake_input} - output_data_dict = {"save_infer_model/scale_0.tmp_0": fake_output} - test_suite.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=5e-1, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_int8", - use_calib_mode=True, - dynamic=True, - tuned=True, - ) - - del test_suite # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_resnet50_trt_fp16.py b/inference/python_api_test/test_class_model/test_resnet50_trt_fp16.py deleted file mode 100644 index d8b8719c67..0000000000 --- a/inference/python_api_test/test_class_model/test_resnet50_trt_fp16.py +++ /dev/null @@ -1,194 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test resnet50 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - resnet50_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/resnet50.tgz" - if not os.path.exists("./resnet50/inference.pdiparams"): - wget.download(resnet50_url, out="./") - tar = tarfile.open("resnet50.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-2 resnet50 outputs with true val - """ - check_model_exist() - - file_path = "./resnet50" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 resnet50 outputs with true val - """ - check_model_exist() - - file_path = "./resnet50" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite1 = InferenceTest() - test_suite1.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - test_suite1.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - tuned=True, - ) - del test_suite1 # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 resnet50 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./resnet50" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"xs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=2e-2, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_resnet50_trt_fp32.py b/inference/python_api_test/test_class_model/test_resnet50_trt_fp32.py deleted file mode 100644 index 30f41ebdad..0000000000 --- a/inference/python_api_test/test_class_model/test_resnet50_trt_fp32.py +++ /dev/null @@ -1,175 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test resnet50 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - resnet50_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/resnet50.tgz" - if not os.path.exists("./resnet50/inference.pdiparams"): - wget.download(resnet50_url, out="./") - tar = tarfile.open("resnet50.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-2 resnet50 outputs with true val - """ - check_model_exist() - - file_path = "./resnet50" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 resnet50 outputs with true val - """ - check_model_exist() - - file_path = "./resnet50" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 resnet50 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./resnet50" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./resnet50/inference.pdmodel", - params_file="./resnet50/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_swin_transformer_trt_fp16.py b/inference/python_api_test/test_class_model/test_swin_transformer_trt_fp16.py deleted file mode 100644 index 71f7ccddd2..0000000000 --- a/inference/python_api_test/test_class_model/test_swin_transformer_trt_fp16.py +++ /dev/null @@ -1,206 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test swin_transformer model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - swin_transformer_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/swin_transformer.tgz" - if not os.path.exists("./swin_transformer/inference.pdiparams"): - wget.download(swin_transformer_url, out="./") - tar = tarfile.open("swin_transformer.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 swin_transformer outputs with true val - """ - check_model_exist() - - file_path = "./swin_transformer" - images_size = 384 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=2e-2, - min_subgraph_size=1, - precision="trt_fp16", - max_batch_size=batch_size, - dynamic=True, - auto_tuned=True, - shape_range_file="./swin_transformer/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 swin_transformer outputs with true val - """ - check_model_exist() - - file_path = "./swin_transformer" - images_size = 384 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - # collect shape for trt - test_suite_c = InferenceTest() - test_suite_c.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - test_suite_c.collect_shape_info( - model_path=file_path, - input_data_dict=input_data_dict, - device="gpu", - ) - - del test_suite_c # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=2e-2, - min_subgraph_size=1, - precision="trt_fp16", - max_batch_size=batch_size, - dynamic=True, - shape_range_file="./swin_transformer/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 swin_transformer multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./swin_transformer" - images_size = 384 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - # collect shape for trt - test_suite_c = InferenceTest() - test_suite_c.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - test_suite_c.collect_shape_info( - model_path=file_path, - input_data_dict=input_data_dict, - device="gpu", - ) - - del test_suite_c # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=2e-2, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_swin_transformer_trt_fp32.py b/inference/python_api_test/test_class_model/test_swin_transformer_trt_fp32.py deleted file mode 100644 index 08a4e0a2d9..0000000000 --- a/inference/python_api_test/test_class_model/test_swin_transformer_trt_fp32.py +++ /dev/null @@ -1,190 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test swin_transformer model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - swin_transformer_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/swin_transformer.tgz" - if not os.path.exists("./swin_transformer/inference.pdiparams"): - wget.download(swin_transformer_url, out="./") - tar = tarfile.open("swin_transformer.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 swin_transformer outputs with true val - """ - check_model_exist() - - file_path = "./swin_transformer" - images_size = 384 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=1e-3, - min_subgraph_size=1, - precision="trt_fp32", - max_batch_size=batch_size, - dynamic=True, - auto_tuned=True, - shape_range_file="./swin_transformer/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 swin_transformer outputs with true val - """ - check_model_exist() - - file_path = "./swin_transformer" - images_size = 384 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - # collect shape for trt - test_suite_c = InferenceTest() - test_suite_c.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - test_suite_c.collect_shape_info( - model_path=file_path, - input_data_dict=input_data_dict, - device="gpu", - ) - - del test_suite_c # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=1e-4, - min_subgraph_size=1, - precision="trt_fp32", - max_batch_size=batch_size, - dynamic=True, - shape_range_file="./swin_transformer/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trtfp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 swin_transformer multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./swin_transformer" - images_size = 384 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./swin_transformer/inference.pdmodel", - params_file="./swin_transformer/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - repeat=1, - delta=1e-4, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_tnt_small_trt_fp16.py b/inference/python_api_test/test_class_model/test_tnt_small_trt_fp16.py deleted file mode 100644 index d33d70a937..0000000000 --- a/inference/python_api_test/test_class_model/test_tnt_small_trt_fp16.py +++ /dev/null @@ -1,214 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test TNT_small model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - tnt_small_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/TNT_small.tgz" - if not os.path.exists("./TNT_small/inference.pdiparams"): - wget.download(tnt_small_url, out="./") - tar = tarfile.open("TNT_small.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 TNT_small outputs with true val - """ - check_model_exist() - - file_path = "./TNT_small" - images_size = 224 - batch_size_pool = [1] - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=1e-2, - max_batch_size=10, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - shape_range_file=file_path + "/shape_range.pbtxt", - delete_pass_list=["trt_skip_layernorm_fuse_pass"], - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 TNT_small outputs with true val - """ - check_model_exist() - - file_path = "./TNT_small" - images_size = 224 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - # collect shape for trt - test_suite_c = InferenceTest() - test_suite_c.load_config( - model_file=file_path + "/inference.pdmodel", - params_file=file_path + "/inference.pdiparams", - ) - test_suite_c.collect_shape_info( - model_path=file_path, - input_data_dict=input_data_dict, - device="gpu", - ) - del test_suite_c - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=1e-2, - max_batch_size=10, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - shape_range_file=file_path + "/shape_range.pbtxt", - delete_pass_list=["trt_skip_layernorm_fuse_pass"], - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16_multi_thread_bz1_precision -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 TNT_small multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./TNT_small" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - # collect shape for trt - test_suite_c = InferenceTest() - test_suite_c.load_config( - model_file=file_path + "/inference.pdmodel", - params_file=file_path + "/inference.pdiparams", - ) - test_suite_c.collect_shape_info( - model_path=file_path, - input_data_dict=input_data_dict, - device="gpu", - ) - del test_suite_c - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=1e-2, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - shape_range_file="./TNT_small/shape_range.pbtxt", - delete_pass_list=["trt_skip_layernorm_fuse_pass"], - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_tnt_small_trt_fp32.py b/inference/python_api_test/test_class_model/test_tnt_small_trt_fp32.py deleted file mode 100644 index 1bdc15fb94..0000000000 --- a/inference/python_api_test/test_class_model/test_tnt_small_trt_fp32.py +++ /dev/null @@ -1,208 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test TNT_small model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - tnt_small_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/TNT_small.tgz" - if not os.path.exists("./TNT_small/inference.pdiparams"): - wget.download(tnt_small_url, out="./") - tar = tarfile.open("TNT_small.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-2 TNT_small outputs with true val - """ - check_model_exist() - - file_path = "./TNT_small" - images_size = 224 - batch_size_pool = [1, 2] - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=10, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - shape_range_file=file_path + "/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 TNT_small outputs with true val - """ - check_model_exist() - - file_path = "./TNT_small" - images_size = 224 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - # collect shape for trt - test_suite_c = InferenceTest() - test_suite_c.load_config( - model_file=file_path + "/inference.pdmodel", - params_file=file_path + "/inference.pdiparams", - ) - test_suite_c.collect_shape_info( - model_path=file_path, - input_data_dict=input_data_dict, - device="gpu", - ) - del test_suite_c - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=10, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - shape_range_file=file_path + "/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 TNT_small multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./TNT_small" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - # collect shape for trt - test_suite_c = InferenceTest() - test_suite_c.load_config( - model_file=file_path + "/inference.pdmodel", - params_file=file_path + "/inference.pdiparams", - ) - test_suite_c.collect_shape_info( - model_path=file_path, - input_data_dict=input_data_dict, - device="gpu", - ) - del test_suite_c - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./TNT_small/inference.pdmodel", - params_file="./TNT_small/inference.pdiparams", - ) - - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - shape_range_file=file_path + "/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_vgg11_trt_fp16.py b/inference/python_api_test/test_class_model/test_vgg11_trt_fp16.py deleted file mode 100644 index 72082a00eb..0000000000 --- a/inference/python_api_test/test_class_model/test_vgg11_trt_fp16.py +++ /dev/null @@ -1,183 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test vgg11 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - vgg11_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/vgg11.tgz" - if not os.path.exists("./vgg11/inference.pdiparams"): - wget.download(vgg11_url, out="./") - tar = tarfile.open("vgg11.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-2 vgg11 outputs with true val - """ - check_model_exist() - - file_path = "./vgg11" - images_size = 224 - batch_size_pool = [1, 2] - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=1e-2, - gpu_mem=3000, - max_batch_size=batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 vgg11 outputs with true val - """ - check_model_exist() - - file_path = "./vgg11" - images_size = 224 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=5e-3, - gpu_mem=3000, - max_batch_size=batch_size, - min_subgraph_size=1, - precision="trt_fp16", - tuned=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 vgg11 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./vgg11" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=1e-3, - gpu_mem=3000, - max_batch_size=batch_size, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_class_model/test_vgg11_trt_fp32.py b/inference/python_api_test/test_class_model/test_vgg11_trt_fp32.py deleted file mode 100644 index fbd575a62a..0000000000 --- a/inference/python_api_test/test_class_model/test_vgg11_trt_fp32.py +++ /dev/null @@ -1,173 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test vgg11 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - vgg11_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/class/vgg11.tgz" - if not os.path.exists("./vgg11/inference.pdiparams"): - wget.download(vgg11_url, out="./") - tar = tarfile.open("vgg11.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-2 vgg11 outputs with true val - """ - check_model_exist() - - file_path = "./vgg11" - images_size = 224 - batch_size_pool = [1, 2] - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 vgg11 outputs with true val - """ - check_model_exist() - - file_path = "./vgg11" - images_size = 224 - batch_size_pool = [1] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory\ - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trtfp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 vgg11 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./vgg11" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./vgg11/inference.pdmodel", - params_file="./vgg11/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_det_model/run.sh b/inference/python_api_test/test_det_model/run.sh index 687dcff1ec..4ed966f97e 100644 --- a/inference/python_api_test/test_det_model/run.sh +++ b/inference/python_api_test/test_det_model/run.sh @@ -2,18 +2,14 @@ export FLAGS_call_stack_level=2 cases="./test_fast_rcnn_mkldnn.py \ ./test_fast_rcnn_gpu.py \ - ./test_fast_rcnn_trt_fp32.py \ ./test_ppyolo_gpu.py \ ./test_ppyolo_mkldnn.py \ ./test_ppyolov2_mkldnn.py \ ./test_solov2_gpu.py \ ./test_solov2_mkldnn.py \ ./test_yolov3_gpu.py \ - ./test_yolov3_mkldnn.py \ - ../test_class_model/test_ViT_base_patch16_224_trt_fp32.py \ - ../test_class_model/test_ViT_small_patch16_224_trt_fp32.py + ./test_yolov3_mkldnn.py " -# The reason for adding ViT_class_cases is to balance task execution time bug=0 echo "============ failed cases =============" >> result.txt diff --git a/inference/python_api_test/test_det_model/run_parallel.sh b/inference/python_api_test/test_det_model/run_parallel.sh index 9d392c21a2..1e506014e6 100644 --- a/inference/python_api_test/test_det_model/run_parallel.sh +++ b/inference/python_api_test/test_det_model/run_parallel.sh @@ -4,21 +4,17 @@ export FLAGS_call_stack_level=2 # ./test_ppyolov2_mkldnn.py \ cases="./test_fast_rcnn_mkldnn.py \ ./test_fast_rcnn_gpu.py \ - ./test_fast_rcnn_trt_fp32.py \ ./test_ppyolo_mkldnn.py \ ./test_ppyolo_gpu.py \ - ./test_ppyolo_trt_fp32.py \ ./test_ppyolov2_gpu.py \ ./test_yolov3_gpu.py \ ./test_yolov3_mkldnn.py \ - ./test_yolov3_trt_fp32.py \ ../test_nlp_model/test_bert_gpu.py \ ../test_nlp_model/test_bert_mkldnn.py \ ../test_nlp_model/test_ernie_gpu.py \ ../test_nlp_model/test_ernie_mkldnn.py \ ../test_ocr_model/test_ocr_det_mv3_db_gpu.py \ ../test_ocr_model/test_ocr_det_mv3_db_mkldnn.py \ - ../test_ocr_model/test_ocr_det_mv3_db_trt_fp32.py \ " bug=0 diff --git a/inference/python_api_test/test_det_model/test_fast_rcnn_trt_fp16.py b/inference/python_api_test/test_det_model/test_fast_rcnn_trt_fp16.py deleted file mode 100644 index 44616fe05f..0000000000 --- a/inference/python_api_test/test_det_model/test_fast_rcnn_trt_fp16.py +++ /dev/null @@ -1,111 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test fast_rcnn model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - fast_rcnn_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.2.2/detection/fast_rcnn.tgz" - if not os.path.exists("./fast_rcnn/model.pdiparams"): - wget.download(fast_rcnn_url, out="./") - tar = tarfile.open("fast_rcnn.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fast_rcnn/model.pdmodel", - params_file="./fast_rcnn/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared mkldnn fast_rcnn batch_size = [1] outputs with true val - """ - check_model_exist() - - file_path = "./fast_rcnn" - images_size = 608 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fast_rcnn/model.pdmodel", - params_file="./fast_rcnn/model.pdiparams", - ) - images_list, images_origin_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det", with_true_data=False - ) - - img = images_origin_list[0:batch_size] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./fast_rcnn/model.pdmodel", - params_file="./fast_rcnn/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=3e-1, - precision="trt_fp16", - min_subgraph_size=1, - dynamic=True, - auto_tuned=True, - det_top_bbox=True, - need_sort=True, - ) - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_det_model/test_fast_rcnn_trt_fp32.py b/inference/python_api_test/test_det_model/test_fast_rcnn_trt_fp32.py deleted file mode 100644 index a512b50a3b..0000000000 --- a/inference/python_api_test/test_det_model/test_fast_rcnn_trt_fp32.py +++ /dev/null @@ -1,110 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test fast_rcnn model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - fast_rcnn_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.2.2/detection/fast_rcnn.tgz" - if not os.path.exists("./fast_rcnn/model.pdiparams"): - wget.download(fast_rcnn_url, out="./") - tar = tarfile.open("fast_rcnn.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fast_rcnn/model.pdmodel", - params_file="./fast_rcnn/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared mkldnn fast_rcnn batch_size = [1] outputs with true val - """ - check_model_exist() - - file_path = "./fast_rcnn" - images_size = 608 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fast_rcnn/model.pdmodel", - params_file="./fast_rcnn/model.pdiparams", - ) - images_list, images_origin_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det", with_true_data=False - ) - - img = images_origin_list[0:batch_size] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./fast_rcnn/model.pdmodel", - params_file="./fast_rcnn/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=3e-5, - precision="trt_fp32", - min_subgraph_size=1, - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_det_model/test_mask_rcnn_trt_fp16.py b/inference/python_api_test/test_det_model/test_mask_rcnn_trt_fp16.py deleted file mode 100644 index 588071b585..0000000000 --- a/inference/python_api_test/test_det_model/test_mask_rcnn_trt_fp16.py +++ /dev/null @@ -1,114 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test mask_rcnn model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - mask_rcnn_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.2.2/detection/mask_rcnn.tgz" - if not os.path.exists("./mask_rcnn/model.pdiparams"): - wget.download(mask_rcnn_url, out="./") - tar = tarfile.open("mask_rcnn.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./mask_rcnn/model.pdmodel", - params_file="./mask_rcnn/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -def test_trt_fp16_more_bz(): - """ - compared trt_fp16 mask_rcnn batch_size = [1] outputs with true val - """ - check_model_exist() - - file_path = "./mask_rcnn" - images_size = 608 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./mask_rcnn/model.pdmodel", - params_file="./mask_rcnn/model.pdiparams", - ) - images_list, images_origin_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det", with_true_data=False - ) - - img = images_origin_list[0:batch_size] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./mask_rcnn/model.pdmodel", - params_file="./mask_rcnn/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=3e-1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - min_subgraph_size=5, - det_top_bbox=True, - need_sort=True, - det_top_bbox_threshold=0.85, - check_output_list=["save_infer_model/scale_0.tmp_0", "save_infer_model/scale_1.tmp_0"], - delete_op_list=["cast_18.tmp_0"], - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_det_model/test_mask_rcnn_trt_fp32.py b/inference/python_api_test/test_det_model/test_mask_rcnn_trt_fp32.py deleted file mode 100644 index e5d2cdb512..0000000000 --- a/inference/python_api_test/test_det_model/test_mask_rcnn_trt_fp32.py +++ /dev/null @@ -1,110 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test mask_rcnn model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - mask_rcnn_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.2.2/detection/mask_rcnn.tgz" - if not os.path.exists("./mask_rcnn/model.pdiparams"): - wget.download(mask_rcnn_url, out="./") - tar = tarfile.open("mask_rcnn.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./mask_rcnn/model.pdmodel", - params_file="./mask_rcnn/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -def test_trt_fp32_more_bz(): - """ - compared trt_fp32 mask_rcnn batch_size = [1] outputs with true val - """ - check_model_exist() - - file_path = "./mask_rcnn" - images_size = 608 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./mask_rcnn/model.pdmodel", - params_file="./mask_rcnn/model.pdiparams", - ) - images_list, images_origin_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det", with_true_data=False - ) - - img = images_origin_list[1 : batch_size + 1] - data = np.array(images_list[1 : batch_size + 1]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./mask_rcnn/model.pdmodel", - params_file="./mask_rcnn/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=3e-5, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - min_subgraph_size=5, - delete_op_list=["cast_18.tmp_0"], - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_det_model/test_ppyolo_trt_fp16.py b/inference/python_api_test/test_det_model/test_ppyolo_trt_fp16.py deleted file mode 100644 index d850870a9d..0000000000 --- a/inference/python_api_test/test_det_model/test_ppyolo_trt_fp16.py +++ /dev/null @@ -1,123 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ppyolo model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ppyolo_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/detection/ppyolo.tgz" - if not os.path.exists("./ppyolo/model.pdiparams"): - wget.download(ppyolo_url, out="./") - tar = tarfile.open("ppyolo.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ppyolo/model.pdmodel", - params_file="./ppyolo/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt_fp16 ppyolo batch_size = [1] outputs with true val - """ - check_model_exist() - - file_path = "./ppyolo" - images_size = 608 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ppyolo/model.pdmodel", - params_file="./ppyolo/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[0:batch_size] - result = npy_list[0 : batch_size * 2] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - test_suite.collect_shape_info(model_path="./ppyolo/", input_data_dict=input_data_dict, device="gpu") - - scale_0 = [] - for batch in range(0, batch_size * 2, 2): - scale_0 = np.concatenate((scale_0, result[batch].flatten()), axis=0) - scale_1 = [] - for batch in range(1, batch_size * 2, 2): - scale_1 = np.concatenate((scale_1, result[batch].flatten()), axis=0) - - # output_data_dict = {"save_infer_model/scale_0.tmp_1": scale_0, "save_infer_model/scale_1.tmp_1": scale_1} - del test_suite.pd_config - - test_suite.load_config( - model_file="./ppyolo/model.pdmodel", - params_file="./ppyolo/model.pdiparams", - ) - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - test_suite.load_config( - model_file="./ppyolo/model.pdmodel", - params_file="./ppyolo/model.pdiparams", - ) - test_suite.trt_more_bz_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - repeat=1, - delta=1, - precision="trt_fp16", - dynamic=True, - shape_range_file="./ppyolo/shape_range.pbtxt", - det_top_bbox=True, - ) diff --git a/inference/python_api_test/test_det_model/test_ppyolo_trt_fp32.py b/inference/python_api_test/test_det_model/test_ppyolo_trt_fp32.py deleted file mode 100644 index 20b76e16ff..0000000000 --- a/inference/python_api_test/test_det_model/test_ppyolo_trt_fp32.py +++ /dev/null @@ -1,127 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ppyolo model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ppyolo_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/detection/ppyolo.tgz" - if not os.path.exists("./ppyolo/model.pdiparams"): - wget.download(ppyolo_url, out="./") - tar = tarfile.open("ppyolo.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ppyolo/model.pdmodel", - params_file="./ppyolo/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt_fp32 ppyolo batch_size = [1] outputs with true val - """ - check_model_exist() - - file_path = "./ppyolo" - images_size = 608 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ppyolo/model.pdmodel", - params_file="./ppyolo/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[0:batch_size] - result = npy_list[0 : batch_size * 2] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - test_suite.collect_shape_info( - model_path="./ppyolo/", - input_data_dict=input_data_dict, - device="gpu", - ) - - scale_0 = [] - for batch in range(0, batch_size * 2, 2): - scale_0 = np.concatenate((scale_0, result[batch].flatten()), axis=0) - scale_1 = [] - for batch in range(1, batch_size * 2, 2): - scale_1 = np.concatenate((scale_1, result[batch].flatten()), axis=0) - - # output_data_dict = {"save_infer_model/scale_0.tmp_1": scale_0, "save_infer_model/scale_1.tmp_1": scale_1} - del test_suite.pd_config - - test_suite.load_config( - model_file="./ppyolo/model.pdmodel", - params_file="./ppyolo/model.pdiparams", - ) - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - test_suite.load_config( - model_file="./ppyolo/model.pdmodel", - params_file="./ppyolo/model.pdiparams", - ) - test_suite.trt_more_bz_test( - input_data_dict, - output_data_dict, - min_subgraph_size=5, - repeat=1, - delta=1, - precision="trt_fp32", - dynamic=True, - shape_range_file="./ppyolo/shape_range.pbtxt", - ) - - del test_suite diff --git a/inference/python_api_test/test_det_model/test_ppyolov2_trt_fp16.py b/inference/python_api_test/test_det_model/test_ppyolov2_trt_fp16.py deleted file mode 100644 index 0e84d4f8cc..0000000000 --- a/inference/python_api_test/test_det_model/test_ppyolov2_trt_fp16.py +++ /dev/null @@ -1,114 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ppyolov2 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ppyolov2_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/detection/ppyolov2.tgz" - if not os.path.exists("./ppyolov2/model.pdiparams"): - wget.download(ppyolov2_url, out="./") - tar = tarfile.open("ppyolov2.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ppyolov2/model.pdmodel", - params_file="./ppyolov2/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared gpu ppyolov2 batch_size = [1] outputs with true val - """ - check_model_exist() - - file_path = "./ppyolov2" - images_size = 640 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ppyolov2/model.pdmodel", - params_file="./ppyolov2/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[0:batch_size] - result = npy_list[0 : (batch_size) * 2] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - - scale_0 = [] - for batch in range(0, batch_size * 2, 2): - scale_0 = np.concatenate((scale_0, result[batch].flatten()), axis=0) - scale_1 = [] - for batch in range(1, batch_size * 2, 2): - scale_1 = np.concatenate((scale_1, result[batch].flatten()), axis=0) - - # output_data_dict = {"save_infer_model/scale_0.tmp_1": scale_0, "save_infer_model/scale_1.tmp_1": scale_1} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - test_suite.load_config( - model_file="./ppyolov2/model.pdmodel", - params_file="./ppyolov2/model.pdiparams", - ) - test_suite.trt_more_bz_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - repeat=1, - delta=2.4, - precision="trt_fp16", - det_top_bbox=True, - det_top_bbox_threshold=0.75, - ) diff --git a/inference/python_api_test/test_det_model/test_ppyolov2_trt_fp32.py b/inference/python_api_test/test_det_model/test_ppyolov2_trt_fp32.py deleted file mode 100644 index dfda01012c..0000000000 --- a/inference/python_api_test/test_det_model/test_ppyolov2_trt_fp32.py +++ /dev/null @@ -1,112 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ppyolov2 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ppyolov2_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/detection/ppyolov2.tgz" - if not os.path.exists("./ppyolov2/model.pdiparams"): - wget.download(ppyolov2_url, out="./") - tar = tarfile.open("ppyolov2.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ppyolov2/model.pdmodel", - params_file="./ppyolov2/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared gpu ppyolov2 batch_size = [1] outputs with true val - """ - check_model_exist() - - file_path = "./ppyolov2" - images_size = 640 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ppyolov2/model.pdmodel", - params_file="./ppyolov2/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[0:batch_size] - result = npy_list[0 : (batch_size) * 2] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - - scale_0 = [] - for batch in range(0, batch_size * 2, 2): - scale_0 = np.concatenate((scale_0, result[batch].flatten()), axis=0) - scale_1 = [] - for batch in range(1, batch_size * 2, 2): - scale_1 = np.concatenate((scale_1, result[batch].flatten()), axis=0) - - # output_data_dict = {"save_infer_model/scale_0.tmp_1": scale_0, "save_infer_model/scale_1.tmp_1": scale_1} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - test_suite.load_config( - model_file="./ppyolov2/model.pdmodel", - params_file="./ppyolov2/model.pdiparams", - ) - test_suite.trt_more_bz_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - repeat=1, - delta=1, - precision="trt_fp32", - ) diff --git a/inference/python_api_test/test_det_model/test_solov2_trt_fp16.py b/inference/python_api_test/test_det_model/test_solov2_trt_fp16.py deleted file mode 100644 index 03f8a8bd47..0000000000 --- a/inference/python_api_test/test_det_model/test_solov2_trt_fp16.py +++ /dev/null @@ -1,121 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test solov2 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - solov2_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.1.3/detection/solov2.tgz" - if not os.path.exists("./solov2/model.pdiparams"): - wget.download(solov2_url, out="./") - tar = tarfile.open("solov2.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./solov2/model.pdmodel", - params_file="./solov2/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared gpu solov2 batch_size = [1] outputs with true val - """ - check_model_exist() - - file_path = "./solov2" - images_size = 608 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./solov2/model.pdmodel", - params_file="./solov2/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[0:batch_size] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./solov2/model.pdmodel", - params_file="./solov2/model.pdiparams", - ) - - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - check_output_list=["shape_40.tmp_0_slice_0"], # select which outputs to check - repeat=1, - delta=4e-4, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - det_top_bbox=True, - delete_op_list=[ - "im_shape_slice_0", - "im_shape_slice_0_slice_1", - "cast_6.tmp_0", - "im_shape_slice_0_slice_0", - "cast_5.tmp_0", - "im_shape_slice_0_slice_2", - "tmp_65", - ], - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_det_model/test_solov2_trt_fp32.py b/inference/python_api_test/test_det_model/test_solov2_trt_fp32.py deleted file mode 100644 index b3c58f14c6..0000000000 --- a/inference/python_api_test/test_det_model/test_solov2_trt_fp32.py +++ /dev/null @@ -1,118 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test solov2 model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - solov2_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.1.3/detection/solov2.tgz" - if not os.path.exists("./solov2/model.pdiparams"): - wget.download(solov2_url, out="./") - tar = tarfile.open("solov2.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./solov2/model.pdmodel", - params_file="./solov2/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared gpu solov2 batch_size = [1] outputs with true val - """ - check_model_exist() - - file_path = "./solov2" - images_size = 640 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./solov2/model.pdmodel", - params_file="./solov2/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[0:batch_size] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./solov2/model.pdmodel", - params_file="./solov2/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - check_output_list=["shape_40.tmp_0_slice_0"], # select which outputs to check - repeat=1, - delta=1e-5, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - delete_op_list=[ - "im_shape_slice_0", - "im_shape_slice_0_slice_1", - "cast_6.tmp_0", - "im_shape_slice_0_slice_0", - "cast_5.tmp_0", - "im_shape_slice_0_slice_2", - "tmp_65", - ], - ) - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_det_model/test_yolov3_trt_fp16.py b/inference/python_api_test/test_det_model/test_yolov3_trt_fp16.py deleted file mode 100644 index bdaffee2a6..0000000000 --- a/inference/python_api_test/test_det_model/test_yolov3_trt_fp16.py +++ /dev/null @@ -1,264 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test yolov3 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - yolov3_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/detection/yolov3.tgz" - if not os.path.exists("./yolov3/model.pdiparams"): - wget.download(yolov3_url, out="./") - tar = tarfile.open("yolov3.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_more_bz_multi_thread(): - """ - compared trt fp32 batch_size=4 yolov3 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./yolov3" - images_size = 608 - batch_size_pool = [4] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[1 : batch_size + 1] - result = npy_list[0 : batch_size * 2] - data = np.array(images_list[1 : batch_size + 1]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - - scale_0 = [] - for batch in range(0, batch_size * 2, 2): - scale_0 = np.concatenate((scale_0, result[batch].flatten()), axis=0) - scale_1 = [] - for batch in range(1, batch_size * 2, 2): - scale_1 = np.concatenate((scale_1, result[batch].flatten()), axis=0) - - # output_data_dict = {"save_infer_model/scale_0.tmp_1": scale_0, "save_infer_model/scale_1.tmp_1": scale_1} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - test_suite.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - test_suite.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=6e-2, - precision="trt_fp16", - ) - - del test_suite # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16_more -def test_trt_fp16_more_bz(): - """ - compared trt fp32 batch_size = [1, 2] yolov3 outputs with true val - """ - check_model_exist() - - file_path = "./yolov3" - images_size = 608 - batch_size_pool = [1, 2] - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[1 : batch_size + 1] - result = npy_list[0 : batch_size * 2] - data = np.array(images_list[1 : batch_size + 1]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - - scale_0 = [] - for batch in range(0, batch_size * 2, 2): - scale_0 = np.concatenate((scale_0, result[batch].flatten()), axis=0) - scale_1 = [] - for batch in range(1, batch_size * 2, 2): - scale_1 = np.concatenate((scale_1, result[batch].flatten()), axis=0) - - # output_data_dict = {"save_infer_model/scale_0.tmp_1": scale_0, "save_infer_model/scale_1.tmp_1": scale_1} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=6e-2, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - det_top_bbox=True, - det_top_bbox_threshold=0.85, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp32 batch_size = [1] yolov3 outputs with true val - """ - check_model_exist() - - file_path = "./yolov3" - images_size = 608 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[1 : batch_size + 1] - result = npy_list[0 : batch_size * 2] - data = np.array(images_list[1 : batch_size + 1]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - - scale_0 = [] - for batch in range(0, batch_size * 2, 2): - scale_0 = np.concatenate((scale_0, result[batch].flatten()), axis=0) - scale_1 = [] - for batch in range(1, batch_size * 2, 2): - scale_1 = np.concatenate((scale_1, result[batch].flatten()), axis=0) - - # output_data_dict = {"save_infer_model/scale_0.tmp_1": scale_0, "save_infer_model/scale_1.tmp_1": scale_1} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - max_batch_size=10, - min_subgraph_size=1, - delta=6e-2, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - det_top_bbox=True, - det_top_bbox_threshold=0.85, - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_det_model/test_yolov3_trt_fp32.py b/inference/python_api_test/test_det_model/test_yolov3_trt_fp32.py deleted file mode 100644 index ac02392180..0000000000 --- a/inference/python_api_test/test_det_model/test_yolov3_trt_fp32.py +++ /dev/null @@ -1,260 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test yolov3 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - yolov3_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/detection/yolov3.tgz" - if not os.path.exists("./yolov3/model.pdiparams"): - wget.download(yolov3_url, out="./") - tar = tarfile.open("yolov3.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.trt_fp32_multi_thread -def test_trtfp32_more_bz_multi_thread(): - """ - compared trt fp32 batch_size=1 yolov3 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./yolov3" - images_size = 608 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[0:batch_size] - result = npy_list[0 : batch_size * 2] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - - scale_0 = [] - for batch in range(0, batch_size * 2, 2): - scale_0 = np.concatenate((scale_0, result[batch].flatten()), axis=0) - scale_1 = [] - for batch in range(1, batch_size * 2, 2): - scale_1 = np.concatenate((scale_1, result[batch].flatten()), axis=0) - - # output_data_dict = {"save_infer_model/scale_0.tmp_1": scale_0, "save_infer_model/scale_1.tmp_1": scale_1} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - test_suite.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - test_suite.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - repeat=1, - delta=1e-4, - precision="trt_fp32", - ) - - del test_suite # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trtfp32_more_bz(): - """ - compared trt fp32 batch_size = [1, 2] yolov3 outputs with true val - """ - check_model_exist() - - file_path = "./yolov3" - images_size = 608 - batch_size_pool = [1, 2] - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[0:batch_size] - result = npy_list[0 : batch_size * 2] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - - scale_0 = [] - for batch in range(0, batch_size * 2, 2): - scale_0 = np.concatenate((scale_0, result[batch].flatten()), axis=0) - scale_1 = [] - for batch in range(1, batch_size * 2, 2): - scale_1 = np.concatenate((scale_1, result[batch].flatten()), axis=0) - - # output_data_dict = {"save_infer_model/scale_0.tmp_1": scale_0, "save_infer_model/scale_1.tmp_1": scale_1} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - max_batch_size=10, - delta=1e-4, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1,5,10 yolov3 outputs with true val - """ - check_model_exist() - - file_path = "./yolov3" - images_size = 608 - batch_size_pool = [1] - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - images_list, images_origin_list, npy_list = test_suite.get_images_npy( - file_path, images_size, center=False, model_type="det" - ) - - img = images_origin_list[0:batch_size] - result = npy_list[0 : batch_size * 2] - data = np.array(images_list[0:batch_size]).astype("float32") - scale_factor_pool = [] - for batch in range(batch_size): - scale_factor = ( - np.array([images_size * 1.0 / img[batch].shape[0], images_size * 1.0 / img[batch].shape[1]]) - .reshape((1, 2)) - .astype(np.float32) - ) - scale_factor_pool.append(scale_factor) - scale_factor_pool = np.array(scale_factor_pool).reshape((batch_size, 2)) - im_shape_pool = [] - for batch in range(batch_size): - im_shape = np.array([images_size, images_size]).reshape((1, 2)).astype(np.float32) - im_shape_pool.append(im_shape) - im_shape_pool = np.array(im_shape_pool).reshape((batch_size, 2)) - input_data_dict = {"im_shape": im_shape_pool, "image": data, "scale_factor": scale_factor_pool} - - scale_0 = [] - for batch in range(0, batch_size * 2, 2): - scale_0 = np.concatenate((scale_0, result[batch].flatten()), axis=0) - scale_1 = [] - for batch in range(1, batch_size * 2, 2): - scale_1 = np.concatenate((scale_1, result[batch].flatten()), axis=0) - - # output_data_dict = {"save_infer_model/scale_0.tmp_1": scale_0, "save_infer_model/scale_1.tmp_1": scale_1} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./yolov3/model.pdmodel", - params_file="./yolov3/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - repeat=1, - max_batch_size=10, - min_subgraph_size=1, - delta=1e-4, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_int8_model/README.md b/inference/python_api_test/test_int8_model/README.md index cf17d7079a..eb1a5c54ba 100644 --- a/inference/python_api_test/test_int8_model/README.md +++ b/inference/python_api_test/test_int8_model/README.md @@ -11,30 +11,6 @@ sh prepare.sh ``` -## Paddle Inference TensorRT测试 -- INT8 - -```shell -sh run_trt_int8.sh > eval_trt_int8_acc.log 2>&1 & -``` - -收集重要log信息： -```shell -grep -i Benchmark eval_trt_int8_acc.log -``` - -- FP16 - -```shell -sh run_trt_fp16.sh > eval_trt_fp16_acc.log 2>&1 & -``` - -收集重要log信息： -```shell -grep -i Benchmark eval_trt_fp16_acc.log -``` - - ## Paddle Inference MKLDNN测试 - INT8 diff --git a/inference/python_api_test/test_int8_model/backend/__init__.py b/inference/python_api_test/test_int8_model/backend/__init__.py index 746b619791..94ef0edb97 100644 --- a/inference/python_api_test/test_int8_model/backend/__init__.py +++ b/inference/python_api_test/test_int8_model/backend/__init__.py @@ -15,6 +15,5 @@ """ from .paddle_inference import PaddleInferenceEngine -from .tensorrt import TensorRTEngine from .monitor import Monitor from .onnxruntime import ONNXRuntimeEngine diff --git a/inference/python_api_test/test_int8_model/backend/onnxruntime.py b/inference/python_api_test/test_int8_model/backend/onnxruntime.py index 073687d704..0f08faa01c 100644 --- a/inference/python_api_test/test_int8_model/backend/onnxruntime.py +++ b/inference/python_api_test/test_int8_model/backend/onnxruntime.py @@ -13,7 +13,6 @@ # See the License for the specific language governing permissions and # limitations under the License. """ -import os import numpy as np import onnxruntime as ort @@ -26,27 +25,17 @@ class ONNXRuntimeEngine(object): def __init__( self, onnx_model_file, - precision="fp32", - use_trt=False, use_mkldnn=False, device="CPU", - min_subgraph_size=3, save_optimized_model=False, ): """set AnalysisConfig, generate AnalysisPredictor Args: onnx_model_file (str): root path of ONNX model. - precision (str): mode of running(fp32/fp16/int8). - use_trt (bool): whether use TensorRT or not. use_mkldnn (bool): whether use MKLDNN or not in CPU. device (str): Choose the device you want to run, it can be: CPU/GPU, default is CPU. - min_subgraph_size (int): min subgraph size in trt. save_optimized_model (bool): whether save optimized model to debug. """ - if device != "GPU" and use_trt: - raise ValueError( - "Predict by TensorRT mode: {}, expect device=='GPU', but device == {}".format(precision, device) - ) sess_options = ort.SessionOptions() if device == "CPU": if use_mkldnn: @@ -54,45 +43,18 @@ def __init__( else: providers = ["CPUExecutionProvider"] elif device == "GPU": - if use_trt: - providers = [ - ( - "TensorrtExecutionProvider", - { - "device_id": 0, - "trt_max_workspace_size": 1073741824, - "trt_min_subgraph_size": min_subgraph_size, - "trt_fp16_enable": True if precision == "fp16" else False, - "trt_int8_enable": True if precision == "int8" else False, - # below two files are used for ort-trt int8! - "trt_int8_calibration_table_name": os.path.dirname(onnx_model_file) + "/calibration.cache", - "trt_int8_use_native_calibration_table": True, - }, - ), - ( - "CUDAExecutionProvider", - { - "device_id": 0, - "arena_extend_strategy": "kNextPowerOfTwo", - "gpu_mem_limit": 2 * 1024 * 1024 * 1024, - "cudnn_conv_algo_search": "EXHAUSTIVE", - "do_copy_in_default_stream": True, - }, - ), - ] - else: - providers = [ - ( - "CUDAExecutionProvider", - { - "device_id": 0, - "arena_extend_strategy": "kNextPowerOfTwo", - "cudnn_conv_algo_search": "EXHAUSTIVE", - "do_copy_in_default_stream": True, - }, - ), - "CPUExecutionProvider", - ] + providers = [ + ( + "CUDAExecutionProvider", + { + "device_id": 0, + "arena_extend_strategy": "kNextPowerOfTwo", + "cudnn_conv_algo_search": "EXHAUSTIVE", + "do_copy_in_default_stream": True, + }, + ), + "CPUExecutionProvider", + ] if save_optimized_model: sess_options.optimized_model_filepath = "./optimize_model.onnx" diff --git a/inference/python_api_test/test_int8_model/backend/paddle_inference.py b/inference/python_api_test/test_int8_model/backend/paddle_inference.py index 5b41843ad7..765b009c71 100644 --- a/inference/python_api_test/test_int8_model/backend/paddle_inference.py +++ b/inference/python_api_test/test_int8_model/backend/paddle_inference.py @@ -35,32 +35,19 @@ def __init__( model_filename="model.pdmodel", params_filename="model.pdiparams", precision="fp32", - use_trt=False, use_l3=False, use_mkldnn=False, - batch_size=1, device="CPU", - min_subgraph_size=3, - use_dynamic_shape=False, cpu_threads=1, ): """set AnalysisConfig, generate AnalysisPredictor Args: model_dir (str): root path of model.pdmodel and model.pdiparams. precision (str): mode of running(fp32/fp16/int8). - use_trt (bool): whether use TensorRT or not. use_mkldnn (bool): whether use MKLDNN or not in CPU. - batch_size (int): Batch size of infer sample. device (str): Choose the device you want to run, it can be: CPU/GPU, default is CPU. - min_subgraph_size (int): min subgraph size in trt. - use_dynamic_shape (bool): use dynamic shape or not. cpu_threads (int): num of thread when use CPU. """ - self.rerun_flag = False - if device != "GPU" and use_trt: - raise ValueError( - "Predict by TensorRT mode: {}, expect device=='GPU', but device == {}".format(precision, device) - ) config = Config(os.path.join(model_dir, model_filename), os.path.join(model_dir, params_filename)) if device == "GPU": # initial GPU memory(M), device ID @@ -95,41 +82,6 @@ def __init__( if precision == "bf16": config.enable_mkldnn_bfloat16() - if use_trt: - if precision == "bf16": - print("paddle trt does not support bf16, switching to fp16.") - precision = "fp16" - - precision_map = { - "int8": Config.Precision.Int8, - "fp32": Config.Precision.Float32, - "fp16": Config.Precision.Half, - } - assert precision in precision_map.keys() - - if use_dynamic_shape: - dynamic_shape_file = os.path.join(model_dir, "dynamic_shape.txt") - if os.path.exists(dynamic_shape_file): - config.enable_tuned_tensorrt_dynamic_shape() - print("trt set dynamic shape done!") - else: - # In order to avoid memory overflow when collecting dynamic shapes, it is changed to use CPU. - config.disable_gpu() - config.set_cpu_math_library_num_threads(10) - config.collect_shape_range_info(dynamic_shape_file) - print("Start collect dynamic shape...") - self.rerun_flag = True - - if not self.rerun_flag: - config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=precision_map[precision], - use_static=True, - use_calib_mode=False, - ) - # enable shared memory config.enable_memory_optim() self.predictor = create_predictor(config) diff --git a/inference/python_api_test/test_int8_model/backend/tensorrt.py b/inference/python_api_test/test_int8_model/backend/tensorrt.py deleted file mode 100644 index a0d08d163b..0000000000 --- a/inference/python_api_test/test_int8_model/backend/tensorrt.py +++ /dev/null @@ -1,356 +0,0 @@ -""" -# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved. -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -""" - -import sys -import os -import copy -import numpy as np - -import tensorrt as trt - -try: - import pycuda.driver as cuda - import pycuda.autoinit -except ModuleNotFoundError as e: - print(e.msg) - print("CUDA might not be installed. TensorRT cannot be used.") - -EXPLICIT_BATCH = 1 << (int)(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH) -EXPLICIT_PRECISION = 1 << (int)(trt.NetworkDefinitionCreationFlag.EXPLICIT_PRECISION) - - -class LoadCalibrator(trt.IInt8EntropyCalibrator2): - """ - Load calibration.cache - Args: - calibration_files(List[str]): List of image filenames to use for INT8 Calibration - cache_file(str): Name of file to read/write calibration cache from/to. - batch_size(int): Number of images to pass through in one batch during calibration - input_shape(Tuple[int]): Tuple of integers defining the shape of input to the model (Default: (3, 224, 224)) - """ - - def __init__(self, calibration_loader=None, cache_file="calibration.cache", max_calib_size=32): - super().__init__() - self.calibration_loader = calibration_loader - self.cache_file = cache_file - self.max_calib_size = max_calib_size - if calibration_loader: - self.batch = next(self.calibration_loader()) - self.batch_size = self.batch.shape[0] - self.device_input = cuda.mem_alloc(self.batch.nbytes) - else: - self.batch_size = 1 - self.batch_id = 0 - - def get_batch(self, names): - """ - calibration data loader - """ - assert self.calibration_loader, "calibration_loader is None, Please set correct calibration_loader." - try: - # Assume self.batches is a generator that provides batch data. - batch = next(self.calibration_loader()) - print("Calibration images pre-processed: {:}/{:}".format(self.batch_id, self.max_calib_size)) - self.batch_id += 1 - assert self.batch_id <= self.max_calib_size - - # Assume that self.device_input is a device buffer allocated by the constructor. - cuda.memcpy_htod(self.device_input, batch) - return [int(self.device_input)] - except StopIteration: - # When we're out of batches, we return either [] or None. - # This signals to TensorRT that there is no calibration data remaining. - print( - "[Note] The calibration process is complete, the calibration file is being saved, " - "please wait and do not kill the process." - ) - return None - - def get_batch_size(self): - """ - get batch size - """ - return self.batch_size - - def read_calibration_cache(self): - """ - read calibration cache file - """ - # If there is a cache, use it instead of calibrating again. Otherwise, implicitly return None. - if os.path.exists(self.cache_file): - with open(self.cache_file, "rb") as f: - print("Using calibration cache to save time: {:}".format(self.cache_file)) - return f.read() - - def write_calibration_cache(self, cache): - """ - write calibration cache file - """ - with open(self.cache_file, "wb") as f: - print("Caching calibration data for future use: {:}".format(self.cache_file)) - f.write(cache) - - -def get_int8_calibrator(calib_cache, calibration_loader, max_calib_size): - """ - The instance of get int8 calibration file. - """ - # Use calibration cache if it exists - if calib_cache and os.path.exists(calib_cache): - print("==> Skipping calibration files, using calibration cache: {:}".format(calib_cache)) - # Use calibration files from validation dataset if no cache exists - else: - print("Not exist calibration cache file, and it will run calibration.") - if not calib_cache: - calib_cache = "calibration.cache" - if not calibration_loader: - raise ValueError( - "ERROR: calibration dataloader requested, but no `calibration_loader` or calibration files provided." - ) - - int8_calibrator = LoadCalibrator( - calibration_loader=calibration_loader, cache_file=calib_cache, max_calib_size=max_calib_size - ) - return int8_calibrator - - -def remove_initializer_from_input(ori_model): - """ - remove initializer from input - """ - model = copy.deepcopy(ori_model) - if model.ir_version < 4: - print("Model with ir_version below 4 requires to include initilizer in graph input") - return - - inputs = model.graph.input - name_to_input = {} - for input in inputs: - name_to_input[input.name] = input - - for initializer in model.graph.initializer: - if initializer.name in name_to_input: - inputs.remove(name_to_input[initializer.name]) - return model - - -# Simple helper data class that's a little nicer to use than a 2-tuple. -class HostDeviceMem(object): - """ - HostDeviceMem instance - """ - - def __init__(self, host_mem, device_mem): - self.host = host_mem - self.device = device_mem - if host_mem: - self.nbytes = host_mem.nbytes - else: - self.nbytes = 0 - - def __str__(self): - return "Host:\n" + str(self.host) + "\nDevice:\n" + str(self.device) - - def __repr__(self): - return self.__str__() - - -class TensorRTEngine: - """ - TensorRT instance - """ - - def __init__( - self, - onnx_model_file, - shape_info=None, - max_batch_size=None, - precision="fp32", - engine_file_path=None, - calibration_cache_file="calibration.cache", - max_calibration_size=32, - calibration_loader=None, - verbose=False, - ): - self.max_batch_size = 1 if max_batch_size is None else max_batch_size - precision = precision.lower() - if precision == "bf16": - print("trt does not support bf16, switching to fp16") - precision = "fp16" - assert precision in [ - "fp32", - "fp16", - "int8", - ], "precision must be fp32, fp16 or int8, but your precision is: {}".format(precision) - - use_int8 = precision == "int8" - use_fp16 = precision == "fp16" - TRT_LOGGER = trt.Logger() - if verbose: - TRT_LOGGER = trt.Logger(trt.Logger.VERBOSE) - if engine_file_path is not None and os.path.exists(engine_file_path): - # If a serialized engine exists, use it instead of building an engine. - print("[TRT Backend] Reading engine from file {}".format(engine_file_path)) - with open(engine_file_path, "rb") as f, trt.Runtime(TRT_LOGGER) as runtime: - self.engine = runtime.deserialize_cuda_engine(f.read()) - else: - builder = trt.Builder(TRT_LOGGER) - config = builder.create_builder_config() - network = None - - if use_int8 and not builder.platform_has_fast_int8: - print("[TRT Backend] INT8 not supported on this platform.") - if use_fp16 and not builder.platform_has_fast_fp16: - print("[TRT Backend] FP16 not supported on this platform.") - - if use_int8 and builder.platform_has_fast_int8: - print("[TRT Backend] Use INT8.") - network = builder.create_network(EXPLICIT_BATCH | EXPLICIT_PRECISION) - - config.int8_calibrator = get_int8_calibrator( - calibration_cache_file, calibration_loader, max_calibration_size - ) - - config.set_flag(trt.BuilderFlag.INT8) - elif use_fp16 and builder.platform_has_fast_fp16: - print("[TRT Backend] Use FP16.") - network = builder.create_network(EXPLICIT_BATCH) - config.set_flag(trt.BuilderFlag.FP16) - else: - print("[TRT Backend] Use FP32.") - network = builder.create_network(EXPLICIT_BATCH) - parser = trt.OnnxParser(network, TRT_LOGGER) - runtime = trt.Runtime(TRT_LOGGER) - config.max_workspace_size = 1 << 30 - - import onnx - - print("[TRT Backend] Loading ONNX model ...") - onnx_model = onnx_model_file - if not isinstance(onnx_model_file, onnx.ModelProto): - onnx_model = onnx.load(onnx_model_file) - onnx_model = remove_initializer_from_input(onnx_model) - if not parser.parse(onnx_model.SerializeToString()): - for error in range(parser.num_errors): - print(parser.get_error(error)) - raise Exception("ERROR: Failed to parse the ONNX file.") - - if shape_info is None: - builder.max_batch_size = 1 - for i in range(len(onnx_model.graph.input)): - input_shape = [x.dim_value for x in onnx_model.graph.input[0].type.tensor_type.shape.dim] - for s in input_shape: - assert ( - s > 0 - ), "In static shape mode, the input of onnx model should be fixed, but now it's {}".format( - onnx_model.graph.input[i] - ) - else: - max_batch_size = 1 - if shape_info is not None: - assert ( - len(shape_info) == network.num_inputs - ), "Length of shape_info: {} is not same with length of model input: {}".format( - len(shape_info), network.num_inputs - ) - profile = builder.create_optimization_profile() - for k, v in shape_info.items(): - if v[2][0] > max_batch_size: - max_batch_size = v[2][0] - print("[TRT Backend] optimize shape: ", k, v[0], v[1], v[2]) - profile.set_shape(k, v[0], v[1], v[2]) - config.add_optimization_profile(profile) - if max_batch_size > self.max_batch_size: - self.max_batch_size = max_batch_size - builder.max_batch_size = self.max_batch_size - - print("[TRT Backend] Completed parsing of ONNX file.") - print("[TRT Backend] Building an engine from onnx model may take a while...") - plan = builder.build_serialized_network(network, config) - print("[TRT Backend] Start Creating Engine.") - self.engine = runtime.deserialize_cuda_engine(plan) - print("[TRT Backend] Completed Creating Engine.") - if engine_file_path is not None: - with open(engine_file_path, "wb") as f: - f.write(self.engine.serialize()) - - self.context = self.engine.create_execution_context() - if shape_info is not None: - self.context.active_optimization_profile = 0 - self.stream = cuda.Stream() - self.bindings = [] - self.inputs = [] - self.outputs = [] - for binding in self.engine: - self.bindings.append(0) - if self.engine.binding_is_input(binding): - self.inputs.append(HostDeviceMem(None, None)) - else: - self.outputs.append(HostDeviceMem(None, None)) - - print("[TRT Backend] Completed TensorRTEngine init ...") - - def prepare_data(self, input_data): - """ - Prepare data - """ - assert len(self.inputs) == len( - input_data - ), "Length of input_data: {} is not same with length of input: {}".format(len(input_data), len(self.inputs)) - - self._allocate_buffers(input_data) - - # Transfer input data to the GPU. - [cuda.memcpy_htod_async(inp.device, inp.host, self.stream) for inp in self.inputs] - - def run(self): - """ - Run inference. - """ - self.context.execute_async_v2(bindings=self.bindings, stream_handle=self.stream.handle) - # Transfer predictions back from the GPU. - [cuda.memcpy_dtoh_async(out.host, out.device, self.stream) for out in self.outputs] - # Synchronize the stream - self.stream.synchronize() - # Return only the host outputs. - return [out.host for out in self.outputs] - - def _allocate_buffers(self, input_data): - input_idx = 0 - output_idx = 0 - for binding in self.engine: - idx = self.engine.get_binding_index(binding) - if self.engine.binding_is_input(binding): - if not input_data[input_idx].flags["C_CONTIGUOUS"]: - input_data[input_idx] = np.ascontiguousarray(input_data[input_idx]) - self.context.set_binding_shape(idx, (input_data[input_idx].shape)) - self.inputs[input_idx].host = input_data[input_idx] - nbytes = input_data[input_idx].nbytes - if self.inputs[input_idx].nbytes < nbytes: - self.inputs[input_idx].nbytes = nbytes - self.inputs[input_idx].device = cuda.mem_alloc(nbytes) - self.bindings[idx] = int(self.inputs[input_idx].device) - input_idx += 1 - else: - dtype = trt.nptype(self.engine.get_binding_dtype(binding)) - shape = self.context.get_binding_shape(idx) - self.outputs[output_idx].host = np.ascontiguousarray(np.empty(shape, dtype=dtype)) - nbytes = self.outputs[output_idx].host.nbytes - if self.outputs[output_idx].nbytes < nbytes: - self.outputs[output_idx].nbytes = nbytes - self.outputs[output_idx].device = cuda.mem_alloc(self.outputs[output_idx].host.nbytes) - self.bindings[idx] = int(self.outputs[output_idx].device) - output_idx += 1 diff --git a/inference/python_api_test/test_int8_model/base_nv_trt_fp16.py b/inference/python_api_test/test_int8_model/base_nv_trt_fp16.py deleted file mode 100644 index b4ff5f9f10..0000000000 --- a/inference/python_api_test/test_int8_model/base_nv_trt_fp16.py +++ /dev/null @@ -1,438 +0,0 @@ -""" -nv_trt_fp16 base -""" - -nv_trt_fp16 = { - "PPYOLOE_PLUS": { - "model_name": "PPYOLOE_PLUS", - "batch_size": 1, - "jingdu": { - "value": 0.5600042618262268, - "unit": "mAP", - "th": 0.01, - }, - "xingneng": { - "value": 3.2, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 2433.3140599999997, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 645.0, - "unit": "MB", - "th": 0.05, - }, - }, - "PicoDet": { - "model_name": "PicoDet", - "batch_size": 1, - "jingdu": { - "value": 0.393377893664232, - "unit": "mAP", - "th": 0.01, - }, - "xingneng": { - "value": 1.6599999999999997, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 2424.47658, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 611.0, - "unit": "MB", - "th": 0.05, - }, - }, - "YOLOv5s": { - "model_name": "YOLOv5s", - "batch_size": 1, - "jingdu": { - "value": 0.475003852545204, - "unit": "mAP", - "th": 0.01, - }, - "xingneng": { - "value": 3.88, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1334.86012, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 299.0, - "unit": "MB", - "th": 0.05, - }, - }, - "YOLOv6s": { - "model_name": "YOLOv6s", - "batch_size": 1, - "jingdu": { - "value": 0.6171771559112594, - "unit": "mAP", - "th": 0.01, - }, - "xingneng": { - "value": 3.2, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1318.9125, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 307.0, - "unit": "MB", - "th": 0.05, - }, - }, - "YOLOv7": { - "model_name": "YOLOv7", - "batch_size": 1, - "jingdu": { - "value": 0.5972319038861243, - "unit": "mAP", - "th": 0.01, - }, - "xingneng": { - "value": 8.48, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1342.5508000000002, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 403.0, - "unit": "MB", - "th": 0.05, - }, - }, - "ResNet_vd": { - "model_name": "ResNet_vd", - "batch_size": 1, - "jingdu": { - "value": 0.7950049950049951, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 1.1, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1339.4, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 4842.69376, - "unit": "MB", - "th": 0.05, - }, - }, - "MobileNetV3_large": { - "model_name": "MobileNetV3_large", - "batch_size": 1, - "jingdu": { - "value": 0.7402597402597403, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 0.7, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1311.4, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 4634.945319999999, - "unit": "MB", - "th": 0.05, - }, - }, - "PPLCNetV2": { - "model_name": "PPLCNetV2", - "batch_size": 1, - "jingdu": { - "value": 0.7702297702297702, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 0.54, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1311.4, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 4619.91562, - "unit": "MB", - "th": 0.05, - }, - }, - "PPHGNet_tiny": { - "model_name": "PPHGNet_tiny", - "batch_size": 1, - "jingdu": { - "value": 0.8095904095904096, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 1.2, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1328.2, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 4705.87656, - "unit": "MB", - "th": 0.05, - }, - }, - "EfficientNetB0": { - "model_name": "EfficientNetB0", - "batch_size": 1, - "jingdu": { - "value": 0.7632367632367633, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 1.0, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1313.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 4629.269539999999, - "unit": "MB", - "th": 0.05, - }, - }, - "PP-HumanSeg-Lite": { - "model_name": "PP-HumanSeg-Lite", - "batch_size": 1, - "jingdu": { - "value": 0.369783895654616, - "unit": "mIoU", - "th": 0.01, - }, - "xingneng": { - "value": 1.1800000000000002, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1387.02892, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 263.0, - "unit": "MB", - "th": 0.05, - }, - }, - "PP-Liteseg": { - "model_name": "PP-Liteseg", - "batch_size": 1, - "jingdu": { - "value": 0.7496189272981876, - "unit": "mIoU", - "th": 0.01, - }, - "xingneng": { - "value": 10.320000000000002, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1534.21794, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 457.0, - "unit": "MB", - "th": 0.05, - }, - }, - "HRNet": { - "model_name": "HRNet", - "batch_size": 1, - "jingdu": { - "value": 0.7852055365324859, - "unit": "mIoU", - "th": 0.01, - }, - "xingneng": { - "value": 37.72, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1538.2250000000001, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 517.0, - "unit": "MB", - "th": 0.05, - }, - }, - "UNet": { - "model_name": "UNet", - "batch_size": 1, - "jingdu": { - "value": 0.6455812617332316, - "unit": "mIoU", - "th": 0.01, - }, - "xingneng": { - "value": 82.26000000000002, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1522.3336, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 1437.0, - "unit": "MB", - "th": 0.05, - }, - }, - "Deeplabv3-ResNet50": { - "model_name": "Deeplabv3-ResNet50", - "batch_size": 1, - "jingdu": { - "value": 0.7893610650102942, - "unit": "mIoU", - "th": 0.01, - }, - "xingneng": { - "value": 97.97999999999999, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 2172.31872, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 1061.0, - "unit": "MB", - "th": 0.05, - }, - }, - "ERNIE_3.0-Medium": { - "model_name": "ERNIE_3.0-Medium", - "batch_size": 32, - "jingdu": { - "value": 0.6035681186283597, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 33.92, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1187.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 3021.94452, - "unit": "MB", - "th": 0.05, - }, - }, - "PP-MiniLM": { - "model_name": "PP-MiniLM", - "batch_size": 32, - "jingdu": { - "value": 0.5857738646895273, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 34.044000000000004, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 899.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 1716.6960800000002, - "unit": "MB", - "th": 0.05, - }, - }, - "BERT_Base": { - "model_name": "BERT_Base", - "batch_size": 1, - "jingdu": { - "value": 0.2671497327351065, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 1.886, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 859.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 3687.3656200000005, - "unit": "MB", - "th": 0.05, - }, - }, -} diff --git a/inference/python_api_test/test_int8_model/base_nv_trt_int8.py b/inference/python_api_test/test_int8_model/base_nv_trt_int8.py deleted file mode 100644 index 0279fd9bd0..0000000000 --- a/inference/python_api_test/test_int8_model/base_nv_trt_int8.py +++ /dev/null @@ -1,438 +0,0 @@ -""" -nv_trt_int8 base -""" - -nv_trt_int8 = { - "PPYOLOE_PLUS": { - "model_name": "PPYOLOE_PLUS", - "batch_size": 1, - "jingdu": { - "value": 0.5584375545585372, - "unit": "mAP", - "th": 0.01, - }, - "xingneng": { - "value": 3.8200000000000003, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 2447.5328, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 637.0, - "unit": "MB", - "th": 0.05, - }, - }, - "PicoDet": { - "model_name": "PicoDet", - "batch_size": 1, - "jingdu": { - "value": 0.3561621621721951, - "unit": "mAP", - "th": 0.01, - }, - "xingneng": { - "value": 1.4, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 2416.29454, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 607.0, - "unit": "MB", - "th": 0.05, - }, - }, - "YOLOv5s": { - "model_name": "YOLOv5s", - "batch_size": 1, - "jingdu": { - "value": 0.4631868442433045, - "unit": "mAP", - "th": 0.01, - }, - "xingneng": { - "value": 3.5200000000000005, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1342.1921799999998, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 297.0, - "unit": "MB", - "th": 0.05, - }, - }, - "YOLOv6s": { - "model_name": "YOLOv6s", - "batch_size": 1, - "jingdu": { - "value": 0.5794493709431752, - "unit": "mAP", - "th": 0.01, - }, - "xingneng": { - "value": 2.2399999999999998, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1320.4117199999998, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 297.0, - "unit": "MB", - "th": 0.05, - }, - }, - "YOLOv7": { - "model_name": "YOLOv7", - "batch_size": 1, - "jingdu": { - "value": 0.6047272452708244, - "unit": "mAP", - "th": 0.01, - }, - "xingneng": { - "value": 5.9, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1347.91564, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 371.0, - "unit": "MB", - "th": 0.05, - }, - }, - "ResNet_vd": { - "model_name": "ResNet_vd", - "batch_size": 1, - "jingdu": { - "value": 0.7754245754245754, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 0.8, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1307.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 4865.91716, - "unit": "MB", - "th": 0.05, - }, - }, - "MobileNetV3_large": { - "model_name": "MobileNetV3_large", - "batch_size": 1, - "jingdu": { - "value": 0.3354645354645355, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 0.6, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1289.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 4670.06094, - "unit": "MB", - "th": 0.05, - }, - }, - "PPLCNetV2": { - "model_name": "PPLCNetV2", - "batch_size": 1, - "jingdu": { - "value": 0.7572427572427572, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 0.4, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1287.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 4644.83126, - "unit": "MB", - "th": 0.05, - }, - }, - "PPHGNet_tiny": { - "model_name": "PPHGNet_tiny", - "batch_size": 1, - "jingdu": { - "value": 0.8041958041958042, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 0.8, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1299.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 4732.16798, - "unit": "MB", - "th": 0.05, - }, - }, - "EfficientNetB0": { - "model_name": "EfficientNetB0", - "batch_size": 1, - "jingdu": { - "value": 0.26073926073926074, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 0.9, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1289.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 4699.13048, - "unit": "MB", - "th": 0.05, - }, - }, - "PP-HumanSeg-Lite": { - "model_name": "PP-HumanSeg-Lite", - "batch_size": 1, - "jingdu": { - "value": 0.369783895654616, - "unit": "mIoU", - "th": 0.01, - }, - "xingneng": { - "value": 0.9, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1379.22654, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 257.8, - "unit": "MB", - "th": 0.05, - }, - }, - "PP-Liteseg": { - "model_name": "PP-Liteseg", - "batch_size": 1, - "jingdu": { - "value": 0.7402814977550732, - "unit": "mIoU", - "th": 0.01, - }, - "xingneng": { - "value": 11.52, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1529.8281200000001, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 543.0, - "unit": "MB", - "th": 0.05, - }, - }, - "HRNet": { - "model_name": "HRNet", - "batch_size": 1, - "jingdu": { - "value": 0.7749321005466953, - "unit": "mIoU", - "th": 0.01, - }, - "xingneng": { - "value": 27.46, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1528.0820200000003, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 627.0, - "unit": "MB", - "th": 0.05, - }, - }, - "UNet": { - "model_name": "UNet", - "batch_size": 1, - "jingdu": { - "value": 0.6281644433416818, - "unit": "mIoU", - "th": 0.01, - }, - "xingneng": { - "value": 41.059999999999995, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1541.5281200000002, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 983.0, - "unit": "MB", - "th": 0.05, - }, - }, - "Deeplabv3-ResNet50": { - "model_name": "Deeplabv3-ResNet50", - "batch_size": 1, - "jingdu": { - "value": 0.7791989132464726, - "unit": "mIoU", - "th": 0.01, - }, - "xingneng": { - "value": 41.84, - "unit": "ms", - "th": 0.05, - }, - "cpu_mem": { - "value": 1524.17972, - "unit": "MB", - "th": 0.05, - }, - "gpu_mem": { - "value": 617.0, - "unit": "MB", - "th": 0.05, - }, - }, - "ERNIE_3.0-Medium": { - "model_name": "ERNIE_3.0-Medium", - "batch_size": 32, - "jingdu": { - "value": 0.6533827618164968, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 16.442, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1421.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 3492.01642, - "unit": "MB", - "th": 0.05, - }, - }, - "PP-MiniLM": { - "model_name": "PP-MiniLM", - "batch_size": 32, - "jingdu": { - "value": 0.6607506950880444, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 20.012, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1298.2, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 3492.9086, - "unit": "MB", - "th": 0.05, - }, - }, - "BERT_Base": { - "model_name": "BERT_Base", - "batch_size": 1, - "jingdu": { - "value": 0.0, - "unit": "acc", - "th": 0.01, - }, - "xingneng": { - "value": 2.9800000000000004, - "unit": "ms", - "th": 0.05, - }, - "gpu_mem": { - "value": 1091.0, - "unit": "MB", - "th": 0.05, - }, - "cpu_mem": { - "value": 3834.2695200000003, - "unit": "MB", - "th": 0.05, - }, - }, -} diff --git a/inference/python_api_test/test_int8_model/base_trt_fp16.py b/inference/python_api_test/test_int8_model/base_trt_fp16.py deleted file mode 100644 index 1b65404e3e..0000000000 --- a/inference/python_api_test/test_int8_model/base_trt_fp16.py +++ /dev/null @@ -1,258 +0,0 @@ -""" -trt_fp16 base values -""" - -trt_fp16 = { - "PPYOLOE": { - "model_name": "PPYOLOE", - "jingdu": { - "value": 0.5135882081820193, - "unit": "mAP", - "th": 0.05, - }, - "xingneng": { - "value": 273.6, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "PicoDet": { - "model_name": "PicoDet", - "jingdu": { - "value": 0.300434412153292, - "unit": "mAP", - "th": 0.05, - }, - "xingneng": { - "value": 17.4, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "YOLOv5s": { - "model_name": "YOLOv5s", - "jingdu": { - "value": 0.37574151469621125, - "unit": "mAP", - "th": 0.05, - }, - "xingneng": { - "value": 40.5, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "YOLOv6s": { - "model_name": "YOLOv6s", - "jingdu": { - "value": 0.42524875891435443, - "unit": "mAP", - "th": 0.05, - }, - "xingneng": { - "value": 58.7, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "YOLOv7": { - "model_name": "YOLOv7", - "jingdu": { - "value": 0.5106915816882776, - "unit": "mAP", - "th": 0.05, - }, - "xingneng": { - "value": 136.2, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "ResNet_vd": { - "model_name": "ResNet_vd", - "jingdu": { - "value": 0.79046, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 13.2, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "MobileNetV3_large": { - "model_name": "MobileNetV3_large", - "jingdu": { - "value": 0.74958, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 5.2, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "PPLCNetV2": { - "model_name": "PPLCNetV2", - "jingdu": { - "value": 0.76868, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 5.1, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "PPHGNet_tiny": { - "model_name": "PPHGNet_tiny", - "jingdu": { - "value": 0.79594, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 12.4, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "EfficientNetB0": { - "model_name": "EfficientNetB0", - "jingdu": { - "value": 0.77026, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 9.8, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "PP-HumanSeg-Lite": { - "model_name": "PP-HumanSeg-Lite", - "jingdu": { - "value": 0.960031583569334, - "unit": "mIoU", - "th": 0.05, - }, - "xingneng": { - "value": 41.5, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "PP-Liteseg": { - "model_name": "PP-Liteseg", - "jingdu": { - "value": 0.7703976119566152, - "unit": "mIoU", - "th": 0.05, - }, - "xingneng": { - "value": 419.6, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "HRNet": { - "model_name": "HRNet", - "jingdu": { - "value": 0.7896978097502604, - "unit": "mIoU", - "th": 0.05, - }, - "xingneng": { - "value": 737.4, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "UNet": { - "model_name": "UNet", - "jingdu": { - "value": 0.649965905161135, - "unit": "mIoU", - "th": 0.05, - }, - "xingneng": { - "value": 2234.3, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "Deeplabv3-ResNet50": { - "model_name": "Deeplabv3-ResNet50", - "jingdu": { - "value": 0.7990287567610845, - "unit": "mIoU", - "th": 0.05, - }, - "xingneng": { - "value": 2806.4, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "ERNIE_3.0-Medium": { - "model_name": "ERNIE_3.0-Medium", - "jingdu": { - "value": 0.7534754402224282, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 187.05, - "unit": "ms", - "batch_size": 32, - "th": 0.05, - }, - }, - "PP-MiniLM": { - "model_name": "PP-MiniLM", - "jingdu": { - "value": 0.7402687673772012, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 180.81, - "unit": "ms", - "batch_size": 32, - "th": 0.05, - }, - }, - "BERT_Base": { - "model_name": "BERT_Base", - "jingdu": { - "value": 0.6006530974238766, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 52.91, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, -} diff --git a/inference/python_api_test/test_int8_model/base_trt_int8.py b/inference/python_api_test/test_int8_model/base_trt_int8.py deleted file mode 100644 index 51b7cfe6ea..0000000000 --- a/inference/python_api_test/test_int8_model/base_trt_int8.py +++ /dev/null @@ -1,258 +0,0 @@ -""" -trt_int8 base values -""" - -trt_int8 = { - "PPYOLOE": { - "model_name": "PPYOLOE", - "jingdu": { - "value": 0.008505799229272469, - "unit": "mAP", - "th": 0.05, - }, - "xingneng": { - "value": 284.9, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "PicoDet": { - "model_name": "PicoDet", - "jingdu": { - "value": 0.29576267147717544, - "unit": "mAP", - "th": 0.05, - }, - "xingneng": { - "value": 15.6, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "YOLOv5s": { - "model_name": "YOLOv5s", - "jingdu": { - "value": 0.337513986405508, - "unit": "mAP", - "th": 0.05, - }, - "xingneng": { - "value": 41.9, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "YOLOv6s": { - "model_name": "YOLOv6s", - "jingdu": { - "value": 0.38167538696759734, - "unit": "mAP", - "th": 0.05, - }, - "xingneng": { - "value": 36.3, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "YOLOv7": { - "model_name": "YOLOv7", - "jingdu": { - "value": 0.4599616751537943, - "unit": "mAP", - "th": 0.05, - }, - "xingneng": { - "value": 101.8, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "ResNet_vd": { - "model_name": "ResNet_vd", - "jingdu": { - "value": 0.78542, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 6.6, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "MobileNetV3_large": { - "model_name": "MobileNetV3_large", - "jingdu": { - "value": 0.70114, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 4.8, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "PPLCNetV2": { - "model_name": "PPLCNetV2", - "jingdu": { - "value": 0.75986, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 3.8, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "PPHGNet_tiny": { - "model_name": "PPHGNet_tiny", - "jingdu": { - "value": 0.77626, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 8.0, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "EfficientNetB0": { - "model_name": "EfficientNetB0", - "jingdu": { - "value": 0.75366, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 9.6, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "PP-HumanSeg-Lite": { - "model_name": "PP-HumanSeg-Lite", - "jingdu": { - "value": 0.9596980417424789, - "unit": "mIoU", - "th": 0.05, - }, - "xingneng": { - "value": 42.2, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "PP-Liteseg": { - "model_name": "PP-Liteseg", - "jingdu": { - "value": 0.6646508698054427, - "unit": "mIoU", - "th": 0.05, - }, - "xingneng": { - "value": 375.9, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "HRNet": { - "model_name": "HRNet", - "jingdu": { - "value": 0.7899464457999261, - "unit": "mIoU", - "th": 0.05, - }, - "xingneng": { - "value": 532.6, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "UNet": { - "model_name": "UNet", - "jingdu": { - "value": 0.6434970135618086, - "unit": "mIoU", - "th": 0.05, - }, - "xingneng": { - "value": 1105.8, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "Deeplabv3-ResNet50": { - "model_name": "Deeplabv3-ResNet50", - "jingdu": { - "value": 0.7900994083314681, - "unit": "mIoU", - "th": 0.05, - }, - "xingneng": { - "value": 861.7, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, - "ERNIE_3.0-Medium": { - "model_name": "ERNIE_3.0-Medium", - "jingdu": { - "value": 0.6809545875810936, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 102.71, - "unit": "ms", - "batch_size": 32, - "th": 0.05, - }, - }, - "PP-MiniLM": { - "model_name": "PP-MiniLM", - "jingdu": { - "value": 0.6899907321594069, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 115.12, - "unit": "ms", - "batch_size": 32, - "th": 0.05, - }, - }, - "BERT_Base": { - "model_name": "BERT_Base", - "jingdu": { - "value": 0.051546658541685234, - "unit": "acc", - "th": 0.05, - }, - "xingneng": { - "value": 18.94, - "unit": "ms", - "batch_size": 1, - "th": 0.05, - }, - }, -} diff --git a/inference/python_api_test/test_int8_model/convert_onnx.sh b/inference/python_api_test/test_int8_model/convert_onnx.sh deleted file mode 100644 index c17f368e36..0000000000 --- a/inference/python_api_test/test_int8_model/convert_onnx.sh +++ /dev/null @@ -1,128 +0,0 @@ -python -m pip install paddle2onnx==1.0.3 - -# ================================ FP32 ====================================== -# PPYOLOE+ no nms -paddle2onnx --model_dir=models/ppyoloe_plus_crn_s_80e_coco_no_nms/ --save_file=models/ppyoloe_plus_crn_s_80e_coco_no_nms/ppyoloe_plus_crn_s_80e_coco_no_nms.onnx --model_filename=model.pdmodel --params_filename=model.pdiparams -# PicoDet no nms -paddle2onnx --model_dir=models/picodet_s_416_coco_npu_no_postprocess/ --save_file=models/picodet_s_416_coco_npu_no_postprocess/picodet_s_416_coco_npu_no_postprocess.onnx --model_filename=model.pdmodel --params_filename=model.pdiparams -# YOLOv5s -wget https://paddle-slim-models.bj.bcebos.com/act/yolov5s.onnx -mv yolov5s.onnx models/yolov5s_infer -# YOLOv6s -wget https://paddle-slim-models.bj.bcebos.com/act/yolov6s.onnx -mv yolov6s.onnx models/yolov6s_infer -# YOLOv7 -wget https://paddle-slim-models.bj.bcebos.com/act/yolov7.onnx -mv yolov7.onnx models/yolov7_infer -# PP-HumanSeg-Lite -python utils/paddle_infer_shape.py --model_dir=models/ppseg_lite_portrait_398x224_with_softmax/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_dir=models/pp_humanseg_fp32 --input_shape_dict="{'x':[1, 3, 398, 224]}" -paddle2onnx --model_dir=models/pp_humanseg_fp32 --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/pp_humanseg_fp32/pp_humanseg_fp32.onnx -# PP-Liteseg -python utils/paddle_infer_shape.py --model_dir=models/RES-paddle2-PPLIteSegSTDC1/ --model_filename=model --params_filename=params --save_dir=models/pp_liteseg_fp32 --input_shape_dict="{'x':[1, 3, 1024, 2048]}" -paddle2onnx --model_dir=models/pp_liteseg_fp32 --model_filename=model --params_filename=params --save_file=models/pp_liteseg_fp32/pp_liteseg_fp32.onnx -# HRNet -python utils/paddle_infer_shape.py --model_dir=models/RES-paddle2-HRNetW18-Seg/ --model_filename=model --params_filename=params --save_dir=models/hrnet_fp32 --input_shape_dict="{'x':[1, 3, 1024, 2048]}" -paddle2onnx --model_dir=models/hrnet_fp32 --model_filename=model --params_filename=params --save_file=models/hrnet_fp32/hrnet_fp32.onnx -# UNet -python utils/paddle_infer_shape.py --model_dir=models/RES-paddle2-UNet/ --model_filename=model --params_filename=params --save_dir=models/unet_fp32 --input_shape_dict="{'x':[1, 3, 1024, 2048]}" -paddle2onnx --model_dir=models/unet_fp32 --model_filename=model --params_filename=params --save_file=models/unet_fp32/unet_fp32.onnx -# Deeplabv3-ResNet50 -python utils/paddle_infer_shape.py --model_dir=models/RES-paddle2-Deeplabv3-ResNet50/ --model_filename=model --params_filename=params --save_dir=models/Deeplabv3_ResNet50_fp32 --input_shape_dict="{'x':[1, 3, 1024, 2048]}" -paddle2onnx --model_dir=models/Deeplabv3_ResNet50_fp32 --model_filename=model --params_filename=params --save_file=models/Deeplabv3_ResNet50_fp32/Deeplabv3_ResNet50_fp32.onnx -# models/ResNet50_vd_infer/ -cd models/ResNet50_vd_infer/ -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx -cd - -# models/MobileNetV3_large_x1_0_infer/ -cd models/MobileNetV3_large_x1_0_infer/ -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx -cd - -# models/PPHGNet_tiny_infer/ -cd models/PPHGNet_tiny_infer/ -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx -cd - -# models/PPLCNetV2_base_infer/ -cd models/PPLCNetV2_base_infer/ -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx -cd - -# models/EfficientNetB0_infer/ -cd models/EfficientNetB0_infer/ -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx -cd - -## nlp -# models/AFQMC -cd models/AFQMC -paddle2onnx --model_dir ./ --model_filename infer.pdmodel --params_filename infer.pdiparams --save_file model.onnx -cd - -# models/afqmc -cd models/afqmc -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx -cd - -# models/x2paddle_cola -cd models/x2paddle_cola -paddle2onnx --model_dir ./ --model_filename model.pdmodel --params_filename model.pdiparams --save_file model.onnx -cd - - - - -# ================================ INT8 ====================================== -# PPYOLOE+ no nms -paddle2onnx --model_dir=models/ppyoloe_plus_crn_s_80e_coco_no_nms_quant/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/ppyoloe_plus_crn_s_80e_coco_no_nms_quant/ppyoloe_plus_crn_s_80e_coco_no_nms_quant.onnx --deploy_backend='tensorrt' --save_calibration_file=models/ppyoloe_plus_crn_s_80e_coco_no_nms_quant/calibration.cache -# PicoDet no nms -paddle2onnx --model_dir=models/picodet_s_416_coco_npu_no_postprocess_quant/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/picodet_s_416_coco_npu_no_postprocess_quant/picodet_s_416_coco_npu_no_postprocess_quant.onnx --deploy_backend='tensorrt' --save_calibration_file=models/picodet_s_416_coco_npu_no_postprocess_quant/calibration.cache -# YOLOv5s -paddle2onnx --model_dir=models/yolov5s_quant/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/yolov5s_quant/yolov5s_quant.onnx --deploy_backend='tensorrt' --save_calibration_file=models/yolov5s_quant/calibration.cache -# YOLOv6s -paddle2onnx --model_dir=models/yolov6s_quant/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/yolov6s_quant/yolov6s_quant.onnx --deploy_backend='tensorrt' --save_calibration_file=models/yolov6s_quant/calibration.cache -# YOLOv7 -paddle2onnx --model_dir=models/yolov7_quant/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/yolov7_quant/yolov7_quant.onnx --deploy_backend='tensorrt' --save_calibration_file=models/yolov7_quant/calibration.cache -# PP-HumanSeg-Lite -python utils/paddle_infer_shape.py --model_dir=models/pp_humanseg_qat/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_dir=models/pp_humanseg_int8 --input_shape_dict="{'x':[1, 3, 398, 224]}" -paddle2onnx --model_dir=models/pp_humanseg_int8 --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/pp_humanseg_int8/pp_humanseg_int8.onnx --deploy_backend='tensorrt' --save_calibration_file=models/pp_humanseg_int8/calibration.cache -# PP-Liteseg -python utils/paddle_infer_shape.py --model_dir=models/pp_liteseg_qat/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_dir=models/pp_liteseg_int8 --input_shape_dict="{'x':[1, 3, 1024, 2048]}" -paddle2onnx --model_dir=models/pp_liteseg_int8/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/pp_liteseg_int8/pp_liteseg_int8.onnx --deploy_backend='tensorrt' --save_calibration_file=models/pp_liteseg_int8/calibration.cache -# HRNet -python utils/paddle_infer_shape.py --model_dir=models/hrnet_qat/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_dir=models/hrnet_int8 --input_shape_dict="{'x':[1, 3, 1024, 2048]}" -paddle2onnx --model_dir=models/hrnet_int8 --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/hrnet_int8/hrnet_int8.onnx --deploy_backend='tensorrt' --save_calibration_file=models/hrnet_int8/calibration.cache -# UNet -python utils/paddle_infer_shape.py --model_dir=models/unet_qat/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_dir=models/unet_int8 --input_shape_dict="{'x':[1, 3, 1024, 2048]}" -paddle2onnx --model_dir=models/unet_int8 --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/unet_int8/unet_int8.onnx --deploy_backend='tensorrt' --save_calibration_file=models/unet_int8/calibration.cache -# Deeplabv3-ResNet50 -python utils/paddle_infer_shape.py --model_dir=models/deeplabv3_qat/ --model_filename=model.pdmodel --params_filename=model.pdiparams --save_dir=models/deeplabv3_int8 --input_shape_dict="{'x':[1, 3, 1024, 2048]}" -paddle2onnx --model_dir=models/deeplabv3_int8 --model_filename=model.pdmodel --params_filename=model.pdiparams --save_file=models/deeplabv3_int8/deeplabv3_int8.onnx --deploy_backend='tensorrt' --save_calibration_file=models/deeplabv3_int8/calibration.cache - -## classification -# models/ResNet50_vd_QAT -cd models/ResNet50_vd_QAT/ -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx --deploy_backend tensorrt -cd - -# models/MobileNetV3_large_x1_0_QAT/ -cd models/MobileNetV3_large_x1_0_QAT/ -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx --deploy_backend tensorrt -cd - -# models/PPLCNetV2_base_QAT/ -cd models/PPLCNetV2_base_QAT/ -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx --deploy_backend tensorrt -cd - -# models/PPHGNet_tiny_QAT/ -cd models/PPHGNet_tiny_QAT/ -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx --deploy_backend tensorrt -cd - -# models/EfficientNetB0_QAT/ -cd models/EfficientNetB0_QAT/ -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx --deploy_backend tensorrt -cd - -## nlp -# models/save_ernie3_afqmc_new_cablib -cd models/save_ernie3_afqmc_new_cablib -paddle2onnx --model_dir ./ --model_filename infer.pdmodel --params_filename infer.pdiparams --save_file model.onnx --deploy_backend tensorrt -cd - -# models/save_ppminilm_afqmc_new_calib -cd models/save_ppminilm_afqmc_new_calib -paddle2onnx --model_dir ./ --model_filename inference.pdmodel --params_filename inference.pdiparams --save_file model.onnx --deploy_backend tensorrt -cd - -# models/x2paddle_cola_new_calib -cd models/x2paddle_cola_new_calib -paddle2onnx --model_dir ./ --model_filename model.pdmodel --params_filename model.pdiparams --save_file model.onnx --deploy_backend tensorrt -cd - diff --git a/inference/python_api_test/test_int8_model/get_benchmark_info.py b/inference/python_api_test/test_int8_model/get_benchmark_info.py index 1ae6633c78..dec62d0edf 100644 --- a/inference/python_api_test/test_int8_model/get_benchmark_info.py +++ b/inference/python_api_test/test_int8_model/get_benchmark_info.py @@ -13,10 +13,6 @@ import base_mkldnn_fp32 import base_mkldnn_int8 -import base_trt_fp16 -import base_trt_int8 -import base_nv_trt_fp16 -import base_nv_trt_int8 import mail_report import write_db @@ -51,17 +47,9 @@ def get_runtime_info(log_file): def get_base_info(mode): """ 从base文件中读取base数据 - mode: trt_int8 trt_fp16 mkldnn_int8 mkldnn_fp32 + mode: mkldnn_int8 mkldnn_fp32 """ - if mode == "trt_int8": - base_res = base_trt_int8.trt_int8 - elif mode == "trt_fp16": - base_res = base_trt_fp16.trt_fp16 - elif mode == "nv_trt_int8": - base_res = base_nv_trt_int8.nv_trt_int8 - elif mode == "nv_trt_fp16": - base_res = base_nv_trt_fp16.nv_trt_fp16 - elif mode == "mkldnn_int8": + if mode == "mkldnn_int8": base_res = base_mkldnn_int8.mkldnn_int8 elif mode == "mkldnn_fp32": base_res = base_mkldnn_fp32.mkldnn_fp32 @@ -350,8 +338,8 @@ def res2db(env, benchmark_res, mode_list, metric_list): "model_name": model, "batch_size": info["batch_size"], "fp_mode": "int8", - "use_trt": True, - "use_mkldnn": False, + "use_trt": False, + "use_mkldnn": True, "jingdu": info["jingdu"]["value"], "jingdu_unit": info["jingdu"]["unit"], "ips": info["xingneng"]["value"], @@ -414,7 +402,6 @@ def run(): benchmark_res = {} diff_res = {} - diff_res_nv = {} for mode in mode_list: log_file = "eval_{}_acc.log".format(mode) _current = get_runtime_info(log_file) @@ -422,24 +409,15 @@ def run(): _base = get_base_info(mode) _diff = compare_diff(_base, _current, metric_list) diff_res.setdefault(mode, _diff) - if mode in ["trt_int8", "trt_fp16"]: - _base_nv = get_base_info("nv_" + mode) - _diff_nv = compare_diff(_base_nv, _current, metric_list) - diff_res_nv.setdefault(mode, _diff_nv) res_base, tongji_base = res_summary(diff_res, mode_list, metric_list) - res_nv, tongji_nv = res_summary(diff_res_nv, list(diff_res_nv.keys()), metric_list) res = { "base": res_base, - "NV-TRT": res_nv, } tongji = { "base": tongji_base, - "NV-TRT": tongji_nv, } jingping_list = ["base"] - if "trt_int8" in mode_list: - jingping_list.append("NV-TRT") env_str = "环境: " env_str += "docker: " diff --git a/inference/python_api_test/test_int8_model/requirements.txt b/inference/python_api_test/test_int8_model/requirements.txt index 9f7833e6c0..aae399e014 100644 --- a/inference/python_api_test/test_int8_model/requirements.txt +++ b/inference/python_api_test/test_int8_model/requirements.txt @@ -1,8 +1,6 @@ paddledet>=2.5.0 paddleseg==2.5.0 paddlenlp>=2.3.0 -pycuda -tensorrt onnx GPUtil psutil diff --git a/inference/python_api_test/test_int8_model/requirements_no_cuda.txt b/inference/python_api_test/test_int8_model/requirements_no_cuda.txt index 6ffe27df53..58ec7a4fca 100644 --- a/inference/python_api_test/test_int8_model/requirements_no_cuda.txt +++ b/inference/python_api_test/test_int8_model/requirements_no_cuda.txt @@ -6,4 +6,3 @@ onnx GPUtil psutil pynvml -tensorrt diff --git a/inference/python_api_test/test_int8_model/run.sh b/inference/python_api_test/test_int8_model/run.sh index 0f20daa3f5..ae59f4821e 100644 --- a/inference/python_api_test/test_int8_model/run.sh +++ b/inference/python_api_test/test_int8_model/run.sh @@ -11,10 +11,6 @@ do echo "==========START ${mode}=========" cp -r models.bak models - if [[ ${mode} =~ "trt_int8" ]] || [[ ${mode} =~ "trt_fp16" ]] - then - bash run_${mode}.sh > eval_${mode}_acc.log.tmp 2>&1 - fi bash run_${mode}.sh > eval_${mode}_acc.log 2>&1 rm -rf models done diff --git a/inference/python_api_test/test_int8_model/run_nv_trt_fp16.sh b/inference/python_api_test/test_int8_model/run_nv_trt_fp16.sh deleted file mode 100644 index 3a46c2c7e8..0000000000 --- a/inference/python_api_test/test_int8_model/run_nv_trt_fp16.sh +++ /dev/null @@ -1,71 +0,0 @@ -export CUDA_VISIBLE_DEVICES=0 -export FLAGS_call_stack_level=2 -PYTHON="python" - -# PPYOLOE+ trt fp16 -echo "[Benchmark] Run PPYOLOE+ trt fp16" -$PYTHON test_ppyoloe_infer.py --model_path=models/ppyoloe_plus_crn_s_80e_coco_no_nms/ppyoloe_plus_crn_s_80e_coco_no_nms.onnx --reader_config=configs/ppyoloe_plus_reader.yml --deploy_backend=tensorrt --precision=fp16 --model_name=PPYOLOE_PLUS --exclude_nms -# PicoDet trt fp16 -echo "[Benchmark] Run PicoDet trt fp16" -$PYTHON test_ppyoloe_infer.py --model_path=models/picodet_s_416_coco_npu_no_postprocess/picodet_s_416_coco_npu_no_postprocess.onnx --reader_config=configs/picodet_reader.yml --deploy_backend=tensorrt --precision=fp16 --model_name=PicoDet --img_shape=416 --exclude_nms -# YOLOv5s trt fp16 -echo "[Benchmark] Run YOLOv5s trt fp16" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov5s_infer/yolov5s.onnx --deploy_backend=tensorrt --precision=fp16 --model_name=YOLOv5s -# YOLOv6s trt fp16 -echo "[Benchmark] Run YOLOv6s trt fp16" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov6s_infer/yolov6s.onnx --deploy_backend=tensorrt --precision=fp16 --model_name=YOLOv6s -# YOLOv7 trt fp16 -echo "[Benchmark] Run YOLOv7 trt fp16" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov7_infer/yolov7.onnx --deploy_backend=tensorrt --precision=fp16 --model_name=YOLOv7 - - -# ResNet_vd trt fp16 -rm -rf model_fp16_model.trt -echo "[Benchmark] Run ResNet_vd trt fp16" -$PYTHON test_image_classification_infer.py --model_path=models/ResNet50_vd_infer/model.onnx --deploy_backend=tensorrt --input_name=inputs --precision=fp16 --model_name=ResNet_vd -rm -rf model_fp16_model.trt -# MobileNetV3_large trt fp16 -echo "[Benchmark] Run MobileNetV3_large trt fp16" -$PYTHON test_image_classification_infer.py --model_path=models/MobileNetV3_large_x1_0_infer/model.onnx --deploy_backend=tensorrt --input_name=inputs --precision=fp16 --model_name=MobileNetV3_large -rm -rf model_fp16_model.trt -# PPLCNetV2 trt fp16 -echo "[Benchmark] Run PPLCNetV2 trt fp16" -$PYTHON test_image_classification_infer.py --model_path=models/PPLCNetV2_base_infer/model.onnx --deploy_backend=tensorrt --precision=fp16 --model_name=PPLCNetV2 -rm -rf model_fp16_model.trt -# PPHGNet_tiny trt fp16 -echo "[Benchmark] Run PPHGNet_tiny trt fp16" -$PYTHON test_image_classification_infer.py --model_path=models/PPHGNet_tiny_infer/model.onnx --deploy_backend=tensorrt --precision=fp16 --model_name=PPHGNet_tiny -rm -rf model_fp16_model.trt -# EfficientNetB0 trt fp16 -echo "[Benchmark] Run EfficientNetB0 trt fp16" -$PYTHON test_image_classification_infer.py --model_path=models/EfficientNetB0_infer/model.onnx --deploy_backend=tensorrt --precision=fp16 --model_name=EfficientNetB0 -rm -rf model_fp16_model.trt - - -# PP-HumanSeg-Lite trt fp16 -echo "[Benchmark] Run PP-HumanSeg-Lite trt fp16" -$PYTHON test_segmentation_infer.py --model_path=models/pp_humanseg_fp32/pp_humanseg_fp32.onnx --dataset='human' --dataset_config=configs/humanseg_dataset.yaml --deploy_backend=tensorrt --precision=fp16 --model_name=PP-HumanSeg-Lite -# PP-Liteseg trt fp16 -echo "[Benchmark] Run PP-Liteseg trt fp16" -$PYTHON test_segmentation_infer.py --model_path=models/pp_liteseg_fp32/pp_liteseg_fp32.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=fp16 --model_name=PP-Liteseg -# HRNet trt fp16 -echo "[Benchmark] Run HRNet trt fp16" -$PYTHON test_segmentation_infer.py --model_path=models/hrnet_fp32/hrnet_fp32.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=fp16 --model_name=HRNet -# UNet trt fp16 -echo "[Benchmark] Run UNet trt fp16" -$PYTHON test_segmentation_infer.py --model_path=models/unet_fp32/unet_fp32.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=fp16 --model_name=UNet -# Deeplabv3-ResNet50 trt fp16 -echo "[Benchmark] Run Deeplabv3-ResNet50 trt fp16" -$PYTHON test_segmentation_infer.py --model_path=models/Deeplabv3_ResNet50_fp32/Deeplabv3_ResNet50_fp32.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=fp16 --model_name=Deeplabv3-ResNet50 - - -# nlp -# models/AFQMC -rm -rf model_fp16_model.trt -$PYTHON test_nlp_infer.py --model_path=models/AFQMC/model.onnx --deploy_backend=tensorrt --task_name='afqmc' --precision=fp16 --model_name=ERNIE_3.0-Medium -# models/afqmc -$PYTHON test_nlp_infer.py --model_path=models/afqmc/model.onnx --deploy_backend=tensorrt --task_name='afqmc' --precision=fp16 --model_name=PP-MiniLM -rm -rf model_fp16_model.trt -# models/x2paddle_cola -$PYTHON test_bert_infer.py --model_path=models/x2paddle_cola/model.onnx --deploy_backend=tensorrt --precision=fp16 --batch_size=1 --model_name=BERT_Base -rm -rf model_fp16_model.trt diff --git a/inference/python_api_test/test_int8_model/run_nv_trt_fp32.sh b/inference/python_api_test/test_int8_model/run_nv_trt_fp32.sh deleted file mode 100644 index b09c4791da..0000000000 --- a/inference/python_api_test/test_int8_model/run_nv_trt_fp32.sh +++ /dev/null @@ -1,35 +0,0 @@ -export CUDA_VISIBLE_DEVICES=0 -export FLAGS_call_stack_level=2 -PYTHON="python" - -# PPYOLOE+ trt fp32 -echo "[Benchmark] Run PPYOLOE+ trt fp32" -$PYTHON test_ppyoloe_infer.py --model_path=models/ppyoloe_plus_crn_s_80e_coco_no_nms/ppyoloe_plus_crn_s_80e_coco_no_nms.onnx --reader_config=configs/ppyoloe_plus_reader.yml --deploy_backend=tensorrt --precision=fp32 --model_name=PPYOLOE_PLUS --exclude_nms -# PicoDet trt fp32 -echo "[Benchmark] Run PicoDet trt fp32" -$PYTHON test_ppyoloe_infer.py --model_path=models/picodet_s_416_coco_npu_no_postprocess/picodet_s_416_coco_npu_no_postprocess.onnx --reader_config=configs/picodet_reader.yml --deploy_backend=tensorrt --precision=fp32 --model_name=PicoDet --img_shape=416 --exclude_nms -# YOLOv5s trt fp32 -echo "[Benchmark] Run YOLOv5s trt fp32" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov5s_infer/yolov5s.onnx --deploy_backend=tensorrt --precision=fp32 --model_name=YOLOv5s -# YOLOv6s trt fp32 -echo "[Benchmark] Run YOLOv6s trt fp32" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov6s_infer/yolov6s.onnx --deploy_backend=tensorrt --precision=fp32 --model_name=YOLOv6s -# YOLOv7 trt fp32 -echo "[Benchmark] Run YOLOv7 trt fp32" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov7_infer/yolov7.onnx --deploy_backend=tensorrt --precision=fp32 --model_name=YOLOv7 - -# PP-HumanSeg-Lite trt fp32 -echo "[Benchmark] Run PP-HumanSeg-Lite trt fp32" -$PYTHON test_segmentation_infer.py --model_path=models/pp_humanseg_fp32/pp_humanseg_fp32.onnx --dataset='human' --dataset_config=configs/humanseg_dataset.yaml --deploy_backend=tensorrt --precision=fp32 --model_name=PP-HumanSeg-Lite -# PP-Liteseg trt fp32 -echo "[Benchmark] Run PP-Liteseg trt fp32" -$PYTHON test_segmentation_infer.py --model_path=models/pp_liteseg_fp32/pp_liteseg_fp32.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=fp32 --model_name=ppliteseg -# HRNet trt fp32 -echo "[Benchmark] Run HRNet trt fp32" -$PYTHON test_segmentation_infer.py --model_path=models/hrnet_fp32/hrnet_fp32.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=fp32 --model_name=HRNet -# UNet trt fp32 -echo "[Benchmark] Run UNet trt fp32" -$PYTHON test_segmentation_infer.py --model_path=models/unet_fp32/unet_fp32.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=fp32 --model_name=UNet -# Deeplabv3-ResNet50 trt fp32 -echo "[Benchmark] Run Deeplabv3-ResNet50 trt fp32" -$PYTHON test_segmentation_infer.py --model_path=models/Deeplabv3_ResNet50_fp32/Deeplabv3_ResNet50_fp32.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=fp32 --model_name=Deeplabv3-ResNet50 diff --git a/inference/python_api_test/test_int8_model/run_nv_trt_int8.sh b/inference/python_api_test/test_int8_model/run_nv_trt_int8.sh deleted file mode 100644 index df78c26468..0000000000 --- a/inference/python_api_test/test_int8_model/run_nv_trt_int8.sh +++ /dev/null @@ -1,73 +0,0 @@ -export CUDA_VISIBLE_DEVICES=0 -export FLAGS_call_stack_level=2 -# Add this to reduce cpu memory! -export CUDA_MODULE_LOADING=LAZY -PYTHON="python" - -# PPYOLOE+ trt int8 -echo "[Benchmark] Run PPYOLOE+ trt int8" -$PYTHON test_ppyoloe_infer.py --model_path=models/ppyoloe_plus_crn_s_80e_coco_no_nms_quant/ppyoloe_plus_crn_s_80e_coco_no_nms_quant.onnx --reader_config=configs/ppyoloe_plus_reader.yml --deploy_backend=tensorrt --precision=int8 --model_name=PPYOLOE_PLUS --calibration_file=models/ppyoloe_plus_crn_s_80e_coco_no_nms_quant/calibration.cache --exclude_nms -# PicoDet trt int8 -echo "[Benchmark] Run PicoDet trt int8" -$PYTHON test_ppyoloe_infer.py --model_path=models/picodet_s_416_coco_npu_no_postprocess_quant/picodet_s_416_coco_npu_no_postprocess_quant.onnx --reader_config=configs/picodet_reader.yml --deploy_backend=tensorrt --precision=int8 --model_name=PicoDet --calibration_file=models/picodet_s_416_coco_npu_no_postprocess_quant/calibration.cache --img_shape=416 --exclude_nms -# YOLOv5s trt int8 -echo "[Benchmark] Run YOLOv5s trt int8" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov5s_quant/yolov5s_quant.onnx --deploy_backend=tensorrt --precision=int8 --model_name=YOLOv5s --calibration_file=models/yolov5s_quant/calibration.cache -# YOLOv6s trt int8 -echo "[Benchmark] Run YOLOv6s trt int8" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov6s_quant/yolov6s_quant.onnx --deploy_backend=tensorrt --precision=int8 --model_name=YOLOv6s --calibration_file=models/yolov6s_quant/calibration.cache -# YOLOv7 trt int8 -echo "[Benchmark] Run YOLOv7 trt int8" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov7_quant/yolov7_quant.onnx --deploy_backend=tensorrt --precision=int8 --model_name=YOLOv7 --calibration_file=models/yolov7_quant/calibration.cache - -# ResNet_vd trt int8 -rm -rf model_int8_model.trt -echo "[Benchmark] Run ResNet_vd trt int8" -$PYTHON test_image_classification_infer.py --model_path=models/ResNet50_vd_QAT/model.onnx --deploy_backend=tensorrt --input_name=inputs --precision=int8 --calibration_file=models/ResNet50_vd_QAT/calibration.cache --model_name=ResNet_vd -rm -rf model_int8_model.trt -# MobileNetV3_large trt int8 -echo "[Benchmark] Run MobileNetV3_large trt int8" -$PYTHON test_image_classification_infer.py --model_path=models/MobileNetV3_large_x1_0_QAT/model.onnx --deploy_backend=tensorrt --input_name=inputs --precision=int8 --calibration_file=models/MobileNetV3_large_x1_0_QAT/calibration.cache --model_name=MobileNetV3_large -rm -rf model_int8_model.trt -# PPLCNetV2 trt int8 -echo "[Benchmark] Run PPLCNetV2 trt int8" -$PYTHON test_image_classification_infer.py --model_path=models/PPLCNetV2_base_QAT/model.onnx --deploy_backend=tensorrt --precision=int8 --calibration_file=models/PPLCNetV2_base_QAT/calibration.cache --model_name=PPLCNetV2 -rm -rf model_int8_model.trt -# PPHGNet_tiny trt int8 -echo "[Benchmark] Run PPHGNet_tiny trt int8" -$PYTHON test_image_classification_infer.py --model_path=models/PPHGNet_tiny_QAT/model.onnx --deploy_backend=tensorrt --precision=int8 --calibration_file=models/PPHGNet_tiny_QAT/calibration.cache --model_name=PPHGNet_tiny -rm -rf model_int8_model.trt -# EfficientNetB0 trt int8 -echo "[Benchmark] Run EfficientNetB0 trt int8" -$PYTHON test_image_classification_infer.py --model_path=models/EfficientNetB0_QAT/model.onnx --deploy_backend=tensorrt --precision=int8 --calibration_file=models/EfficientNetB0_QAT/calibration.cache --model_name=EfficientNetB0 -rm -rf model_int8_model.trt - -# PP-HumanSeg-Lite trt int8 -echo "[Benchmark] Run PP-HumanSeg-Lite trt int8" -$PYTHON test_segmentation_infer.py --model_path=models/pp_humanseg_int8/pp_humanseg_int8.onnx --dataset='human' --dataset_config=configs/humanseg_dataset.yaml --deploy_backend=tensorrt --precision=int8 --model_name=PP-HumanSeg-Lite --calibration_file=models/pp_humanseg_int8/calibration.cache -# PP-Liteseg trt int8 -echo "[Benchmark] Run PP-Liteseg trt int8" -$PYTHON test_segmentation_infer.py --model_path=models/pp_liteseg_int8/pp_liteseg_int8.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=int8 --model_name=PP-Liteseg --calibration_file=models/pp_liteseg_int8/calibration.cache -# HRNet trt int8 -echo "[Benchmark] Run HRNet trt int8" -$PYTHON test_segmentation_infer.py --model_path=models/hrnet_int8/hrnet_int8.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=int8 --model_name=HRNet --calibration_file=models/hrnet_int8/calibration.cache -# UNet trt int8 -echo "[Benchmark] Run UNet trt int8" -$PYTHON test_segmentation_infer.py --model_path=models/unet_int8/unet_int8.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=int8 --model_name=UNet --calibration_file=models/unet_int8/calibration.cache -# Deeplabv3-ResNet50 trt int8 -echo "[Benchmark] Run Deeplabv3-ResNet50 trt int8" -$PYTHON test_segmentation_infer.py --model_path=models/deeplabv3_int8/deeplabv3_int8.onnx --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --deploy_backend=tensorrt --precision=int8 --model_name=Deeplabv3-ResNet50 --calibration_file=models/deeplabv3_int8/calibration.cache - -# ERNIE 3.0-Medium nv-trt int8 -rm -rf model_int8_model.trt -echo "[Benchmark] Run NV-TRT ERNIE 3.0-Medium trt int8" -$PYTHON test_nlp_infer.py --model_path=models/save_ernie3_afqmc_new_cablib/model.onnx --deploy_backend=tensorrt --task_name='afqmc' --precision=int8 --calibration_file=models/save_ernie3_afqmc_new_cablib/calibration.cache --model_name=ERNIE_3.0-Medium -rm -rf model_int8_model.trt -# PP-MiniLM nv-trt int8 -echo "[Benchmark] Run NV-TRT PP-MiniLM trt int8" -$PYTHON test_nlp_infer.py --model_path=models/save_ppminilm_afqmc_new_calib/model.onnx --deploy_backend=tensorrt --task_name='afqmc' --precision=int8 --calibration_file=models/save_ppminilm_afqmc_new_calib/calibration.cache --model_name=PP-MiniLM -rm -rf model_int8_model.trt -# BERT Base nv-trt int8 -echo "[Benchmark] Run NV-TRT BERT Base trt int8" -$PYTHON test_bert_infer.py --model_path=models/x2paddle_cola_new_calib/model.onnx --precision=int8 --batch_size=1 --deploy_backend=tensorrt --calibration_file=./models/x2paddle_cola_new_calib/calibration.cache --model_name=BERT_Base -rm -rf model_int8_model.trt diff --git a/inference/python_api_test/test_int8_model/run_ocr.sh b/inference/python_api_test/test_int8_model/run_ocr.sh deleted file mode 100644 index cf62cdb186..0000000000 --- a/inference/python_api_test/test_int8_model/run_ocr.sh +++ /dev/null @@ -1,15 +0,0 @@ -export PYTHONPATH=$PWD/PaddleOCR:$PYTHONPATH - -# 测速 -# paddle2onnx --model_dir=models/PPOCRV3_det_QAT --model_filename=inference.pdmodel --params_filename=inference.pdiparams --save_file=models/PPOCRV3_det_QAT/model.onnx --save_calibration_file=models/PPOCRV3_det_QAT/calibration.cache --deploy_backend='tensorrt' -python test_ocr_infer.py --model_type='det' --model_path="./models/PPOCRV3_det_QAT" --model_filename="inference.pdmodel" --params_filename="inference.pdiparams" --image_file="test.jpg" --device='GPU' --use_trt=True --precision='int8' --benchmark=True --deploy_backend='paddle_inference' - -# 测精度 -# python test_ocr_infer.py --model_path="./PPOCRV3_det_QAT" --model_filename="inference.pdmodel" --params_filename="inference.pdiparams" --dataset_config="./configs/ppocrv3_det.yaml" --device='GPU' --use_trt=True --precision='int8' --deploy_backend='paddle_inference' - - -# 测速 ocr_rec -python test_ocr_infer.py --model_type='rec' --model_path="./PPOCRV3_rec_QAT" --model_filename="inference.pdmodel" --params_filename="inference.pdiparams" --image_file="test.jpg" --device='GPU' --use_trt=True --precision='int8' --benchmark=True --deploy_backend='paddle_inference' - -# 测精度 ocr_rec -# python test_ocr_infer.py --model_path="./PPOCRV3_rec_QAT" --model_filename="inference.pdmodel" --params_filename="inference.pdiparams" --dataset_config="./configs/ppocrv3_rec.yaml" --device='GPU' --use_trt=True --precision='int8' --deploy_backend='paddle_inference' diff --git a/inference/python_api_test/test_int8_model/run_trt_fp16.sh b/inference/python_api_test/test_int8_model/run_trt_fp16.sh deleted file mode 100644 index bddd458628..0000000000 --- a/inference/python_api_test/test_int8_model/run_trt_fp16.sh +++ /dev/null @@ -1,69 +0,0 @@ -export CUDA_VISIBLE_DEVICES=0 -export FLAGS_call_stack_level=2 -# Add this to reduce cpu memory! -export CUDA_MODULE_LOADING=LAZY -PYTHON="python" - -# PPYOLOE trt fp16 -echo "[Benchmark] Run PPYOLOE trt fp16" -$PYTHON test_ppyoloe_infer.py --model_path=models/ppyoloe_crn_l_300e_coco --reader_config=configs/ppyoloe_reader.yml --use_trt=True --precision=fp16 --model_name=PPYOLOE -# PPYOLOE+ trt fp16 -echo "[Benchmark] Run PPYOLOE+ trt fp16" -$PYTHON test_ppyoloe_infer.py --model_path=models/ppyoloe_plus_crn_s_80e_coco_no_nms --reader_config=configs/ppyoloe_plus_reader.yml --use_trt=True --precision=fp16 --model_name=PPYOLOE_PLUS --exclude_nms -# PicoDet trt fp16 -### echo "[Benchmark] Run PicoDet trt fp16" -### $PYTHON test_ppyoloe_infer.py --model_path=models/picodet_s_416_coco_npu --reader_config=configs/picodet_reader.yml --use_trt=True --precision=fp16 --model_name=PicoDet -# PicoDet no nms trt fp16 -echo "[Benchmark] Run PicoDet no nms trt fp16" -$PYTHON test_ppyoloe_infer.py --model_path=models/picodet_s_416_coco_npu_no_postprocess --reader_config=configs/picodet_reader.yml --use_trt=True --precision=fp16 --model_name=PicoDet --exclude_nms -# YOLOv5s trt fp16 -echo "[Benchmark] Run YOLOv5s trt fp16" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov5s_infer --use_trt=True --precision=fp16 --model_name=YOLOv5s -# YOLOv6s trt fp16 -echo "[Benchmark] Run YOLOv6s trt fp16" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov6s_infer --use_trt=True --precision=fp16 --model_name=YOLOv6s -# YOLOv7 trt fp16 -echo "[Benchmark] Run YOLOv7 trt fp16" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov7_infer --use_trt=True --precision=fp16 --model_name=YOLOv7 - -# ResNet_vd trt fp16 -echo "[Benchmark] Run ResNet_vd trt fp16" -$PYTHON test_image_classification_infer.py --model_path=models/ResNet50_vd_infer --use_trt=True --precision=fp16 --use_gpu=True --model_name=ResNet_vd -# MobileNetV3_large trt fp16 -echo "[Benchmark] Run MobileNetV3_large trt fp16" -$PYTHON test_image_classification_infer.py --model_path=models/MobileNetV3_large_x1_0_infer --use_trt=True --precision=fp16 --use_gpu=True --model_name=MobileNetV3_large -# PPLCNetV2 trt fp16 -echo "[Benchmark] Run PPLCNetV2 trt fp16" -$PYTHON test_image_classification_infer.py --model_path=models/PPLCNetV2_base_infer --use_trt=True --precision=fp16 --use_gpu=True --model_name=PPLCNetV2 -# PPHGNet_tiny trt fp16 -echo "[Benchmark] Run PPHGNet_tiny trt fp16" -$PYTHON test_image_classification_infer.py --model_path=models/PPHGNet_tiny_infer --use_trt=True --precision=fp16 --use_gpu=True --model_name=PPHGNet_tiny -# EfficientNetB0 trt fp16 -echo "[Benchmark] Run EfficientNetB0 trt fp16" -$PYTHON test_image_classification_infer.py --model_path=models/EfficientNetB0_infer --use_trt=True --precision=fp16 --use_gpu=True --model_name=EfficientNetB0 - -# PP-HumanSeg-Lite trt fp16 -echo "[Benchmark] Run PP-HumanSeg-Lite trt fp16" -$PYTHON test_segmentation_infer.py --model_path=models/ppseg_lite_portrait_398x224_with_softmax --dataset='human' --dataset_config=configs/humanseg_dataset.yaml --use_trt=True --precision=fp16 --model_name=PP-HumanSeg-Lite -# PP-Liteseg trt fp16 -echo "[Benchmark] Run PP-Liteseg trt fp16" -$PYTHON test_segmentation_infer.py --model_path=models/RES-paddle2-PPLIteSegSTDC1 --model_filename=model --params_filename=params --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=fp16 --model_name=PP-Liteseg -# HRNet trt fp16 -echo "[Benchmark] Run HRNet trt fp16" -$PYTHON test_segmentation_infer.py --model_path=models/RES-paddle2-HRNetW18-Seg --model_filename=model --params_filename=params --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=fp16 --model_name=HRNet -# UNet trt fp16 -echo "[Benchmark] Run UNet trt fp16" -$PYTHON test_segmentation_infer.py --model_path=models/RES-paddle2-UNet --model_filename=model --params_filename=params --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=fp16 --model_name=UNet -# Deeplabv3-ResNet50 trt fp16 -echo "[Benchmark] Run Deeplabv3-ResNet50 trt fp16" -$PYTHON test_segmentation_infer.py --model_path=models/RES-paddle2-Deeplabv3-ResNet50 --model_filename=model --params_filename=params --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=fp16 --model_name=Deeplabv3-ResNet50 - -# ERNIE 3.0-Medium trt fp16 -echo "[Benchmark] Run ERNIE 3.0-Medium trt fp16" -$PYTHON test_nlp_infer.py --model_path=models/AFQMC --model_filename=infer.pdmodel --params_filename=infer.pdiparams --task_name='afqmc' --use_trt --precision=fp16 --model_name=ERNIE_3.0-Medium -# PP-MiniLM trt fp16 -echo "[Benchmark] Run PP-MiniLM trt fp16" -$PYTHON test_nlp_infer.py --model_path=models/afqmc --task_name='afqmc' --use_trt --precision=fp16 --model_name=PP-MiniLM -# BERT Base trt fp16 -echo "[Benchmark] Run BERT Base trt fp16" -$PYTHON test_bert_infer.py --model_path=models/x2paddle_cola --use_trt --precision=fp16 --batch_size=1 --model_name=BERT_Base diff --git a/inference/python_api_test/test_int8_model/run_trt_fp32.sh b/inference/python_api_test/test_int8_model/run_trt_fp32.sh deleted file mode 100644 index d34ea618de..0000000000 --- a/inference/python_api_test/test_int8_model/run_trt_fp32.sh +++ /dev/null @@ -1,67 +0,0 @@ -export CUDA_VISIBLE_DEVICES=0 -export FLAGS_call_stack_level=2 -PYTHON="python" - -# PPYOLOE trt fp32 -echo "[Benchmark] Run PPYOLOE trt fp32" -$PYTHON test_ppyoloe_infer.py --model_path=models/ppyoloe_crn_l_300e_coco --reader_config=configs/ppyoloe_reader.yml --use_trt=True --precision=fp32 --model_name=PPYOLOE -# PPYOLOE+ trt fp32 -echo "[Benchmark] Run PPYOLOE+ trt fp32" -$PYTHON test_ppyoloe_infer.py --model_path=models/ppyoloe_plus_crn_s_80e_coco_no_nms --reader_config=configs/ppyoloe_plus_reader.yml --use_trt=True --precision=fp32 --model_name=PPYOLOE_PLUS --exclude_nms -# PicoDet trt fp32 -### echo "[Benchmark] Run PicoDet trt fp32" -### $PYTHON test_ppyoloe_infer.py --model_path=models/picodet_s_416_coco_npu --reader_config=configs/picodet_reader.yml --use_trt=True --precision=fp32 --model_name=PicoDet -# PicoDet no nms trt fp32 -echo "[Benchmark] Run PicoDet no nms trt fp32" -$PYTHON test_ppyoloe_infer.py --model_path=models/picodet_s_416_coco_npu_no_postprocess --reader_config=configs/picodet_reader.yml --use_trt=True --precision=fp32 --model_name=PicoDet --exclude_nms -# YOLOv5s trt fp32 -echo "[Benchmark] Run YOLOv5s trt fp32" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov5s_infer --use_trt=True --precision=fp32 --model_name=YOLOv5s -# YOLOv6s trt fp32 -echo "[Benchmark] Run YOLOv6s trt fp32" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov6s_infer --use_trt=True --precision=fp32 --model_name=YOLOv6s -# YOLOv7 trt fp32 -echo "[Benchmark] Run YOLOv7 trt fp32" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov7_infer --use_trt=True --precision=fp32 --model_name=YOLOv7 - -# ResNet_vd trt fp32 -echo "[Benchmark] Run ResNet_vd trt fp32" -$PYTHON test_image_classification_infer.py --model_path=models/ResNet50_vd_infer --use_trt=True --use_gpu=True --model_name=ResNet_vd -# MobileNetV3_large trt fp32 -echo "[Benchmark] Run MobileNetV3_large trt fp32" -$PYTHON test_image_classification_infer.py --model_path=models/MobileNetV3_large_x1_0_infer --use_trt=True --use_gpu=True --model_name=MobileNetV3_large -# PPLCNetV2 trt fp32 -echo "[Benchmark] Run PPLCNetV2 trt fp32" -$PYTHON test_image_classification_infer.py --model_path=models/PPLCNetV2_base_infer --use_trt=True --use_gpu=True --model_name=PPLCNetV2 -# PPHGNet_tiny trt fp32 -echo "[Benchmark] Run PPHGNet_tiny trt fp32" -$PYTHON test_image_classification_infer.py --model_path=models/PPHGNet_tiny_infer --use_trt=True --use_gpu=True --model_name=PPHGNet_tiny -# EfficientNetB0 trt fp32 -echo "[Benchmark] Run EfficientNetB0 trt fp32" -$PYTHON test_image_classification_infer.py --model_path=models/EfficientNetB0_infer --use_trt=True --use_gpu=True --model_name=EfficientNetB0 - -# PP-HumanSeg-Lite trt fp32 -echo "[Benchmark] Run PP-HumanSeg-Lite trt fp32" -$PYTHON test_segmentation_infer.py --model_path=models/ppseg_lite_portrait_398x224_with_softmax --dataset='human' --dataset_config=configs/humanseg_dataset.yaml --use_trt=True --precision=fp32 --model_name=PP-HumanSeg-Lite -# PP-Liteseg trt fp32 -echo "[Benchmark] Run PP-Liteseg trt fp32" -$PYTHON test_segmentation_infer.py --model_path=models/RES-paddle2-PPLIteSegSTDC1 --model_filename=model --params_filename=params --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=fp32 --model_name=PP-Liteseg -# HRNet trt fp32 -echo "[Benchmark] Run HRNet trt fp32" -$PYTHON test_segmentation_infer.py --model_path=models/RES-paddle2-HRNetW18-Seg --model_filename=model --params_filename=params --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=fp32 --model_name=HRNet -# UNet trt fp32 -echo "[Benchmark] Run UNet trt fp32" -$PYTHON test_segmentation_infer.py --model_path=models/RES-paddle2-UNet --model_filename=model --params_filename=params --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=fp32 --model_name=UNet -# Deeplabv3-ResNet50 trt fp32 -echo "[Benchmark] Run Deeplabv3-ResNet50 trt fp32" -$PYTHON test_segmentation_infer.py --model_path=models/RES-paddle2-Deeplabv3-ResNet50 --model_filename=model --params_filename=params --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=fp32 --model_name=Deeplabv3-ResNet50 - -# ERNIE 3.0-Medium trt fp32 -echo "[Benchmark] Run ERNIE 3.0-Medium trt fp32" -$PYTHON test_nlp_infer.py --model_path=models/AFQMC --model_filename=infer.pdmodel --params_filename=infer.pdiparams --task_name='afqmc' --use_trt --precision=fp32 --model_name=ERNIE_3.0-Medium -# PP-MiniLM trt fp32 -echo "[Benchmark] Run PP-MiniLM trt fp32" -$PYTHON test_nlp_infer.py --model_path=models/afqmc --task_name='afqmc' --use_trt --precision=fp32 --model_name=PP-MiniLM -# BERT Base trt fp32 -echo "[Benchmark] Run BERT Base trt fp32" -$PYTHON test_bert_infer.py --model_path=models/x2paddle_cola --use_trt --precision=fp32 --batch_size=1 --model_name=BERT_Base diff --git a/inference/python_api_test/test_int8_model/run_trt_int8.sh b/inference/python_api_test/test_int8_model/run_trt_int8.sh deleted file mode 100644 index 1781035b2d..0000000000 --- a/inference/python_api_test/test_int8_model/run_trt_int8.sh +++ /dev/null @@ -1,69 +0,0 @@ -export CUDA_VISIBLE_DEVICES=0 -export FLAGS_call_stack_level=2 -# Add this to reduce cpu memory! -export CUDA_MODULE_LOADING=LAZY -PYTHON="python" - -# PPYOLOE trt int8 -echo "[Benchmark] Run PPYOLOE trt int8" -$PYTHON test_ppyoloe_infer.py --model_path=models/ppyoloe_crn_l_300e_coco_quant --reader_config=configs/ppyoloe_reader.yml --use_trt=True --precision=int8 --model_name=PPYOLOE -# PPYOLOE+ trt int8 -echo "[Benchmark] Run PPYOLOE+ trt int8" -$PYTHON test_ppyoloe_infer.py --model_path=models/ppyoloe_plus_crn_s_80e_coco_no_nms_quant --reader_config=configs/ppyoloe_plus_reader.yml --use_trt=True --precision=int8 --model_name=PPYOLOE_PLUS --exclude_nms -# PicoDet trt int8 -### echo "[Benchmark] Run PicoDet trt int8" -### $PYTHON test_ppyoloe_infer.py --model_path=models/picodet_s_416_coco_npu_quant --reader_config=configs/picodet_reader.yml --use_trt=True --precision=int8 --model_name=PicoDet -# PicoDet no nms trt int8 -echo "[Benchmark] Run PicoDet no nms trt int8" -$PYTHON test_ppyoloe_infer.py --model_path=models/picodet_s_416_coco_npu_no_postprocess_quant --reader_config=configs/picodet_reader.yml --use_trt=True --precision=int8 --model_name=PicoDet --exclude_nms -# YOLOv5s trt int8 -echo "[Benchmark] Run YOLOv5s trt int8" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov5s_quant --use_trt=True --precision=int8 --model_name=YOLOv5s -# YOLOv6s trt int8 -echo "[Benchmark] Run YOLOv6s trt int8" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov6s_quant --use_trt=True --precision=int8 --model_name=YOLOv6s -# YOLOv7 trt int8 -echo "[Benchmark] Run YOLOv7 trt int8" -$PYTHON test_yolo_series_infer.py --model_path=models/yolov7_quant --use_trt=True --precision=int8 --model_name=YOLOv7 - -# ResNet_vd trt int8 -echo "[Benchmark] Run ResNet_vd trt int8" -$PYTHON test_image_classification_infer.py --model_path=models/ResNet50_vd_QAT --use_trt=True --precision=int8 --use_gpu=True --model_name=ResNet_vd -# MobileNetV3_large trt int8 -echo "[Benchmark] Run MobileNetV3_large trt int8" -$PYTHON test_image_classification_infer.py --model_path=models/MobileNetV3_large_x1_0_QAT --use_trt=True --precision=int8 --use_gpu=True --model_name=MobileNetV3_large -# PPLCNetV2 trt int8 -echo "[Benchmark] Run PPLCNetV2 trt int8" -$PYTHON test_image_classification_infer.py --model_path=models/PPLCNetV2_base_QAT --use_trt=True --precision=int8 --use_gpu=True --model_name=PPLCNetV2 -# PPHGNet_tiny trt int8 -echo "[Benchmark] Run PPHGNet_tiny trt int8" -$PYTHON test_image_classification_infer.py --model_path=models/PPHGNet_tiny_QAT --use_trt=True --precision=int8 --use_gpu=True --model_name=PPHGNet_tiny -# EfficientNetB0 trt int8 -echo "[Benchmark] Run EfficientNetB0 trt int8" -$PYTHON test_image_classification_infer.py --model_path=models/EfficientNetB0_QAT --use_trt=True --precision=int8 --use_gpu=True --model_name=EfficientNetB0 - -# PP-HumanSeg-Lite trt int8 -echo "[Benchmark] Run PP-HumanSeg-Lite trt int8" -$PYTHON test_segmentation_infer.py --model_path=models/pp_humanseg_qat --dataset='human' --dataset_config=configs/humanseg_dataset.yaml --use_trt=True --precision=int8 --model_name=PP-HumanSeg-Lite -# PP-Liteseg trt int8 -echo "[Benchmark] Run PP-Liteseg trt int8" -$PYTHON test_segmentation_infer.py --model_path=models/pp_liteseg_qat --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=int8 --model_name=PP-Liteseg -# HRNet trt int8 -echo "[Benchmark] Run HRNet trt int8" -$PYTHON test_segmentation_infer.py --model_path=models/hrnet_qat --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=int8 --model_name=HRNet -# UNet trt int8 -echo "[Benchmark] Run UNet trt int8" -$PYTHON test_segmentation_infer.py --model_path=models/unet_qat --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=int8 --model_name=UNet -# Deeplabv3-ResNet50 trt int8 -echo "[Benchmark] Run Deeplabv3-ResNet50 trt int8" -$PYTHON test_segmentation_infer.py --model_path=models/deeplabv3_qat --dataset='cityscape' --dataset_config=configs/cityscapes_1024x512_scale1.0.yml --use_trt=True --precision=int8 --model_name=Deeplabv3-ResNet50 - -# ERNIE 3.0-Medium trt int8 -echo "[Benchmark] Run ERNIE 3.0-Medium trt int8" -$PYTHON test_nlp_infer.py --model_path=models/save_ernie3_afqmc_new_cablib --model_filename=infer.pdmodel --params_filename=infer.pdiparams --task_name='afqmc' --use_trt --precision=int8 --model_name=ERNIE_3.0-Medium -# PP-MiniLM MKLDNN trt int8 -echo "[Benchmark] Run PP-MiniLM trt int8" -$PYTHON test_nlp_infer.py --model_path=models/save_ppminilm_afqmc_new_calib --task_name='afqmc' --use_trt --precision=int8 --model_name=PP-MiniLM -# BERT Base MKLDNN trt int8 -echo "[Benchmark] Run BERT Base trt int8" -$PYTHON test_bert_infer.py --model_path=models/x2paddle_cola_new_calib --use_trt --precision=int8 --batch_size=1 --model_name=BERT_Base diff --git a/inference/python_api_test/test_int8_model/test_bert_infer.py b/inference/python_api_test/test_int8_model/test_bert_infer.py index 967c60aaa6..5a1d4d4fde 100644 --- a/inference/python_api_test/test_int8_model/test_bert_infer.py +++ b/inference/python_api_test/test_int8_model/test_bert_infer.py @@ -89,7 +89,6 @@ def argsparser(): default="GPU", help="Choose the device you want to run, it can be: CPU/GPU/XPU, default is GPU", ) - parser.add_argument("--use_dynamic_shape", type=bool, default=True, help="Whether use dynamic shape or not.") parser.add_argument( "--batch_size", default=32, @@ -109,11 +108,6 @@ def argsparser(): type=int, help="Warmup steps for performance test.", ) - parser.add_argument( - "--use_trt", - action="store_true", - help="Whether to use inference engin TensorRT.", - ) parser.add_argument("--use_l3", type=bool, default=False, help="Whether use L3_cache or not.") parser.add_argument( "--precision", @@ -128,9 +122,8 @@ def argsparser(): "--deploy_backend", type=str, default="paddle_inference", - help="deploy backend, it can be: `paddle_inference`, `tensorrt`, `onnxruntime`", + help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) - parser.add_argument("--calibration_file", type=str, default=None, help="quant onnx model calibration cache file.") parser.add_argument("--model_name", type=str, default="", help="model_name for benchmark") return parser @@ -311,45 +304,17 @@ def main(FLAGS): model_filename=FLAGS.model_filename, params_filename=FLAGS.params_filename, precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_l3=FLAGS.use_l3, use_mkldnn=FLAGS.use_mkldnn, - batch_size=FLAGS.batch_size, device=FLAGS.device, - min_subgraph_size=5, - use_dynamic_shape=FLAGS.use_dynamic_shape, cpu_threads=FLAGS.cpu_threads, ) - elif FLAGS.deploy_backend == "tensorrt": - from backend.tensorrt import TensorRTEngine - - model_name = os.path.split(FLAGS.model_path)[-1].rstrip(".onnx") - token_dir = os.path.dirname(FLAGS.model_path) - engine_file = "{}_{}_model.trt".format(model_name, FLAGS.precision) - print(engine_file) - predictor = TensorRTEngine( - onnx_model_file=FLAGS.model_path, - shape_info={ - "x0": [[1, 128], [1, 128], [1, 128]], - "x1": [[1, 128], [1, 128], [1, 128]], - "x2": [[1, 128], [1, 128], [1, 128]], - }, - max_batch_size=FLAGS.batch_size, - precision=FLAGS.precision, - engine_file_path=engine_file, - calibration_cache_file=FLAGS.calibration_file, - verbose=False, - ) elif FLAGS.deploy_backend == "onnxruntime": from backend.onnxruntime import ONNXRuntimeEngine - model_name = os.path.split(FLAGS.model_path)[-1].rstrip(".onnx") token_dir = os.path.dirname(FLAGS.model_path) - engine_file = "{}_{}_model.trt".format(model_name, FLAGS.precision) predictor = ONNXRuntimeEngine( onnx_model_file=FLAGS.model_path, - precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_mkldnn=FLAGS.use_mkldnn, device=FLAGS.device, ) @@ -380,7 +345,6 @@ def main(FLAGS): if __name__ == "__main__": - # If the device is not set to cpu, the nv-trt will report an error when executing paddle.set_device("cpu") parser = argsparser() FLAGS = parser.parse_args() diff --git a/inference/python_api_test/test_int8_model/test_image_classification_infer.py b/inference/python_api_test/test_int8_model/test_image_classification_infer.py index f370e7f72c..bcd58cde01 100644 --- a/inference/python_api_test/test_int8_model/test_image_classification_infer.py +++ b/inference/python_api_test/test_int8_model/test_image_classification_infer.py @@ -45,7 +45,6 @@ def argsparser(): parser.add_argument("--use_mkldnn", type=bool, default=False, help="Whether to use mkldnn") parser.add_argument("--cpu_num_threads", type=int, default=10, help="Number of cpu threads") parser.add_argument("--precision", type=str, default="paddle", help="mode of running(fp32/fp16/int8)") - parser.add_argument("--use_trt", type=bool, default=False, help="Whether to use tensorrt") parser.add_argument("--use_l3", type=bool, default=False, help="Whether use L3_cache or not.") parser.add_argument("--gpu_mem", type=int, default=8000, help="GPU memory") parser.add_argument( @@ -56,19 +55,11 @@ def argsparser(): ) parser.add_argument("--cpu_threads", type=int, default=1, help="Num of cpu threads.") parser.add_argument("--ir_optim", type=bool, default=True) - parser.add_argument("--use_dynamic_shape", type=bool, default=True, help="Whether use dynamic shape or not.") - parser.add_argument("--calibration_file", type=str, default=None, help="quant onnx model calibration cache file.") parser.add_argument( "--deploy_backend", type=str, default="paddle_inference", - help="deploy backend, it can be: `paddle_inference`, `tensorrt`, `onnxruntime`", - ) - parser.add_argument( - "--input_name", - type=str, - default="x", - help="input name of image classification model, this is only used by nv-trt", + help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) parser.add_argument( "--small_data", @@ -89,18 +80,6 @@ def eval_reader(data_dir, batch_size, crop_size, resize_size): return val_loader -def reader_wrapper(reader, input_field="inputs"): - """ - reader wrapper func - """ - - def gen(): - for batch_id, (image, label) in enumerate(reader): - yield np.array(image).astype(np.float32) - - return gen - - def eval(predictor, FLAGS): """ eval func @@ -126,11 +105,7 @@ def eval(predictor, FLAGS): use_xpu = True if FLAGS.device == "XPU" else False monitor = Monitor(0, use_gpu, 0, use_xpu) - - rerun_flag = True if hasattr(predictor, "rerun_flag") and predictor.rerun_flag else False - # in collect shape mode ,we do not start monitor! - if not rerun_flag: - monitor.start() + monitor.start() for batch_id, (image, label) in enumerate(val_loader): image = np.array(image) # classfication model usually having only one input @@ -169,9 +144,6 @@ def eval(predictor, FLAGS): if batch_id % 100 == 0: print("Eval iter:", batch_id) sys.stdout.flush() - if rerun_flag: - return - monitor.stop() monitor_result = monitor.output() @@ -192,12 +164,8 @@ def eval(predictor, FLAGS): result = np.mean(np.array(results), axis=0) fp_message = FLAGS.precision print_msg = "Paddle-Inference-GPU" - if FLAGS.use_trt and FLAGS.deploy_backend == "paddle_inference": - print_msg = "using Paddle-TensorRT" - elif FLAGS.use_mkldnn: + if FLAGS.use_mkldnn: print_msg = "using Paddle-MKLDNN" - elif FLAGS.deploy_backend == "tensorrt": - print_msg = "using NV-TensorRT" time_avg = predict_time / sample_nums print( "[Benchmark]{}\t{}\tbatch size: {}.Inference time(ms): min={}, max={}, avg={}".format( @@ -279,56 +247,20 @@ def main(FLAGS): model_filename=FLAGS.model_filename, params_filename=FLAGS.params_filename, precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_l3=FLAGS.use_l3, use_mkldnn=FLAGS.use_mkldnn, - batch_size=FLAGS.batch_size, device=FLAGS.device, - min_subgraph_size=3, - use_dynamic_shape=FLAGS.use_dynamic_shape, cpu_threads=FLAGS.cpu_threads, ) - elif FLAGS.deploy_backend == "tensorrt": - from backend.tensorrt import TensorRTEngine - - model_name = os.path.split(FLAGS.model_path)[-1].rstrip(".onnx") - engine_file = "{}_{}_model.trt".format(model_name, FLAGS.precision) - print(engine_file) - predictor = TensorRTEngine( - onnx_model_file=FLAGS.model_path, - shape_info={FLAGS.input_name: [[1, 3, 224, 224], [1, 3, 224, 224], [1, 3, 224, 224]]}, - max_batch_size=FLAGS.batch_size, - precision=FLAGS.precision, - engine_file_path=engine_file, - calibration_cache_file=FLAGS.calibration_file, - calibration_loader=reader_wrapper( - eval_reader( - FLAGS.data_path, - batch_size=FLAGS.batch_size, - crop_size=FLAGS.img_size, - resize_size=FLAGS.resize_size, - ) - ), - verbose=False, - ) elif FLAGS.deploy_backend == "onnxruntime": from backend.onnxruntime import ONNXRuntimeEngine - model_name = os.path.split(FLAGS.model_path)[-1].rstrip(".onnx") - engine_file = "{}_{}_model.trt".format(model_name, FLAGS.precision) - print(engine_file) predictor = ONNXRuntimeEngine( onnx_model_file=FLAGS.model_path, - precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_mkldnn=FLAGS.use_mkldnn, device=FLAGS.device, ) eval(predictor, FLAGS) - rerun_flag = True if hasattr(predictor, "rerun_flag") and predictor.rerun_flag else False - if rerun_flag: - print("***** Collect dynamic shape done, Please rerun the program to get correct results. *****") - return if __name__ == "__main__": diff --git a/inference/python_api_test/test_int8_model/test_nlp_infer.py b/inference/python_api_test/test_int8_model/test_nlp_infer.py index b231dcc60c..1a339af846 100644 --- a/inference/python_api_test/test_int8_model/test_nlp_infer.py +++ b/inference/python_api_test/test_int8_model/test_nlp_infer.py @@ -94,11 +94,6 @@ def argsparser(): type=int, help="Warmup steps for performance test.", ) - parser.add_argument( - "--use_trt", - action="store_true", - help="Whether to use inference engin TensorRT.", - ) parser.add_argument("--use_l3", type=bool, default=False, help="Whether use L3_cache or not.") parser.add_argument( "--precision", @@ -113,16 +108,14 @@ def argsparser(): default="GPU", help="Choose the device you want to run, it can be: CPU/GPU/XPU, default is GPU", ) - parser.add_argument("--use_dynamic_shape", type=bool, default=True, help="Whether use dynamic shape or not.") parser.add_argument("--use_mkldnn", type=bool, default=False, help="Whether use mkldnn or not.") parser.add_argument("--cpu_threads", type=int, default=1, help="Num of cpu threads.") parser.add_argument( "--deploy_backend", type=str, default="paddle_inference", - help="deploy backend, it can be: `paddle_inference`, `tensorrt`, `onnxruntime`", + help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) - parser.add_argument("--calibration_file", type=str, default=None, help="quant onnx model calibration cache file.") parser.add_argument("--model_name", type=str, default="", help="model_name for benchmark") return parser @@ -339,44 +332,17 @@ def main(FLAGS): model_filename=FLAGS.model_filename, params_filename=FLAGS.params_filename, precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_l3=FLAGS.use_l3, use_mkldnn=FLAGS.use_mkldnn, - batch_size=FLAGS.batch_size, device=FLAGS.device, - min_subgraph_size=3, - use_dynamic_shape=FLAGS.use_dynamic_shape, cpu_threads=FLAGS.cpu_threads, ) - elif FLAGS.deploy_backend == "tensorrt": - from backend.tensorrt import TensorRTEngine - - model_name = os.path.split(FLAGS.model_path)[-1].rstrip(".onnx") - token_dir = os.path.dirname(FLAGS.model_path) - engine_file = "{}_{}_model.trt".format(model_name, FLAGS.precision) - print(engine_file) - predictor = TensorRTEngine( - onnx_model_file=FLAGS.model_path, - shape_info={ - "input_ids": [[28, 37], [32, 51], [32, 128]], - "token_type_ids": [[28, 37], [32, 51], [32, 128]], - }, - max_batch_size=FLAGS.batch_size, - precision=FLAGS.precision, - engine_file_path=engine_file, - calibration_cache_file=FLAGS.calibration_file, - verbose=False, - ) elif FLAGS.deploy_backend == "onnxruntime": from backend.onnxruntime import ONNXRuntimeEngine - model_name = os.path.split(FLAGS.model_path)[-1].rstrip(".onnx") token_dir = os.path.dirname(FLAGS.model_path) - engine_file = "{}_{}_model.trt".format(model_name, FLAGS.precision) predictor = ONNXRuntimeEngine( onnx_model_file=FLAGS.model_path, - precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_mkldnn=FLAGS.use_mkldnn, device=FLAGS.device, ) @@ -389,13 +355,9 @@ def main(FLAGS): ): fn(samples) WrapperPredictor(predictor).eval(dev_ds, tokenizer, batchify_fn, FLAGS) - rerun_flag = True if hasattr(predictor, "rerun_flag") and predictor.rerun_flag else False - if rerun_flag: - print("***** Collect dynamic shape done, Please rerun the program to get correct results. *****") if __name__ == "__main__": - # If the device is not set to cpu, the nv-trt will report an error when executing paddle.set_device("cpu") parser = argsparser() FLAGS = parser.parse_args() diff --git a/inference/python_api_test/test_int8_model/test_ocr_infer.py b/inference/python_api_test/test_int8_model/test_ocr_infer.py index d5c824aa8f..e35bb9ddd5 100644 --- a/inference/python_api_test/test_int8_model/test_ocr_infer.py +++ b/inference/python_api_test/test_int8_model/test_ocr_infer.py @@ -25,7 +25,7 @@ import paddle from paddle.inference import create_predictor, PrecisionType from paddle.inference import Config as PredictConfig -from backend import PaddleInferenceEngine, TensorRTEngine, Monitor +from backend import PaddleInferenceEngine, Monitor from ppocr.data import create_operators, transform, build_dataloader from ppocr.postprocess import build_post_process @@ -112,19 +112,7 @@ def resize_norm_img_svtr(image_file, image_shape=[3, 48, 320]): return resized_image -def reader_wrapper(reader, input_field="image"): - """ - reader wrapper func - """ - - def gen(): - for data in reader: - yield np.array(data[0]).astype(np.float32) - - return gen - - -def predict_image(predictor, rerun_flag=False): +def predict_image(predictor): """ predict image func """ @@ -148,9 +136,6 @@ def predict_image(predictor, rerun_flag=False): for i in range(warmup): predictor.run() - if rerun_flag: - return - monitor = Monitor(0) monitor.start() predict_time = 0.0 @@ -210,7 +195,7 @@ def predict_image(predictor, rerun_flag=False): # eval is not correct -def eval(args, predictor, rerun_flag=False): +def eval(args, predictor): """ eval func """ @@ -224,8 +209,7 @@ def eval(args, predictor, rerun_flag=False): repeats = len(val_loader) monitor = Monitor(0) - if not rerun_flag: - monitor.start() + monitor.start() predict_time = 0.0 time_min = float("inf") time_max = float("-inf") @@ -245,9 +229,6 @@ def eval(args, predictor, rerun_flag=False): time_max = max(time_max, timed) predict_time += timed - if rerun_flag: - return - batch_numpy = [] for item in batch: batch_numpy.append(np.array(item)) @@ -318,64 +299,19 @@ def main(args): """ main func """ - - val_loader = None - if args.image_file: - if args.model_type == "det": - data = preprocess_det(args.image_file, args.det_limit_side_len, args.det_limit_type) - img, shape_list = data - else: - img = resize_norm_img_svtr(args.image_file) - img = np.expand_dims(img, axis=0) - val_loader = [[img]] - else: - # DataLoader need run on cpu - config = load_config(args.dataset_config) - devices = paddle.set_device("cpu") - val_loader = build_dataloader(config, "Eval", devices, logger) - - predictor = None - if args.deploy_backend == "paddle_inference": - predictor = PaddleInferenceEngine( - model_dir=args.model_path, - model_filename=args.model_filename, - params_filename=args.params_filename, - precision=args.precision, - use_trt=args.use_trt, - use_mkldnn=args.use_mkldnn, - batch_size=args.batch_size, - device=args.device, - min_subgraph_size=3, - use_dynamic_shape=args.use_dynamic_shape, - cpu_threads=args.cpu_threads, - ) - elif args.deploy_backend == "tensorrt": - model_name = os.path.join(args.model_path, args.model_filename) - print(model_name) - engine_file = "{}_{}.trt".format(args.precision, args.batch_size) - predictor = TensorRTEngine( - onnx_model_file=model_name, - shape_info={ - "x": [[1, 3, 100, 100], [1, 3, 800, 800], [1, 3, 1600, 1600]], - }, - max_batch_size=args.batch_size, - precision=args.precision, - engine_file_path=engine_file, - calibration_cache_file=args.calibration_file, - calibration_loader=reader_wrapper(val_loader), - verbose=False, - ) - if predictor is None: - return - rerun_flag = True if hasattr(predictor, "rerun_flag") and predictor.rerun_flag else False - + predictor = PaddleInferenceEngine( + model_dir=args.model_path, + model_filename=args.model_filename, + params_filename=args.params_filename, + precision=args.precision, + use_mkldnn=args.use_mkldnn, + device=args.device, + cpu_threads=args.cpu_threads, + ) if args.image_file: - predict_image(predictor, rerun_flag) + predict_image(predictor) else: - eval(args, predictor, rerun_flag) - - if rerun_flag: - print("***** Collect dynamic shape done, Please rerun the program to get correct results. *****") + eval(args, predictor) if __name__ == "__main__": @@ -386,7 +322,6 @@ def main(args): parser.add_argument("--image_file", type=str, default=None, help="Image path to be processed.") parser.add_argument("--dataset_config", type=str, default=None, help="path of dataset config.") parser.add_argument("--benchmark", type=bool, default=False, help="Whether to run benchmark or not.") - parser.add_argument("--use_trt", type=bool, default=False, help="Whether to use tensorrt engine or not.") parser.add_argument( "--device", type=str, @@ -401,22 +336,11 @@ def main(args): choices=["fp32", "fp16", "int8"], help="The precision of inference. It can be 'fp32', 'fp16' or 'int8'. Default is 'fp16'.", ) - parser.add_argument( - "--deploy_backend", - type=str, - default="paddle_inference", - choices=["paddle_inference", "tensorrt"], - help="deploy backend, it can be: `paddle`, `tensorrt`, `onnxruntime`", - ) - parser.add_argument("--calibration_file", type=str, default="calibration.cache") - parser.add_argument("--use_dynamic_shape", type=bool, default=True, help="Whether use dynamic shape or not.") parser.add_argument("--batch_size", type=int, default=1, help="Batch size of model input.") parser.add_argument("--use_mkldnn", type=bool, default=False, help="Whether use mkldnn or not.") parser.add_argument("--cpu_threads", type=int, default=1, help="Num of cpu threads.") parser.add_argument("--det_limit_side_len", type=float, default=960) parser.add_argument("--det_limit_type", type=str, default="max") - parser.add_argument("--max_batch_size", type=int, default=10) - parser.add_argument("--min_subgraph_size", type=int, default=15) parser.add_argument("--model_type", type=str, default="det") parser.add_argument("--model_name", type=str, default="", help="model name for benchmark") args = parser.parse_args() diff --git a/inference/python_api_test/test_int8_model/test_ppyoloe_infer.py b/inference/python_api_test/test_int8_model/test_ppyoloe_infer.py index 5671cbbb06..0b49ed3869 100644 --- a/inference/python_api_test/test_int8_model/test_ppyoloe_infer.py +++ b/inference/python_api_test/test_int8_model/test_ppyoloe_infer.py @@ -37,14 +37,13 @@ def argsparser(): parser = argparse.ArgumentParser() parser.add_argument("--model_path", type=str, help="inference model filepath") parser.add_argument("--reader_config", type=str, default=None, help="path of datset and reader config.") - parser.add_argument("--use_trt", type=bool, default=False, help="Whether use TensorRT or not.") parser.add_argument("--use_l3", type=bool, default=False, help="Whether use L3_cache or not.") parser.add_argument("--precision", type=str, default="paddle", help="mode of running(fp32/fp16/int8)") parser.add_argument( "--deploy_backend", type=str, default="paddle_inference", - help="deploy backend, it can be: `paddle_inference`, `tensorrt`, `onnxruntime`", + help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) parser.add_argument( "--device", @@ -52,13 +51,11 @@ def argsparser(): default="GPU", help="Choose the device you want to run, it can be: CPU/GPU/XPU, default is GPU", ) - parser.add_argument("--use_dynamic_shape", type=bool, default=True, help="Whether use dynamic shape or not.") parser.add_argument("--use_mkldnn", type=bool, default=False, help="Whether use mkldnn or not.") parser.add_argument("--cpu_threads", type=int, default=1, help="Num of cpu threads.") parser.add_argument("--img_shape", type=int, default=640, help="input_size") parser.add_argument("--model_name", type=str, default="", help="model_name for benchmark") parser.add_argument("--exclude_nms", action="store_true", default=False, help="Whether exclude nms or not.") - parser.add_argument("--calibration_file", type=str, default=None, help="quant onnx model calibration cache file.") parser.add_argument("--small_data", action="store_true", default=False, help="Whether use small data to eval.") return parser @@ -90,7 +87,7 @@ def get_current_memory_mb(): return round(cpu_mem, 4), round(gpu_mem, 4) -def eval(predictor, val_loader, metric, rerun_flag=False): +def eval(predictor, val_loader, metric): """ eval main func """ @@ -105,9 +102,7 @@ def eval(predictor, val_loader, metric, rerun_flag=False): use_xpu = True if FLAGS.device == "XPU" else False monitor = Monitor(0, use_gpu, 0, use_xpu) - - if not rerun_flag: - monitor.start() + monitor.start() for batch_id, data in enumerate(val_loader): data_all = {k: np.array(v) for k, v in data.items()} if FLAGS.exclude_nms: @@ -128,8 +123,6 @@ def eval(predictor, val_loader, metric, rerun_flag=False): time_min = min(time_min, timed) time_max = max(time_max, timed) predict_time += timed - if rerun_flag: - return if FLAGS.exclude_nms and "PPYOLOE" in FLAGS.model_name: postprocess = PPYOLOEPostProcess(score_threshold=0.01, nms_threshold=0.6) res = postprocess(outs[0], data_all["scale_factor"]) @@ -248,49 +241,19 @@ def main(): predictor = PaddleInferenceEngine( model_dir=FLAGS.model_path, precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_l3=FLAGS.use_l3, use_mkldnn=FLAGS.use_mkldnn, - batch_size=FLAGS.batch_size, device=FLAGS.device, - min_subgraph_size=3, - use_dynamic_shape=FLAGS.use_dynamic_shape, cpu_threads=FLAGS.cpu_threads, ) - elif FLAGS.deploy_backend == "tensorrt": - from backend.tensorrt import TensorRTEngine - - model_name = os.path.split(FLAGS.model_path)[-1].rstrip(".onnx") - engine_file = "{}_{}_model.trt".format(model_name, FLAGS.precision) - predictor = TensorRTEngine( - onnx_model_file=FLAGS.model_path, - max_batch_size=FLAGS.batch_size, - precision=FLAGS.precision, - engine_file_path=engine_file, - shape_info={ - "image": [ - [1, 3, FLAGS.img_shape, FLAGS.img_shape], - [1, 3, FLAGS.img_shape, FLAGS.img_shape], - [1, 3, FLAGS.img_shape, FLAGS.img_shape], - ], - }, - calibration_cache_file=FLAGS.calibration_file, - verbose=False, - ) elif FLAGS.deploy_backend == "onnxruntime": from backend.onnxruntime import ONNXRuntimeEngine predictor = ONNXRuntimeEngine( onnx_model_file=FLAGS.model_path, - precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_mkldnn=FLAGS.use_mkldnn, device=FLAGS.device, ) - else: - raise ValueError("deploy_backend not support {}".format(FLAGS.deploy_backend)) - - rerun_flag = True if hasattr(predictor, "rerun_flag") and predictor.rerun_flag else False if FLAGS.small_data: dataset = reader_cfg["TestDataset"] @@ -301,10 +264,7 @@ def main(): clsid2catid = {v: k for k, v in dataset.catid2clsid.items()} anno_file = dataset.get_anno() metric = COCOMetric(anno_file=anno_file, clsid2catid=clsid2catid, IouType="bbox") - eval(predictor, val_loader, metric, rerun_flag=rerun_flag) - - if rerun_flag: - print("***** Collect dynamic shape done, Please rerun the program to get correct results. *****") + eval(predictor, val_loader, metric) if __name__ == "__main__": diff --git a/inference/python_api_test/test_int8_model/test_segmentation_infer.py b/inference/python_api_test/test_int8_model/test_segmentation_infer.py index 2a926bbda1..9d7486f24d 100644 --- a/inference/python_api_test/test_int8_model/test_segmentation_infer.py +++ b/inference/python_api_test/test_int8_model/test_segmentation_infer.py @@ -16,7 +16,6 @@ import argparse import time -import os import sys import cv2 import numpy as np @@ -48,11 +47,10 @@ def argsparser(): "--deploy_backend", type=str, default="paddle_inference", - help="deploy backend, it can be: `paddle_inference`, `tensorrt`, `onnxruntime`", + help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) parser.add_argument("--dataset_config", type=str, default=None, help="path of dataset config.") parser.add_argument("--benchmark", type=bool, default=False, help="Whether to run benchmark or not.") - parser.add_argument("--use_trt", type=bool, default=False, help="Whether to use tensorrt engine or not.") parser.add_argument("--use_l3", type=bool, default=False, help="Whether use L3_cache or not.") parser.add_argument( "--device", @@ -70,13 +68,12 @@ def argsparser(): ) parser.add_argument("--use_mkldnn", type=bool, default=False, help="Whether use mkldnn or not.") parser.add_argument("--cpu_threads", type=int, default=1, help="Num of cpu threads.") - parser.add_argument("--calibration_file", type=str, default=None, help="quant onnx model calibration cache file.") parser.add_argument("--model_name", type=str, default="", help="model_name for benchmark") parser.add_argument("--small_data", action="store_true", default=False, help="Whether use small data to eval.") return parser -def eval(predictor, loader, eval_dataset, rerun_flag): +def eval(predictor, loader, eval_dataset): """ eval mIoU func """ @@ -93,9 +90,7 @@ def eval(predictor, loader, eval_dataset, rerun_flag): use_xpu = True if FLAGS.device == "XPU" else False monitor = Monitor(0, use_gpu, 0, use_xpu) - - if not rerun_flag: - monitor.start() + monitor.start() print("Start evaluating (total_samples: {}, total_iters: {}).".format(FLAGS.total_samples, FLAGS.sample_nums)) for batch_id, data in enumerate(loader): @@ -120,9 +115,6 @@ def eval(predictor, loader, eval_dataset, rerun_flag): time_min = min(time_min, timed) time_max = max(time_max, timed) predict_time += timed - if rerun_flag: - return - logit = reverse_transform(paddle.to_tensor(outs[0]), trans_info, mode="bilinear") pred = paddle.to_tensor(logit) if len(pred.shape) == 4: # for humanseg model whose prediction is distribution but not class id @@ -250,48 +242,21 @@ def main(): model_filename=FLAGS.model_filename, params_filename=FLAGS.params_filename, precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_l3=FLAGS.use_l3, use_mkldnn=FLAGS.use_mkldnn, - batch_size=FLAGS.batch_size, device=FLAGS.device, - min_subgraph_size=3, - use_dynamic_shape=True, cpu_threads=FLAGS.cpu_threads, ) - elif FLAGS.deploy_backend == "tensorrt": - from backend.tensorrt import TensorRTEngine - - model_name = os.path.split(FLAGS.model_path)[-1].rstrip(".onnx") - engine_file = "{}_{}_model.trt".format(model_name, FLAGS.precision) - predictor = TensorRTEngine( - onnx_model_file=FLAGS.model_path, - shape_info=None, - max_batch_size=FLAGS.batch_size, - precision=FLAGS.precision, - engine_file_path=engine_file, - calibration_cache_file=FLAGS.calibration_file, - verbose=False, - ) elif FLAGS.deploy_backend == "onnxruntime": from backend.onnxruntime import ONNXRuntimeEngine predictor = ONNXRuntimeEngine( onnx_model_file=FLAGS.model_path, - precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_mkldnn=FLAGS.use_mkldnn, device=FLAGS.device, ) - else: - raise ValueError("deploy_backend not support {}".format(FLAGS.deploy_backend)) - - rerun_flag = True if hasattr(predictor, "rerun_flag") and predictor.rerun_flag else False - - eval(predictor, eval_loader, eval_dataset, rerun_flag) - if rerun_flag: - print("***** Collect dynamic shape done, Please rerun the program to get correct results. *****") + eval(predictor, eval_loader, eval_dataset) if __name__ == "__main__": diff --git a/inference/python_api_test/test_int8_model/test_yolo_series_infer.py b/inference/python_api_test/test_int8_model/test_yolo_series_infer.py index 9c81035412..c3dbcb6247 100644 --- a/inference/python_api_test/test_int8_model/test_yolo_series_infer.py +++ b/inference/python_api_test/test_int8_model/test_yolo_series_infer.py @@ -44,10 +44,8 @@ def argsparser(): "--deploy_backend", type=str, default="paddle_inference", - help="deploy backend, it can be: `paddle_inference`, `tensorrt`, `onnxruntime`", + help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) - parser.add_argument("--use_dynamic_shape", type=bool, default=True, help="Whether use dynamic shape or not.") - parser.add_argument("--use_trt", type=bool, default=False, help="Whether use TensorRT or not.") parser.add_argument("--use_l3", type=bool, default=False, help="Whether use L3_cache or not.") parser.add_argument("--precision", type=str, default="paddle", help="mode of running(fp32/fp16/int8)") parser.add_argument( @@ -59,7 +57,6 @@ def argsparser(): parser.add_argument("--batch_size", type=int, default=1, help="Batch size of model input.") parser.add_argument("--use_mkldnn", type=bool, default=False, help="Whether use mkldnn or not.") parser.add_argument("--cpu_threads", type=int, default=1, help="Num of cpu threads.") - parser.add_argument("--calibration_file", type=str, default=None, help="quant onnx model calibration cache file.") parser.add_argument("--model_name", type=str, default="", help="model name for benchmark") parser.add_argument("--small_data", action="store_true", default=False, help="Whether use small data to eval.") return parser @@ -92,19 +89,7 @@ def get_current_memory_mb(): return round(cpu_mem, 4), round(gpu_mem, 4) -def reader_wrapper(reader, input_field="image"): - """ - reader wrapper func - """ - - def gen(): - for data in reader: - yield np.array(data[input_field]).astype(np.float32) - - return gen - - -def eval(predictor, val_loader, anno_file, rerun_flag=False): +def eval(predictor, val_loader, anno_file): """ eval main func """ @@ -121,9 +106,7 @@ def eval(predictor, val_loader, anno_file, rerun_flag=False): use_xpu = True if FLAGS.device == "XPU" else False monitor = Monitor(0, use_gpu, 0, use_xpu) - - if not rerun_flag: - monitor.start() + monitor.start() for batch_id, data in enumerate(val_loader): data_all = {k: np.array(v) for k, v in data.items()} @@ -142,8 +125,6 @@ def eval(predictor, val_loader, anno_file, rerun_flag=False): time_min = min(time_min, timed) time_max = max(time_max, timed) predict_time += timed - if rerun_flag: - return postprocess = YOLOPostProcess(score_threshold=0.001, nms_threshold=0.65, multi_label=True) res = postprocess(np.array(outs), data_all["scale_factor"]) bboxes_list.append(res["bbox"]) @@ -244,49 +225,21 @@ def main(): predictor = PaddleInferenceEngine( model_dir=FLAGS.model_path, precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_l3=FLAGS.use_l3, use_mkldnn=FLAGS.use_mkldnn, - batch_size=FLAGS.batch_size, device=FLAGS.device, - min_subgraph_size=3, - use_dynamic_shape=FLAGS.use_dynamic_shape, cpu_threads=FLAGS.cpu_threads, ) - elif FLAGS.deploy_backend == "tensorrt": - from backend.tensorrt import TensorRTEngine - - model_name = os.path.split(FLAGS.model_path)[-1].rstrip(".onnx") - engine_file = "{}_{}_model.trt".format(model_name, FLAGS.precision) - predictor = TensorRTEngine( - onnx_model_file=FLAGS.model_path, - shape_info=None, - max_batch_size=FLAGS.batch_size, - precision=FLAGS.precision, - engine_file_path=engine_file, - calibration_cache_file=FLAGS.calibration_file, - calibration_loader=reader_wrapper(val_loader), - verbose=False, - ) elif FLAGS.deploy_backend == "onnxruntime": from backend.onnxruntime import ONNXRuntimeEngine predictor = ONNXRuntimeEngine( onnx_model_file=FLAGS.model_path, - precision=FLAGS.precision, - use_trt=FLAGS.use_trt, use_mkldnn=FLAGS.use_mkldnn, device=FLAGS.device, ) - else: - raise ValueError("deploy_backend not support {}".format(FLAGS.deploy_backend)) - - rerun_flag = True if hasattr(predictor, "rerun_flag") and predictor.rerun_flag else False - - eval(predictor, val_loader, anno_file, rerun_flag=rerun_flag) - if rerun_flag: - print("***** Collect dynamic shape done, Please rerun the program to get correct results. *****") + eval(predictor, val_loader, anno_file) if __name__ == "__main__": diff --git a/inference/python_api_test/test_int8_model/xly.sh b/inference/python_api_test/test_int8_model/xly.sh index ca29ec73e6..c89dcd4865 100644 --- a/inference/python_api_test/test_int8_model/xly.sh +++ b/inference/python_api_test/test_int8_model/xly.sh @@ -3,16 +3,13 @@ set -ex DOCKER_NAME="test_infer_slim" -# DOCKER_IMAGE="paddlepaddle/paddle_manylinux_devel:cuda11.1-cudnn8.1-gcc82-trt7" -# PADDLE_WHL="https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release_GpuAll_LinuxCentos_Gcc82_Cuda11.1_cudnn8.1.1_trt8406_Py38_Compile_H/latest/paddlepaddle_gpu-0.0.0-cp38-cp38-linux_x86_64.whl" DOCKER_IMAGE=${DOCKER_IMAGE:-registry.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda11.2-cudnn8.1-trt8.0-gcc8.2} -#PADDLE_WHL=${PADDLE_WHL:-https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release-GpuAll-Centos-Gcc82-Cuda112-Cudnn82-Trt8034-Py38-Compile/latest/paddlepaddle_gpu-0.0.0-cp38-cp38-linux_x86_64.whl} -PADDLE_WHL=${PADDLE_WHL:-https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-GpuAll-LinuxCentos-Gcc82-Cuda112-Trton-Py38-Compile/latest/paddlepaddle_gpu-0.0.0-cp38-cp38-linux_x86_64.whl} +PADDLE_WHL=${PADDLE_WHL:-https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-GpuAll-LinuxCentos-Gcc82-Cuda112-Trtoff-Py38-Compile/latest/paddlepaddle_gpu-0.0.0-cp38-cp38-linux_x86_64.whl} FRAME=${FRAME:-paddle} FRAME_BRANCH=${FRAME_BRANCH:-develop} FRAME_VERSION=${FRAME_VERSION:-0.0.0} DEVICE=${DEVICE:-T4} -MODE=${MODE:-trt_int8,trt_fp16,mkldnn_int8,mkldnn_fp32} +MODE=${MODE:-mkldnn_int8,mkldnn_fp32} METRIC=${METRIC:-jingdu,xingneng,cpu_mem,gpu_mem} export CUDA_SO="$(\ls -d /usr/lib64/libcuda* | xargs -I{} echo '-v {}:{}') $(\ls -d /usr/lib64/libnvidia* | xargs -I{} echo '-v {}:{}')" @@ -52,11 +49,6 @@ export LD_LIBRARY_PATH=/opt/_internal/cpython-3.8.0/lib/:${LD_LIBRARY_PATH} export PATH=/opt/_internal/cpython-3.8.0/bin/:${PATH} export PYTHON_FLAGS="-DPYTHON_EXECUTABLE:FILEPATH=/opt/_internal/cpython-3.8.0/bin/python3.8 -DPYTHON_INCLUDE_DIR:PATH=/opt/_internal/cpython-3.8.0/include/python3.8 -DPYTHON_LIBRARIES:FILEPATH=/opt/_internal/cpython-3.8.0/lib/libpython3.so" -wget https://paddle-qa.bj.bcebos.com/tools/TensorRT-8.4.0.6.tgz -tar -zxf TensorRT-8.4.0.6.tgz -export LD_LIBRARY_PATH=${PWD}/TensorRT-8.4.0.6/lib/:${LD_LIBRARY_PATH} - - python -m pip install --retries 50 --upgrade pip -i https://mirror.baidu.com/pypi/simple python -m pip config set global.index-url https://mirror.baidu.com/pypi/simple; @@ -66,7 +58,6 @@ pip install paddledet\>=2.4.0 pip install paddleseg==2.5.0 pip install paddlenlp\>=2.3.0 pip install opencv-python -pip install pycuda pip install onnx pip install GPUtil pip install psutil @@ -77,9 +68,6 @@ pip install onnxruntime pip install -U ${PADDLE_WHL} -pip install nvidia-pyindex -pip install nvidia-cublas-cu11 -pip install nvidia-tensorrt pip install openpyxl pip install pymysql pip install bce-python-sdk @@ -91,7 +79,7 @@ SAVE_FILE=${DT}_${FRAME}_${FRAME_BRANCH/\//-}_${PADDLE_COMMIT}_${DEVICE}.xlsx PYTHON_VERSION=${PYTHON_VERSION:-3.8} CUDA_VERSION=${CUDA_VERSION:-11.2} CUDNN_VERSION=${CUDNN_VERSION:-8.2} -TRT_VERSION=${TRT_VERSION:-8} +TRT_VERSION=${TRT_VERSION:--} GPU=${DEVICE} CPU="-" diff --git a/inference/python_api_test/test_nlp_model/run.sh b/inference/python_api_test/test_nlp_model/run.sh index 2e42003ea9..fcc716b4f1 100644 --- a/inference/python_api_test/test_nlp_model/run.sh +++ b/inference/python_api_test/test_nlp_model/run.sh @@ -2,13 +2,9 @@ export FLAGS_call_stack_level=2 cases="./test_bert_gpu.py \ ./test_bert_mkldnn.py \ - ./test_bert_trt_fp32.py \ ./test_ernie_gpu.py \ ./test_ernie_mkldnn.py \ - ./test_ernie_trt_fp32.py \ ./test_lac_gpu.py \ - ./test_lac_trt_fp32.py \ - ./test_lac_trt_fp16.py \ " bug=0 diff --git a/inference/python_api_test/test_nlp_model/test_AFQMC_PTQ_trt_int8.py b/inference/python_api_test/test_nlp_model/test_AFQMC_PTQ_trt_int8.py deleted file mode 100644 index ef1a7b067c..0000000000 --- a/inference/python_api_test/test_nlp_model/test_AFQMC_PTQ_trt_int8.py +++ /dev/null @@ -1,542 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test AFQMC_PTQ model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest, clip_model_extra_op - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - model_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.2.2/nlp/AFQMC_PTQ_1.tgz" - if not os.path.exists("./AFQMC_PTQ_1/strategy_1/__model__"): - wget.download(model_url, out="./") - tar = tarfile.open("AFQMC_PTQ_1.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config(model_path="./AFQMC_PTQ_1/strategy_1/") - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_int8 -def test_trt_int8_bz1(): - """ - compared trt int8 batch_size=1 bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config(model_path="./AFQMC_PTQ_1/strategy_1/") - - src_ids = ( - np.array( - [ - 1, - 654, - 21, - 778, - 291, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 17963, - 9, - 358, - 567, - 1179, - 17963, - 532, - 537, - 358, - 386, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - 691, - 736, - 1431, - 1137, - 1279, - 779, - 12049, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 12043, - 2, - 459, - 335, - 263, - 65, - 129, - 37, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 42, - 42, - 42, - 2, - 51, - 23, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - input_ids = np.tile(src_ids, (40, 1)) - sent_ids = ( - np.array( - [ - 0, - 0, - 0, - 0, - 0, - 0, - 0, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - token_type_ids = np.tile(sent_ids, (40, 1)) - - output_data_path = "./AFQMC_PTQ_1/trt_int8" - output_data_dict = test_suite.get_output_data(output_data_path) - - input_data_dict = {"token_type_ids": token_type_ids, "input_ids": input_ids} - test_suite.collect_shape_info(model_path="./AFQMC_PTQ_1/strategy_1/", input_data_dict=input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_path="./AFQMC_PTQ_1/strategy_1/") - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - delta=4e-1, - max_batch_size=40, - use_static=False, - precision="trt_int8", - dynamic=True, - shape_range_file="./AFQMC_PTQ_1/strategy_1/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_int8_multi_thread -def test_trt_int8_bz1_multi_thread(): - """ - compared trt int8 batch_size=1 multi_thread bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config(model_path="./AFQMC_PTQ_1/strategy_1/") - - src_ids = ( - np.array( - [ - 1, - 654, - 21, - 778, - 291, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 17963, - 9, - 358, - 567, - 1179, - 17963, - 532, - 537, - 358, - 386, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - 691, - 736, - 1431, - 1137, - 1279, - 779, - 12049, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 12043, - 2, - 459, - 335, - 263, - 65, - 129, - 37, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 42, - 42, - 42, - 2, - 51, - 23, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - input_ids = np.tile(src_ids, (40, 1)) - sent_ids = ( - np.array( - [ - 0, - 0, - 0, - 0, - 0, - 0, - 0, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - token_type_ids = np.tile(sent_ids, (40, 1)) - - output_data_path = "./AFQMC_PTQ_1/trt_int8" - output_data_dict = test_suite.get_output_data(output_data_path) - - input_data_dict = {"token_type_ids": token_type_ids, "input_ids": input_ids} - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_path="./AFQMC_PTQ_1/strategy_1/") - set_dynamic_shape(test_suite2.pd_config) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - delta=1e-1, - use_static=False, - precision="trt_int8", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_nlp_model/test_AFQMC_base_trt_fp16.py b/inference/python_api_test/test_nlp_model/test_AFQMC_base_trt_fp16.py deleted file mode 100644 index 469bc9e6bd..0000000000 --- a/inference/python_api_test/test_nlp_model/test_AFQMC_base_trt_fp16.py +++ /dev/null @@ -1,574 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test AFQMC_base model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - model_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.2.2/nlp/AFQMC_base.tgz" - if not os.path.exists("./AFQMC_base/inference.pdmodel"): - wget.download(model_url, out="./") - tar = tarfile.open("AFQMC_base.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_bz1(): - """ - compared trt fp16 batch_size = [1] AFQMC_base outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - - src_ids = ( - np.array( - [ - 1, - 654, - 21, - 778, - 291, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 17963, - 9, - 358, - 567, - 1179, - 17963, - 532, - 537, - 358, - 386, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - 691, - 736, - 1431, - 1137, - 1279, - 779, - 12049, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 12043, - 2, - 459, - 335, - 263, - 65, - 129, - 37, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 42, - 42, - 42, - 2, - 51, - 23, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - input_ids = np.tile(src_ids, (40, 1)) - sent_ids = ( - np.array( - [ - 0, - 0, - 0, - 0, - 0, - 0, - 0, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - token_type_ids = np.tile(sent_ids, (40, 1)) - - input_data_dict = {"token_type_ids": token_type_ids, "input_ids": input_ids} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite.pd_config - test_suite.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./AFQMC_base/", - input_data_dict=input_data_dict, - device="gpu", - ) - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - test_suite2.pd_config.exp_disable_tensorrt_ops(["elementwise_sub"]) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - delta=1e-2, - max_batch_size=40, - use_static=False, - precision="trt_fp16", - dynamic=True, - shape_range_file="./AFQMC_base/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 multi_thread bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - - src_ids = ( - np.array( - [ - 1, - 654, - 21, - 778, - 291, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 17963, - 9, - 358, - 567, - 1179, - 17963, - 532, - 537, - 358, - 386, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - 691, - 736, - 1431, - 1137, - 1279, - 779, - 12049, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 12043, - 2, - 459, - 335, - 263, - 65, - 129, - 37, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 42, - 42, - 42, - 2, - 51, - 23, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - input_ids = np.tile(src_ids, (40, 1)) - sent_ids = ( - np.array( - [ - 0, - 0, - 0, - 0, - 0, - 0, - 0, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - token_type_ids = np.tile(sent_ids, (40, 1)) - - input_data_dict = {"token_type_ids": token_type_ids, "input_ids": input_ids} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite.pd_config - test_suite.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./AFQMC_base/", - input_data_dict=input_data_dict, - device="gpu", - ) - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - test_suite2.pd_config.exp_disable_tensorrt_ops(["elementwise_sub"]) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - delta=1e-2, - use_static=False, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_nlp_model/test_AFQMC_base_trt_fp32.py b/inference/python_api_test/test_nlp_model/test_AFQMC_base_trt_fp32.py deleted file mode 100644 index 1b6e89998f..0000000000 --- a/inference/python_api_test/test_nlp_model/test_AFQMC_base_trt_fp32.py +++ /dev/null @@ -1,574 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test AFQMC_base model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - model_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.2.2/nlp/AFQMC_base.tgz" - if not os.path.exists("./AFQMC_base/inference.pdmodel"): - wget.download(model_url, out="./") - tar = tarfile.open("AFQMC_base.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_bz1(): - """ - compared trt fp32 batch_size = [1] AFQMC_base outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - - src_ids = ( - np.array( - [ - 1, - 654, - 21, - 778, - 291, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 17963, - 9, - 358, - 567, - 1179, - 17963, - 532, - 537, - 358, - 386, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - 691, - 736, - 1431, - 1137, - 1279, - 779, - 12049, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 12043, - 2, - 459, - 335, - 263, - 65, - 129, - 37, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 42, - 42, - 42, - 2, - 51, - 23, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - input_ids = np.tile(src_ids, (40, 1)) - sent_ids = ( - np.array( - [ - 0, - 0, - 0, - 0, - 0, - 0, - 0, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - token_type_ids = np.tile(sent_ids, (40, 1)) - - input_data_dict = {"token_type_ids": token_type_ids, "input_ids": input_ids} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite.pd_config - test_suite.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./AFQMC_base/", - input_data_dict=input_data_dict, - device="gpu", - ) - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - test_suite2.pd_config.exp_disable_tensorrt_ops(["elementwise_sub"]) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - delta=1e-5, - max_batch_size=40, - use_static=False, - precision="trt_fp32", - dynamic=True, - shape_range_file="./AFQMC_base/shape_range.pbtxt", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 multi_thread bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - - src_ids = ( - np.array( - [ - 1, - 654, - 21, - 778, - 291, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 17963, - 9, - 358, - 567, - 1179, - 17963, - 532, - 537, - 358, - 386, - 21, - 2, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - 691, - 736, - 1431, - 1137, - 1279, - 779, - 12049, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 12043, - 2, - 459, - 335, - 263, - 65, - 129, - 37, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 42, - 42, - 42, - 2, - 51, - 23, - 654, - 8, - 778, - 326, - 778, - 45, - 291, - 751, - 85, - 155, - 76, - 2, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - input_ids = np.tile(src_ids, (40, 1)) - sent_ids = ( - np.array( - [ - 0, - 0, - 0, - 0, - 0, - 0, - 0, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - 1, - ] - + [1] * 31 - ) - .astype(np.int64) - .reshape(1, 128) - ) - token_type_ids = np.tile(sent_ids, (40, 1)) - - input_data_dict = {"token_type_ids": token_type_ids, "input_ids": input_ids} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite.pd_config - test_suite.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - test_suite.collect_shape_info( - model_path="./AFQMC_base/", - input_data_dict=input_data_dict, - device="gpu", - ) - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./AFQMC_base/inference.pdmodel", - params_file="./AFQMC_base/inference.pdiparams", - ) - test_suite2.pd_config.exp_disable_tensorrt_ops(["elementwise_sub"]) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - min_subgraph_size=1, - delta=1e-5, - use_static=False, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_nlp_model/test_bert_trt_fp16.py b/inference/python_api_test/test_nlp_model/test_bert_trt_fp16.py deleted file mode 100644 index 597795bc76..0000000000 --- a/inference/python_api_test/test_nlp_model/test_bert_trt_fp16.py +++ /dev/null @@ -1,121 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test bert model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - bert_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.1.2/nlp/bert.tgz" - if not os.path.exists("./bert/inference.pdiparams"): - wget.download(bert_url, out="./") - tar = tarfile.open("bert.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./bert/inference.pdmodel", - params_file="./bert/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_trt_fp16_bz1(): - """ - compared trt fp16 batch_size=1 bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./bert/inference.pdmodel", - params_file="./bert/inference.pdiparams", - ) - input_ids = np.load("./bert/input_ids.npy").astype("int64") - token_type_ids = np.load("./bert/token_type_ids.npy").astype("int64") - - input_data_dict = {"input_ids": np.array([input_ids]), "token_type_ids": np.array([token_type_ids])} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./bert/inference.pdmodel", - params_file="./bert/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=3e-3, - max_batch_size=1, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 multi_thread bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./bert/inference.pdmodel", - params_file="./bert/inference.pdiparams", - ) - input_ids = np.load("./bert/input_ids.npy").astype("int64") - token_type_ids = np.load("./bert/token_type_ids.npy").astype("int64") - - input_data_dict = {"input_ids": np.array([input_ids]), "token_type_ids": np.array([token_type_ids])} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./bert/inference.pdmodel", - params_file="./bert/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=1e-5, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_nlp_model/test_bert_trt_fp32.py b/inference/python_api_test/test_nlp_model/test_bert_trt_fp32.py deleted file mode 100644 index 36df918133..0000000000 --- a/inference/python_api_test/test_nlp_model/test_bert_trt_fp32.py +++ /dev/null @@ -1,121 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test bert model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - bert_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.1.2/nlp/bert.tgz" - if not os.path.exists("./bert/inference.pdiparams"): - wget.download(bert_url, out="./") - tar = tarfile.open("bert.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./bert/inference.pdmodel", - params_file="./bert/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_trt_fp32_bz1(): - """ - compared trt fp32 batch_size=1 bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./bert/inference.pdmodel", - params_file="./bert/inference.pdiparams", - ) - input_ids = np.load("./bert/input_ids.npy").astype("int64") - token_type_ids = np.load("./bert/token_type_ids.npy").astype("int64") - - input_data_dict = {"input_ids": np.array([input_ids]), "token_type_ids": np.array([token_type_ids])} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./bert/inference.pdmodel", - params_file="./bert/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.0002, - max_batch_size=1, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 multi_thread bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./bert/inference.pdmodel", - params_file="./bert/inference.pdiparams", - ) - input_ids = np.load("./bert/input_ids.npy").astype("int64") - token_type_ids = np.load("./bert/token_type_ids.npy").astype("int64") - - input_data_dict = {"input_ids": np.array([input_ids]), "token_type_ids": np.array([token_type_ids])} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./bert/inference.pdmodel", - params_file="./bert/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=1e-5, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_nlp_model/test_ernie_trt_fp16.py b/inference/python_api_test/test_nlp_model/test_ernie_trt_fp16.py deleted file mode 100644 index 6875e97843..0000000000 --- a/inference/python_api_test/test_nlp_model/test_ernie_trt_fp16.py +++ /dev/null @@ -1,121 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ernie model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ernie_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.1.2/nlp/ernie.tgz" - if not os.path.exists("./ernie/inference.pdiparams"): - wget.download(ernie_url, out="./") - tar = tarfile.open("ernie.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ernie/inference.pdmodel", - params_file="./ernie/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_trt_fp16_bz1(): - """ - compared trt fp16 batch_size=1 ernie outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ernie/inference.pdmodel", - params_file="./ernie/inference.pdiparams", - ) - input_ids = np.load("./ernie/input_ids.npy").astype("int64") - token_type_ids = np.load("./ernie/token_type_ids.npy").astype("int64") - - input_data_dict = {"input_ids": np.array([input_ids]), "token_type_ids": np.array([token_type_ids])} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ernie/inference.pdmodel", - params_file="./ernie/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=3e-3, - max_batch_size=1, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 multi_thread ernie outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ernie/inference.pdmodel", - params_file="./ernie/inference.pdiparams", - ) - input_ids = np.load("./ernie/input_ids.npy").astype("int64") - token_type_ids = np.load("./ernie/token_type_ids.npy").astype("int64") - - input_data_dict = {"input_ids": np.array([input_ids]), "token_type_ids": np.array([token_type_ids])} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ernie/inference.pdmodel", - params_file="./ernie/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=1e-5, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_nlp_model/test_ernie_trt_fp32.py b/inference/python_api_test/test_nlp_model/test_ernie_trt_fp32.py deleted file mode 100644 index 8d00ab0f5a..0000000000 --- a/inference/python_api_test/test_nlp_model/test_ernie_trt_fp32.py +++ /dev/null @@ -1,121 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ernie model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ernie_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.1.2/nlp/ernie.tgz" - if not os.path.exists("./ernie/inference.pdiparams"): - wget.download(ernie_url, out="./") - tar = tarfile.open("ernie.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ernie/inference.pdmodel", - params_file="./ernie/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_trt_fp32_bz1(): - """ - compared trt fp32 batch_size=1 ernie outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ernie/inference.pdmodel", - params_file="./ernie/inference.pdiparams", - ) - input_ids = np.load("./ernie/input_ids.npy").astype("int64") - token_type_ids = np.load("./ernie/token_type_ids.npy").astype("int64") - - input_data_dict = {"input_ids": np.array([input_ids]), "token_type_ids": np.array([token_type_ids])} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ernie/inference.pdmodel", - params_file="./ernie/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=0.0002, - max_batch_size=1, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 multi_thread ernie outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ernie/inference.pdmodel", - params_file="./ernie/inference.pdiparams", - ) - input_ids = np.load("./ernie/input_ids.npy").astype("int64") - token_type_ids = np.load("./ernie/token_type_ids.npy").astype("int64") - - input_data_dict = {"input_ids": np.array([input_ids]), "token_type_ids": np.array([token_type_ids])} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ernie/inference.pdmodel", - params_file="./ernie/inference.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=1e-5, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_nlp_model/test_ernie_varlen_trt_fp16.py b/inference/python_api_test/test_nlp_model/test_ernie_varlen_trt_fp16.py deleted file mode 100644 index 256340d116..0000000000 --- a/inference/python_api_test/test_nlp_model/test_ernie_varlen_trt_fp16.py +++ /dev/null @@ -1,199 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ernie_model_4 model -""" - -import os -import sys -import logging -import tarfile -import wget -import pytest -import numpy as np -import paddle.inference as paddle_infer - -from paddle.inference import Config -from paddle.inference import create_predictor -from paddle.inference import PrecisionType -from paddle.inference import InternalUtils - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest -from test_case.image_preprocess import sig_fig_compare - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ernie_model_4_url = "https://paddle-qa.bj.bcebos.com/inference_model/unknown/nlp/ernie_model_4.tgz" - if not os.path.exists("./ernie_model_4/__model__"): - wget.download(ernie_model_4_url, out="./") - tar = tarfile.open("ernie_model_4.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config(model_path="./ernie_model_4") - test_suite.config_test() - - -def init_predictor(model_path): - """ - Args: - model_path (str): Path to the TensorRT model - Returns: - Predictor: Returns a TensorRT model predictor object - """ - config = Config(model_path) - config.enable_memory_optim() - config.enable_use_gpu(1000, 0, PrecisionType.Float32) - - config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=5, - precision_mode=PrecisionType.Half, - use_static=False, - use_calib_mode=False, - ) - - min_batch = 1 - max_batch = 10 - min_single_seq_len = 1 - max_single_seq_len = 384 - opt_single_seq_len = 384 - min_batch_seq_len = 1 - max_batch_seq_len = 3840 - opt_batch_seq_len = 3840 - - input_name0 = "read_file_0.tmp_0" - input_name1 = "read_file_0.tmp_1" - input_name2 = "read_file_0.tmp_2" - input_name3 = "read_file_0.tmp_4" - - min_shape = [min_batch_seq_len] - max_shape = [max_batch_seq_len] - opt_shape = [opt_batch_seq_len] - - config.set_trt_dynamic_shape_info( - { - input_name0: min_shape, - input_name1: min_shape, - input_name2: [1], - input_name3: [min_batch, min_single_seq_len, 1], - }, - { - input_name0: max_shape, - input_name1: max_shape, - input_name2: [max_batch + 1], - input_name3: [max_batch, max_single_seq_len, 1], - }, - { - input_name0: opt_shape, - input_name1: opt_shape, - input_name2: [max_batch + 1], - input_name3: [max_batch, opt_single_seq_len, 1], - }, - ) - - config.enable_tensorrt_varseqlen() - InternalUtils.set_transformer_posid(config, input_name2) - InternalUtils.set_transformer_maskid(config, input_name3) - - predictor = create_predictor(config) - return predictor - - -def run(predictor, delta): - """ - Runs model prediction and compares the prediction results with the real values within a tolerance threshold. - Args: - predictor (Predictor): A model predictor object. - delta (float): A tolerance threshold for comparison. - Returns: - None - """ - run_batch = 10 - seq_len = 384 - run_seq_len = run_batch * seq_len - max_seq_len = seq_len - i0 = np.ones(run_seq_len, dtype=np.int64) - i1 = np.zeros(run_seq_len, dtype=np.int64) - i2 = np.array([0, 384, 768, 1152, 1536, 1920, 2304, 2688, 3072, 3456, 3840], dtype=np.int64) - i3 = np.ones([run_batch, max_seq_len, 1], dtype=float) - - input_names = predictor.get_input_names() - - input_tensor0 = predictor.get_input_handle(input_names[0]) - input_tensor0.copy_from_cpu(i0) - - input_tensor1 = predictor.get_input_handle(input_names[1]) - input_tensor1.copy_from_cpu(i1) - - input_tensor2 = predictor.get_input_handle(input_names[2]) - input_tensor2.copy_from_cpu(i2) - - input_tensor3 = predictor.get_input_handle(input_names[3]) - input_tensor3.copy_from_cpu(i3) - - # do the inference - predictor.run() - - # get out data from output tensor - output_names = predictor.get_output_names() - - output_data_dict = {} - for i, name in enumerate(output_names): - output_tensor = predictor.get_output_handle(name) - output_data = output_tensor.copy_to_cpu() - output_data_dict[name] = output_data - - output_data_array = np.array(output_data_dict["save_infer_model/scale_0"]) - - if "win" in sys.platform: - npy_path = ".\\ernie_model_4\\output_data_truth_val.npy" - else: - npy_path = "./ernie_model_4/output_data_truth_val.npy" - - output_data_truth_val = np.load(npy_path, allow_pickle=True) - truth_val_dict = output_data_truth_val.item() - truth_val_array = truth_val_dict["save_infer_model/scale_0"] - diff = sig_fig_compare(output_data_array, truth_val_array, delta) - - -# skip test on trt_ver < 7.2 platform -ver = paddle_infer.get_trt_compile_version() -if ver[0] * 1000 + ver[1] * 100 + ver[2] * 10 < 7200: - trt_skip = pytest.mark.skip(reason="Varlen only support trt_fp16, trt_int8 on trt_ver > 7.2") -else: - trt_skip = pytest.mark.none - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -@trt_skip -def test_trt_fp16(): - """ - compared trt_fp16 ernie_varlen outputs with true val - """ - check_model_exist() - - if "win" in sys.platform: - model_path = ".\\ernie_model_4" - else: - model_path = "./ernie_model_4" - - pred = init_predictor(model_path) - run(pred, delta=1e-3) diff --git a/inference/python_api_test/test_nlp_model/test_lac_trt_fp16.py b/inference/python_api_test/test_nlp_model/test_lac_trt_fp16.py deleted file mode 100644 index bacc05e434..0000000000 --- a/inference/python_api_test/test_nlp_model/test_lac_trt_fp16.py +++ /dev/null @@ -1,84 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test lac model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - lac_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.2.2/nlp/lac.tgz" - if not os.path.exists("./lac/inference.pdiparams"): - wget.download(lac_url, out="./") - tar = tarfile.open("lac.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./lac/inference.pdmodel", - params_file="./lac/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_lac_trt_fp16(): - """ - compared trt_fp16 lac outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./lac/inference.pdmodel", - params_file="./lac/inference.pdiparams", - ) - in1 = np.random.randint(0, 100, (1, 20)).astype(np.int64) - in2 = np.array([20]).astype(np.int64) - input_data_dict = {"token_ids": in1, "length": in2} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./lac/inference.pdmodel", - params_file="./lac/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=1e-5, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - min_subgraph_size=1, - # see comments in test_lac_trt_fp32.py - delete_op_list=["transpose_2.tmp_0_slice_0"], - ) diff --git a/inference/python_api_test/test_nlp_model/test_lac_trt_fp32.py b/inference/python_api_test/test_nlp_model/test_lac_trt_fp32.py deleted file mode 100644 index 2880912f97..0000000000 --- a/inference/python_api_test/test_nlp_model/test_lac_trt_fp32.py +++ /dev/null @@ -1,89 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test lac model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - lac_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.2.2/nlp/lac.tgz" - if not os.path.exists("./lac/inference.pdiparams"): - wget.download(lac_url, out="./") - tar = tarfile.open("lac.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./lac/inference.pdmodel", - params_file="./lac/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_lac_trt_fp32(): - """ - compared trt_fp32 lac outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./lac/inference.pdmodel", - params_file="./lac/inference.pdiparams", - ) - in1 = np.random.randint(0, 100, (1, 20)).astype(np.int64) - in2 = np.array([20]).astype(np.int64) - input_data_dict = {"token_ids": in1, "length": in2} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="cpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./lac/inference.pdmodel", - params_file="./lac/inference.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=1e-5, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - min_subgraph_size=1, - # transpose_2.tmp_0_slice_0 is a slice op's output name, forbid this slice op into paddle-trt - # because it's EndsTensorList is max_0.tmp_0, there is - # another tensorrt_engine who has a input called max_0.tmp_0 too. - # see below comments - delete_op_list=["transpose_2.tmp_0_slice_0"], - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_ocr_model/run.sh b/inference/python_api_test/test_ocr_model/run.sh index a9dffdda40..bd10b2400d 100644 --- a/inference/python_api_test/test_ocr_model/run.sh +++ b/inference/python_api_test/test_ocr_model/run.sh @@ -1,8 +1,7 @@ [[ -n $1 ]] && export CUDA_VISIBLE_DEVICES=$1 export FLAGS_call_stack_level=2 cases="./test_ocr_det_mv3_db_gpu.py \ - ./test_ocr_det_mv3_db_mkldnn.py \ - ./test_ocr_det_mv3_db_trt_fp32.py + ./test_ocr_det_mv3_db_mkldnn.py " bug=0 diff --git a/inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_gpu.py b/inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_gpu.py index f6a1ab27ac..01409b2d2d 100644 --- a/inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_gpu.py +++ b/inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_gpu.py @@ -71,7 +71,7 @@ def test_disable_gpu(): @pytest.mark.gpu_more def test_gpu_more_bz(): """ - compared trt fp32 batch_size=1,2 ocr_det_mv3_db outputs with true val + compared gpu batch_size=1,2 ocr_det_mv3_db outputs with true val """ check_model_exist() @@ -115,7 +115,7 @@ def test_gpu_more_bz(): @pytest.mark.gpu_more def test_gpu_mixed_precision_bz1(): """ - compared trt fp32 batch_size=1 ocr_det_mv3_db outputs with true val + compared gpu batch_size=1 ocr_det_mv3_db outputs with true val """ check_model_exist() @@ -159,7 +159,7 @@ def test_gpu_mixed_precision_bz1(): @pytest.mark.gpu_more def test_jetson_gpu_more_bz(): """ - compared trt fp32 more batch_size ocr_det_mv3_db outputs with true val + compared gpu more batch_size ocr_det_mv3_db outputs with true val """ check_model_exist() diff --git a/inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_trt_fp16.py b/inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_trt_fp16.py deleted file mode 100644 index c2ddbd93be..0000000000 --- a/inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_trt_fp16.py +++ /dev/null @@ -1,357 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ocr_det_mv3_db model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ocr_det_mv3_db_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.1.1/ocr/ocr_det_mv3_db.tgz" - if not os.path.exists("./ocr_det_mv3_db/inference.pdiparams"): - wget.download(ocr_det_mv3_db_url, out="./") - tar = tarfile.open("ocr_det_mv3_db.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-2 ocr_det_mv3_db outputs with true val - """ - check_model_exist() - - file_path = "./ocr_det_mv3_db" - images_size = 640 - batch_size_pool = [1, 2] - max_batch_size = 2 - names = [ - "x", - "conv2d_92.tmp_0", - "conv2d_91.tmp_0", - "conv2d_59.tmp_0", - "nearest_interp_v2_1.tmp_0", - "nearest_interp_v2_2.tmp_0", - "conv2d_124.tmp_0", - "nearest_interp_v2_3.tmp_0", - "nearest_interp_v2_4.tmp_0", - "nearest_interp_v2_5.tmp_0", - "elementwise_add_7", - "nearest_interp_v2_0.tmp_0", - ] - min_input_shape = [ - [1, 3, 50, 50], - [1, 120, 20, 20], - [1, 24, 10, 10], - [1, 96, 20, 20], - [1, 256, 10, 10], - [1, 256, 20, 20], - [1, 256, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 56, 2, 2], - [1, 256, 2, 2], - ] - - max_input_shape = [ - [max_batch_size, 3, 2000, 2000], - [max_batch_size, 120, 400, 400], - [max_batch_size, 24, 200, 200], - [max_batch_size, 96, 400, 400], - [max_batch_size, 256, 200, 200], - [max_batch_size, 256, 400, 400], - [max_batch_size, 256, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 56, 400, 400], - [max_batch_size, 256, 400, 400], - ] - - opt_input_shape = [ - [1, 3, 640, 640], - [1, 120, 160, 160], - [1, 24, 80, 80], - [1, 96, 160, 160], - [1, 256, 80, 80], - [1, 256, 160, 160], - [1, 256, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 56, 40, 40], - [1, 256, 40, 40], - ] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - test_suite2.trt_more_bz_dynamic_test( - input_data_dict, - output_data_dict, - gpu_mem=5000, - max_batch_size=max_batch_size, - repeat=1, - delta=9e-2, - names=names, - min_input_shape=min_input_shape, - max_input_shape=max_input_shape, - opt_input_shape=opt_input_shape, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 ocr_det_mv3_db outputs with true val - """ - check_model_exist() - - file_path = "./ocr_det_mv3_db" - images_size = 640 - batch_size_pool = [1] - max_batch_size = 2 - names = [ - "x", - "conv2d_92.tmp_0", - "conv2d_91.tmp_0", - "conv2d_59.tmp_0", - "nearest_interp_v2_1.tmp_0", - "nearest_interp_v2_2.tmp_0", - "conv2d_124.tmp_0", - "nearest_interp_v2_3.tmp_0", - "nearest_interp_v2_4.tmp_0", - "nearest_interp_v2_5.tmp_0", - "elementwise_add_7", - "nearest_interp_v2_0.tmp_0", - ] - min_input_shape = [ - [1, 3, 50, 50], - [1, 120, 20, 20], - [1, 24, 10, 10], - [1, 96, 20, 20], - [1, 256, 10, 10], - [1, 256, 20, 20], - [1, 256, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 56, 2, 2], - [1, 256, 2, 2], - ] - - max_input_shape = [ - [max_batch_size, 3, 2000, 2000], - [max_batch_size, 120, 400, 400], - [max_batch_size, 24, 200, 200], - [max_batch_size, 96, 400, 400], - [max_batch_size, 256, 200, 200], - [max_batch_size, 256, 400, 400], - [max_batch_size, 256, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 56, 400, 400], - [max_batch_size, 256, 400, 400], - ] - - opt_input_shape = [ - [1, 3, 640, 640], - [1, 120, 160, 160], - [1, 24, 80, 80], - [1, 96, 160, 160], - [1, 256, 80, 80], - [1, 256, 160, 160], - [1, 256, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 56, 40, 40], - [1, 256, 40, 40], - ] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - test_suite2.trt_more_bz_dynamic_test( - input_data_dict, - output_data_dict, - gpu_mem=5000, - max_batch_size=max_batch_size, - repeat=1, - delta=9e-2, - names=names, - min_input_shape=min_input_shape, - max_input_shape=max_input_shape, - opt_input_shape=opt_input_shape, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trtfp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 ocr_det_mv3_db multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ocr_det_mv3_db" - images_size = 640 - batch_size = 1 - max_batch_size = 1 - names = [ - "x", - "conv2d_92.tmp_0", - "conv2d_91.tmp_0", - "conv2d_59.tmp_0", - "nearest_interp_v2_1.tmp_0", - "nearest_interp_v2_2.tmp_0", - "conv2d_124.tmp_0", - "nearest_interp_v2_3.tmp_0", - "nearest_interp_v2_4.tmp_0", - "nearest_interp_v2_5.tmp_0", - "elementwise_add_7", - "nearest_interp_v2_0.tmp_0", - ] - min_input_shape = [ - [1, 3, 50, 50], - [1, 120, 20, 20], - [1, 24, 10, 10], - [1, 96, 20, 20], - [1, 256, 10, 10], - [1, 256, 20, 20], - [1, 256, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 56, 2, 2], - [1, 256, 2, 2], - ] - - max_input_shape = [ - [max_batch_size, 3, 2000, 2000], - [max_batch_size, 120, 400, 400], - [max_batch_size, 24, 200, 200], - [max_batch_size, 96, 400, 400], - [max_batch_size, 256, 200, 200], - [max_batch_size, 256, 400, 400], - [max_batch_size, 256, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 56, 400, 400], - [max_batch_size, 256, 400, 400], - ] - - opt_input_shape = [ - [1, 3, 640, 640], - [1, 120, 160, 160], - [1, 24, 80, 80], - [1, 96, 160, 160], - [1, 256, 80, 80], - [1, 256, 160, 160], - [1, 256, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 56, 40, 40], - [1, 256, 40, 40], - ] - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - test_suite2.trt_dynamic_multi_thread_test( - input_data_dict, - output_data_dict, - gpu_mem=5000, - max_batch_size=max_batch_size, - repeat=1, - delta=9e-2, - names=names, - min_input_shape=min_input_shape, - max_input_shape=max_input_shape, - opt_input_shape=opt_input_shape, - precision="trt_fp16", - ) diff --git a/inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_trt_fp32.py b/inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_trt_fp32.py deleted file mode 100644 index 88b0670738..0000000000 --- a/inference/python_api_test/test_ocr_model/test_ocr_det_mv3_db_trt_fp32.py +++ /dev/null @@ -1,359 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test ocr_det_mv3_db model -""" - -import os -import sys -import logging -import tarfile -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - ocr_det_mv3_db_url = "https://paddle-qa.bj.bcebos.com/inference_model_clipped/2.1.1/ocr/ocr_det_mv3_db.tgz" - if not os.path.exists("./ocr_det_mv3_db/inference.pdiparams"): - wget.download(ocr_det_mv3_db_url, out="./") - tar = tarfile.open("ocr_det_mv3_db.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_dynamic_multi_thread(): - """ - compared trt fp32 batch_size=1 ocr_det_mv3_db multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ocr_det_mv3_db" - images_size = 640 - batch_size = 1 - max_batch_size = 1 - names = [ - "x", - "conv2d_92.tmp_0", - "conv2d_91.tmp_0", - "conv2d_59.tmp_0", - "nearest_interp_v2_1.tmp_0", - "nearest_interp_v2_2.tmp_0", - "conv2d_124.tmp_0", - "nearest_interp_v2_3.tmp_0", - "nearest_interp_v2_4.tmp_0", - "nearest_interp_v2_5.tmp_0", - "elementwise_add_7", - "nearest_interp_v2_0.tmp_0", - ] - min_input_shape = [ - [1, 3, 50, 50], - [1, 120, 20, 20], - [1, 24, 10, 10], - [1, 96, 20, 20], - [1, 256, 10, 10], - [1, 256, 20, 20], - [1, 256, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 56, 2, 2], - [1, 256, 2, 2], - ] - - max_input_shape = [ - [max_batch_size, 3, 2000, 2000], - [max_batch_size, 120, 400, 400], - [max_batch_size, 24, 200, 200], - [max_batch_size, 96, 400, 400], - [max_batch_size, 256, 200, 200], - [max_batch_size, 256, 400, 400], - [max_batch_size, 256, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 56, 400, 400], - [max_batch_size, 256, 400, 400], - ] - - opt_input_shape = [ - [1, 3, 640, 640], - [1, 120, 160, 160], - [1, 24, 80, 80], - [1, 96, 160, 160], - [1, 256, 80, 80], - [1, 256, 160, 160], - [1, 256, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 56, 40, 40], - [1, 256, 40, 40], - ] - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - test_suite2.trt_dynamic_multi_thread_test( - input_data_dict, - output_data_dict, - gpu_mem=5000, - max_batch_size=max_batch_size, - repeat=1, - delta=2e-5, - names=names, - min_input_shape=min_input_shape, - max_input_shape=max_input_shape, - opt_input_shape=opt_input_shape, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trtfp32_more_bz_dynamic_bz(): - """ - compared trt fp32 batch_size=1,2 ocr_det_mv3_db outputs with true val - """ - check_model_exist() - - file_path = "./ocr_det_mv3_db" - images_size = 640 - batch_size_pool = [1, 2] - max_batch_size = 2 - names = [ - "x", - "conv2d_92.tmp_0", - "conv2d_91.tmp_0", - "conv2d_59.tmp_0", - "nearest_interp_v2_1.tmp_0", - "nearest_interp_v2_2.tmp_0", - "conv2d_124.tmp_0", - "nearest_interp_v2_3.tmp_0", - "nearest_interp_v2_4.tmp_0", - "nearest_interp_v2_5.tmp_0", - "elementwise_add_7", - "nearest_interp_v2_0.tmp_0", - ] - min_input_shape = [ - [1, 3, 50, 50], - [1, 120, 20, 20], - [1, 24, 10, 10], - [1, 96, 20, 20], - [1, 256, 10, 10], - [1, 256, 20, 20], - [1, 256, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 56, 2, 2], - [1, 256, 2, 2], - ] - - max_input_shape = [ - [max_batch_size, 3, 2000, 2000], - [max_batch_size, 120, 400, 400], - [max_batch_size, 24, 200, 200], - [max_batch_size, 96, 400, 400], - [max_batch_size, 256, 200, 200], - [max_batch_size, 256, 400, 400], - [max_batch_size, 256, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 56, 400, 400], - [max_batch_size, 256, 400, 400], - ] - - opt_input_shape = [ - [1, 3, 640, 640], - [1, 120, 160, 160], - [1, 24, 80, 80], - [1, 96, 160, 160], - [1, 256, 80, 80], - [1, 256, 160, 160], - [1, 256, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 56, 40, 40], - [1, 256, 40, 40], - ] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - - test_suite2.trt_more_bz_dynamic_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - repeat=1, - delta=2e-5, - names=names, - min_input_shape=min_input_shape, - max_input_shape=max_input_shape, - opt_input_shape=opt_input_shape, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_jetson_trtfp32_more_bz_dynamic_bz(): - """ - compared trt fp32 batch_size=1 ocr_det_mv3_db outputs with true val - """ - check_model_exist() - - file_path = "./ocr_det_mv3_db" - images_size = 640 - batch_size_pool = [1] - max_batch_size = 5 - names = [ - "x", - "conv2d_92.tmp_0", - "conv2d_91.tmp_0", - "conv2d_59.tmp_0", - "nearest_interp_v2_1.tmp_0", - "nearest_interp_v2_2.tmp_0", - "conv2d_124.tmp_0", - "nearest_interp_v2_3.tmp_0", - "nearest_interp_v2_4.tmp_0", - "nearest_interp_v2_5.tmp_0", - "elementwise_add_7", - "nearest_interp_v2_0.tmp_0", - ] - min_input_shape = [ - [1, 3, 50, 50], - [1, 120, 20, 20], - [1, 24, 10, 10], - [1, 96, 20, 20], - [1, 256, 10, 10], - [1, 256, 20, 20], - [1, 256, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 56, 2, 2], - [1, 256, 2, 2], - ] - - max_input_shape = [ - [max_batch_size, 3, 2000, 2000], - [max_batch_size, 120, 400, 400], - [max_batch_size, 24, 200, 200], - [max_batch_size, 96, 400, 400], - [max_batch_size, 256, 200, 200], - [max_batch_size, 256, 400, 400], - [max_batch_size, 256, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 56, 400, 400], - [max_batch_size, 256, 400, 400], - ] - - opt_input_shape = [ - [1, 3, 640, 640], - [1, 120, 160, 160], - [1, 24, 80, 80], - [1, 96, 160, 160], - [1, 256, 80, 80], - [1, 256, 160, 160], - [1, 256, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 56, 40, 40], - [1, 256, 40, 40], - ] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", - params_file="./ocr_det_mv3_db/inference.pdiparams", - ) - - test_suite2.trt_more_bz_dynamic_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - repeat=1, - delta=2e-5, - names=names, - min_input_shape=min_input_shape, - max_input_shape=max_input_shape, - opt_input_shape=opt_input_shape, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_seg_model/test_deeplabv3_trt_fp16.py b/inference/python_api_test/test_seg_model/test_deeplabv3_trt_fp16.py deleted file mode 100644 index 50bdcb73dd..0000000000 --- a/inference/python_api_test/test_seg_model/test_deeplabv3_trt_fp16.py +++ /dev/null @@ -1,194 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test deeplabv3 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - deeplabv3_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/seg/deeplabv3.tgz" - if not os.path.exists("./deeplabv3/model.pdiparams"): - wget.download(deeplabv3_url, out="./") - tar = tarfile.open("deeplabv3.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 deeplabv3 outputs with true val - """ - check_model_exist() - - file_path = "./deeplabv3" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 2 - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 deeplabv3 outputs with true val - """ - check_model_exist() - - file_path = "./deeplabv3" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite1 = InferenceTest() - test_suite1.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - test_suite1.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - tuned=True, - ) - del test_suite1 # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 deeplabv3 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./deeplabv3" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=2e-2, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_seg_model/test_deeplabv3_trt_fp32.py b/inference/python_api_test/test_seg_model/test_deeplabv3_trt_fp32.py deleted file mode 100644 index f7eb306ad9..0000000000 --- a/inference/python_api_test/test_seg_model/test_deeplabv3_trt_fp32.py +++ /dev/null @@ -1,175 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test deeplabv3 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - deeplabv3_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/seg/deeplabv3.tgz" - if not os.path.exists("./deeplabv3/model.pdiparams"): - wget.download(deeplabv3_url, out="./") - tar = tarfile.open("deeplabv3.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-2 deeplabv3 outputs with true val - """ - check_model_exist() - - file_path = "./deeplabv3" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 deeplabv3 outputs with true val - """ - check_model_exist() - - file_path = "./deeplabv3" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 deeplabv3 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./deeplabv3" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./deeplabv3/model.pdmodel", - params_file="./deeplabv3/model.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_seg_model/test_fcn_hrnetw18_trt_fp16.py b/inference/python_api_test/test_seg_model/test_fcn_hrnetw18_trt_fp16.py deleted file mode 100644 index 8e96facb28..0000000000 --- a/inference/python_api_test/test_seg_model/test_fcn_hrnetw18_trt_fp16.py +++ /dev/null @@ -1,190 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test fcn_hrnetw18 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - fcn_hrnetw18_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/seg/fcn_hrnetw18.tgz" - if not os.path.exists("./fcn_hrnetw18/model.pdiparams"): - wget.download(fcn_hrnetw18_url, out="./") - tar = tarfile.open("fcn_hrnetw18.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 fcn_hrnetw18 outputs with true val - """ - check_model_exist() - - file_path = "./fcn_hrnetw18" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 2 - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 fcn_hrnetw18 outputs with true val - """ - check_model_exist() - - file_path = "./fcn_hrnetw18" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite1 = InferenceTest() - test_suite1.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - test_suite1.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - tuned=True, - ) - del test_suite1 # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 fcn_hrnetw18 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./fcn_hrnetw18" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=2e-2, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_seg_model/test_fcn_hrnetw18_trt_fp32.py b/inference/python_api_test/test_seg_model/test_fcn_hrnetw18_trt_fp32.py deleted file mode 100644 index de82cf2682..0000000000 --- a/inference/python_api_test/test_seg_model/test_fcn_hrnetw18_trt_fp32.py +++ /dev/null @@ -1,172 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test fcn_hrnetw18 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - fcn_hrnetw18_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/seg/fcn_hrnetw18.tgz" - if not os.path.exists("./fcn_hrnetw18/model.pdiparams"): - wget.download(fcn_hrnetw18_url, out="./") - tar = tarfile.open("fcn_hrnetw18.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 fcn_hrnetw18 outputs with true val - """ - check_model_exist() - - file_path = "./fcn_hrnetw18" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 2 - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 fcn_hrnetw18 outputs with true val - """ - check_model_exist() - - file_path = "./fcn_hrnetw18" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 fcn_hrnetw18 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./fcn_hrnetw18" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./fcn_hrnetw18/model.pdmodel", - params_file="./fcn_hrnetw18/model.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_seg_model/test_pp_humanseg_lite_trt_fp16.py b/inference/python_api_test/test_seg_model/test_pp_humanseg_lite_trt_fp16.py deleted file mode 100644 index 906d6d72fe..0000000000 --- a/inference/python_api_test/test_seg_model/test_pp_humanseg_lite_trt_fp16.py +++ /dev/null @@ -1,194 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test pp_humanseg_lite model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - pp_humanseg_lite_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/seg/pp_humanseg_lite.tgz" - if not os.path.exists("./pp_humanseg_lite/model.pdiparams"): - wget.download(pp_humanseg_lite_url, out="./") - tar = tarfile.open("pp_humanseg_lite.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-2 pp_humanseg_lite outputs with true val - """ - check_model_exist() - - file_path = "./pp_humanseg_lite" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=3, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 pp_humanseg_lite outputs with true val - """ - check_model_exist() - - file_path = "./pp_humanseg_lite" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite1 = InferenceTest() - test_suite1.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - test_suite1.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - tuned=True, - ) - del test_suite1 # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 pp_humanseg_lite multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./pp_humanseg_lite" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=2e-2, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_seg_model/test_pp_humanseg_lite_trt_fp32.py b/inference/python_api_test/test_seg_model/test_pp_humanseg_lite_trt_fp32.py deleted file mode 100644 index fd43ce2490..0000000000 --- a/inference/python_api_test/test_seg_model/test_pp_humanseg_lite_trt_fp32.py +++ /dev/null @@ -1,175 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test pp_humanseg_lite model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - pp_humanseg_lite_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/seg/pp_humanseg_lite.tgz" - if not os.path.exists("./pp_humanseg_lite/model.pdiparams"): - wget.download(pp_humanseg_lite_url, out="./") - tar = tarfile.open("pp_humanseg_lite.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-2 pp_humanseg_lite outputs with true val - """ - check_model_exist() - - file_path = "./pp_humanseg_lite" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 pp_humanseg_lite outputs with true val - """ - check_model_exist() - - file_path = "./pp_humanseg_lite" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 pp_humanseg_lite multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./pp_humanseg_lite" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_humanseg_lite/model.pdmodel", - params_file="./pp_humanseg_lite/model.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_seg_model/test_pp_liteseg_stdc1_trt_fp16.py b/inference/python_api_test/test_seg_model/test_pp_liteseg_stdc1_trt_fp16.py deleted file mode 100644 index 55072f73dc..0000000000 --- a/inference/python_api_test/test_seg_model/test_pp_liteseg_stdc1_trt_fp16.py +++ /dev/null @@ -1,190 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test pp_liteseg_stdc1 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - pp_liteseg_stdc1_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/seg/pp_liteseg_stdc1.tgz" - if not os.path.exists("./pp_liteseg_stdc1/model.pdiparams"): - wget.download(pp_liteseg_stdc1_url, out="./") - tar = tarfile.open("pp_liteseg_stdc1.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 pp_liteseg_stdc1 outputs with true val - """ - check_model_exist() - - file_path = "./pp_liteseg_stdc1" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 pp_liteseg_stdc1 outputs with true val - """ - check_model_exist() - - file_path = "./pp_liteseg_stdc1" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite1 = InferenceTest() - test_suite1.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - test_suite1.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - tuned=True, - ) - del test_suite1 # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=2e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 pp_liteseg_stdc1 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./pp_liteseg_stdc1" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=2e-2, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_seg_model/test_pp_liteseg_stdc1_trt_fp32.py b/inference/python_api_test/test_seg_model/test_pp_liteseg_stdc1_trt_fp32.py deleted file mode 100644 index d13a99aaf2..0000000000 --- a/inference/python_api_test/test_seg_model/test_pp_liteseg_stdc1_trt_fp32.py +++ /dev/null @@ -1,171 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test pp_liteseg_stdc1 model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - pp_liteseg_stdc1_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/seg/pp_liteseg_stdc1.tgz" - if not os.path.exists("./pp_liteseg_stdc1/model.pdiparams"): - wget.download(pp_liteseg_stdc1_url, out="./") - tar = tarfile.open("pp_liteseg_stdc1.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 pp_liteseg_stdc1 outputs with true val - """ - check_model_exist() - - file_path = "./pp_liteseg_stdc1" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 pp_liteseg_stdc1 outputs with true val - """ - check_model_exist() - - file_path = "./pp_liteseg_stdc1" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 pp_liteseg_stdc1 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./pp_liteseg_stdc1" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./pp_liteseg_stdc1/model.pdmodel", - params_file="./pp_liteseg_stdc1/model.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_seg_model/test_unet_trt_fp16.py b/inference/python_api_test/test_seg_model/test_unet_trt_fp16.py deleted file mode 100644 index 5f4c6d52fa..0000000000 --- a/inference/python_api_test/test_seg_model/test_unet_trt_fp16.py +++ /dev/null @@ -1,194 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test unet model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - unet_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/seg/unet.tgz" - if not os.path.exists("./unet/model.pdiparams"): - wget.download(unet_url, out="./") - tar = tarfile.open("unet.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 unet outputs with true val - """ - check_model_exist() - - file_path = "./unet" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 2 - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=3e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_jetson_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1 unet outputs with true val - """ - check_model_exist() - - file_path = "./unet" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite1 = InferenceTest() - test_suite1.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - test_suite1.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=5e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - tuned=True, - ) - del test_suite1 # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - delta=5e-2, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp16", - dynamic=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 unet multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./unet" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - delta=2e-2, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory diff --git a/inference/python_api_test/test_seg_model/test_unet_trt_fp32.py b/inference/python_api_test/test_seg_model/test_unet_trt_fp32.py deleted file mode 100644 index 01127d694b..0000000000 --- a/inference/python_api_test/test_seg_model/test_unet_trt_fp32.py +++ /dev/null @@ -1,175 +0,0 @@ -# -*- coding: utf-8 -*- -# encoding=utf-8 vi:ts=4:sw=4:expandtab:ft=python -""" -test unet model -""" - -import os -import sys -import logging -import tarfile -import shutil -import six -import wget -import pytest -import numpy as np - -# pylint: disable=wrong-import-position -sys.path.append("..") -from test_case import InferenceTest - - -# pylint: enable=wrong-import-position - - -def check_model_exist(): - """ - check model exist - """ - unet_url = "https://paddle-qa.bj.bcebos.com/inference_model/2.6/seg/unet.tgz" - if not os.path.exists("./unet/model.pdiparams"): - wget.download(unet_url, out="./") - tar = tarfile.open("unet.tgz") - tar.extractall() - tar.close() - - -def test_config(): - """ - test combined model config - """ - check_model_exist() - test_suite = InferenceTest() - test_suite.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - test_suite.config_test() - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-2 unet outputs with true val - """ - check_model_exist() - - file_path = "./unet" - images_size = 224 - batch_size_pool = [1, 2] - max_batch_size = 2 - for batch_size in batch_size_pool: - try: - shutil.rmtree(f"{file_path}/_opt_cache") # delete trt serialized cache - except Exception as e: - print("no need to delete trt serialized cache") - - test_suite = InferenceTest() - test_suite.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.jetson -@pytest.mark.trt_fp32_more_bz_precision -def test_jetson_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1 unet outputs with true val - """ - check_model_exist() - - file_path = "./unet" - images_size = 224 - batch_size_pool = [1] - max_batch_size = 1 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - test_suite2.trt_more_bz_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - min_subgraph_size=1, - precision="trt_fp32", - dynamic=True, - auto_tuned=True, - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 unet multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./unet" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./unet/model.pdmodel", - params_file="./unet/model.pdiparams", - ) - test_suite2.trt_bz1_multi_thread_test( - input_data_dict, - output_data_dict, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/models/AutomaticTestSystem/ModelsTestFramework.py b/models/AutomaticTestSystem/ModelsTestFramework.py index 22afe0c0ac..34ba178618 100644 --- a/models/AutomaticTestSystem/ModelsTestFramework.py +++ b/models/AutomaticTestSystem/ModelsTestFramework.py @@ -271,11 +271,6 @@ def __init__(self): exit_code = repo_result[0] output = repo_result[1] assert exit_code == 0, "pretrain_models configure failed! log information:%s" % output - if (platform.system() == "Windows") or (platform.system() == "Linux"): - repo_result = subprocess.getstatusoutput(cmd) - exit_code = repo_result[0] - output = repo_result[1] - assert exit_code == 0, "tensorRT dynamic shape configure failed! log information:%s" % output class TestOcrModelFunction: @@ -541,7 +536,7 @@ def test_ocr_export_model(self, use_gpu): allure_step(cmd, output) exit_check_fucntion(exit_code, output, "export_model") - def test_ocr_rec_predict(self, use_gpu, use_tensorrt, enable_mkldnn): + def test_ocr_rec_predict(self, use_gpu, enable_mkldnn): """ test_ocr_rec_predict """ @@ -558,7 +553,7 @@ def test_ocr_rec_predict(self, use_gpu, use_tensorrt, enable_mkldnn): algorithm, rec_char_dict_path, use_gpu, - use_tensorrt, + False, enable_mkldnn, ) elif self.category == "det": @@ -568,14 +563,14 @@ def test_ocr_rec_predict(self, use_gpu, use_tensorrt, enable_mkldnn): self.model, algorithm, use_gpu, - use_tensorrt, + False, enable_mkldnn, ) elif self.category == "table": cmd = self.testcase_yml["cmd"][self.category]["predict"] % ( self.model, use_gpu, - use_tensorrt, + False, enable_mkldnn, ) elif self.category == "sr": @@ -584,14 +579,14 @@ def test_ocr_rec_predict(self, use_gpu, use_tensorrt, enable_mkldnn): self.model, sr_image_shape, use_gpu, - use_tensorrt, + False, enable_mkldnn, ) elif (self.category == "kie/vi_layoutxlm") or (self.category == "e2e"): cmd = self.testcase_yml["cmd"][self.category]["predict"] % ( self.model, use_gpu, - use_tensorrt, + False, enable_mkldnn, ) elif self.category == "picodet/legacy_model/application/layout_analysis": @@ -605,7 +600,7 @@ def test_ocr_rec_predict(self, use_gpu, use_tensorrt, enable_mkldnn): cmd = self.testcase_yml["cmd"][self.category]["predict_SLANet"] % ( self.model, use_gpu, - use_tensorrt, + False, enable_mkldnn, ) @@ -810,7 +805,7 @@ def test_3D_export_model(self, use_gpu): allure_step(cmd, output) exit_check_fucntion(exit_code, output, "export_model") - def test_3D_predict_python(self, use_gpu, use_trt): + def test_3D_predict_python(self, use_gpu): """ test_3D_predict_python """ @@ -823,17 +818,6 @@ def test_3D_predict_python(self, use_gpu, use_trt): --params_file exported_model/%s/inference.pdiparams --image %s --use_gpu" % (self.model, self.model, infer_image) ) - if use_trt is True: - cmd = ( - "cd Paddle3D; python deploy/smoke/python/infer.py \ - --model_file exported_model/%s/inference.pdmodel\ - --params_file exported_model/%s/inference.pdiparams --image %s --collect_dynamic_shape_info \ - --dynamic_shape_file %s/shape_info.txt; \ - python deploy/smoke/python/infer.py --model_file exported_model/%s/inference.pdmodel \ - --params_file exported_model/%s/inference.pdiparams --image %s --use_gpu \ - --use_trt --dynamic_shape_file %s/shape_info.txt;" - % (self.model, self.model, infer_image, self.model, self.model, self.model, infer_image, self.model) - ) if paddle.is_compiled_with_cuda() is False: cmd = ( "cd Paddle3D; python deploy/smoke/python/infer.py \ diff --git a/models/AutomaticTestSystem/test_3D_acc.py b/models/AutomaticTestSystem/test_3D_acc.py index 43c5b91132..7c35056f25 100644 --- a/models/AutomaticTestSystem/test_3D_acc.py +++ b/models/AutomaticTestSystem/test_3D_acc.py @@ -165,32 +165,7 @@ def test_3D_accuracy_predict_python(yml_name, use_gpu): category = get_category(yml_name) model = Test3DModelFunction(model=model_name, yml=yml_name, category=category) - model.test_3D_predict_python(use_gpu, False) - - -@allure.story("predict") -@pytest.mark.parametrize("yml_name", get_model_list()) -@pytest.mark.parametrize("use_gpu", [True]) -def test_3D_accuracy_predict_python_trt(yml_name, use_gpu): - """ - test_3D_accuracy_predict_python_trt - """ - model_name = os.path.splitext(os.path.basename(yml_name))[0] - hardware = "_TensorRT" - allure.dynamic.title(model_name + hardware + "_predict") - allure.dynamic.description("预测库python预测") - pytest.skip("not supported for tensorRT predict") - if paddle.is_compiled_with_cuda() is False: - pytest.skip("CPU not supported for tensorRT predict") - category = get_category(yml_name) - if (category == "pointpillars") or (category == "centerpoint") or (category == "squeezesegv3"): - pytest.skip("not supoorted for tensorRT predict") - if sys.platform == "darwin": - pytest.skip("mac skip tensorRT predict") - - category = get_category(yml_name) - model = Test3DModelFunction(model=model_name, yml=yml_name, category=category) - model.test_3D_predict_python(use_gpu, True) + model.test_3D_predict_python(use_gpu) @allure.story("train") diff --git a/models/AutomaticTestSystem/test_ocr_acc.py b/models/AutomaticTestSystem/test_ocr_acc.py index 953dc94565..4b29a7833c 100644 --- a/models/AutomaticTestSystem/test_ocr_acc.py +++ b/models/AutomaticTestSystem/test_ocr_acc.py @@ -174,32 +174,7 @@ def test_ocr_accuracy_predict_mkl(yml_name, enable_mkldnn): category = r.group(1) print(category) model = TestOcrModelFunction(model=model_name, yml=yml_name, category=category) - model.test_ocr_rec_predict(False, 0, enable_mkldnn) - - -@allure.story("predict") -@pytest.mark.parametrize("yml_name", get_model_list()) -@pytest.mark.parametrize("use_tensorrt", [True, False]) -def test_ocr_accuracy_predict_trt(yml_name, use_tensorrt): - """ - test_ocr_accuracy_predict_trt - """ - model_name = os.path.splitext(os.path.basename(yml_name))[0] - if use_tensorrt is True: - hardware = "_tensorRT" - else: - hardware = "_GPU" - allure.dynamic.title(model_name + hardware + "_predict") - allure.dynamic.description("预测库预测") - - if model_name == "re_vi_layoutxlm_xfund_zh": - pytest.skip("not supported") - - r = re.search("/(.*)/", yml_name) - category = r.group(1) - print(category) - model = TestOcrModelFunction(model=model_name, yml=yml_name, category=category) - model.test_ocr_rec_predict(True, use_tensorrt, 0) + model.test_ocr_rec_predict(False, enable_mkldnn) def test_ocr_accuracy_predict_recovery(): diff --git a/models/PaddleDetection/CE/inference_script/detection_inference.sh b/models/PaddleDetection/CE/inference_script/detection_inference.sh index d4e3ae9762..4fb9e4929c 100644 --- a/models/PaddleDetection/CE/inference_script/detection_inference.sh +++ b/models/PaddleDetection/CE/inference_script/detection_inference.sh @@ -22,11 +22,9 @@ MACHINE_TYPE=`uname -m` echo "MACHINE_TYPE: "${MACHINE_TYPE} config_list='ppyolo_r50vd_dcn_1x_coco ppyolov2_r50vd_dcn_365e_coco yolov3_darknet53_270e_coco solov2_r50_fpn_1x_coco faster_rcnn_r50_fpn_1x_coco mask_rcnn_r50_1x_coco s2anet_conv_2x_dota ssd_mobilenet_v1_300_120e_voc ttfnet_darknet53_1x_coco fcos_r50_fpn_1x_coco' config_list_cpp='ppyolo_r50vd_dcn_1x_coco ppyolov2_r50vd_dcn_365e_coco yolov3_darknet53_270e_coco faster_rcnn_r50_fpn_1x_coco mask_rcnn_r50_1x_coco s2anet_conv_2x_dota ssd_mobilenet_v1_300_120e_voc ttfnet_darknet53_1x_coco fcos_r50_fpn_1x_coco' -config_skip_trt8='ppyolo_r50vd_dcn_1x_coco ppyolov2_r50vd_dcn_365e_coco solov2_r50_fpn_1x_coco faster_rcnn_r50_fpn_1x_coco mask_rcnn_r50_1x_coco ttfnet_darknet53_1x_coco fcos_r50_fpn_1x_coco s2anet_conv_2x_dota' config_skip_bs2='solov2_r50_fpn_1x_coco mask_rcnn_r50_1x_coco s2anet_conv_2x_dota' config_skip_video='mask_rcnn_r50_1x_coco' config_s2anet='s2anet_conv_2x_dota' -mode_list='trt_fp32 trt_fp16 trt_int8 paddle' err_sign=false print_result_python(){ if [ $? -ne 0 ];then @@ -42,14 +40,14 @@ print_result_python(){ echo -e "${config}_${mode},python_infer,SUCCESS" fi } -python_trt(){ +python_gpu(){ + mode=paddle python deploy/python/infer.py \ --model_dir=./inference_model/${config} \ --image_file=${image} \ --device=GPU \ - --run_mode=${mode} \ + --run_mode=paddle \ --threshold=0.5 \ - --trt_calib_mode=${trt_calib_mode} \ --output_dir=python_infer_output/${config}_${mode} >logs/${config}_${mode}.log 2>&1 print_result_python } @@ -109,19 +107,7 @@ python tools/export_model.py \ -c configs/${model} \ --output_dir=inference_model \ -o weights=https://paddledet.bj.bcebos.com/models/${config}.pdparams -for mode in ${mode_list} -do -if [[ ${mode} == 'trt_int8' ]];then - trt_calib_mode=True -else - trt_calib_mode=False -fi -if [[ ${mode} == 'trt_int8' ]] && [[ -n `echo "${config_skip_trt8}" | grep -w "${config}"` ]];then - echo -e "***skip trt_int8 for ${config}" -else - python_trt -fi -done +python_gpu python_cpu python_mkldnn if [[ -n `echo "${config_skip_bs2}" | grep -w "${config}"` ]];then @@ -143,16 +129,12 @@ tar -xvf paddle_inference.tgz mv paddle_inference_install_dir paddle_inference sed -i "s|/path/to/paddle_inference|../paddle_inference|g" scripts/build.sh sed -i "s|WITH_GPU=OFF|WITH_GPU=ON|g" scripts/build.sh -sed -i "s|WITH_TENSORRT=OFF|WITH_TENSORRT=ON|g" scripts/build.sh sed -i "s|CUDA_LIB=/path/to/cuda/lib|CUDA_LIB=/usr/local/cuda/lib64|g" scripts/build.sh if [[ "$MACHINE_TYPE" == "aarch64" ]] then sed -i "s|WITH_MKL=ON|WITH_MKL=OFF|g" scripts/build.sh -sed -i "s|TENSORRT_INC_DIR=/path/to/tensorrt/include|TENSORRT_INC_DIR=/usr/include/aarch64-linux-gnu|g" scripts/build.sh -sed -i "s|TENSORRT_LIB_DIR=/path/to/tensorrt/lib|TENSORRT_LIB_DIR=/usr/lib/aarch64-linux-gnu|g" scripts/build.sh sed -i "s|CUDNN_LIB=/path/to/cudnn/lib|CUDNN_LIB=/usr/lib/aarch64-linux-gnu|g" scripts/build.sh else -sed -i "s|TENSORRT_LIB_DIR=/path/to/tensorrt/lib|TENSORRT_LIB_DIR=/usr/local/TensorRT6-cuda10.1-cudnn7/lib|g" scripts/build.sh sed -i "s|CUDNN_LIB=/path/to/cudnn/lib|CUDNN_LIB=/usr/lib/x86_64-linux-gnu|g" scripts/build.sh fi sh scripts/build.sh @@ -171,8 +153,9 @@ print_result_cpp(){ echo -e "${config}_${mode},cpp_infer,SUCCESS" fi } -cpp_trt(){ - ./deploy/cpp/build/main --model_dir=inference_model/${config} --image_file=${image} --output_dir=cpp_infer_output/${config}_${mode} --device=GPU --run_mode=${mode} --threshold=0.5 --trt_calib_mode=${trt_calib_mode} >logs_cpp/${config}_${mode}.log 2>&1 +cpp_gpu(){ + mode=paddle + ./deploy/cpp/build/main --model_dir=inference_model/${config} --image_file=${image} --output_dir=cpp_infer_output/${config}_${mode} --device=GPU --run_mode=paddle --threshold=0.5 >logs_cpp/${config}_${mode}.log 2>&1 print_result_cpp } cpp_cpu(){ @@ -201,19 +184,7 @@ image=demo/000000570688.jpg if [[ -n `echo "${config_s2anet}" | grep -w "${config}"` ]];then image=demo/P0072__1.0__0___0.png fi -for mode in ${mode_list} -do -if [[ ${mode} == 'trt_int8' ]];then - trt_calib_mode=True -else - trt_calib_mode=False -fi -if [[ ${mode} == 'trt_int8' ]] && [[ -n `echo "${config_skip_trt8}" | grep -w "${config}"` ]];then - echo -e "***skip trt_int8 for ${config}" -else - cpp_trt -fi -done +cpp_gpu cpp_cpu cpp_mkldnn if [[ -n `echo "${config_skip_bs2}" | grep -w "${config}"` ]];then diff --git a/models/PaddleMIX/CE/ppdiffusers/deploy/controlnet/infer.sh b/models/PaddleMIX/CE/ppdiffusers/deploy/controlnet/infer.sh index 25d1662628..9a6f55bcc8 100644 --- a/models/PaddleMIX/CE/ppdiffusers/deploy/controlnet/infer.sh +++ b/models/PaddleMIX/CE/ppdiffusers/deploy/controlnet/infer.sh @@ -52,77 +52,6 @@ else fi echo "*******ppdiffusers/deploy/controlnet controlnet_inference_inpaint end***********" -# tensorrt -# tune -# -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-canny/ \ - --scheduler "ddim" \ - --backend paddle \ - --device gpu \ - --task_name all \ - --width 512 \ - --height 512 \ - --inference_steps 5 \ - --tune True \ - --use_fp16 False) 2>&1 | tee ${log_dir}/controlnet_inference_tensorrt_tune.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_tune success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_tune fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_tune end***********" - - -# text2img -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-canny/ \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name text2img) 2>&1 | tee ${log_dir}/controlnet_inference_tensorrt_text2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_text2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_text2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_text2img end***********" - -# img2img -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-canny/ \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name img2img) 2>&1 | tee ${log_dir}/controlnet_inference_tensorrt_img2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_img2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_img2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_img2img end***********" - -# inpaint -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-canny/ \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name inpaint_legacy) 2>&1 | tee ${log_dir}/controlnet_inference_tensorrt_inpaint.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_inpaint success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_inpaint fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet controlnet_inference_tensorrt_inpaint end***********" echo exit_code:${exit_code} exit ${exit_code} diff --git a/models/PaddleMIX/CE/ppdiffusers/deploy/ipadapter_sd15/sd15_infer.sh b/models/PaddleMIX/CE/ppdiffusers/deploy/ipadapter_sd15/sd15_infer.sh index ef030edeec..d266854ede 100644 --- a/models/PaddleMIX/CE/ppdiffusers/deploy/ipadapter_sd15/sd15_infer.sh +++ b/models/PaddleMIX/CE/ppdiffusers/deploy/ipadapter_sd15/sd15_infer.sh @@ -63,84 +63,6 @@ else fi echo "*******ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_inpaint end***********" -# tensorrt -# tune -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-ipadapter/ \ - --scheduler "ddim" \ - --backend paddle \ - --device gpu \ - --task_name all \ - --width 512 \ - --height 512 \ - --inference_steps 5 \ - --tune True \ - --use_fp16 False) 2>&1 | tee ${log_dir}/ipadapter_sd15_inference_tune.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tune success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tune fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tune end***********" - -# text2img -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-ipadapter/ \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --width 512 \ - --height 512 \ - --inference_steps 50 \ - --task_name text2img) 2>&1 | tee ${log_dir}/ipadapter_sd15_inference_tensorrt_text2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tensorrt_text2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tensorrt_text2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tensorrt_text2img end***********" - -# img2img -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-ipadapter/ \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --width 512 \ - --height 512 \ - --inference_steps 50 \ - --task_name img2img) 2>&1 | tee ${log_dir}/ipadapter_sd15_inference_tensorrt_img2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tensorrt_img2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tensorrt_img2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tensorrt_img2img end***********" - -# inpaint -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-ipadapter/ \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --width 512 \ - --height 512 \ - --inference_steps 50 \ - --task_name inpaint_legacy) 2>&1 | tee ${log_dir}/ipadapter_sd15_inference_tensorrt_inpaint.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tensorrt_inpaint success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tensorrt_inpaint fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/ipadapter_sd15 ipadapter_sd15_inference_tensorrt_inpaint end***********" echo exit_code:${exit_code} exit ${exit_code} diff --git a/models/PaddleMIX/CE/ppdiffusers/deploy/ipadapter_sdxl/sdxl_infer.sh b/models/PaddleMIX/CE/ppdiffusers/deploy/ipadapter_sdxl/sdxl_infer.sh index 8c71e20a04..dfc6a1d0dc 100644 --- a/models/PaddleMIX/CE/ppdiffusers/deploy/ipadapter_sdxl/sdxl_infer.sh +++ b/models/PaddleMIX/CE/ppdiffusers/deploy/ipadapter_sdxl/sdxl_infer.sh @@ -62,83 +62,6 @@ else fi echo "*******ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_inpaint end***********" -# tensorrt -# tune -(python infer.py \ - --model_dir static_model/stable-diffusion-xl-base-1.0-ipadapter/ \ - --scheduler "ddim" \ - --backend paddle \ - --device gpu \ - --task_name all \ - --width 512 \ - --height 512 \ - --inference_steps 5 \ - --tune True \ - --use_fp16 False) 2>&1 | tee ${log_dir}/ipadapter_sdxl_inference_tune.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tune success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tune fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tune end***********" - -# text2img -(python infer.py \ - --model_dir static_model/stable-diffusion-xl-base-1.0-ipadapter \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --width 512 \ - --height 512 \ - --inference_steps 50 \ - --device gpu --task_name text2img) 2>&1 | tee ${log_dir}/ipadapter_sdxl_inference_tensorrt_text2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tensorrt_text2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tensorrt_text2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tensorrt_text2img end***********" - -# img2img -(python infer.py \ - --model_dir static_model/stable-diffusion-xl-base-1.0-ipadapter \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --width 512 \ - --height 512 \ - --inference_steps 50 \ - --task_name img2img) 2>&1 | tee ${log_dir}/ipadapter_sdxl_inference_tensorrt_img2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tensorrt_img2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tensorrt_img2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tensorrt_img2img end***********" - -# inpaint -(python infer.py \ - --model_dir static_model/stable-diffusion-xl-base-1.0-ipadapter/ \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --width 512 \ - --height 512 \ - --inference_steps 50 \ - --task_name inpaint) 2>&1 | tee ${log_dir}/ipadapter_sdxl_inference_tensorrt_inpaint.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tensorrt_inpaint success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tensorrt_inpaint fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sdxl ipadapter_sdxl_inference_tensorrt_inpaint end***********" echo exit_code:${exit_code} exit ${exit_code} diff --git a/models/PaddleMIX/CE/ppdiffusers/deploy/script/start.sh b/models/PaddleMIX/CE/ppdiffusers/deploy/script/start.sh index 5f5797a489..be90dcb1b3 100644 --- a/models/PaddleMIX/CE/ppdiffusers/deploy/script/start.sh +++ b/models/PaddleMIX/CE/ppdiffusers/deploy/script/start.sh @@ -30,16 +30,6 @@ fi echo "*******ppdiffusers/deploy/controlnet end***********" -# controlnet_tensorrt -(bash scripts/benchmark_paddle_deploy_tensorrt.sh) 2>&1 | tee ${log_dir}/controlnet_tensorrt.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet_tensorrt success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet_tensorrt fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet_tensorrt end***********" cd .. @@ -56,16 +46,6 @@ else fi echo "*******ppdiffusers/deploy/ipadapter_sd15 end***********" -# ipadapter/sd15 tensorrt -(bash scripts/benchmark_paddle_deploy_tensorrt.sh) 2>&1 | tee ${log_dir}/ipadapter_sd15_tensorrt.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/ipadapter_sd15_tensorrt success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/ipadapter_sd15_tensorrt fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/ipadapter_sd15_tensorrt end***********" cd ../../ # ipadapter sdxl @@ -80,16 +60,6 @@ else fi echo "*******ppdiffusers/deploy/ipadapter_sdxl end***********" -# ipadapter sdxl tensorrt -(bash scripts/benchmark_paddle_deploy_tensorrt.sh) 2>&1 | tee ${log_dir}/ipadapter_sdxl_tensorrt.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/ipadapter_sdxl_tensorrt success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/ipadapter_sdxl_tensorrt fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/ipadapter_sdxl_tensorrt end***********" cd ../../ # sd15 @@ -105,16 +75,6 @@ fi echo "*******ppdiffusers/deploy/sd15 end***********" -# sd15_tensorrt -(bash scripts/benchmark_paddle_deploy_tensorrt.sh) 2>&1 | tee ${log_dir}/sd15_tensorrt.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sd15_tensorrt success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sd15_tensorrt fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sd15_tensorrt end***********" cd .. # sdxl @@ -130,16 +90,6 @@ fi echo "*******ppdiffusers/deploy/sdxl end***********" -# sdxl_tensorrt -(bash scripts/benchmark_paddle_deploy_tensorrt.sh) 2>&1 | tee ${log_dir}/sdxl_tensorrt.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sdxl_tensorrt success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sdxl_tensorrt fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sdxl_tensorrt end***********" cd .. # sd3 diff --git a/models/PaddleMIX/CE/ppdiffusers/deploy/script/start_hand.sh b/models/PaddleMIX/CE/ppdiffusers/deploy/script/start_hand.sh index 753560128d..e293e4dad1 100644 --- a/models/PaddleMIX/CE/ppdiffusers/deploy/script/start_hand.sh +++ b/models/PaddleMIX/CE/ppdiffusers/deploy/script/start_hand.sh @@ -26,8 +26,6 @@ for subdir in */; do cp -f ../test_*.sh . bash test_paddle.sh > ${log_dir}/${subdir}_paddle.log 2>&1 exit_code=$((exit_code + $?)) - bash test_paddle_tensorrt.sh > ${log_dir}/${subdir}_paddle_tensorrt.log 2>&1 - exit_code=$((exit_code + $?)) cd .. fi done @@ -41,10 +39,8 @@ for subdir in */; do cp -f ../test_*.sh . bash test_paddle.sh > ${log_dir}/ipadapter_${subdir}_paddle.log 2>&1 exit_code=$((exit_code + $?)) - bash test_paddle_tensorrt.sh > ${log_dir}/ipadapter_${subdir}_paddle_tensorrt.log 2>&1 - exit_code=$((exit_code + $?)) cd .. fi done echo exit_code:${exit_code} -exit ${exit_code} \ No newline at end of file +exit ${exit_code} diff --git a/models/PaddleMIX/CE/ppdiffusers/deploy/script/test_paddle_tensorrt.sh b/models/PaddleMIX/CE/ppdiffusers/deploy/script/test_paddle_tensorrt.sh deleted file mode 100644 index bad3ac333b..0000000000 --- a/models/PaddleMIX/CE/ppdiffusers/deploy/script/test_paddle_tensorrt.sh +++ /dev/null @@ -1,50 +0,0 @@ -# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved. -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -#!/bin/bash - -export USE_PPXFORMERS=False -export FLAGS_set_to_1d=1 - -# Define the base path to the static model directory -base_model_dir="static_model" -model_dir=$(find $base_model_dir -mindepth 1 -maxdepth 1 -type d | head -n 1) - -if [ -z "$model_dir" ]; then - echo "No model directory found under $base_model_dir, starting to export static model..." - - if sh scripts/export.sh; then - echo "Model exported successfully." - model_dir=$(find $base_model_dir -mindepth 1 -maxdepth 1 -type d | head -n 1) - else - echo "Failed to export model." - exit 1 - fi -else - echo "Using model directory: $model_dir" -fi - -# Tune static model shape info -echo "############### egnore this info - Begin ###############" -python infer.py --model_dir $model_dir --scheduler "ddim" --backend paddle_tensorrt --device gpu --task_name all --width 512 --height 512 --inference_steps 50 --tune True --use_fp16 False --benchmark_steps 3 -echo "############### egnore this info - End #################" - -# Inference with FP16 -echo "Running inference with FP16..." -python infer.py --model_dir $model_dir --scheduler "ddim" --backend paddle_tensorrt --device gpu --task_name all --width 512 --height 512 --inference_steps 50 --tune False --use_fp16 True --benchmark_steps 10 - -mv ./results-paddle_tensorrt ./results-paddle_tensorrt-fp16 -# Inference with FP32 -echo "Running inference with FP32..." -python infer.py --model_dir $model_dir --scheduler "ddim" --backend paddle_tensorrt --device gpu --task_name all --width 512 --height 512 --inference_steps 50 --tune False --use_fp16 False --benchmark_steps 10 \ No newline at end of file diff --git a/models/PaddleMIX/CE/ppdiffusers/deploy/sd15/infer.sh b/models/PaddleMIX/CE/ppdiffusers/deploy/sd15/infer.sh index c6396dc132..d48549765a 100644 --- a/models/PaddleMIX/CE/ppdiffusers/deploy/sd15/infer.sh +++ b/models/PaddleMIX/CE/ppdiffusers/deploy/sd15/infer.sh @@ -27,8 +27,6 @@ else echo "ppdiffusers/deploy/sd15 sd15_inference_text2img fail" >>"${log_dir}/ce_res.log" fi echo "*******ppdiffusers/deploy/sd15 sd15_inference_text2img end***********" -python infer.py --model_dir static_model/stable-diffusion-v1-5 --scheduler "ddim" --backend paddle_tensorrt --device gpu --task_name all --width 512 --height 512 --inference_steps 30 --tune True --use_fp16 False --benchmark_steps 3 - # img2img (python infer.py \ --model_dir static_model/stable-diffusion-v1-5/ \ @@ -61,75 +59,6 @@ else fi echo "*******ppdiffusers/deploy/sd15 sd15_inference_inpaint end***********" -# tensorrt -#tune -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5 \ - --scheduler "ddim" \ - --backend paddle \ - --device gpu \ - --task_name all \ - --width 512 \ - --height 512 \ - --inference_steps 5 \ - --tune True \ - --use_fp16 False) 2>&1 | tee ${log_dir}/sd15_inference_tensorrt_tune.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sd15 sd15_inference_tensorrt_tune success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sd15 sd15_inference_tensorrt_tune fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sd15 sd15_inference_tensorrt_tune end***********" - -# text2img -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5 \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name text2img) 2>&1 | tee ${log_dir}/sd15_inference_tensorrt_text2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sd15 sd15_inference_tensorrt_text2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sd15 sd15_inference_tensorrt_text2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sd15 sd15_inference_tensorrt_text2img end***********" - -# img2img -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5 \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name img2img) 2>&1 | tee ${log_dir}/sd15_inference_tensorrt_img2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sd15 sd15_inference_tensorrt_img2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sd15 sd15_inference_tensorrt_img2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sd15 sd15_inference_tensorrt_img2img end***********" - -# inpaint -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5 \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name inpaint_legacy) 2>&1 | tee ${log_dir}/sd15_inference_tensorrt_inpaint.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sd15 sd15_inference_tensorrt_inpaint success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sd15 sd15_inference_tensorrt_inpaint fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sd15 sd15_inference_tensorrt_inpaint end***********" echo exit_code:${exit_code} exit ${exit_code} diff --git a/models/PaddleMIX/CE/ppdiffusers/deploy/sdxl/infer.sh b/models/PaddleMIX/CE/ppdiffusers/deploy/sdxl/infer.sh index f0654ebff5..eebdd6ec62 100644 --- a/models/PaddleMIX/CE/ppdiffusers/deploy/sdxl/infer.sh +++ b/models/PaddleMIX/CE/ppdiffusers/deploy/sdxl/infer.sh @@ -57,75 +57,6 @@ else fi echo "*******ppdiffusers/deploy/sdxl sdxl_inference_inpaint end***********" -# tensorrt -#tune -(python infer.py \ - --model_dir static_model/stable-diffusion-xl-base-1.0 \ - --scheduler "ddim" \ - --backend paddle \ - --device gpu \ - --task_name all \ - --width 512 \ - --height 512 \ - --inference_steps 5 \ - --tune True \ - --use_fp16 False) 2>&1 | tee ${log_dir}/sdxl_inference_tune.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sdxl sdxl_inference_tune success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sdxl sdxl_inference_tune fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sdxl sdxl_inference_tune end***********" - -# text2img -(python infer.py \ - --model_dir static_model/stable-diffusion-xl-base-1.0 \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name text2img) 2>&1 | tee ${log_dir}/sdxl_inference_tensorrt_text2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sdxl sdxl_inference_tensorrt_text2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sdxl sdxl_inference_tensorrt_text2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sdxl sdxl_inference_tensorrt_text2img end***********" - -# img2img -(python infer.py \ - --model_dir static_model/stable-diffusion-xl-base-1.0 \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name img2img) 2>&1 | tee ${log_dir}/sdxl_inference_tensorrt_img2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sdxl sdxl_inference_tensorrt_img2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sdxl sdxl_inference_tensorrt_img2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sdxl sdxl_inference_tensorrt_img2img end***********" - -# inpaint -(python infer.py \ - --model_dir static_model/stable-diffusion-xl-base-1.0 \ - --scheduler "ddim" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name inpaint) 2>&1 | tee ${log_dir}/sdxl_inference_tensorrt_inpaint.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sdxl sdxl_inference_tensorrt_inpaint success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sdxl sdxl_inference_tensorrt_inpaint fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sdxl sdxl_inference_tensorrt_inpaint end***********" echo exit_code:${exit_code} exit ${exit_code} diff --git a/models/PaddleMIX/CE/ppdiffusers/deploy/svd/infer.sh b/models/PaddleMIX/CE/ppdiffusers/deploy/svd/infer.sh index 7c10964646..1d097df744 100644 --- a/models/PaddleMIX/CE/ppdiffusers/deploy/svd/infer.sh +++ b/models/PaddleMIX/CE/ppdiffusers/deploy/svd/infer.sh @@ -34,46 +34,6 @@ fi echo "*******ppdiffusers/deploy/svd svd_inference_text2video end***********" -# tensorrt -#tune -(python infer.py \ - --model_dir static_model/stable-video-diffusion-img2vid-xt \ - --scheduler "euler" \ - --backend paddle \ - --device gpu \ - --task_name all \ - --width 256 \ - --height 256 \ - --inference_steps 5 \ - --tune True \ - --use_fp16 False) 2>&1 | tee ${log_dir}/svd_inference_tune.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/svd svd_inference_tune success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/svd svd_inference_tune fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/svd svd_inference_tune end***********" - -# text2img -(python infer.py \ - --model_dir static_model/stable-video-diffusion-img2vid-xt \ - --scheduler "euler" \ - --backend paddle_tensorrt \ - --device gpu \ - --width 256 \ - --height 256 \ - --inference_steps 25 \ - --task_name img2video) 2>&1 | tee ${log_dir}/svd_inference_tensorrt_text2video.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/svd svd_inference_tensorrt_text2video success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/svd svd_inference_tensorrt_text2video fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/svd svd_inference_tensorrt_text2video end***********" echo exit_code:${exit_code} exit ${exit_code} diff --git a/models/PaddleMIX/CE2/ppdiffusers/sdxl/dynamic2static.sh b/models/PaddleMIX/CE2/ppdiffusers/sdxl/dynamic2static.sh index c9ba66355e..4d68fb4159 100644 --- a/models/PaddleMIX/CE2/ppdiffusers/sdxl/dynamic2static.sh +++ b/models/PaddleMIX/CE2/ppdiffusers/sdxl/dynamic2static.sh @@ -85,22 +85,6 @@ else fi echo "*******ppdiffusers/deploy/sdxl sdxl_test_image_diff_inpaint end***********" -# paddle_tensorrt -(python infer.py \ - --model_dir static_model/stable-diffusion-xl-base-1.0 \ - --scheduler "preconfig-euler-ancestral" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name all \ - --infer_op raw) 2>&1 | tee ${log_dir}/sdxl_inference_paddle_tensorrt.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/sdxl sdxl_inference_paddle_tensorrt success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/sdxl sdxl_inference_paddle_tensorrt fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/sdxl sdxl_inference_paddle_tensorrt end***********" echo exit_code:${exit_code} exit ${exit_code} diff --git a/models/PaddleMIX/CE2/ppdiffusers/stable_diffusion_controlnet/dynamic2static.sh b/models/PaddleMIX/CE2/ppdiffusers/stable_diffusion_controlnet/dynamic2static.sh index ce901b1805..673418c5fa 100644 --- a/models/PaddleMIX/CE2/ppdiffusers/stable_diffusion_controlnet/dynamic2static.sh +++ b/models/PaddleMIX/CE2/ppdiffusers/stable_diffusion_controlnet/dynamic2static.sh @@ -74,91 +74,6 @@ else fi echo "*******ppdiffusers/deploy/controlnet sd_controlnet_infer_inpaint_legacy_control end***********" -# paddle_tensorrt -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-canny \ - --scheduler "preconfig-euler-ancestral" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name text2img_control) 2>&1 | tee ${log_dir}/paddle_tensorrt_sd_controlnet_infer_inpaint_legacy_control.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet paddle_tensorrt sd_controlnet_infer_inpaint_legacy_control success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet paddle_tensorrt sd_controlnet_infer_inpaint_legacy_control fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet paddle_tensorrt sd_infer_inpaint_legacy_control end***********" - -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-canny \ - --scheduler "preconfig-euler-ancestral" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name img2img_control) 2>&1 | tee ${log_dir}/paddle_tensorrt_sd_controlnet_infer_img2img_control.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet paddle_tensorrt paddle_tensorrt_sd_controlnet_infer_img2img_control success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet paddle_tensorrt paddle_tensorrt_sd_controlnet_infer_img2img_control fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet paddle_tensorrt paddle_tensorrt_sd_controlnet_infer_img2img_control end***********" - -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5-canny \ - --scheduler "preconfig-euler-ancestral" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name inpaint_legacy_control) 2>&1 | tee ${log_dir}/paddle_tensorrt_sd_controlnet_inpaint_legacy_control.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet paddle_tensorrt paddle_tensorrt_sd_controlnet_inpaint_legacy_control success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet paddle_tensorrt paddle_tensorrt_sd_controlnet_inpaint_legacy_control fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet paddle_tensorrt paddle_tensorrt_sd_controlnet_inpaint_legacy_control end***********" - -(python ../utils/test_image_diff.py \ - --source_image ./infer_op_raw_fp16/text2img_control.png \ - --target_image https://paddlenlp.bj.bcebos.com/models/community/baicai/sd15_controlnet_infer_op_raw_fp16/text2img_control.png) 2>&1 | tee ${log_dir}/sd_controlnet_test_image_diff_text2img.log -python ${cur_path}/annalyse_log_tool.py \ - --file_path ${log_dir}/sd_controlnet_test_image_diff_text2img.log -tmp_exit_code=$? -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet sd_controlnet_test_image_diff_text2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet sd_controlnet_test_image_diff_text2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet sd_controlnet_test_image_diff_text2img end***********" - -(python ../utils/test_image_diff.py \ - --source_image ./infer_op_raw_fp16/img2img_control.png \ - --target_image https://paddlenlp.bj.bcebos.com/models/community/baicai/sd15_controlnet_infer_op_raw_fp16/img2img_control.png) 2>&1 | tee ${log_dir}/sd_controlnet_test_image_diff_img2img.log -python ${cur_path}/annalyse_log_tool.py --file_path ${log_dir}/sd_controlnet_test_image_diff_img2img.log -tmp_exit_code=$? -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet sd_controlnet_test_image_diff_img2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet sd_controlnet_test_image_diff_img2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet sd_controlnet_test_image_diff_img2img end***********" - -(python ../utils/test_image_diff.py \ - --source_image ./infer_op_raw_fp16/inpaint_legacy_control.png \ - --target_image https://paddlenlp.bj.bcebos.com/models/community/baicai/sd15_controlnet_infer_op_raw_fp16/inpaint_legacy_control.png) 2>&1 | tee ${log_dir}/sd_controlnet_test_image_diff_inpaint.log -python ${cur_path}/annalyse_log_tool.py --file_path ${log_dir}/sd_controlnet_test_image_diff_inpaint.log -tmp_exit_code=$? -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy/controlnet sd_controlnet_test_image_diff_inpaint success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy/controlnet sd_controlnet_test_image_diff_inpaint fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy/controlnet sd_controlnet_test_image_diff_inpaint end***********" echo exit_code:${exit_code} exit ${exit_code} diff --git a/models/PaddleMIX/CE2/ppdiffusers/stable_diffusion_deploy/dynamic2static.sh b/models/PaddleMIX/CE2/ppdiffusers/stable_diffusion_deploy/dynamic2static.sh index a9abac2ca5..7449ac4f70 100644 --- a/models/PaddleMIX/CE2/ppdiffusers/stable_diffusion_deploy/dynamic2static.sh +++ b/models/PaddleMIX/CE2/ppdiffusers/stable_diffusion_deploy/dynamic2static.sh @@ -72,90 +72,6 @@ else fi echo "*******ppdiffusers/deploy paddle sd_deploy_infer_inpaint_legacy end***********" -# paddle_tensorrt -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5 \ - --scheduler "preconfig-euler-ancestral" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name text2img) 2>&1 | tee ${log_dir}/paddle_tensorrt_sd_infer_text2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy paddle_tensorrt sd_deploy_infer_text2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy paddle_tensorrt sd_deploy_infer_text2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy paddle_tensorrt sd_deploy_infer_text2img end***********" - -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5 \ - --scheduler "preconfig-euler-ancestral" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name img2img) 2>&1 | tee ${log_dir}/paddle_tensorrt_sd_deploy_infer_img2img.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy paddle_tensorrt sd_deploy_infer_img2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy paddle_tensorrt sd_deploy_infer_img2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy paddle_tensorrt sd_deploy_infer_img2img end***********" - -(python infer.py \ - --model_dir static_model/stable-diffusion-v1-5 \ - --scheduler "preconfig-euler-ancestral" \ - --backend paddle_tensorrt \ - --device gpu \ - --task_name inpaint_legacy) 2>&1 | tee ${log_dir}/paddle_tensorrt_sd_deploy_infer_inpaint_legacy.log -tmp_exit_code=${PIPESTATUS[0]} -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy paddle_tensorrt sd_deploy_infer_inpaint_legacy success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy paddle_tensorrt sd_deploy_infer_inpaint_legacy fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy paddle_tensorrt sd_deploy_infer_inpaint_legacy end***********" - -(python ./utils/test_image_diff.py \ - --source_image ./infer_op_raw_fp16/text2img.png \ - --target_image https://paddlenlp.bj.bcebos.com/models/community/baicai/sd15_infer_op_raw_fp16/text2img.png) 2>&1 | tee ${log_dir}/sd_deploy_test_image_diff_text2img.log -python ${cur_path}/annalyse_log_tool.py --file_path ${log_dir}/sd_deploy_test_image_diff_text2img.log -tmp_exit_code=$? -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy sd_deploy_test_image_diff_text2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy sd_deploy_test_image_diff_text2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy sd_deploy_test_image_diff_text2img end***********" - -(python ./utils/test_image_diff.py \ - --source_image ./infer_op_raw_fp16/img2img.png \ - --target_image https://paddlenlp.bj.bcebos.com/models/community/baicai/sd15_infer_op_raw_fp16/img2img.png) 2>&1 | tee ${log_dir}/sd_deploy_test_image_diff_img2img.log -python ${cur_path}/annalyse_log_tool.py --file_path ${log_dir}/sd_deploy_test_image_diff_img2img.log -tmp_exit_code=$? -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy sd_deploy_test_image_diff_img2img success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy sd_deploy_test_image_diff_img2img fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy sd_deploy_test_image_diff_img2img end***********" - -(python ./utils/test_image_diff.py \ - --source_image ./infer_op_raw_fp16/inpaint_legacy.png \ - --target_image https://paddlenlp.bj.bcebos.com/models/community/baicai/sd15_infer_op_raw_fp16/inpaint_legacy.png) 2>&1 | tee ${log_dir}/sd_deploy_test_image_diff_inpaint.log -python ${cur_path}/annalyse_log_tool.py --file_path ${log_dir}/sd_deploy_test_image_diff_inpaint.log -tmp_exit_code=$? -exit_code=$(($exit_code + ${tmp_exit_code})) -if [ ${tmp_exit_code} -eq 0 ]; then - echo "ppdiffusers/deploy sd_deploy_test_image_diff_inpaint success" >>"${log_dir}/ce_res.log" -else - echo "ppdiffusers/deploy sd_deploy_test_image_diff_inpaint fail" >>"${log_dir}/ce_res.log" -fi -echo "*******ppdiffusers/deploy sd_deploy_test_image_diff_inpaint end***********" echo exit_code:${exit_code} exit ${exit_code} diff --git a/models/PaddleVideo/CI/test_video.sh b/models/PaddleVideo/CI/test_video.sh index 4d2ed28251..366ab6b851 100644 --- a/models/PaddleVideo/CI/test_video.sh +++ b/models/PaddleVideo/CI/test_video.sh @@ -83,28 +83,11 @@ INFER(){ --use_tensorrt=False >log/${model}/${model}_infer.log 2>&1 print_result } -TRT(){ - mode=trt - python tools/predict.py \ - --input_file data/k400/videos/abseiling/_UtLXOVn5Jk_000083_000093.mp4 \ - --config ${config} \ - --model_file inference/${model}/${model}.pdmodel \ - --params_file inference/${model}/${model}.pdiparams \ - --use_gpu=True \ - --use_tensorrt=True \ - --batch_size=${trt_bs} >log/${model}/${model}_trt.log 2>&1 - print_result -} model_list='TSM ppTSN' for model in ${model_list} do typeset -l model_small model_small=${model} -if [[ ${model} == 'TSM' ]];then -trt_bs=8 -else -trt_bs=2 -fi config=`cat model_list_video | grep ${model_small}` cd log mkdir ${model} @@ -113,7 +96,6 @@ TRAIN EVAL EXPORT INFER -TRT done if [ "${err_sign}" = true ];then exit 1 diff --git a/models/paddlecv/test_paddlecv.py b/models/paddlecv/test_paddlecv.py index 31f64343d7..0d4cc14fa9 100644 --- a/models/paddlecv/test_paddlecv.py +++ b/models/paddlecv/test_paddlecv.py @@ -39,7 +39,7 @@ def setup_module(): @allure.story("paddlecv_gpu_predict") @pytest.mark.parametrize("model_name", get_model_list()) -@pytest.mark.parametrize("run_mode", ["paddle", "trt_fp32", "trt_fp16", "trt_int8"]) +@pytest.mark.parametrize("run_mode", ["paddle"]) def test_paddlecv_gpu_predict(model_name, run_mode): """ test_paddlecv_gpu_predict diff --git a/models_restruct/PaddleClas/base/ImageNet_base.yaml b/models_restruct/PaddleClas/base/ImageNet_base.yaml index b893a1d2c7..1723f0706a 100644 --- a/models_restruct/PaddleClas/base/ImageNet_base.yaml +++ b/models_restruct/PaddleClas/base/ImageNet_base.yaml @@ -326,16 +326,6 @@ predict: - -o Global.use_gpu=${set_cuda_flag} - -o Global.output_dir=output/${qa_yaml_name}_predict_trained_mkldnn - -o Global.enable_mkldnn=True - - - name: function_trt - path: deploy - cmd: python python/predict_cls.py -c configs/inference_cls.yaml - params: - - -o Global.infer_imgs="./images" - - -o Global.inference_model_dir=${predict_trained_model} - - -o Global.use_gpu=${set_cuda_flag} - - -o Global.output_dir=output/${qa_yaml_name}_predict_trained_trt - - -o Global.use_tensorrt=True - name: trained path: deploy @@ -365,21 +355,6 @@ predict: base: "[11, 11, 11, 11]" threshold: 0 evaluation: "=" - - - name: trained_trt - path: deploy - cmd: python python/predict_cls.py -c configs/inference_cls.yaml - params: - - -o Global.infer_imgs="./images" - - -o Global.inference_model_dir=${predict_trained_model} - - -o Global.use_gpu=${set_cuda_flag} - - -o Global.output_dir=output/${qa_yaml_name}_predict_trained_trt - - -o Global.use_tensorrt=True - result: - class_ids: - base: "[11, 11, 11, 11]" - threshold: 0 - evaluation: "=" - name: pretrained path: deploy diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PPHGNet^PPHGNet_base.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PPHGNet^PPHGNet_base.yaml index 58c7dd15af..8f7d435607 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PPHGNet^PPHGNet_base.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PPHGNet^PPHGNet_base.yaml @@ -29,9 +29,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - linux_convergence: base: ./base/ImageNet_base.yaml train: @@ -73,9 +70,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - windows_cpu: base: ./base/ImageNet_base.yaml train: diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PPLCNetV2^PPLCNetV2_base.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PPLCNetV2^PPLCNetV2_base.yaml index 0234fd9375..a8fc2e822e 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PPLCNetV2^PPLCNetV2_base.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PPLCNetV2^PPLCNetV2_base.yaml @@ -46,8 +46,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - name: pretrained @@ -104,8 +102,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PeleeNet^PeleeNet.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PeleeNet^PeleeNet.yaml index 58c7dd15af..8f7d435607 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PeleeNet^PeleeNet.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^PeleeNet^PeleeNet.yaml @@ -29,9 +29,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - linux_convergence: base: ./base/ImageNet_base.yaml train: @@ -73,9 +70,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - windows_cpu: base: ./base/ImageNet_base.yaml train: diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50.yaml index ea58adcadc..86625b9f0f 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50.yaml @@ -35,8 +35,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - name: pretrained @@ -87,8 +85,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O1.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O1.yaml index e70a439b54..25d7a51050 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O1.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O1.yaml @@ -47,8 +47,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O1_ultra.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O1_ultra.yaml index cdea267c84..39a775558a 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O1_ultra.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O1_ultra.yaml @@ -55,9 +55,6 @@ case: - name: trained_mkldnn cmd: python python/predict_cls.py -c configs/inference_cls_ch4.yaml - - - name: trained_trt - cmd: python python/predict_cls.py -c configs/inference_cls_ch4.yaml - name: pretrained cmd: python python/predict_cls.py -c configs/inference_cls_ch4.yaml diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O2_ultra.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O2_ultra.yaml index 3ca5a2ce92..d9e17823fb 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O2_ultra.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^ResNet^ResNet50_amp_O2_ultra.yaml @@ -47,9 +47,6 @@ case: - name: trained_mkldnn cmd: python python/predict_cls.py -c configs/inference_cls_ch4.yaml - - - name: trained_trt - cmd: python python/predict_cls.py -c configs/inference_cls_ch4.yaml - name: pretrained cmd: python python/predict_cls.py -c configs/inference_cls_ch4.yaml diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_base_patch4_window12_384.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_base_patch4_window12_384.yaml index 32aac80baf..1d9f6bebf3 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_base_patch4_window12_384.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_base_patch4_window12_384.yaml @@ -47,8 +47,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_large_patch4_window12_384.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_large_patch4_window12_384.yaml index 32aac80baf..1d9f6bebf3 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_large_patch4_window12_384.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_large_patch4_window12_384.yaml @@ -47,8 +47,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_large_patch4_window7_224.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_large_patch4_window7_224.yaml index 32aac80baf..1d9f6bebf3 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_large_patch4_window7_224.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_large_patch4_window7_224.yaml @@ -47,8 +47,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_small_patch4_window7_224.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_small_patch4_window7_224.yaml index 18a1a1d074..c5337cf016 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_small_patch4_window7_224.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_small_patch4_window7_224.yaml @@ -35,8 +35,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_tiny_patch4_window7_224.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_tiny_patch4_window7_224.yaml index f8db9d0805..2327e05e40 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_tiny_patch4_window7_224.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^SwinTransformer^SwinTransformer_tiny_patch4_window7_224.yaml @@ -35,8 +35,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_base.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_base.yaml index dea249758c..bd9d617ef1 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_base.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_base.yaml @@ -84,8 +84,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_large.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_large.yaml index dea249758c..bd9d617ef1 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_large.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_large.yaml @@ -84,8 +84,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_small.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_small.yaml index dea249758c..bd9d617ef1 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_small.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^alt_gvt_small.yaml @@ -84,8 +84,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_base.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_base.yaml index dea249758c..bd9d617ef1 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_base.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_base.yaml @@ -84,8 +84,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_large.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_large.yaml index 4a038321e1..a1a2145964 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_large.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_large.yaml @@ -90,8 +90,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_small.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_small.yaml index a5eeaf89a9..c2aa0d91de 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_small.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^Twins^pcpvt_small.yaml @@ -82,8 +82,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - name: pretrained diff --git a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^VAN^VAN_B0.yaml b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^VAN^VAN_B0.yaml index b082fc8360..390802a5d6 100644 --- a/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^VAN^VAN_B0.yaml +++ b/models_restruct/PaddleClas/cases/ppcls^configs^ImageNet^VAN^VAN_B0.yaml @@ -23,9 +23,6 @@ case: name: trained - name: trained_mkldnn - - - name: trained_trt - linux_convergence: base: ./base/ImageNet_base.yaml train: @@ -67,9 +64,6 @@ case: name: function - name: function_mkldnn - - - name: function_trt - windows_cpu: base: ./base/ImageNet_base.yaml train: diff --git a/models_restruct/PaddleDetection/base/keypoint_base.yml b/models_restruct/PaddleDetection/base/keypoint_base.yml index c36b42d588..3415dd1dd5 100644 --- a/models_restruct/PaddleDetection/base/keypoint_base.yml +++ b/models_restruct/PaddleDetection/base/keypoint_base.yml @@ -153,36 +153,6 @@ predict: base: 0 threshold: 0 evaluation: "=" - - - name: trt_fp32 - cmd: python deploy/python/keypoint_infer.py - params: - - --model_dir=inference_model/${model} - - --image_file=demo/000000570688.jpg - - --device=GPU - - --run_mode=trt_fp32 - - --threshold=0.5 - - --output_dir=python_infer_trtfp32_output/${model}/ - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" - - - name: trt_fp16 - cmd: python deploy/python/keypoint_infer.py - params: - - --model_dir=inference_model/${model} - - --image_file=demo/000000570688.jpg - - --device=GPU - - --run_mode=trt_fp16 - - --threshold=0.5 - - --output_dir=python_infer_trtfp16_output/${model}/ - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" - name: paddle2onnx cmd: paddle2onnx diff --git a/models_restruct/PaddleDetection/base/mot_base.yml b/models_restruct/PaddleDetection/base/mot_base.yml index 185fabbb20..81961ca84d 100644 --- a/models_restruct/PaddleDetection/base/mot_base.yml +++ b/models_restruct/PaddleDetection/base/mot_base.yml @@ -154,36 +154,6 @@ predict: base: 0 threshold: 0 evaluation: "=" - - - name: trt_fp32 - cmd: python deploy/pptracking/python/mot_jde_infer.py - params: - - --model_dir=inference_model/${model} - - --image_file=demo/000000570688.jpg - - --device=GPU - - --run_mode=trt_fp32 - - --threshold=0.5 - - --output_dir=python_infer_trtfp32_output/${model}/ - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" - - - name: trt_fp16 - cmd: python deploy/pptracking/python/mot_jde_infer.py - params: - - --model_dir=inference_model/${model} - - --image_file=demo/000000570688.jpg - - --device=GPU - - --run_mode=trt_fp16 - - --threshold=0.5 - - --output_dir=python_infer_trtfp16_output/${model}/ - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" - name: paddle2onnx cmd: paddle2onnx diff --git a/models_restruct/PaddleDetection/base/normal_base.yml b/models_restruct/PaddleDetection/base/normal_base.yml index 77e82e5170..d5dd3f6b6e 100644 --- a/models_restruct/PaddleDetection/base/normal_base.yml +++ b/models_restruct/PaddleDetection/base/normal_base.yml @@ -158,36 +158,6 @@ predict: base: 0 threshold: 0 evaluation: "=" - - - name: trt_fp32 - cmd: python deploy/python/infer.py - params: - - --model_dir=inference_model/${model} - - --image_file=demo/000000570688.jpg - - --device=GPU - - --run_mode=trt_fp32 - - --threshold=0.5 - - --output_dir=python_infer_trtfp32_output/${model}/ - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" - - - name: trt_fp16 - cmd: python deploy/python/infer.py - params: - - --model_dir=inference_model/${model} - - --image_file=demo/000000570688.jpg - - --device=GPU - - --run_mode=trt_fp16 - - --threshold=0.5 - - --output_dir=python_infer_trtfp16_output/${model}/ - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" - name: paddle2onnx cmd: paddle2onnx diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_lcnet_1_5x_416_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_lcnet_1_5x_416_coco.yml index c3da7d3be3..97e3cccf78 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_lcnet_1_5x_416_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_lcnet_1_5x_416_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_s_320_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_s_320_coco.yml index c3da7d3be3..97e3cccf78 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_s_320_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_s_320_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_320_coco_lcnet.yml b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_320_coco_lcnet.yml index c3da7d3be3..97e3cccf78 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_320_coco_lcnet.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_320_coco_lcnet.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_mbv3_large_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_mbv3_large_coco.yml index 935996bc50..7011fa6957 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_mbv3_large_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_mbv3_large_coco.yml @@ -45,10 +45,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_1x_coco.yml index 063027a453..bf897770d8 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_1x_coco.yml @@ -45,10 +45,6 @@ case: name: python # - # name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 # - # name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_tiny_650e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_tiny_650e_coco.yml index c3da7d3be3..97e3cccf78 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_tiny_650e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_tiny_650e_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolov2_r50vd_dcn_365e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolov2_r50vd_dcn_365e_coco.yml index 5a32064a7a..719d33dc89 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolov2_r50vd_dcn_365e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolov2_r50vd_dcn_365e_coco.yml @@ -47,10 +47,6 @@ case: name: python # - # name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 # - # name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_s_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_s_300e_coco.yml index c3da7d3be3..97e3cccf78 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_s_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_s_300e_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_s_80e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_s_80e_coco.yml index c3da7d3be3..97e3cccf78 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_s_80e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_s_80e_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r50_fpn_1x_coco.yml index 0cf3815fb1..4de36d7313 100644 --- a/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r50_fpn_1x_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 # - # name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^smalldet^ppyoloe_plus_sod_crn_l_80e_coco.yml b/models_restruct/PaddleDetection/cases/configs^smalldet^ppyoloe_plus_sod_crn_l_80e_coco.yml index c3da7d3be3..97e3cccf78 100644 --- a/models_restruct/PaddleDetection/cases/configs^smalldet^ppyoloe_plus_sod_crn_l_80e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^smalldet^ppyoloe_plus_sod_crn_l_80e_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^vitdet^ppyoloe_vit_base_csppan_cae_36e_coco.yml b/models_restruct/PaddleDetection/cases/configs^vitdet^ppyoloe_vit_base_csppan_cae_36e_coco.yml index c3da7d3be3..97e3cccf78 100644 --- a/models_restruct/PaddleDetection/cases/configs^vitdet^ppyoloe_vit_base_csppan_cae_36e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^vitdet^ppyoloe_vit_base_csppan_cae_36e_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/test.yml b/models_restruct/PaddleDetection/cases/test.yml index c3da7d3be3..97e3cccf78 100644 --- a/models_restruct/PaddleDetection/cases/test.yml +++ b/models_restruct/PaddleDetection/cases/test.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/test_configs^ppyolo^ppyolo_mbv3_large_coco.yml b/models_restruct/PaddleDetection/cases/test_configs^ppyolo^ppyolo_mbv3_large_coco.yml index 935996bc50..7011fa6957 100644 --- a/models_restruct/PaddleDetection/cases/test_configs^ppyolo^ppyolo_mbv3_large_coco.yml +++ b/models_restruct/PaddleDetection/cases/test_configs^ppyolo^ppyolo_mbv3_large_coco.yml @@ -45,10 +45,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/test_configs^ppyolo^ppyolo_r50vd_dcn_1x_coco.yml b/models_restruct/PaddleDetection/cases/test_configs^ppyolo^ppyolo_r50vd_dcn_1x_coco.yml index 063027a453..bf897770d8 100644 --- a/models_restruct/PaddleDetection/cases/test_configs^ppyolo^ppyolo_r50vd_dcn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/test_configs^ppyolo^ppyolo_r50vd_dcn_1x_coco.yml @@ -45,10 +45,6 @@ case: name: python # - # name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 # - # name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/diy_build/PaddleDetection_Build.py b/models_restruct/PaddleDetection/diy_build/PaddleDetection_Build.py index a173c632da..7d70ab728a 100644 --- a/models_restruct/PaddleDetection/diy_build/PaddleDetection_Build.py +++ b/models_restruct/PaddleDetection/diy_build/PaddleDetection_Build.py @@ -214,8 +214,6 @@ def build_paddledetection(self): wget.download(paddle_inference) os.system("tar -xf paddle_inference.tgz") os.system('sed -i "s|WITH_GPU=OFF|WITH_GPU=ON|g" scripts/build.sh') - os.system('sed -i "s|/path/to/tensorrt/lib|/usr/local/TensorRT-7.0.0.11/lib|g" scripts/build.sh') - os.system('sed -i "s|/path/to/tensorrt/include|/usr/local/TensorRT-7.0.0.11/include|g" scripts/build.sh') os.system('sed -i "s|CUDA_LIB=/path/to/cuda/lib|CUDA_LIB=/usr/local/cuda/lib64|g" scripts/build.sh') os.system('sed -i "s|/path/to/paddle_inference|../paddle_inference|g" scripts/build.sh') os.system('sed -i "s|CUDNN_LIB=/path/to/cudnn/lib|CUDNN_LIB=/usr/lib/x86_64-linux-gnu|g" scripts/build.sh') diff --git a/models_restruct/PaddleNLP/cases/model_zoo^ernie-3.0-fastdepoly.yaml b/models_restruct/PaddleNLP/cases/model_zoo^ernie-3.0-fastdepoly.yaml index 89798078d8..1f5b77af33 100644 --- a/models_restruct/PaddleNLP/cases/model_zoo^ernie-3.0-fastdepoly.yaml +++ b/models_restruct/PaddleNLP/cases/model_zoo^ernie-3.0-fastdepoly.yaml @@ -15,15 +15,6 @@ case: params: - --model_dir ../../best_models/afqmc/export - --device cpu --backend paddle - - - name: Qunt-GPU - path: slm/model_zoo/ernie-3.0/deploy/python/ - cmd: python seq_cls_infer.py - params: - - --model_dir ../../best_models/afqmc/width_mult_0.75/mse16_1/ - - --device gpu - - --backend tensorrt - - --model_prefix int8 - name: Qunt-CPU path: slm/model_zoo/ernie-3.0/deploy/python/ diff --git a/models_restruct/PaddleNLP/cases/model_zoo^ernie-tiny-fastdeploy.yaml b/models_restruct/PaddleNLP/cases/model_zoo^ernie-tiny-fastdeploy.yaml index a0d114fd33..39b92a1d24 100644 --- a/models_restruct/PaddleNLP/cases/model_zoo^ernie-tiny-fastdeploy.yaml +++ b/models_restruct/PaddleNLP/cases/model_zoo^ernie-tiny-fastdeploy.yaml @@ -112,14 +112,6 @@ case: name: fastdeploy_python_cpu path: slm/model_zoo/ernie-tiny/deploy/python cmd: python infer_demo.py --device cpu --backend paddle --model_dir ../../output/BS64_LR5e-5_EPOCHS30 --slot_label_path ../../data/slot_label.txt --intent_label_path ../../data/intent_label.txt - - - name: deploy_python_compress_gpu - path: slm/model_zoo/ernie-tiny/deploy/python - cmd: python infer_demo.py --device gpu --backend paddle_tensorrt --model_prefix int8 --model_dir ../../output/BS64_LR5e-5_EPOCHS30/ --slot_label_path ../../data/slot_label.txt --intent_label_path ../../data/intent_label.txt - - - name: deploy_python_compress_cpu - path: slm/model_zoo/ernie-tiny/deploy/python - cmd: python infer_demo.py --device cpu --backend paddle_tensorrt --model_prefix int8 --model_dir ../../output/BS64_LR5e-5_EPOCHS30/ --slot_label_path ../../data/slot_label.txt --intent_label_path ../../data/intent_label.txt - name: deploy_cpp_prepare path: slm/model_zoo/ernie-tiny/deploy/cpp @@ -136,15 +128,6 @@ case: name: deploy_cpp_cpu path: slm/model_zoo/ernie-tiny/deploy/cpp/build cmd: ./infer_demo --device cpu --backend paddle --model_dir ../../../output/BS64_LR5e-5_EPOCHS30 --slot_label_path ../../../data/slot_label.txt --intent_label_path ../../../data/intent_label.txt - - - name: deploy_cpp_compress_gpu - path: slm/model_zoo/ernie-tiny/deploy/cpp/build - cmd: ./infer_demo --device gpu --backend paddle_tensorrt --model_prefix int8 --model_dir ../../../output/BS64_LR5e-5_EPOCHS30 --slot_label_path ../../../data/slot_label.txt --intent_label_path ../../../data/intent_label.txt - - - name: deploy_cpp_compress_cpu - path: slm/model_zoo/ernie-tiny/deploy/cpp/build - cmd: ./infer_demo --device cpu --backend paddle_tensorrt --model_prefix int8 --model_dir ../../../output/BS64_LR5e-5_EPOCHS30 --slot_label_path ../../../data/slot_label.txt --intent_label_path ../../../data/intent_label.txt - windows: train: skipped eval: skipped diff --git a/models_restruct/PaddleOCR/base/ocr_cls_base.yaml b/models_restruct/PaddleOCR/base/ocr_cls_base.yaml index 5b103fc464..f0a33d01a9 100755 --- a/models_restruct/PaddleOCR/base/ocr_cls_base.yaml +++ b/models_restruct/PaddleOCR/base/ocr_cls_base.yaml @@ -163,15 +163,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python tools/infer/predict_cls.py - params: - - --image_dir="doc/imgs_words_en/word_10.png" - - --cls_model_dir="./models_inference/"${qa_yaml_name} - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False api: name: trained_C_plus_plus_GPU path: deploy/cpp_infer diff --git a/models_restruct/PaddleOCR/base/ocr_cls_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_cls_base_pretrained.yaml index a6b02b85ee..09ece540aa 100755 --- a/models_restruct/PaddleOCR/base/ocr_cls_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_cls_base_pretrained.yaml @@ -163,15 +163,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python tools/infer/predict_cls.py - params: - - --image_dir="doc/imgs_words_en/word_10.png" - - --cls_model_dir="./models_inference/"${qa_yaml_name} - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - name: pretrained cmd: python tools/infer/predict_cls.py @@ -190,15 +181,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: pretrained_tensorRT - cmd: python tools/infer/predict_cls.py - params: - - --image_dir="doc/imgs_words_en/word_10.png" - - --cls_model_dir="./models_inference/"${model} - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False api: name: trained_C_plus_plus_GPU path: deploy/cpp_infer diff --git a/models_restruct/PaddleOCR/base/ocr_det_base.yaml b/models_restruct/PaddleOCR/base/ocr_det_base.yaml index 96d92b59c8..aa200916dd 100755 --- a/models_restruct/PaddleOCR/base/ocr_det_base.yaml +++ b/models_restruct/PaddleOCR/base/ocr_det_base.yaml @@ -129,17 +129,6 @@ predict: - --use_tensorrt=False - --enable_mkldnn=True - --det_model_dir=./models_inference/${qa_yaml_name} - - - name: trained_tensorRT - cmd: python tools/infer/predict_det.py - params: - - --image_dir=doc/imgs_en/img_10.jpg - - --det_algorithm=${algorithm} - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - - --det_model_dir=./models_inference/${qa_yaml_name} - api: name: trained_C_plus_plus_GPU path: deploy/cpp_infer diff --git a/models_restruct/PaddleOCR/base/ocr_det_base_distill.yaml b/models_restruct/PaddleOCR/base/ocr_det_base_distill.yaml index c068ddfebf..f310787200 100755 --- a/models_restruct/PaddleOCR/base/ocr_det_base_distill.yaml +++ b/models_restruct/PaddleOCR/base/ocr_det_base_distill.yaml @@ -129,17 +129,6 @@ predict: - --use_tensorrt=False - --enable_mkldnn=True - --det_model_dir=./models_inference/${qa_yaml_name}/Student - - - name: trained_tensorRT - cmd: python tools/infer/predict_det.py - params: - - --image_dir=doc/imgs_en/img_10.jpg - - --det_algorithm=${algorithm} - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - - --det_model_dir=./models_inference/${qa_yaml_name}/Student - api: name: trained_C_plus_plus_GPU path: deploy/cpp_infer diff --git a/models_restruct/PaddleOCR/base/ocr_det_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_det_base_pretrained.yaml index 82b721b765..7ab9c5f0cc 100755 --- a/models_restruct/PaddleOCR/base/ocr_det_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_det_base_pretrained.yaml @@ -159,16 +159,6 @@ predict: - --use_tensorrt=False - --enable_mkldnn=True - --det_model_dir=./models_inference/${qa_yaml_name} - - - name: trained_tensorRT - cmd: python tools/infer/predict_det.py - params: - - --image_dir=doc/imgs_en/img_10.jpg - - --det_algorithm=${algorithm} - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - - --det_model_dir=./models_inference/${qa_yaml_name} - name: pretrained cmd: python tools/infer/predict_det.py @@ -189,17 +179,6 @@ predict: - --use_tensorrt=False - --enable_mkldnn=True - --det_model_dir=./models_inference/${model} - - - name: pretrained_tensorRT - cmd: python tools/infer/predict_det.py - params: - - --image_dir=doc/imgs_en/img_10.jpg - - --det_algorithm=${algorithm} - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - - --det_model_dir=./models_inference/${model} - api: name: pretrained_C_plus_plus_GPU path: deploy/cpp_infer diff --git a/models_restruct/PaddleOCR/base/ocr_e2e_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_e2e_base_pretrained.yaml index 5a3d90dae1..8bca67d507 100755 --- a/models_restruct/PaddleOCR/base/ocr_e2e_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_e2e_base_pretrained.yaml @@ -185,16 +185,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python tools/infer/predict_e2e.py - params: - - --image_dir="./doc/imgs_en/img623.jpg" - - --e2e_model_dir="./models_inference/"${qa_yaml_name} - - --e2e_algorithm=PGNet - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - name: pretrained cmd: python tools/infer/predict_e2e.py @@ -215,18 +205,3 @@ predict: - --use_gpu=$False - --use_tensorrt=False - --enable_mkldnn=True - - - name: pretrained_tensorRT - cmd: python tools/infer/predict_e2e.py - params: - - --image_dir="./doc/imgs_en/img623.jpg" - - --e2e_model_dir="./models_inference/"${model} - - --e2e_algorithm=PGNet - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" diff --git a/models_restruct/PaddleOCR/base/ocr_kie_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_kie_base_pretrained.yaml index eb4be31bc0..ed5e2cc8dc 100755 --- a/models_restruct/PaddleOCR/base/ocr_kie_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_kie_base_pretrained.yaml @@ -162,19 +162,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python ppstructure/kie/predict_${kie_token}.py - params: - - --kie_algorithm=LayoutXLM - - --ser_model_dir="./models_inference/"${qa_yaml_name} - - --image_dir=./ppstructure/docs/kie/input/zh_val_42.jpg - - --ser_dict_path=./train_data/XFUND/class_list_xfun.txt - - --vis_font_path=./doc/fonts/simfang.ttf - - --ocr_order_method="tb-yx" - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - name: pretrained cmd: python ppstructure/kie/predict_${kie_token}.py @@ -201,21 +188,3 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: pretrained_tensorRT - cmd: python ppstructure/kie/predict_${kie_token}.py - params: - - --kie_algorithm=LayoutXLM - - --ser_model_dir=./models_inference/${model} - - --image_dir=./ppstructure/docs/kie/input/zh_val_42.jpg - - --ser_dict_path=./train_data/XFUND/class_list_xfun.txt - - --vis_font_path=./doc/fonts/simfang.ttf - - --ocr_order_method=tb-yx - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" diff --git a/models_restruct/PaddleOCR/base/ocr_rec_base.yaml b/models_restruct/PaddleOCR/base/ocr_rec_base.yaml index 94f7892f4c..dd4ce19374 100755 --- a/models_restruct/PaddleOCR/base/ocr_rec_base.yaml +++ b/models_restruct/PaddleOCR/base/ocr_rec_base.yaml @@ -152,20 +152,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python tools/infer/predict_rec.py - params: - - --image_dir="./doc/imgs_words_en/word_336.png" - - --rec_model_dir="./models_inference/"${qa_yaml_name} - - --rec_image_shape=${image_shape} - - --rec_algorithm=${algorithm} - - --rec_char_dict_path=${rec_dict} - - --use_space_char=False - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - api: name: trained_C_plus_plus_GPU path: deploy/cpp_infer diff --git a/models_restruct/PaddleOCR/base/ocr_rec_base_distill.yaml b/models_restruct/PaddleOCR/base/ocr_rec_base_distill.yaml index 20ceb98dc8..7283daa515 100755 --- a/models_restruct/PaddleOCR/base/ocr_rec_base_distill.yaml +++ b/models_restruct/PaddleOCR/base/ocr_rec_base_distill.yaml @@ -158,21 +158,3 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python tools/infer/predict_rec.py - params: - - --image_dir="./doc/imgs_words_en/word_336.png" - - --rec_model_dir="./models_inference/"${qa_yaml_name}/Student - - --rec_image_shape=${image_shape} - - --rec_algorithm=${algorithm} - - --rec_char_dict_path=${rec_dict} - - --use_space_char=False - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" diff --git a/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained.yaml index f63abcc88a..3da219f535 100755 --- a/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained.yaml @@ -181,19 +181,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python tools/infer/predict_rec.py - params: - - --image_dir="./doc/imgs_words_en/word_336.png" - - --rec_model_dir="./models_inference/"${qa_yaml_name} - - --rec_image_shape=${image_shape} - - --rec_algorithm=${algorithm} - - --rec_char_dict_path=${rec_dict} - - --use_space_char=False - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - name: pretrained cmd: python tools/infer/predict_rec.py @@ -220,20 +207,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: pretrained_tensorRT - cmd: python tools/infer/predict_rec.py - params: - - --image_dir=./doc/imgs_words_en/word_336.png - - --rec_model_dir=./models_inference/${model} - - --rec_image_shape=${image_shape} - - --rec_algorithm=${algorithm} - - --rec_char_dict_path=${rec_dict} - - --use_space_char=False - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - api: name: pretrained_C_plus_plus_GPU path: deploy/cpp_infer diff --git a/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained_distill.yaml b/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained_distill.yaml index 4fa3634b6c..3480b2254d 100755 --- a/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained_distill.yaml +++ b/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained_distill.yaml @@ -187,19 +187,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python tools/infer/predict_rec.py - params: - - --image_dir="./doc/imgs_words_en/word_336.png" - - --rec_model_dir="./models_inference/"${qa_yaml_name}/Student - - --rec_image_shape=${image_shape} - - --rec_algorithm=${algorithm} - - --rec_char_dict_path=${rec_dict} - - --use_space_char=False - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - name: pretrained cmd: python tools/infer/predict_rec.py @@ -226,21 +213,3 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: pretrained_tensorRT - cmd: python tools/infer/predict_rec.py - params: - - --image_dir=./doc/imgs_words_en/word_336.png - - --rec_model_dir=./models_inference/${model}/Student - - --rec_image_shape=${image_shape} - - --rec_algorithm=${algorithm} - - --rec_char_dict_path=${rec_dict} - - --use_space_char=False - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" diff --git a/models_restruct/PaddleOCR/base/ocr_sr_base.yaml b/models_restruct/PaddleOCR/base/ocr_sr_base.yaml index f7e60df87b..da86fa2aca 100755 --- a/models_restruct/PaddleOCR/base/ocr_sr_base.yaml +++ b/models_restruct/PaddleOCR/base/ocr_sr_base.yaml @@ -99,18 +99,3 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python tools/infer/predict_sr.py - params: - - --image_dir="./doc/imgs_words_en/word_52.png" - - --sr_model_dir="./models_inference/"${qa_yaml_name} - - --sr_image_shape=${image_shape} - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" diff --git a/models_restruct/PaddleOCR/base/ocr_sr_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_sr_base_pretrained.yaml index 4147b1d1bd..958a887f3e 100755 --- a/models_restruct/PaddleOCR/base/ocr_sr_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_sr_base_pretrained.yaml @@ -125,16 +125,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python tools/infer/predict_sr.py - params: - - --image_dir="./doc/imgs_words_en/word_52.png" - - --sr_model_dir="./models_inference/"${qa_yaml_name} - - --sr_image_shape=${image_shape} - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - name: pretrained cmd: python tools/infer/predict_sr.py @@ -155,18 +145,3 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: pretrained_tensorRT - cmd: python tools/infer/predict_sr.py - params: - - --image_dir="./doc/imgs_words_en/word_52.png" - - --sr_model_dir="./models_inference/"${model} - - --sr_image_shape=${image_shape} - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" diff --git a/models_restruct/PaddleOCR/base/ocr_table_base.yaml b/models_restruct/PaddleOCR/base/ocr_table_base.yaml index 9cdc697a0e..ac2e3838ad 100755 --- a/models_restruct/PaddleOCR/base/ocr_table_base.yaml +++ b/models_restruct/PaddleOCR/base/ocr_table_base.yaml @@ -157,20 +157,3 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python ppstructure/table/predict_structure.py - params: - - --image_dir="ppstructure/docs/table/table.jpg" - - --table_model_dir="./models_inference/"${qa_yaml_name} - - --table_char_dict_path=${rec_dict} - - --table_max_len=488 - - --vis_font_path=./doc/fonts/simfang.ttf - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" diff --git a/models_restruct/PaddleOCR/base/ocr_table_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_table_base_pretrained.yaml index b09a99fa27..088f866ea7 100755 --- a/models_restruct/PaddleOCR/base/ocr_table_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_table_base_pretrained.yaml @@ -186,18 +186,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: pretrained_tensorRT - cmd: python ppstructure/table/predict_structure.py - params: - - --image_dir="ppstructure/docs/table/table.jpg" - - --table_model_dir="./models_inference/"${model} - - --table_char_dict_path=${rec_dict} - - --table_max_len=488 - - --vis_font_path=./doc/fonts/simfang.ttf - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - name: trained cmd: python ppstructure/table/predict_structure.py @@ -222,18 +210,6 @@ predict: - --use_gpu=False - --use_tensorrt=False - --enable_mkldnn=True - - - name: trained_tensorRT - cmd: python ppstructure/table/predict_structure.py - params: - - --image_dir="ppstructure/docs/table/table.jpg" - - --table_model_dir="./models_inference/"${qa_yaml_name} - - --table_char_dict_path=${rec_dict} - - --table_max_len=488 - - --vis_font_path=./doc/fonts/simfang.ttf - - --use_gpu=True - - --use_tensorrt=True - - --enable_mkldnn=False - name: pretrained_C_plus_plus_GPU path: deploy/cpp_infer diff --git a/models_restruct/PaddleOCR/tools/start.py b/models_restruct/PaddleOCR/tools/start.py index 6e10fd18db..6c7d21c682 100755 --- a/models_restruct/PaddleOCR/tools/start.py +++ b/models_restruct/PaddleOCR/tools/start.py @@ -188,8 +188,6 @@ def gengrate_test_case(self): " name: trained" + os.linesep, " -" + os.linesep, " name: trained_mkldnn" + os.linesep, - " -" + os.linesep, - " name: trained_tensorRT" + os.linesep, " windows:" + os.linesep, " base: ./base/ocr_" + self.category + "_base_distill.yaml" + os.linesep, " windows_cpu:" + os.linesep, @@ -233,11 +231,7 @@ def gengrate_test_case(self): " -" + os.linesep, " name: trained_mkldnn" + os.linesep, " -" + os.linesep, - " name: trained_tensorRT" + os.linesep, - " -" + os.linesep, " name: pretrained_mkldnn" + os.linesep, - " -" + os.linesep, - " name: pretrained_tensorRT" + os.linesep, " windows:" + os.linesep, " base: ./base/ocr_" + self.category + "_base_pretrained.yaml" + os.linesep, " train:" + os.linesep, @@ -261,11 +255,7 @@ def gengrate_test_case(self): " -" + os.linesep, " name: trained_mkldnn" + os.linesep, " -" + os.linesep, - " name: trained_tensorRT" + os.linesep, - " -" + os.linesep, " name: pretrained_mkldnn" + os.linesep, - " -" + os.linesep, - " name: pretrained_tensorRT" + os.linesep, " windows_cpu:" + os.linesep, " base: ./base/ocr_" + self.category + "_base_pretrained.yaml" + os.linesep, " train:" + os.linesep, @@ -338,8 +328,6 @@ def gengrate_test_case(self): " name: trained" + os.linesep, " -" + os.linesep, " name: trained_mkldnn" + os.linesep, - " -" + os.linesep, - " name: trained_tensorRT" + os.linesep, " windows:" + os.linesep, " base: ./base/ocr_" + self.category + "_base.yaml" + os.linesep, " train:" + os.linesep, @@ -358,8 +346,6 @@ def gengrate_test_case(self): " name: trained" + os.linesep, " -" + os.linesep, " name: trained_mkldnn" + os.linesep, - " -" + os.linesep, - " name: trained_tensorRT" + os.linesep, " windows_cpu:" + os.linesep, " base: ./base/ocr_" + self.category + "_base.yaml" + os.linesep, " train:" + os.linesep, diff --git a/models_restruct/PaddleSeg/base/normal_base.yml b/models_restruct/PaddleSeg/base/normal_base.yml index 9e84ac6e9f..c3ce444f2d 100644 --- a/models_restruct/PaddleSeg/base/normal_base.yml +++ b/models_restruct/PaddleSeg/base/normal_base.yml @@ -174,36 +174,6 @@ predict: base: 0 threshold: 0 evaluation: "=" - - - name: trt_fp32 - cmd: python deploy/python/infer.py - params: - - --config ./inference_model/${model}/deploy.yaml - - --image_path ./demo/${image} - - --save_dir ./python_infer_output/${model}/ - - --precision=fp32 - - --device=gpu - - --use_trt=True - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" - - - name: trt_fp16 - cmd: python deploy/python/infer.py - params: - - --config ./inference_model/${model}/deploy.yaml - - --image_path ./demo/${image} - - --save_dir ./python_infer_output/${model}/ - - --precision=fp16 - - --device=gpu - - --use_trt=True - result: - exit_code: - base: 0 - threshold: 0 - evaluation: "=" - name: paddle2onnx cmd: paddle2onnx diff --git a/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet101_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet101_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet101_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet101_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet101_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet101_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet101_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet101_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet50_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ann^ann_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^attention_unet^attention_unet_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^attention_unet^attention_unet_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^attention_unet^attention_unet_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^attention_unet^attention_unet_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^bisenet^bisenet_cityscapes_1024x1024_160k.yml b/models_restruct/PaddleSeg/cases/configs^bisenet^bisenet_cityscapes_1024x1024_160k.yml index 1c76f3ef59..d50352ab06 100644 --- a/models_restruct/PaddleSeg/cases/configs^bisenet^bisenet_cityscapes_1024x1024_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^bisenet^bisenet_cityscapes_1024x1024_160k.yml @@ -43,10 +43,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^bisenetv1^bisenetv1_resnet18_os8_cityscapes_1024x512_160k.yml b/models_restruct/PaddleSeg/cases/configs^bisenetv1^bisenetv1_resnet18_os8_cityscapes_1024x512_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^bisenetv1^bisenetv1_resnet18_os8_cityscapes_1024x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^bisenetv1^bisenetv1_resnet18_os8_cityscapes_1024x512_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ccnet^ccnet_resnet101_os8_cityscapes_769x769_60k.yml b/models_restruct/PaddleSeg/cases/configs^ccnet^ccnet_resnet101_os8_cityscapes_769x769_60k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ccnet^ccnet_resnet101_os8_cityscapes_769x769_60k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ccnet^ccnet_resnet101_os8_cityscapes_769x769_60k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^danet^danet_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^danet^danet_resnet50_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^danet^danet_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^danet^danet_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^danet^danet_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^danet^danet_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^danet^danet_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^danet^danet_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ddrnet^ddrnet23_cityscapes_1024x1024_120k.yml b/models_restruct/PaddleSeg/cases/configs^ddrnet^ddrnet23_cityscapes_1024x1024_120k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ddrnet^ddrnet23_cityscapes_1024x1024_120k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ddrnet^ddrnet23_cityscapes_1024x1024_120k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^decoupled_segnet^decoupledsegnet_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^decoupled_segnet^decoupledsegnet_resnet50_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^decoupled_segnet^decoupledsegnet_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^decoupled_segnet^decoupledsegnet_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^decoupled_segnet^decoupledsegnet_resnet50_os8_cityscapes_832x832_80k.yml b/models_restruct/PaddleSeg/cases/configs^decoupled_segnet^decoupledsegnet_resnet50_os8_cityscapes_832x832_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^decoupled_segnet^decoupledsegnet_resnet50_os8_cityscapes_832x832_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^decoupled_segnet^decoupledsegnet_resnet50_os8_cityscapes_832x832_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet101_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet101_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet101_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet101_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet101_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet101_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet101_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet101_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet50_os8_cityscapes_1024x512_80k.yml index 8a4c2cada6..4a1467a6f4 100644 --- a/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet50_os8_cityscapes_1024x512_80k.yml @@ -43,10 +43,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^deeplabv3^deeplabv3_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_cityscapes_769x769_80k.yml b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_cityscapes_769x769_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_cityscapes_769x769_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_cityscapes_769x769_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet101_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_cityscapes_1024x512_80k.yml index e07d00d7f0..1577fd9e4d 100644 --- a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_cityscapes_1024x512_80k_rmiloss.yml b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_cityscapes_1024x512_80k_rmiloss.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_cityscapes_1024x512_80k_rmiloss.yml +++ b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_cityscapes_1024x512_80k_rmiloss.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_voc12aug_512x512_40k.yml index 13062ae355..29ee5a7ea9 100644 --- a/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^deeplabv3p^deeplabv3p_resnet50_os8_voc12aug_512x512_40k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^dmnet^dmnet_resnet101_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^dmnet^dmnet_resnet101_os8_cityscapes_1024x512_80k.yml index 6cf5523426..e6ebd4d39f 100644 --- a/models_restruct/PaddleSeg/cases/configs^dmnet^dmnet_resnet101_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^dmnet^dmnet_resnet101_os8_cityscapes_1024x512_80k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet101_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet101_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet101_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet101_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet101_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet101_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet101_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet101_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet50_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^dnlnet^dnlnet_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet101_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet101_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet101_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet101_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet101_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet101_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet101_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet101_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet50_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^emanet^emanet_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^encnet^encnet_resnet101_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^encnet^encnet_resnet101_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^encnet^encnet_resnet101_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^encnet^encnet_resnet101_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^enet^enet_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^enet^enet_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^enet^enet_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^enet^enet_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^espnet^espnet_cityscapes_1024x512_120k.yml b/models_restruct/PaddleSeg/cases/configs^espnet^espnet_cityscapes_1024x512_120k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^espnet^espnet_cityscapes_1024x512_120k.yml +++ b/models_restruct/PaddleSeg/cases/configs^espnet^espnet_cityscapes_1024x512_120k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^espnetv1^espnetv1_cityscapes_1024x512_120k.yml b/models_restruct/PaddleSeg/cases/configs^espnetv1^espnetv1_cityscapes_1024x512_120k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^espnetv1^espnetv1_cityscapes_1024x512_120k.yml +++ b/models_restruct/PaddleSeg/cases/configs^espnetv1^espnetv1_cityscapes_1024x512_120k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^fastfcn^fastfcn_resnet50_os8_ade20k_480x480_120k.yml b/models_restruct/PaddleSeg/cases/configs^fastfcn^fastfcn_resnet50_os8_ade20k_480x480_120k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^fastfcn^fastfcn_resnet50_os8_ade20k_480x480_120k.yml +++ b/models_restruct/PaddleSeg/cases/configs^fastfcn^fastfcn_resnet50_os8_ade20k_480x480_120k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^fastscnn^fastscnn_cityscapes_1024x1024_160k.yml b/models_restruct/PaddleSeg/cases/configs^fastscnn^fastscnn_cityscapes_1024x1024_160k.yml index 178c6798ca..6ac88c23d9 100644 --- a/models_restruct/PaddleSeg/cases/configs^fastscnn^fastscnn_cityscapes_1024x1024_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^fastscnn^fastscnn_cityscapes_1024x1024_160k.yml @@ -43,10 +43,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw18_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw18_cityscapes_1024x512_80k.yml index 93ab1402d5..49503949a0 100644 --- a/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw18_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw18_cityscapes_1024x512_80k.yml @@ -45,10 +45,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw18_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw18_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw18_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw18_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw48_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw48_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw48_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw48_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw48_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw48_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw48_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^fcn^fcn_hrnetw48_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet101_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet101_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet101_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet101_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet101_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet101_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet101_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet101_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet50_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^gcnet^gcnet_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_ade20k_520x520_150k.yml b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_ade20k_520x520_150k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_ade20k_520x520_150k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_ade20k_520x520_150k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet101_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_ade20k_520x520_150k.yml b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_ade20k_520x520_150k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_ade20k_520x520_150k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_ade20k_520x520_150k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ginet^ginet_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^glore^glore_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^glore^glore_resnet50_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^glore^glore_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^glore^glore_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^glore^glore_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^glore^glore_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^glore^glore_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^glore^glore_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^hardnet^hardnet_cityscapes_1024x1024_160k.yml b/models_restruct/PaddleSeg/cases/configs^hardnet^hardnet_cityscapes_1024x1024_160k.yml index ef304cbcc2..9f12e54a4d 100644 --- a/models_restruct/PaddleSeg/cases/configs^hardnet^hardnet_cityscapes_1024x1024_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^hardnet^hardnet_cityscapes_1024x1024_160k.yml @@ -43,10 +43,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^hrnet_w48_contrast^HRNet_W48_contrast_cityscapes_1024x512_60k.yml b/models_restruct/PaddleSeg/cases/configs^hrnet_w48_contrast^HRNet_W48_contrast_cityscapes_1024x512_60k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^hrnet_w48_contrast^HRNet_W48_contrast_cityscapes_1024x512_60k.yml +++ b/models_restruct/PaddleSeg/cases/configs^hrnet_w48_contrast^HRNet_W48_contrast_cityscapes_1024x512_60k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet101_os8_cityscapes_769x769_80k.yml b/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet101_os8_cityscapes_769x769_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet101_os8_cityscapes_769x769_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet101_os8_cityscapes_769x769_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet101_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet101_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet101_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet101_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet50_os8_cityscapes_769x769_80k.yml b/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet50_os8_cityscapes_769x769_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet50_os8_cityscapes_769x769_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet50_os8_cityscapes_769x769_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^isanet^isanet_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k_large_kernel.yml b/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k_large_kernel.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k_large_kernel.yml +++ b/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k_large_kernel.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k_os32.yml b/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k_os32.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k_os32.yml +++ b/models_restruct/PaddleSeg/cases/configs^lraspp^lraspp_mobilenetv3_cityscapes_1024x512_80k_os32.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_ghostnet_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_ghostnet_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_ghostnet_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_ghostnet_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_litehrnet18_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_litehrnet18_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_litehrnet18_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_litehrnet18_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_mobilenetv2_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_mobilenetv2_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_mobilenetv2_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_mobilenetv2_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_mobilenetv3_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_mobilenetv3_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_mobilenetv3_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_mobilenetv3_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_shufflenetv2_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_shufflenetv2_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_shufflenetv2_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^mobileseg^mobileseg_shufflenetv2_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw18_cityscapes_1024x512_160k.yml b/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw18_cityscapes_1024x512_160k.yml index 13062ae355..29ee5a7ea9 100644 --- a/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw18_cityscapes_1024x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw18_cityscapes_1024x512_160k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw18_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw18_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw18_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw18_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw48_cityscapes_1024x512_160k.yml b/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw48_cityscapes_1024x512_160k.yml index 13062ae355..29ee5a7ea9 100644 --- a/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw48_cityscapes_1024x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw48_cityscapes_1024x512_160k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw48_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw48_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw48_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^ocrnet^ocrnet_hrnetw48_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pfpn^pfpn_resnet101_os8_cityscapes_512x1024_40k.yml b/models_restruct/PaddleSeg/cases/configs^pfpn^pfpn_resnet101_os8_cityscapes_512x1024_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pfpn^pfpn_resnet101_os8_cityscapes_512x1024_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pfpn^pfpn_resnet101_os8_cityscapes_512x1024_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet101_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet101_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet101_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet101_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet50_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pointrend^pointrend_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_camvid_960x720_10k.yml b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_camvid_960x720_10k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_camvid_960x720_10k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_camvid_960x720_10k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale0.5_160k.yml b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale0.5_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale0.5_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale0.5_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale0.75_160k.yml b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale0.75_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale0.75_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale0.75_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale1.0_160k.yml b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale1.0_160k.yml index 13062ae355..29ee5a7ea9 100644 --- a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale1.0_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale1.0_160k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_camvid_960x720_10k.yml b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_camvid_960x720_10k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_camvid_960x720_10k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_camvid_960x720_10k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.5_160k.yml b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.5_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.5_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.5_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.75_160k.yml b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.75_160k.yml index 13062ae355..29ee5a7ea9 100644 --- a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.75_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.75_160k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale1.0_160k.yml b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale1.0_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale1.0_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale1.0_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet101_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet101_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet101_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet101_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet101_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet101_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet101_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet101_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet50_os8_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet50_os8_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet50_os8_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet50_os8_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet50_os8_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^pspnet^pspnet_resnet50_os8_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^rtformer^rtformer_slim_ade20k_512x512_160k.yml b/models_restruct/PaddleSeg/cases/configs^rtformer^rtformer_slim_ade20k_512x512_160k.yml index 6cf5523426..e6ebd4d39f 100644 --- a/models_restruct/PaddleSeg/cases/configs^rtformer^rtformer_slim_ade20k_512x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^rtformer^rtformer_slim_ade20k_512x512_160k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^rtformer^rtformer_slim_cityscapes_1024x512_120k.yml b/models_restruct/PaddleSeg/cases/configs^rtformer^rtformer_slim_cityscapes_1024x512_120k.yml index 6cf5523426..e6ebd4d39f 100644 --- a/models_restruct/PaddleSeg/cases/configs^rtformer^rtformer_slim_cityscapes_1024x512_120k.yml +++ b/models_restruct/PaddleSeg/cases/configs^rtformer^rtformer_slim_cityscapes_1024x512_120k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml b/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b1_cityscapes_1024x1024_160k.yml b/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b1_cityscapes_1024x1024_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b1_cityscapes_1024x1024_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b1_cityscapes_1024x1024_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b2_cityscapes_1024x1024_160k.yml b/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b2_cityscapes_1024x1024_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b2_cityscapes_1024x1024_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b2_cityscapes_1024x1024_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b3_cityscapes_1024x1024_160k.yml b/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b3_cityscapes_1024x1024_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b3_cityscapes_1024x1024_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b3_cityscapes_1024x1024_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b4_cityscapes_1024x1024_160k.yml b/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b4_cityscapes_1024x1024_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b4_cityscapes_1024x1024_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^segformer^segformer_b4_cityscapes_1024x1024_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^segmenter^segmenter_vit_small_linear_ade20k_512x512_160k.yml b/models_restruct/PaddleSeg/cases/configs^segmenter^segmenter_vit_small_linear_ade20k_512x512_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^segmenter^segmenter_vit_small_linear_ade20k_512x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^segmenter^segmenter_vit_small_linear_ade20k_512x512_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^segmenter^segmenter_vit_small_mask_ade20k_512x512_160k.yml b/models_restruct/PaddleSeg/cases/configs^segmenter^segmenter_vit_small_mask_ade20k_512x512_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^segmenter^segmenter_vit_small_mask_ade20k_512x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^segmenter^segmenter_vit_small_mask_ade20k_512x512_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^sfnet^sfnet_resnet18_os8_cityscapes_1024x1024_80k.yml b/models_restruct/PaddleSeg/cases/configs^sfnet^sfnet_resnet18_os8_cityscapes_1024x1024_80k.yml index 90bd2de910..89ad6546c3 100644 --- a/models_restruct/PaddleSeg/cases/configs^sfnet^sfnet_resnet18_os8_cityscapes_1024x1024_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^sfnet^sfnet_resnet18_os8_cityscapes_1024x1024_80k.yml @@ -43,10 +43,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^sfnet^sfnet_resnet50_os8_cityscapes_1024x1024_80k.yml b/models_restruct/PaddleSeg/cases/configs^sfnet^sfnet_resnet50_os8_cityscapes_1024x1024_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^sfnet^sfnet_resnet50_os8_cityscapes_1024x1024_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^sfnet^sfnet_resnet50_os8_cityscapes_1024x1024_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc1_seg_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc1_seg_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc1_seg_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc1_seg_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc1_seg_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc1_seg_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc1_seg_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc1_seg_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc2_seg_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc2_seg_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc2_seg_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc2_seg_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc2_seg_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc2_seg_voc12aug_512x512_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc2_seg_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^stdcseg^stdc2_seg_voc12aug_512x512_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^topformer^topformer_small_ade20k_512x512_160k.yml b/models_restruct/PaddleSeg/cases/configs^topformer^topformer_small_ade20k_512x512_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^topformer^topformer_small_ade20k_512x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^topformer^topformer_small_ade20k_512x512_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^topformer^topformer_tiny_ade20k_512x512_160k.yml b/models_restruct/PaddleSeg/cases/configs^topformer^topformer_tiny_ade20k_512x512_160k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^topformer^topformer_tiny_ade20k_512x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^topformer^topformer_tiny_ade20k_512x512_160k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw18_small_cityscapes_1024x512_120k_bs3.yml b/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw18_small_cityscapes_1024x512_120k_bs3.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw18_small_cityscapes_1024x512_120k_bs3.yml +++ b/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw18_small_cityscapes_1024x512_120k_bs3.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw18_small_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw18_small_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw18_small_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw18_small_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw48_cityscapes_1024x512_120k_bs3.yml b/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw48_cityscapes_1024x512_120k_bs3.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw48_cityscapes_1024x512_120k_bs3.yml +++ b/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw48_cityscapes_1024x512_120k_bs3.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw48_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw48_cityscapes_1024x512_80k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw48_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/configs^uhrnet^fcn_uhrnetw48_cityscapes_1024x512_80k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^unet^unet_cityscapes_1024x512_160k.yml b/models_restruct/PaddleSeg/cases/configs^unet^unet_cityscapes_1024x512_160k.yml index bd5e0c402f..ed5b3c2764 100644 --- a/models_restruct/PaddleSeg/cases/configs^unet^unet_cityscapes_1024x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/configs^unet^unet_cityscapes_1024x512_160k.yml @@ -43,10 +43,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/configs^upernet^upernet_resnet101_os8_cityscapes_512x1024_40k.yml b/models_restruct/PaddleSeg/cases/configs^upernet^upernet_resnet101_os8_cityscapes_512x1024_40k.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/configs^upernet^upernet_resnet101_os8_cityscapes_512x1024_40k.yml +++ b/models_restruct/PaddleSeg/cases/configs^upernet^upernet_resnet101_os8_cityscapes_512x1024_40k.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/fix_input.yml b/models_restruct/PaddleSeg/cases/fix_input.yml index 6cf5523426..e6ebd4d39f 100644 --- a/models_restruct/PaddleSeg/cases/fix_input.yml +++ b/models_restruct/PaddleSeg/cases/fix_input.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/prim^configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml b/models_restruct/PaddleSeg/cases/prim^configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml index e00c095925..4a3082564b 100644 --- a/models_restruct/PaddleSeg/cases/prim^configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml +++ b/models_restruct/PaddleSeg/cases/prim^configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/static^configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml b/models_restruct/PaddleSeg/cases/static^configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml index b1558230f2..0d11338234 100644 --- a/models_restruct/PaddleSeg/cases/static^configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml +++ b/models_restruct/PaddleSeg/cases/static^configs^segformer^segformer_b0_cityscapes_1024x1024_160k.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/test.yml b/models_restruct/PaddleSeg/cases/test.yml index 64bbf82232..e5e7d10c62 100644 --- a/models_restruct/PaddleSeg/cases/test.yml +++ b/models_restruct/PaddleSeg/cases/test.yml @@ -33,10 +33,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/test_configs^deeplabv3p^deeplabv3p_resnet50_os8_voc12aug_512x512_40k.yml b/models_restruct/PaddleSeg/cases/test_configs^deeplabv3p^deeplabv3p_resnet50_os8_voc12aug_512x512_40k.yml index 013fa72313..b1fc966724 100644 --- a/models_restruct/PaddleSeg/cases/test_configs^deeplabv3p^deeplabv3p_resnet50_os8_voc12aug_512x512_40k.yml +++ b/models_restruct/PaddleSeg/cases/test_configs^deeplabv3p^deeplabv3p_resnet50_os8_voc12aug_512x512_40k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/test_configs^fcn^fcn_hrnetw18_cityscapes_1024x512_80k.yml b/models_restruct/PaddleSeg/cases/test_configs^fcn^fcn_hrnetw18_cityscapes_1024x512_80k.yml index 013fa72313..b1fc966724 100644 --- a/models_restruct/PaddleSeg/cases/test_configs^fcn^fcn_hrnetw18_cityscapes_1024x512_80k.yml +++ b/models_restruct/PaddleSeg/cases/test_configs^fcn^fcn_hrnetw18_cityscapes_1024x512_80k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/test_configs^ocrnet^ocrnet_hrnetw18_cityscapes_1024x512_160k.yml b/models_restruct/PaddleSeg/cases/test_configs^ocrnet^ocrnet_hrnetw18_cityscapes_1024x512_160k.yml index 13062ae355..29ee5a7ea9 100644 --- a/models_restruct/PaddleSeg/cases/test_configs^ocrnet^ocrnet_hrnetw18_cityscapes_1024x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/test_configs^ocrnet^ocrnet_hrnetw18_cityscapes_1024x512_160k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/test_configs^ocrnet^ocrnet_hrnetw48_cityscapes_1024x512_160k.yml b/models_restruct/PaddleSeg/cases/test_configs^ocrnet^ocrnet_hrnetw48_cityscapes_1024x512_160k.yml index 13062ae355..29ee5a7ea9 100644 --- a/models_restruct/PaddleSeg/cases/test_configs^ocrnet^ocrnet_hrnetw48_cityscapes_1024x512_160k.yml +++ b/models_restruct/PaddleSeg/cases/test_configs^ocrnet^ocrnet_hrnetw48_cityscapes_1024x512_160k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/test_configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale1.0_160k.yml b/models_restruct/PaddleSeg/cases/test_configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale1.0_160k.yml index 013fa72313..b1fc966724 100644 --- a/models_restruct/PaddleSeg/cases/test_configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale1.0_160k.yml +++ b/models_restruct/PaddleSeg/cases/test_configs^pp_liteseg^pp_liteseg_stdc1_cityscapes_1024x512_scale1.0_160k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleSeg/cases/test_configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.75_160k.yml b/models_restruct/PaddleSeg/cases/test_configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.75_160k.yml index 013fa72313..b1fc966724 100644 --- a/models_restruct/PaddleSeg/cases/test_configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.75_160k.yml +++ b/models_restruct/PaddleSeg/cases/test_configs^pp_liteseg^pp_liteseg_stdc2_cityscapes_1024x512_scale0.75_160k.yml @@ -35,10 +35,6 @@ case: name: python - name: mkldnn - - - name: trt_fp32 - - - name: trt_fp16 - name: paddle2onnx - diff --git a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_bert.py b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_bert.py index 694e2792ee..185db63fdb 100644 --- a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_bert.py +++ b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_bert.py @@ -116,121 +116,3 @@ def test_mkldnn(): test_suite2.mkldnn_test(input_data_dict, output_data_dict, delta=1e-5) del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_trt_fp16_bz1(): - """ - compared trt fp16 batch_size=1 bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config(model_file="./bert/inference.pdmodel", params_file="./bert/inference.pdiparams") - data_path = "./bert/data.txt" - images_list = test_suite.get_text_npy(data_path) - - input_data_dict = { - "input_ids": np.array([images_list[0][0]]).astype("int64"), - "token_type_ids": np.array([images_list[0][1]]).astype("int64"), - } - output_data_dict = test_suite.get_truth_val(input_data_dict, device="cpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./bert/inference.pdmodel", params_file="./bert/inference.pdiparams") - test_suite2.trt_more_bz_test(input_data_dict, output_data_dict, delta=1e-5, max_batch_size=1, precision="trt_fp16") - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 multi_thread bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config(model_file="./bert/inference.pdmodel", params_file="./bert/inference.pdiparams") - data_path = "./bert/data.txt" - images_list = test_suite.get_text_npy(data_path) - - input_data_dict = { - "input_ids": np.array([images_list[0][0]]).astype("int64"), - "token_type_ids": np.array([images_list[0][1]]).astype("int64"), - } - output_data_dict = test_suite.get_truth_val(input_data_dict, device="cpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./bert/inference.pdmodel", params_file="./bert/inference.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, delta=1e-5, precision="trt_fp16") - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_trt_fp32_bz1(): - """ - compared trt fp32 batch_size=1 bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config(model_file="./bert/inference.pdmodel", params_file="./bert/inference.pdiparams") - data_path = "./bert/data.txt" - images_list = test_suite.get_text_npy(data_path) - - input_data_dict = { - "input_ids": np.array([images_list[0][0]]).astype("int64"), - "token_type_ids": np.array([images_list[0][1]]).astype("int64"), - } - output_data_dict = test_suite.get_truth_val(input_data_dict, device="cpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./bert/inference.pdmodel", params_file="./bert/inference.pdiparams") - test_suite2.trt_more_bz_test(input_data_dict, output_data_dict, delta=1e-5, max_batch_size=1, precision="trt_fp32") - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 multi_thread bert outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config(model_file="./bert/inference.pdmodel", params_file="./bert/inference.pdiparams") - data_path = "./bert/data.txt" - images_list = test_suite.get_text_npy(data_path) - - input_data_dict = { - "input_ids": np.array([images_list[0][0]]).astype("int64"), - "token_type_ids": np.array([images_list[0][1]]).astype("int64"), - } - output_data_dict = test_suite.get_truth_val(input_data_dict, device="cpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./bert/inference.pdmodel", params_file="./bert/inference.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, delta=1e-5, precision="trt_fp32") - - del test_suite2 # destroy class to save memory diff --git a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_case/infer_test.py b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_case/infer_test.py index 287708425d..e8cf66b228 100644 --- a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_case/infer_test.py +++ b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_case/infer_test.py @@ -248,416 +248,6 @@ def gpu_more_bz_test(self, input_data_dict: dict, output_data_dict: dict, repeat diff <= delta ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - def trt_bz1_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=5, - delta=1e-5, - gpu_mem=1000, - min_subgraph_size=10, - precision="trt_fp32", - use_static=False, - use_calib_mode=False, - ): - """ - test enable_use_gpu() - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - - output_names = predictor.get_output_names() - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data = output_data.flatten() - output_data_truth_val = output_data_dict[output_data_name].flatten() - for j, out_data in enumerate(output_data): - diff = sig_fig_compare(out_data, output_data_truth_val[j]) - assert ( - diff <= delta - ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - - predictor.try_shrink_memory() # try_shrink_memory - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - - output_names = predictor.get_output_names() - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data = output_data.flatten() - output_data_truth_val = output_data_dict[output_data_name].flatten() - for j, out_data in enumerate(output_data): - diff = sig_fig_compare(out_data, output_data_truth_val[j]) - assert ( - diff <= delta - ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - - def trt_more_bz_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - delta=1e-5, - gpu_mem=1000, - max_batch_size=3, - min_subgraph_size=10, - precision="fp32", - use_static=True, - use_calib_mode=False, - dynamic=False, - tuned=False, - ): - """ - test enable_tensorrt_engine() - batch_size = 10 - trt max_batch_size = 10 - precision_mode = fp32,fp16,int8 - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - if dynamic: - if tuned: - self.pd_config.collect_shape_range_info("shape_range.pbtxt") - return 0 - else: - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - self.pd_config.enable_tuned_tensorrt_dynamic_shape("shape_range.pbtxt", True) - else: - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - - output_names = predictor.get_output_names() - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data = output_data.flatten() - output_data_truth_val = output_data_dict[output_data_name].flatten() - for j, out_data in enumerate(output_data): - diff = sig_fig_compare(out_data, output_data_truth_val[j]) - assert ( - diff <= delta - ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - - def trt_more_bz_dynamic_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - delta=1e-5, - gpu_mem=1000, - max_batch_size=10, - names=None, - min_input_shape=None, - max_input_shape=None, - opt_input_shape=None, - min_subgraph_size=10, - precision="trt_fp32", - use_static=False, - use_calib_mode=False, - ): - """ - test enable_tensorrt_engine() - max_batch_size = 1-10 - trt max_batch_size = 10 - precision_mode = fp32,fp16,int8 - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - names(list): input names - min_input_shape(list): TensorRT min input shape - max_input_shape(list): TensorRT max input shape - opt_input_shape(list): TensorRT best input shape - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - - self.pd_config.set_trt_dynamic_shape_info( - {names[i]: min_input_shape[i] for i in range(len(names))}, - {names[i]: max_input_shape[i] for i in range(len(names))}, - {names[i]: opt_input_shape[i] for i in range(len(names))}, - ) - - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - - output_names = predictor.get_output_names() - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data = output_data.flatten() - output_data_truth_val = output_data_dict[output_data_name].flatten() - for j, out_data in enumerate(output_data): - diff = sig_fig_compare(out_data, output_data_truth_val[j]) - assert ( - diff <= delta - ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - predictor.try_shrink_memory() - - def trt_bz1_multi_thread_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - thread_num=2, - delta=1e-5, - gpu_mem=1000, - min_subgraph_size=10, - precision="trt_fp32", - use_static=False, - use_calib_mode=False, - ): - """ - test enable_tensorrt_engine() - batch_size = 1 - trt max_batch_size = 4 - thread_num = 5 - precision_mode = fp32,fp16,int8 - Multithreading TensorRT predictor - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time - thread_num(int): number of threads - delta(float): difference threshold between inference outputs and thruth value - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - predictors = paddle_infer.PredictorPool(self.pd_config, thread_num) - for i in range(thread_num): - record_thread = threading.Thread( - target=self.run_multi_thread_test_predictor, - args=(predictors.retrive(i), input_data_dict, output_data_dict, repeat, delta), - ) - record_thread.start() - record_thread.join() - - def trt_dynamic_multi_thread_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - delta=1e-5, - thread_num=2, - gpu_mem=1000, - max_batch_size=1, - names=None, - min_input_shape=None, - max_input_shape=None, - opt_input_shape=None, - min_subgraph_size=10, - precision="trt_fp32", - use_static=False, - use_calib_mode=False, - ): - """ - test enable_tensorrt_engine() - batch_size = 1 - trt max_batch_size = 1 - thread_num = 2 - precision_mode = fp32,fp16,int8 - Multithreading TensorRT predictor - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time - delta(float): difference threshold between inference outputs and thruth value - names(list): input names - min_input_shape(list): TensorRT min input shape - max_input_shape(list): TensorRT max input shape - opt_input_shape(list): TensorRT best input shape - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - self.pd_config.set_trt_dynamic_shape_info( - {names[i]: min_input_shape[i] for i in range(len(names))}, - {names[i]: max_input_shape[i] for i in range(len(names))}, - {names[i]: opt_input_shape[i] for i in range(len(names))}, - ) - predictors = paddle_infer.PredictorPool(self.pd_config, thread_num) - for i in range(thread_num): - record_thread = threading.Thread( - target=self.run_multi_thread_test_predictor, - args=(predictors.retrive(i), input_data_dict, output_data_dict, repeat, delta), - ) - record_thread.start() - record_thread.join() - - def run_multi_thread_test_predictor( - self, predictor, input_data_dict: dict, output_data_dict: dict, repeat=1, delta=1e-5 - ): - """ - test paddle predictor in multithreaded task - Args: - predictor: paddle inference predictor - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - Returns: - None - """ - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - output_names = predictor.get_output_names() - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data = output_data.flatten() - output_data_truth_val = output_data_dict[output_data_name].flatten() - for j, out_data in enumerate(output_data): - diff = sig_fig_compare(out_data, output_data_truth_val[j]) - assert ( - diff <= delta - ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - def get_gpu_mem(gpu_id=0): """ diff --git a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_deeplabv3.py b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_deeplabv3.py index 90f05a5e7d..38884c7bf1 100644 --- a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_deeplabv3.py +++ b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_deeplabv3.py @@ -97,7 +97,7 @@ def test_config(): def test_gpu_bz1(): """ - compared trt gpu batch_size=1 deeplabv3p_resnet50 outputs with true val + compared gpu batch_size=1 deeplabv3p_resnet50 outputs with true val """ check_model_exist() diff --git a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_ernie.py b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_ernie.py index 7a6f9a4325..30d3050f71 100644 --- a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_ernie.py +++ b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_ernie.py @@ -116,121 +116,3 @@ def test_mkldnn(): test_suite2.mkldnn_test(input_data_dict, output_data_dict, delta=1e-5) del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp16 -def test_trt_fp16_bz1(): - """ - compared trt fp16 batch_size=1 ernie outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config(model_file="./ernie/inference.pdmodel", params_file="./ernie/inference.pdiparams") - data_path = "./ernie/data.txt" - images_list = test_suite.get_text_npy(data_path) - - input_data_dict = { - "input_ids": np.array([images_list[0][0]]).astype("int64"), - "token_type_ids": np.array([images_list[0][1]]).astype("int64"), - } - output_data_dict = test_suite.get_truth_val(input_data_dict, device="cpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./ernie/inference.pdmodel", params_file="./ernie/inference.pdiparams") - test_suite2.trt_more_bz_test(input_data_dict, output_data_dict, delta=1e-5, max_batch_size=1, precision="trt_fp16") - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16_multi_thread -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 multi_thread ernie outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config(model_file="./ernie/inference.pdmodel", params_file="./ernie/inference.pdiparams") - data_path = "./ernie/data.txt" - images_list = test_suite.get_text_npy(data_path) - - input_data_dict = { - "input_ids": np.array([images_list[0][0]]).astype("int64"), - "token_type_ids": np.array([images_list[0][1]]).astype("int64"), - } - output_data_dict = test_suite.get_truth_val(input_data_dict, device="cpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./ernie/inference.pdmodel", params_file="./ernie/inference.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, delta=1e-5, precision="trt_fp16") - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.jetson -@pytest.mark.trt_fp32 -def test_trt_fp32_bz1(): - """ - compared trt fp32 batch_size=1 ernie outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config(model_file="./ernie/inference.pdmodel", params_file="./ernie/inference.pdiparams") - data_path = "./ernie/data.txt" - images_list = test_suite.get_text_npy(data_path) - - input_data_dict = { - "input_ids": np.array([images_list[0][0]]).astype("int64"), - "token_type_ids": np.array([images_list[0][1]]).astype("int64"), - } - output_data_dict = test_suite.get_truth_val(input_data_dict, device="cpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./ernie/inference.pdmodel", params_file="./ernie/inference.pdiparams") - test_suite2.trt_more_bz_test(input_data_dict, output_data_dict, delta=1e-5, max_batch_size=1, precision="trt_fp32") - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 multi_thread ernie outputs with true val - """ - check_model_exist() - - test_suite = InferenceTest() - test_suite.load_config(model_file="./ernie/inference.pdmodel", params_file="./ernie/inference.pdiparams") - data_path = "./ernie/data.txt" - images_list = test_suite.get_text_npy(data_path) - - input_data_dict = { - "input_ids": np.array([images_list[0][0]]).astype("int64"), - "token_type_ids": np.array([images_list[0][1]]).astype("int64"), - } - output_data_dict = test_suite.get_truth_val(input_data_dict, device="cpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./ernie/inference.pdmodel", params_file="./ernie/inference.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, delta=1e-5, precision="trt_fp32") - - del test_suite2 # destroy class to save memory diff --git a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_mobilenetv1.py b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_mobilenetv1.py index b38201353c..3fe0ddb494 100644 --- a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_mobilenetv1.py +++ b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_mobilenetv1.py @@ -59,120 +59,16 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.p1 -@pytest.mark.trt_fp32_more_bz_precision -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-10 MobileNetV1 outputs with true val - """ - check_model_exist() - - file_path = "./MobileNetV1" - images_size = 224 - batch_size_pool = [1, 5, 10] - max_batch_size = 10 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - test_suite2.trt_more_bz_test( - input_data_dict, output_data_dict, max_batch_size=max_batch_size, precision="trt_fp32" - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.p1 -@pytest.mark.trt_fp32_multi_thread_bz1_precision -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 MobileNetV1 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./MobileNetV1" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, precision="trt_fp32") - del test_suite2 # destroy class to save memory -@pytest.mark.p1 -@pytest.mark.trt_fp16_more_bz_precision -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-10 MobileNetV1 outputs with true val - """ - check_model_exist() - file_path = "./MobileNetV1" - images_size = 224 - batch_size_pool = [1, 5, 10] - max_batch_size = 10 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - test_suite2.trt_more_bz_test( - input_data_dict, output_data_dict, delta=1e-2, max_batch_size=max_batch_size, precision="trt_fp16" - ) - del test_suite2 # destroy class to save memory -@pytest.mark.p1 -@pytest.mark.trt_fp16_multi_thread_bz1_precision -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 MobileNetV1 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./MobileNetV1" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, precision="trt_fp16") - - del test_suite2 # destroy class to save memory @pytest.mark.p1 @@ -206,7 +102,7 @@ def test_mkldnn(): @pytest.mark.gpu_bz1_precision def test_gpu_bz1(): """ - compared trt gpu batch_size=1-10 MobileNetV1 outputs with true val + compared gpu batch_size=1-10 MobileNetV1 outputs with true val """ check_model_exist() diff --git a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_ocr.py b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_ocr.py index 2c06e4bbcd..9dcdde0a30 100644 --- a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_ocr.py +++ b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_ocr.py @@ -67,7 +67,7 @@ def test_disable_gpu(): @pytest.mark.gpu_bz1 def test_gpu_bz1(): """ - compared trt fp32 batch_size=1,2 ocr_det_mv3_db outputs with true val + compared gpu batch_size=1,2 ocr_det_mv3_db outputs with true val """ check_model_exist() @@ -100,7 +100,7 @@ def test_gpu_bz1(): @pytest.mark.gpu_more def test_jetson_gpu_more_bz(): """ - compared trt fp32 more batch_size ocr_det_mv3_db outputs with true val + compared gpu more batch_size ocr_det_mv3_db outputs with true val """ check_model_exist() @@ -159,415 +159,3 @@ def test_mkldnn(): test_suite2.mkldnn_test(input_data_dict, output_data_dict, delta=1e-4) del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16 -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-10 ocr_det_mv3_db outputs with true val - """ - check_model_exist() - - file_path = "./ocr_det_mv3_db" - images_size = 640 - batch_size_pool = [1, 5, 10] - max_batch_size = 10 - names = [ - "x", - "conv2d_92.tmp_0", - "conv2d_91.tmp_0", - "conv2d_59.tmp_0", - "nearest_interp_v2_1.tmp_0", - "nearest_interp_v2_2.tmp_0", - "conv2d_124.tmp_0", - "nearest_interp_v2_3.tmp_0", - "nearest_interp_v2_4.tmp_0", - "nearest_interp_v2_5.tmp_0", - "elementwise_add_7", - "nearest_interp_v2_0.tmp_0", - ] - min_input_shape = [ - [1, 3, 50, 50], - [1, 120, 20, 20], - [1, 24, 10, 10], - [1, 96, 20, 20], - [1, 256, 10, 10], - [1, 256, 20, 20], - [1, 256, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 56, 2, 2], - [1, 256, 2, 2], - ] - - max_input_shape = [ - [max_batch_size, 3, 2000, 2000], - [max_batch_size, 120, 400, 400], - [max_batch_size, 24, 200, 200], - [max_batch_size, 96, 400, 400], - [max_batch_size, 256, 200, 200], - [max_batch_size, 256, 400, 400], - [max_batch_size, 256, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 56, 400, 400], - [max_batch_size, 256, 400, 400], - ] - - opt_input_shape = [ - [1, 3, 640, 640], - [1, 120, 160, 160], - [1, 24, 80, 80], - [1, 96, 160, 160], - [1, 256, 80, 80], - [1, 256, 160, 160], - [1, 256, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 56, 40, 40], - [1, 256, 40, 40], - ] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", params_file="./ocr_det_mv3_db/inference.pdiparams" - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", params_file="./ocr_det_mv3_db/inference.pdiparams" - ) - test_suite2.trt_more_bz_dynamic_test( - input_data_dict, - output_data_dict, - gpu_mem=5000, - max_batch_size=max_batch_size, - repeat=1, - delta=9e-2, - names=names, - min_input_shape=min_input_shape, - max_input_shape=max_input_shape, - opt_input_shape=opt_input_shape, - precision="trt_fp16", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp16_multi_thread -def test_trtfp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 ocr_det_mv3_db multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ocr_det_mv3_db" - images_size = 640 - batch_size = 1 - max_batch_size = 1 - names = [ - "x", - "conv2d_92.tmp_0", - "conv2d_91.tmp_0", - "conv2d_59.tmp_0", - "nearest_interp_v2_1.tmp_0", - "nearest_interp_v2_2.tmp_0", - "conv2d_124.tmp_0", - "nearest_interp_v2_3.tmp_0", - "nearest_interp_v2_4.tmp_0", - "nearest_interp_v2_5.tmp_0", - "elementwise_add_7", - "nearest_interp_v2_0.tmp_0", - ] - min_input_shape = [ - [1, 3, 50, 50], - [1, 120, 20, 20], - [1, 24, 10, 10], - [1, 96, 20, 20], - [1, 256, 10, 10], - [1, 256, 20, 20], - [1, 256, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 56, 2, 2], - [1, 256, 2, 2], - ] - - max_input_shape = [ - [max_batch_size, 3, 2000, 2000], - [max_batch_size, 120, 400, 400], - [max_batch_size, 24, 200, 200], - [max_batch_size, 96, 400, 400], - [max_batch_size, 256, 200, 200], - [max_batch_size, 256, 400, 400], - [max_batch_size, 256, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 56, 400, 400], - [max_batch_size, 256, 400, 400], - ] - - opt_input_shape = [ - [1, 3, 640, 640], - [1, 120, 160, 160], - [1, 24, 80, 80], - [1, 96, 160, 160], - [1, 256, 80, 80], - [1, 256, 160, 160], - [1, 256, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 56, 40, 40], - [1, 256, 40, 40], - ] - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", params_file="./ocr_det_mv3_db/inference.pdiparams" - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", params_file="./ocr_det_mv3_db/inference.pdiparams" - ) - test_suite2.trt_dynamic_multi_thread_test( - input_data_dict, - output_data_dict, - gpu_mem=5000, - max_batch_size=max_batch_size, - repeat=1, - delta=9e-2, - names=names, - min_input_shape=min_input_shape, - max_input_shape=max_input_shape, - opt_input_shape=opt_input_shape, - precision="trt_fp16", - ) - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32_multi_thread -def test_trt_fp32_bz1_dynamic_multi_thread(): - """ - compared trt fp32 batch_size=1 ocr_det_mv3_db multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./ocr_det_mv3_db" - images_size = 640 - batch_size = 1 - max_batch_size = 1 - names = [ - "x", - "conv2d_92.tmp_0", - "conv2d_91.tmp_0", - "conv2d_59.tmp_0", - "nearest_interp_v2_1.tmp_0", - "nearest_interp_v2_2.tmp_0", - "conv2d_124.tmp_0", - "nearest_interp_v2_3.tmp_0", - "nearest_interp_v2_4.tmp_0", - "nearest_interp_v2_5.tmp_0", - "elementwise_add_7", - "nearest_interp_v2_0.tmp_0", - ] - min_input_shape = [ - [1, 3, 50, 50], - [1, 120, 20, 20], - [1, 24, 10, 10], - [1, 96, 20, 20], - [1, 256, 10, 10], - [1, 256, 20, 20], - [1, 256, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 56, 2, 2], - [1, 256, 2, 2], - ] - - max_input_shape = [ - [max_batch_size, 3, 2000, 2000], - [max_batch_size, 120, 400, 400], - [max_batch_size, 24, 200, 200], - [max_batch_size, 96, 400, 400], - [max_batch_size, 256, 200, 200], - [max_batch_size, 256, 400, 400], - [max_batch_size, 256, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 56, 400, 400], - [max_batch_size, 256, 400, 400], - ] - - opt_input_shape = [ - [1, 3, 640, 640], - [1, 120, 160, 160], - [1, 24, 80, 80], - [1, 96, 160, 160], - [1, 256, 80, 80], - [1, 256, 160, 160], - [1, 256, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 56, 40, 40], - [1, 256, 40, 40], - ] - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", params_file="./ocr_det_mv3_db/inference.pdiparams" - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", params_file="./ocr_det_mv3_db/inference.pdiparams" - ) - test_suite2.trt_dynamic_multi_thread_test( - input_data_dict, - output_data_dict, - gpu_mem=5000, - max_batch_size=max_batch_size, - repeat=1, - delta=2e-5, - names=names, - min_input_shape=min_input_shape, - max_input_shape=max_input_shape, - opt_input_shape=opt_input_shape, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.win -@pytest.mark.server -@pytest.mark.trt_fp32 -def test_trtfp32_more_bz_dynamic_bz(): - """ - compared trt fp32 batch_size=1,2 ocr_det_mv3_db outputs with true val - """ - check_model_exist() - - file_path = "./ocr_det_mv3_db" - images_size = 640 - batch_size_pool = [1, 5] - max_batch_size = 5 - names = [ - "x", - "conv2d_92.tmp_0", - "conv2d_91.tmp_0", - "conv2d_59.tmp_0", - "nearest_interp_v2_1.tmp_0", - "nearest_interp_v2_2.tmp_0", - "conv2d_124.tmp_0", - "nearest_interp_v2_3.tmp_0", - "nearest_interp_v2_4.tmp_0", - "nearest_interp_v2_5.tmp_0", - "elementwise_add_7", - "nearest_interp_v2_0.tmp_0", - ] - min_input_shape = [ - [1, 3, 50, 50], - [1, 120, 20, 20], - [1, 24, 10, 10], - [1, 96, 20, 20], - [1, 256, 10, 10], - [1, 256, 20, 20], - [1, 256, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 64, 20, 20], - [1, 56, 2, 2], - [1, 256, 2, 2], - ] - - max_input_shape = [ - [max_batch_size, 3, 2000, 2000], - [max_batch_size, 120, 400, 400], - [max_batch_size, 24, 200, 200], - [max_batch_size, 96, 400, 400], - [max_batch_size, 256, 200, 200], - [max_batch_size, 256, 400, 400], - [max_batch_size, 256, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 64, 400, 400], - [max_batch_size, 56, 400, 400], - [max_batch_size, 256, 400, 400], - ] - - opt_input_shape = [ - [1, 3, 640, 640], - [1, 120, 160, 160], - [1, 24, 80, 80], - [1, 96, 160, 160], - [1, 256, 80, 80], - [1, 256, 160, 160], - [1, 256, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 64, 160, 160], - [1, 56, 40, 40], - [1, 256, 40, 40], - ] - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", params_file="./ocr_det_mv3_db/inference.pdiparams" - ) - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"x": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config( - model_file="./ocr_det_mv3_db/inference.pdmodel", params_file="./ocr_det_mv3_db/inference.pdiparams" - ) - - test_suite2.trt_more_bz_dynamic_test( - input_data_dict, - output_data_dict, - max_batch_size=max_batch_size, - repeat=1, - delta=2e-5, - names=names, - min_input_shape=min_input_shape, - max_input_shape=max_input_shape, - opt_input_shape=opt_input_shape, - precision="trt_fp32", - ) - - del test_suite2 # destroy class to save memory diff --git a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_resnet50.py b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_resnet50.py index da78076397..de57404c17 100644 --- a/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_resnet50.py +++ b/tools/test/test-tools/tool-test-dl-algorithm-correctness/tpaddle/test_resnet50.py @@ -59,120 +59,16 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.p1 -@pytest.mark.trt_fp32_more_bz_precision -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-10 resnet50 outputs with true val - """ - check_model_exist() - - file_path = "./resnet50" - images_size = 224 - batch_size_pool = [1, 5, 10] - max_batch_size = 10 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - test_suite2.trt_more_bz_test( - input_data_dict, output_data_dict, max_batch_size=max_batch_size, precision="trt_fp32" - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.p1 -@pytest.mark.trt_fp32_multi_thread_bz1_precision -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 resnet50 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./resnet50" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, precision="trt_fp32") - - del test_suite2 # destroy class to save memory - - -@pytest.mark.p1 -@pytest.mark.trt_fp16_more_bz_precision -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-10 resnet50 outputs with true val - """ - check_model_exist() - file_path = "./resnet50" - images_size = 224 - batch_size_pool = [1, 5, 10] - max_batch_size = 10 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - test_suite2.trt_more_bz_test( - input_data_dict, output_data_dict, delta=1e-2, max_batch_size=max_batch_size, precision="trt_fp16" - ) - del test_suite2 # destroy class to save memory -@pytest.mark.p1 -@pytest.mark.trt_fp16_multi_thread_bz1_precision -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 resnet50 multi_thread outputs with true val - """ - check_model_exist() - file_path = "./resnet50" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, precision="trt_fp16") - del test_suite2 # destroy class to save memory @pytest.mark.p1 @@ -206,7 +102,7 @@ def test_mkldnn(): @pytest.mark.gpu_bz1_precision def test_gpu_bz1(): """ - compared trt gpu batch_size=1-10 resnet50 outputs with true val + compared gpu batch_size=1-10 resnet50 outputs with true val """ check_model_exist() diff --git a/tools/test/test-tools/tool-test-dl-algorithm-performance/tpaddle/test_helper.py b/tools/test/test-tools/tool-test-dl-algorithm-performance/tpaddle/test_helper.py index c71e9e4287..1b34fc1646 100644 --- a/tools/test/test-tools/tool-test-dl-algorithm-performance/tpaddle/test_helper.py +++ b/tools/test/test-tools/tool-test-dl-algorithm-performance/tpaddle/test_helper.py @@ -35,13 +35,10 @@ def parse_args(): parser.add_argument("--model_path", type=str, help="model filename") parser.add_argument("--params_path", type=str, default="", help="parameter filename") - parser.add_argument("--trt_precision", type=str, default="fp32", - help="trt precision, choice = ['fp32', 'fp16', 'int8']") parser.add_argument("--image_shape", type=str, default="3,224,224", help="can only use for one input model(e.g. image classification)") parser.add_argument("--use_gpu", dest="use_gpu", action='store_true') - parser.add_argument("--use_trt", dest="use_trt", action='store_true') parser.add_argument("--use_mkldnn", dest="use_mkldnn", action='store_true') parser.add_argument("--batch_size", type=int, default=1, help="batch size") @@ -52,8 +49,6 @@ def parse_args(): type=int, default=1, help="math_thread_num") - parser.add_argument("--trt_min_subgraph_size", type=int, default=3, - help="tensorrt min_subgraph_size") return parser.parse_args() @@ -64,9 +59,6 @@ def prepare_config(args): Returns: config : paddle inference config """ - trt_precision_map = {"fp32" : paddle_infer.PrecisionType.Float32, - "fp16" : paddle_infer.PrecisionType.Half, - "int8" : paddle_infer.PrecisionType.Int8} if (args.params_path != ""): logger.info("params_path detected, set model with combined model") config = paddle_infer.Config(args.model_path, args.params_path) @@ -74,17 +66,8 @@ def prepare_config(args): logger.info("no params_path detected, set model with uncombined model") config = paddle_infer.Config(args.model_path) - if (args.use_gpu or args.use_trt): + if (args.use_gpu): config.enable_use_gpu(100, 0) - use_calib = True if args.trt_precision == "int8" else False - if (args.use_trt): - logger.info("tensorrt enabled") - config.enable_tensorrt_engine(1 << 30, # workspace_size - args.batch_size, # max_batch_size - args.trt_min_subgraph_size, # min_subgraph_size - trt_precision_map[args.trt_precision], # Precision precision - False, # use_static - use_calib) else: config.disable_gpu() config.set_cpu_math_library_num_threads( @@ -113,11 +96,7 @@ def summary_config(config, args, infer_time : float): logger.info("device: {0}, ir_optim: {1}".format("gpu" if config.use_gpu() else "cpu", config.ir_optim())) # logger.info("enable_memory_optim: {0}".format(config.enable_memory_optim())) - if (config.use_gpu()): - logger.info("enable_tensorrt: {0}".format(config.tensorrt_engine_enabled())) - if (config.tensorrt_engine_enabled()): - logger.info("trt_precision: {0}".format(args.trt_precision)) - else: + if not config.use_gpu(): logger.info("enable_mkldnn: {0}".format(config.mkldnn_enabled())) logger.info("cpu_math_library_num_threads: {0}".format(config.cpu_math_library_num_threads())) logger.info("----------------------- Perf info -----------------------") diff --git a/tools/test/test-tools/tool-test-inference-performance/tpaddle/test_helper.py b/tools/test/test-tools/tool-test-inference-performance/tpaddle/test_helper.py index 8d79ea7427..9deefb4597 100644 --- a/tools/test/test-tools/tool-test-inference-performance/tpaddle/test_helper.py +++ b/tools/test/test-tools/tool-test-inference-performance/tpaddle/test_helper.py @@ -36,13 +36,10 @@ def parse_args(): parser.add_argument("--thread_num", type=int, default=1, help="thread num") parser.add_argument("--params_path", type=str, default="", help="parameter filename") - parser.add_argument("--trt_precision", type=str, default="fp32", - help="trt precision, choice = ['fp32', 'fp16', 'int8']") parser.add_argument("--image_shape", type=str, default="3,224,224", help="can only use for one input model(e.g. image classification)") parser.add_argument("--use_gpu", dest="use_gpu", action='store_true') - parser.add_argument("--use_trt", dest="use_trt", action='store_true') parser.add_argument("--use_mkldnn", dest="use_mkldnn", action='store_true') parser.add_argument("--batch_size", type=int, default=1, help="batch size") @@ -53,8 +50,6 @@ def parse_args(): type=int, default=1, help="math_thread_num") - parser.add_argument("--trt_min_subgraph_size", type=int, default=3, - help="tensorrt min_subgraph_size") return parser.parse_args() @@ -65,9 +60,6 @@ def prepare_config(args): Returns: config : paddle inference config """ - trt_precision_map = {"fp32" : paddle_infer.PrecisionType.Float32, - "fp16" : paddle_infer.PrecisionType.Half, - "int8" : paddle_infer.PrecisionType.Int8} if (args.params_path != ""): logger.info("params_path detected, set model with combined model") config = paddle_infer.Config(args.model_path, args.params_path) @@ -75,17 +67,8 @@ def prepare_config(args): logger.info("no params_path detected, set model with uncombined model") config = paddle_infer.Config(args.model_path) - if (args.use_gpu or args.use_trt): + if (args.use_gpu): config.enable_use_gpu(100, 0) - use_calib = True if args.trt_precision == "int8" else False - if (args.use_trt): - logger.info("tensorrt enabled") - config.enable_tensorrt_engine(1 << 30, # workspace_size - args.batch_size, # max_batch_size - args.trt_min_subgraph_size, # min_subgraph_size - trt_precision_map[args.trt_precision], # Precision precision - False, # use_static - use_calib) else: config.disable_gpu() config.set_cpu_math_library_num_threads( @@ -114,11 +97,7 @@ def summary_config(config, args, infer_time : float): logger.info("device: {0}, ir_optim: {1}".format("gpu" if config.use_gpu() else "cpu", config.ir_optim())) # logger.info("enable_memory_optim: {0}".format(config.enable_memory_optim())) - if (config.use_gpu()): - logger.info("enable_tensorrt: {0}".format(config.tensorrt_engine_enabled())) - if (config.tensorrt_engine_enabled()): - logger.info("trt_precision: {0}".format(args.trt_precision)) - else: + if not config.use_gpu(): logger.info("enable_mkldnn: {0}".format(config.mkldnn_enabled())) logger.info("cpu_math_library_num_threads: {0}".format(config.cpu_math_library_num_threads())) logger.info("----------------------- Perf info -----------------------") diff --git a/tools/test/test-tools/tool-test-inference/tpaddle/test_case/infer_test.py b/tools/test/test-tools/tool-test-inference/tpaddle/test_case/infer_test.py index 8acdef2a56..b3be1b3949 100644 --- a/tools/test/test-tools/tool-test-inference/tpaddle/test_case/infer_test.py +++ b/tools/test/test-tools/tool-test-inference/tpaddle/test_case/infer_test.py @@ -250,416 +250,6 @@ def gpu_more_bz_test(self, input_data_dict: dict, output_data_dict: dict, repeat diff <= delta ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - def trt_bz1_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=5, - delta=1e-5, - gpu_mem=1000, - min_subgraph_size=10, - precision="trt_fp32", - use_static=False, - use_calib_mode=False, - ): - """ - test enable_use_gpu() - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - - output_names = predictor.get_output_names() - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data = output_data.flatten() - output_data_truth_val = output_data_dict[output_data_name].flatten() - for j, out_data in enumerate(output_data): - diff = sig_fig_compare(out_data, output_data_truth_val[j]) - assert ( - diff <= delta - ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - - predictor.try_shrink_memory() # try_shrink_memory - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - - output_names = predictor.get_output_names() - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data = output_data.flatten() - output_data_truth_val = output_data_dict[output_data_name].flatten() - for j, out_data in enumerate(output_data): - diff = sig_fig_compare(out_data, output_data_truth_val[j]) - assert ( - diff <= delta - ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - - def trt_more_bz_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - delta=1e-5, - gpu_mem=1000, - max_batch_size=3, - min_subgraph_size=10, - precision="fp32", - use_static=True, - use_calib_mode=False, - dynamic=False, - tuned=False, - ): - """ - test enable_tensorrt_engine() - batch_size = 10 - trt max_batch_size = 10 - precision_mode = fp32,fp16,int8 - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - if dynamic: - if tuned: - self.pd_config.collect_shape_range_info("shape_range.pbtxt") - return 0 - else: - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - self.pd_config.enable_tuned_tensorrt_dynamic_shape("shape_range.pbtxt", True) - else: - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - - output_names = predictor.get_output_names() - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data = output_data.flatten() - output_data_truth_val = output_data_dict[output_data_name].flatten() - for j, out_data in enumerate(output_data): - diff = sig_fig_compare(out_data, output_data_truth_val[j]) - assert ( - diff <= delta - ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - - def trt_more_bz_dynamic_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - delta=1e-5, - gpu_mem=1000, - max_batch_size=10, - names=None, - min_input_shape=None, - max_input_shape=None, - opt_input_shape=None, - min_subgraph_size=10, - precision="trt_fp32", - use_static=False, - use_calib_mode=False, - ): - """ - test enable_tensorrt_engine() - max_batch_size = 1-10 - trt max_batch_size = 10 - precision_mode = fp32,fp16,int8 - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - names(list): input names - min_input_shape(list): TensorRT min input shape - max_input_shape(list): TensorRT max input shape - opt_input_shape(list): TensorRT best input shape - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=max_batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - - self.pd_config.set_trt_dynamic_shape_info( - {names[i]: min_input_shape[i] for i in range(len(names))}, - {names[i]: max_input_shape[i] for i in range(len(names))}, - {names[i]: opt_input_shape[i] for i in range(len(names))}, - ) - - predictor = paddle_infer.create_predictor(self.pd_config) - - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - - output_names = predictor.get_output_names() - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data = output_data.flatten() - output_data_truth_val = output_data_dict[output_data_name].flatten() - for j, out_data in enumerate(output_data): - diff = sig_fig_compare(out_data, output_data_truth_val[j]) - assert ( - diff <= delta - ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - predictor.try_shrink_memory() - - def trt_bz1_multi_thread_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - thread_num=2, - delta=1e-5, - gpu_mem=1000, - min_subgraph_size=10, - precision="trt_fp32", - use_static=False, - use_calib_mode=False, - ): - """ - test enable_tensorrt_engine() - batch_size = 1 - trt max_batch_size = 4 - thread_num = 5 - precision_mode = fp32,fp16,int8 - Multithreading TensorRT predictor - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time - thread_num(int): number of threads - delta(float): difference threshold between inference outputs and thruth value - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - predictors = paddle_infer.PredictorPool(self.pd_config, thread_num) - for i in range(thread_num): - record_thread = threading.Thread( - target=self.run_multi_thread_test_predictor, - args=(predictors.retrive(i), input_data_dict, output_data_dict, repeat, delta), - ) - record_thread.start() - record_thread.join() - - def trt_dynamic_multi_thread_test( - self, - input_data_dict: dict, - output_data_dict: dict, - repeat=1, - delta=1e-5, - thread_num=2, - gpu_mem=1000, - max_batch_size=1, - names=None, - min_input_shape=None, - max_input_shape=None, - opt_input_shape=None, - min_subgraph_size=10, - precision="trt_fp32", - use_static=False, - use_calib_mode=False, - ): - """ - test enable_tensorrt_engine() - batch_size = 1 - trt max_batch_size = 1 - thread_num = 2 - precision_mode = fp32,fp16,int8 - Multithreading TensorRT predictor - Args: - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time - delta(float): difference threshold between inference outputs and thruth value - names(list): input names - min_input_shape(list): TensorRT min input shape - max_input_shape(list): TensorRT max input shape - opt_input_shape(list): TensorRT best input shape - min_subgraph_size(int): min subgraph size - precision(str): trt precision mode,[fp32,fp16,int8] - use_static(bool): use static - use_calib_mode(bool): use calib mode - Returns: - None - """ - trt_precision_map = { - "trt_fp32": paddle_infer.PrecisionType.Float32, - "trt_fp16": paddle_infer.PrecisionType.Half, - "trt_int8": paddle_infer.PrecisionType.Int8, - } - self.pd_config.enable_use_gpu(gpu_mem, 0) - self.pd_config.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=min_subgraph_size, - precision_mode=trt_precision_map[precision], - use_static=use_static, - use_calib_mode=use_calib_mode, - ) - self.pd_config.set_trt_dynamic_shape_info( - {names[i]: min_input_shape[i] for i in range(len(names))}, - {names[i]: max_input_shape[i] for i in range(len(names))}, - {names[i]: opt_input_shape[i] for i in range(len(names))}, - ) - predictors = paddle_infer.PredictorPool(self.pd_config, thread_num) - for i in range(thread_num): - record_thread = threading.Thread( - target=self.run_multi_thread_test_predictor, - args=(predictors.retrive(i), input_data_dict, output_data_dict, repeat, delta), - ) - record_thread.start() - record_thread.join() - - def run_multi_thread_test_predictor( - self, predictor, input_data_dict: dict, output_data_dict: dict, repeat=1, delta=1e-5 - ): - """ - test paddle predictor in multithreaded task - Args: - predictor: paddle inference predictor - input_data_dict(dict): input data constructed as dictionary - output_data_dict(dict): output data constructed as dictionary - repeat(int): inference repeat time, set to catch gpu mem - delta(float): difference threshold between inference outputs and thruth value - Returns: - None - """ - input_names = predictor.get_input_names() - for _, input_data_name in enumerate(input_names): - input_handle = predictor.get_input_handle(input_data_name) - input_handle.copy_from_cpu(input_data_dict[input_data_name]) - - for i in range(repeat): - predictor.run() - output_names = predictor.get_output_names() - for i, output_data_name in enumerate(output_names): - output_handle = predictor.get_output_handle(output_data_name) - output_data = output_handle.copy_to_cpu() - output_data = output_data.flatten() - output_data_truth_val = output_data_dict[output_data_name].flatten() - for j, out_data in enumerate(output_data): - diff = sig_fig_compare(out_data, output_data_truth_val[j]) - assert ( - diff <= delta - ), f"{out_data} and {output_data_truth_val[j]} significant digits {diff} diff > {delta}" - def get_gpu_mem(gpu_id=0): """ diff --git a/tools/test/test-tools/tool-test-inference/tpaddle/test_mobilenet.py b/tools/test/test-tools/tool-test-inference/tpaddle/test_mobilenet.py index 3ef5d67e8f..29b642c0c7 100644 --- a/tools/test/test-tools/tool-test-inference/tpaddle/test_mobilenet.py +++ b/tools/test/test-tools/tool-test-inference/tpaddle/test_mobilenet.py @@ -59,120 +59,16 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.p1 -@pytest.mark.trt_fp32_more_bz_precision -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-10 MobileNetV1 outputs with true val - """ - check_model_exist() - - file_path = "./MobileNetV1" - images_size = 224 - batch_size_pool = [1, 5, 10] - max_batch_size = 10 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - test_suite2.trt_more_bz_test( - input_data_dict, output_data_dict, max_batch_size=max_batch_size, precision="trt_fp32" - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.p1 -@pytest.mark.trt_fp32_multi_thread_bz1_precision -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 MobileNetV1 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./MobileNetV1" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, precision="trt_fp32") - del test_suite2 # destroy class to save memory -@pytest.mark.p1 -@pytest.mark.trt_fp16_more_bz_precision -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-10 MobileNetV1 outputs with true val - """ - check_model_exist() - file_path = "./MobileNetV1" - images_size = 224 - batch_size_pool = [1, 5, 10] - max_batch_size = 10 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - test_suite2.trt_more_bz_test( - input_data_dict, output_data_dict, delta=1e-2, max_batch_size=max_batch_size, precision="trt_fp16" - ) - del test_suite2 # destroy class to save memory -@pytest.mark.p1 -@pytest.mark.trt_fp16_multi_thread_bz1_precision -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 MobileNetV1 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./MobileNetV1" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./MobileNetV1/model.pdmodel", params_file="./MobileNetV1/model.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, precision="trt_fp16") - - del test_suite2 # destroy class to save memory @pytest.mark.p1 @@ -206,7 +102,7 @@ def test_mkldnn(): @pytest.mark.gpu_bz1_precision def test_gpu_bz1(): """ - compared trt gpu batch_size=1-10 MobileNetV1 outputs with true val + compared gpu batch_size=1-10 MobileNetV1 outputs with true val """ check_model_exist() diff --git a/tools/test/test-tools/tool-test-inference/tpaddle/test_resnet50.py b/tools/test/test-tools/tool-test-inference/tpaddle/test_resnet50.py index ce548d4599..21a544e242 100644 --- a/tools/test/test-tools/tool-test-inference/tpaddle/test_resnet50.py +++ b/tools/test/test-tools/tool-test-inference/tpaddle/test_resnet50.py @@ -59,120 +59,16 @@ def test_disable_gpu(): test_suite.disable_gpu_test(input_data_dict) -@pytest.mark.p1 -@pytest.mark.trt_fp32_more_bz_precision -def test_trt_fp32_more_bz(): - """ - compared trt fp32 batch_size=1-10 resnet50 outputs with true val - """ - check_model_exist() - - file_path = "./resnet50" - images_size = 224 - batch_size_pool = [1, 5, 10] - max_batch_size = 10 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - test_suite2.trt_more_bz_test( - input_data_dict, output_data_dict, max_batch_size=max_batch_size, precision="trt_fp32" - ) - - del test_suite2 # destroy class to save memory - - -@pytest.mark.p1 -@pytest.mark.trt_fp32_multi_thread_bz1_precision -def test_trt_fp32_bz1_multi_thread(): - """ - compared trt fp32 batch_size=1 resnet50 multi_thread outputs with true val - """ - check_model_exist() - - file_path = "./resnet50" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - - del test_suite # destroy class to save memory - - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, precision="trt_fp32") - - del test_suite2 # destroy class to save memory - - -@pytest.mark.p1 -@pytest.mark.trt_fp16_more_bz_precision -def test_trt_fp16_more_bz(): - """ - compared trt fp16 batch_size=1-10 resnet50 outputs with true val - """ - check_model_exist() - file_path = "./resnet50" - images_size = 224 - batch_size_pool = [1, 5, 10] - max_batch_size = 10 - for batch_size in batch_size_pool: - test_suite = InferenceTest() - test_suite.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - test_suite2.trt_more_bz_test( - input_data_dict, output_data_dict, delta=1e-2, max_batch_size=max_batch_size, precision="trt_fp16" - ) - del test_suite2 # destroy class to save memory -@pytest.mark.p1 -@pytest.mark.trt_fp16_multi_thread_bz1_precision -def test_trt_fp16_bz1_multi_thread(): - """ - compared trt fp16 batch_size=1 resnet50 multi_thread outputs with true val - """ - check_model_exist() - file_path = "./resnet50" - images_size = 224 - batch_size = 1 - test_suite = InferenceTest() - test_suite.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - images_list, npy_list = test_suite.get_images_npy(file_path, images_size) - fake_input = np.array(images_list[0:batch_size]).astype("float32") - input_data_dict = {"inputs": fake_input} - output_data_dict = test_suite.get_truth_val(input_data_dict, device="gpu") - del test_suite # destroy class to save memory - test_suite2 = InferenceTest() - test_suite2.load_config(model_file="./resnet50/inference.pdmodel", params_file="./resnet50/inference.pdiparams") - test_suite2.trt_bz1_multi_thread_test(input_data_dict, output_data_dict, precision="trt_fp16") - del test_suite2 # destroy class to save memory @pytest.mark.p1 @@ -206,7 +102,7 @@ def test_mkldnn(): @pytest.mark.gpu_bz1_precision def test_gpu_bz1(): """ - compared trt gpu batch_size=1-10 resnet50 outputs with true val + compared gpu batch_size=1-10 resnet50 outputs with true val """ check_model_exist() From 1ace2839d785c4ba4a7ffb1c1e5bb3d8cb241c78 Mon Sep 17 00:00:00 2001 From: gouzi <530971494@qq.com> Date: Sun, 19 Jul 2026 02:26:20 +0800 Subject: [PATCH 2/5] remove remaining TensorRT usage --- .github/workflows/so_denpency_analyzer.yml | 3 - .../ResNet/CE_ResNet50_train_infer_python.txt | 1 - .../configs/deep_walk/train_infer_python.txt | 1 - .../configs/lightgcn/train_infer_python.txt | 3 +- .../transformer_conv/train_infer_python.txt | 3 +- .../end2end/csp_darknet_det_backbone.py | 3 +- .../end2end/cspresnet_det_backbone.py | 20 +-- .../end2end/mobileone_det_backbone.py | 12 +- .../Det/modeling/backbones/csp_darknet.yml | 1 - .../yaml/Det/modeling/backbones/cspresnet.yml | 1 - .../yaml/Det/modeling/necks/custom_pan.yml | 2 - .../yaml/Det/modeling/necks/yolo_fpn.yml | 1 - .../Det/modeling/backbones/csp_darknet.yml | 1 - .../yaml/Det/modeling/backbones/cspresnet.yml | 1 - .../yaml/Det/modeling/necks/custom_pan.yml | 2 - .../yaml/Det/modeling/necks/yolo_fpn.yml | 1 - inference/benchmark/jetson/LICENSE.md | 7 - inference/benchmark/jetson/README.md | 94 ----------- inference/benchmark/jetson/benchmark.py | 122 -------------- .../jetson/benchmark_csv/nx-benchmarks.csv | 9 -- .../benchmark_csv/tx2-nano-benchmarks.csv | 9 -- .../benchmark_csv/xavier-benchmarks.csv | 9 -- .../benchmark/jetson/install_requirements.sh | 8 - inference/benchmark/jetson/utils/__init__.py | 6 - .../jetson/utils/benchmark_argparser.py | 37 ----- .../benchmark/jetson/utils/download_models.py | 82 ---------- .../jetson/utils/load_store_engine.py | 150 ------------------ .../benchmark/jetson/utils/read_write_data.py | 149 ----------------- .../jetson/utils/run_benchmark_models.py | 47 ------ .../benchmark/jetson/utils/run_xavier_maxn.sh | 10 -- inference/benchmark/jetson/utils/utilities.py | 107 ------------- .../benchmark/python/paddle/fast_rcnn.py | 22 +-- .../benchmark/python/paddle/mobilenetv2.py | 24 +-- .../benchmark/python/paddle/resnet101.py | 22 --- .../benchmark/python/paddle/squeezenet.py | 24 +-- inference/benchmark/python/paddle/vgg16.py | 24 +-- inference/benchmark/python/parse_log.py | 6 +- .../python/tensorflow/clas_keras_benchmark.py | 76 +-------- inference/benchmark/python/torch/README.md | 5 - .../benchmark/python/torch/clas_benchmark.py | 45 ------ .../python/torch/detection_benchmark.py | 42 ----- .../benchmark/python/torch/seg_benchmark.py | 42 +---- .../test_int8_model/get_benchmark_info.py | 12 +- .../test_int8_model/write_db.py | 8 +- .../python_api_test/test_int8_model/xly.sh | 3 +- .../test_new_devices/benchmark.py | 3 - .../python_api_test/test_new_devices/demo.sh | 5 +- .../inference_benchmark_new_devices.sh | 6 +- inference/report/get_gsb.py | 4 +- .../paddle_serving_server/test_server.py | 1 - .../paddle_serving_server/util.py | 1 - models/Paddle2ONNX/Det2ONNX/infer_for_onnx.py | 111 +------------ .../Det2ONNX/key_infer_for_onnx.py | 32 +--- models/Paddle2ONNX/Det2ONNX/utils_for_onnx.py | 12 -- models/Paddle2ONNX/Seg2ONNX/infer_for_onnx.py | 122 +------------- models/PaddleClas/Full_Chain/tipc.sh | 2 - .../module_test/heads/test_FCOSHead.py | 1 - .../module_test/heads/test_FCOSRHead.py | 1 - .../module_test/heads/test_PPYOLOERHead.py | 1 - .../module_test/heads/test_YOLOFHead.py | 1 - models/PaddleGAN/Full_Chain/tipc.sh | 2 - .../script/gather_img_video_to_one_file.sh | 4 +- models/PaddleNLP/CI/ci_case.sh | 2 - .../CE/whole_function/run_PaddleSeg.sh | 2 - models/PaddleVideo/CI/test_video.sh | 3 +- models/ROCM/RocmTestFramework.py | 4 +- .../Paddle3D/diy_build/Paddle3D_Build.py | 29 ++-- .../PaddleClas/base/slim_base.yaml | 1 - .../PaddleClas/tools/get_result.py | 1 - ...rcnn^cascade_mask_rcnn_r50_fpn_1x_coco.yml | 4 - ...cade_mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml | 4 - ...cade_mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml | 4 - ...cade_rcnn^cascade_rcnn_r50_fpn_1x_coco.yml | 4 - ...n^cascade_rcnn_r50_vd_fpn_ssld_1x_coco.yml | 4 - ...n^cascade_rcnn_r50_vd_fpn_ssld_2x_coco.yml | 4 - ...gs^centernet^centernet_dla34_140e_coco.yml | 4 - ...igs^centernet^centernet_mbv1_140e_coco.yml | 4 - ...nternet^centernet_mbv3_large_140e_coco.yml | 4 - ...nternet^centernet_mbv3_small_140e_coco.yml | 4 - ...figs^centernet^centernet_r50_140e_coco.yml | 4 - ...ernet^centernet_shufflenetv2_140e_coco.yml | 4 - ...s^dcn^cascade_rcnn_dcn_r50_fpn_1x_coco.yml | 4 - ...ade_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml | 4 - ...cn^faster_rcnn_dcn_r101_vd_fpn_1x_coco.yml | 4 - ...gs^dcn^faster_rcnn_dcn_r50_fpn_1x_coco.yml | 4 - ...dcn^faster_rcnn_dcn_r50_vd_fpn_1x_coco.yml | 4 - ...dcn^faster_rcnn_dcn_r50_vd_fpn_2x_coco.yml | 4 - ...ter_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml | 4 - ...^dcn^mask_rcnn_dcn_r101_vd_fpn_1x_coco.yml | 4 - ...figs^dcn^mask_rcnn_dcn_r50_fpn_1x_coco.yml | 4 - ...s^dcn^mask_rcnn_dcn_r50_vd_fpn_2x_coco.yml | 4 - ...ask_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml | 4 - ...mable_detr^deformable_detr_r50_1x_coco.yml | 4 - .../cases/configs^detr^detr_r50_1x_coco.yml | 4 - ...configs^face_detection^blazeface_1000e.yml | 4 - ...face_detection^blazeface_fpn_ssh_1000e.yml | 4 - ...s^faster_rcnn^faster_rcnn_r101_1x_coco.yml | 4 - ...ster_rcnn^faster_rcnn_r101_fpn_2x_coco.yml | 4 - ...r_rcnn^faster_rcnn_r101_vd_fpn_1x_coco.yml | 4 - ...r_rcnn^faster_rcnn_r101_vd_fpn_2x_coco.yml | 4 - ...aster_rcnn^faster_rcnn_r34_fpn_1x_coco.yml | 4 - ...er_rcnn^faster_rcnn_r34_vd_fpn_1x_coco.yml | 4 - ...gs^faster_rcnn^faster_rcnn_r50_1x_coco.yml | 4 - ...aster_rcnn^faster_rcnn_r50_fpn_1x_coco.yml | 4 - ...aster_rcnn^faster_rcnn_r50_fpn_2x_coco.yml | 4 - ...faster_rcnn^faster_rcnn_r50_vd_1x_coco.yml | 4 - ...er_rcnn^faster_rcnn_r50_vd_fpn_1x_coco.yml | 4 - ...er_rcnn^faster_rcnn_r50_vd_fpn_2x_coco.yml | 4 - ...nn^faster_rcnn_r50_vd_fpn_ssld_1x_coco.yml | 4 - ...nn^faster_rcnn_r50_vd_fpn_ssld_2x_coco.yml | 4 - ...rcnn^faster_rcnn_swin_tiny_fpn_1x_coco.yml | 4 - ...rcnn^faster_rcnn_swin_tiny_fpn_2x_coco.yml | 4 - ...^faster_rcnn_x101_vd_64x4d_fpn_1x_coco.yml | 4 - ...^faster_rcnn_x101_vd_64x4d_fpn_2x_coco.yml | 4 - .../configs^fcos^fcos_dcn_r50_fpn_1x_coco.yml | 4 - .../configs^fcos^fcos_r50_fpn_1x_coco.yml | 4 - .../configs^fcos^fcos_r50_fpn_iou_1x_coco.yml | 4 - ...os^fcos_r50_fpn_iou_multiscale_2x_coco.yml | 4 - ...s^fcos^fcos_r50_fpn_multiscale_2x_coco.yml | 4 - ...igs^gfl^gfl_r101vd_fpn_mstrain_2x_coco.yml | 4 - .../cases/configs^gfl^gfl_r18vd_1x_coco.yml | 4 - .../cases/configs^gfl^gfl_r34vd_1x_coco.yml | 4 - .../cases/configs^gfl^gfl_r50_fpn_1x_coco.yml | 4 - .../configs^gfl^gflv2_r50_fpn_1x_coco.yml | 4 - ...n^cascade_mask_rcnn_r50_fpn_gn_2x_coco.yml | 4 - ...igs^gn^cascade_rcnn_r50_fpn_gn_2x_coco.yml | 4 - ...figs^gn^faster_rcnn_r50_fpn_gn_2x_coco.yml | 4 - ...onfigs^gn^mask_rcnn_r50_fpn_gn_2x_coco.yml | 4 - ...hrnet^faster_rcnn_hrnetv2p_w18_1x_coco.yml | 4 - ...hrnet^faster_rcnn_hrnetv2p_w18_2x_coco.yml | 4 - ...^higherhrnet^higherhrnet_hrnet_w32_512.yml | 4 - ...rhrnet^higherhrnet_hrnet_w32_512_swahr.yml | 4 - ...^higherhrnet^higherhrnet_hrnet_w32_640.yml | 4 - ...^keypoint^hrnet^dark_hrnet_w32_256x192.yml | 4 - ...^keypoint^hrnet^dark_hrnet_w32_384x288.yml | 4 - ...nfigs^keypoint^hrnet^hrnet_w32_256x192.yml | 4 - ...nfigs^keypoint^hrnet^hrnet_w32_384x288.yml | 4 - ...^lite_hrnet^lite_hrnet_18_256x192_coco.yml | 4 - ...^lite_hrnet^lite_hrnet_18_384x288_coco.yml | 4 - ...^lite_hrnet^lite_hrnet_30_256x192_coco.yml | 4 - ...^lite_hrnet^lite_hrnet_30_384x288_coco.yml | 4 - ...rnet^wider_naive_hrnet_18_256x192_coco.yml | 4 - ...igs^keypoint^tiny_pose^tinypose_128x96.yml | 4 - ...gs^keypoint^tiny_pose^tinypose_256x192.yml | 4 - ...s^mask_rcnn^mask_rcnn_r101_fpn_1x_coco.yml | 4 - ...ask_rcnn^mask_rcnn_r101_vd_fpn_1x_coco.yml | 4 - ...onfigs^mask_rcnn^mask_rcnn_r50_1x_coco.yml | 4 - ...onfigs^mask_rcnn^mask_rcnn_r50_2x_coco.yml | 4 - ...gs^mask_rcnn^mask_rcnn_r50_fpn_1x_coco.yml | 4 - ...gs^mask_rcnn^mask_rcnn_r50_fpn_2x_coco.yml | 4 - ...mask_rcnn^mask_rcnn_r50_vd_fpn_1x_coco.yml | 4 - ...mask_rcnn^mask_rcnn_r50_vd_fpn_2x_coco.yml | 4 - ...rcnn^mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml | 4 - ...rcnn^mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml | 4 - ...nn^mask_rcnn_x101_vd_64x4d_fpn_1x_coco.yml | 4 - ...nn^mask_rcnn_x101_vd_64x4d_fpn_2x_coco.yml | 4 - ...mot^fairmot^fairmot_dla34_30e_1088x608.yml | 4 - ...ot^fairmot_dla34_30e_1088x608_airplane.yml | 4 - ...fairmot_dla34_30e_1088x608_bytetracker.yml | 4 - ...^mot^fairmot^fairmot_dla34_30e_576x320.yml | 4 - ...^mot^fairmot^fairmot_dla34_30e_864x480.yml | 4 - ...mot^fairmot_enhance_dla34_60e_1088x608.yml | 4 - ...fairmot_enhance_hardnet85_30e_1088x608.yml | 4 - ...airmot_hrnetv2_w18_dlafpn_30e_1088x608.yml | 4 - ...fairmot_hrnetv2_w18_dlafpn_30e_576x320.yml | 4 - ...fairmot_hrnetv2_w18_dlafpn_30e_864x480.yml | 4 - ...rmot_dla34_30e_1088x608_headtracking21.yml | 4 - ...igs^mot^jde^jde_darknet53_30e_1088x608.yml | 4 - ...figs^mot^jde^jde_darknet53_30e_576x320.yml | 4 - ...figs^mot^jde^jde_darknet53_30e_864x480.yml | 4 - ...^mcfairmot_dla34_30e_1088x608_visdrone.yml | 4 - ..._1088x608_visdrone_vehicle_bytetracker.yml | 4 - ...netv2_w18_dlafpn_30e_1088x608_visdrone.yml | 4 - ..._1088x608_visdrone_vehicle_bytetracker.yml | 4 - ...2_w18_dlafpn_30e_576x320_bdd100k_mcmot.yml | 4 - ...rnetv2_w18_dlafpn_30e_576x320_visdrone.yml | 4 - ...rnetv2_w18_dlafpn_30e_864x480_visdrone.yml | 4 - ...n^fairmot_dla34_30e_1088x608_pathtrack.yml | 4 - ...dla34_30e_1088x608_visdrone_pedestrian.yml | 4 - ...lafpn_30e_1088x608_visdrone_pedestrian.yml | 4 - ...dlafpn_30e_576x320_visdrone_pedestrian.yml | 4 - ...dlafpn_30e_864x480_visdrone_pedestrian.yml | 4 - ..._dla34_30e_1088x608_bdd100kmot_vehicle.yml | 4 - ...irmot_dla34_30e_1088x608_kitti_vehicle.yml | 4 - ...ot_dla34_30e_1088x608_visdrone_vehicle.yml | 4 - ..._dlafpn_30e_576x320_bdd100kmot_vehicle.yml | 4 - ...18_dlafpn_30e_576x320_visdrone_vehicle.yml | 4 - ...ore_config^picodet_lcnet_1_5x_640_coco.yml | 4 - ...^picodet_mobilenetv3_large_1x_416_coco.yml | 4 - ...model^more_config^picodet_r18_640_coco.yml | 4 - ...onfig^picodet_shufflenetv2_1x_416_coco.yml | 4 - ...icodet^legacy_model^picodet_l_320_coco.yml | 4 - ...icodet^legacy_model^picodet_l_416_coco.yml | 4 - ...icodet^legacy_model^picodet_l_640_coco.yml | 4 - ...icodet^legacy_model^picodet_m_320_coco.yml | 4 - ...icodet^legacy_model^picodet_m_416_coco.yml | 4 - ...icodet^legacy_model^picodet_s_416_coco.yml | 4 - ...nfigs^picodet^picodet_l_320_coco_lcnet.yml | 4 - ...nfigs^picodet^picodet_l_416_coco_lcnet.yml | 4 - ...nfigs^picodet^picodet_l_640_coco_lcnet.yml | 4 - ...nfigs^picodet^picodet_m_320_coco_lcnet.yml | 4 - ...nfigs^picodet^picodet_m_416_coco_lcnet.yml | 4 - ...nfigs^picodet^picodet_s_416_coco_lcnet.yml | 4 - ...configs^picodet^picodet_s_416_coco_npu.yml | 4 - ...figs^picodet^picodet_xs_320_coco_lcnet.yml | 4 - ...figs^picodet^picodet_xs_416_coco_lcnet.yml | 4 - .../configs^ppyolo^ppyolo_mbv3_small_coco.yml | 4 - .../configs^ppyolo^ppyolo_r18vd_coco.yml | 4 - ...onfigs^ppyolo^ppyolo_r50vd_dcn_2x_coco.yml | 4 - .../configs^ppyolo^ppyolo_r50vd_dcn_voc.yml | 4 - ...s^ppyolo^ppyolov2_r101vd_dcn_365e_coco.yml | 4 - ...onfigs^ppyoloe^ppyoloe_crn_l_300e_coco.yml | 4 - ...onfigs^ppyoloe^ppyoloe_crn_m_300e_coco.yml | 4 - ...onfigs^ppyoloe^ppyoloe_crn_s_400e_coco.yml | 4 - ...onfigs^ppyoloe^ppyoloe_crn_x_300e_coco.yml | 4 - ...gs^ppyoloe^ppyoloe_plus_crn_l_80e_coco.yml | 4 - ...gs^ppyoloe^ppyoloe_plus_crn_m_80e_coco.yml | 4 - ...gs^ppyoloe^ppyoloe_plus_crn_x_80e_coco.yml | 4 - ...ppyoloe^voc^ppyoloe_plus_crn_l_30e_voc.yml | 4 - ...ppyoloe^voc^ppyoloe_plus_crn_s_30e_voc.yml | 4 - ...nn_enhance^faster_rcnn_enhance_3x_coco.yml | 4 - ...r_rcnn_res2net50_vb_26w_4s_fpn_1x_coco.yml | 4 - ...k_rcnn_res2net50_vb_26w_4s_fpn_2x_coco.yml | 4 - ...k_rcnn_res2net50_vd_26w_4s_fpn_2x_coco.yml | 4 - ...net^retinanet_r101_distill_r50_2x_coco.yml | 4 - ...s^retinanet^retinanet_r101_fpn_2x_coco.yml | 4 - ...gs^retinanet^retinanet_r50_fpn_2x_coco.yml | 4 - ...configs^rotate^fcosr^fcosr_x50_3x_dota.yml | 4 - ...tate^ppyoloe_r^ppyoloe_r_crn_l_3x_dota.yml | 4 - ...e^ppyoloe_r^ppyoloe_r_crn_l_3x_dota_ms.yml | 4 - ...tate^ppyoloe_r^ppyoloe_r_crn_m_3x_dota.yml | 4 - ...e^ppyoloe_r^ppyoloe_r_crn_m_3x_dota_ms.yml | 4 - ...tate^ppyoloe_r^ppyoloe_r_crn_s_3x_dota.yml | 4 - ...e^ppyoloe_r^ppyoloe_r_crn_s_3x_dota_ms.yml | 4 - ...tate^ppyoloe_r^ppyoloe_r_crn_x_3x_dota.yml | 4 - ...e^ppyoloe_r^ppyoloe_r_crn_x_3x_dota_ms.yml | 4 - .../configs^rotate^s2anet^s2anet_1x_spine.yml | 4 - ...rotate^s2anet^s2anet_alignconv_2x_dota.yml | 4 - ...figs^rotate^s2anet^s2anet_conv_2x_dota.yml | 4 - ...ine^faster_rcnn_r50_fpn_2x_coco_sup010.yml | 4 - ...t^baseline^fcos_r50_fpn_2x_coco_sup005.yml | 4 - ...t^baseline^fcos_r50_fpn_2x_coco_sup010.yml | 4 - ...ine^ppyoloe_plus_crn_s_80e_coco_sup005.yml | 4 - ...ine^ppyoloe_plus_crn_s_80e_coco_sup010.yml | 4 - ...eline^retinanet_r50_fpn_2x_coco_sup010.yml | 4 - ...faster_rcnn_r50_fpn_1x_sniper_visdrone.yml | 4 - ...sniper^faster_rcnn_r50_fpn_1x_visdrone.yml | 4 - ...figs^solov2^solov2_r101_vd_fpn_3x_coco.yml | 4 - ...configs^solov2^solov2_r50_enhance_coco.yml | 4 - .../configs^solov2^solov2_r50_fpn_1x_coco.yml | 4 - .../configs^solov2^solov2_r50_fpn_3x_coco.yml | 4 - ...figs^ssd^ssd_mobilenet_v1_300_120e_voc.yml | 4 - .../configs^ssd^ssd_vgg16_300_240e_voc.yml | 4 - .../configs^ssd^ssdlite_ghostnet_320_coco.yml | 4 - ...figs^ssd^ssdlite_mobilenet_v1_300_coco.yml | 4 - ...sd^ssdlite_mobilenet_v3_large_320_coco.yml | 4 - ...sd^ssdlite_mobilenet_v3_small_320_coco.yml | 4 - .../configs^tood^tood_r50_fpn_1x_coco.yml | 4 - .../cases/configs^ttfnet^pafnet_10x_coco.yml | 4 - ...fnet^pafnet_lite_mobilenet_v3_20x_coco.yml | 4 - ...onfigs^ttfnet^ttfnet_darknet53_1x_coco.yml | 4 - .../configs^yolof^yolof_r50_c5_1x_coco.yml | 4 - ...figs^yolov3^yolov3_darknet53_270e_coco.yml | 4 - ...s^yolov3^yolov3_mobilenet_v1_270e_coco.yml | 4 - ...gs^yolov3^yolov3_mobilenet_v1_270e_voc.yml | 4 - ...gs^yolov3^yolov3_mobilenet_v1_roadsign.yml | 4 - ...ov3^yolov3_mobilenet_v1_ssld_270e_coco.yml | 4 - ...lov3^yolov3_mobilenet_v1_ssld_270e_voc.yml | 4 - ...v3^yolov3_mobilenet_v3_large_270e_coco.yml | 4 - ...ov3^yolov3_mobilenet_v3_large_270e_voc.yml | 4 - ...olov3_mobilenet_v3_large_ssld_270e_voc.yml | 4 - .../configs^yolov3^yolov3_r34_270e_coco.yml | 4 - ...figs^yolov3^yolov3_r50vd_dcn_270e_coco.yml | 4 - ...configs^yolox^yolox_cdn_tiny_300e_coco.yml | 4 - .../configs^yolox^yolox_crn_s_300e_coco.yml | 4 - .../cases/configs^yolox^yolox_l_300e_coco.yml | 4 - .../cases/configs^yolox^yolox_m_300e_coco.yml | 4 - .../configs^yolox^yolox_nano_300e_coco.yml | 4 - .../cases/configs^yolox^yolox_s_300e_coco.yml | 4 - .../configs^yolox^yolox_tiny_300e_coco.yml | 4 - .../cases/configs^yolox^yolox_x_300e_coco.yml | 4 - ...rcnn^faster_rcnn_swin_tiny_fpn_1x_coco.yml | 4 - ...rcnn^faster_rcnn_swin_tiny_fpn_2x_coco.yml | 4 - ...igs^keypoint^tiny_pose^tinypose_128x96.yml | 4 - .../PaddleDetection/cases/test_keypoint.yml | 4 - .../PaddleDetection/cases/test_mot.yml | 4 - models_restruct/PaddleLLM/tools/run.sh | 1 - models_restruct/PaddleLLM/tools/run_build.sh | 1 - .../PaddleOCR/base/ocr_cls_base.yaml | 2 - .../base/ocr_cls_base_pretrained.yaml | 4 - .../PaddleOCR/base/ocr_det_base.yaml | 2 - .../PaddleOCR/base/ocr_det_base_distill.yaml | 2 - .../base/ocr_det_base_pretrained.yaml | 4 - .../base/ocr_e2e_base_pretrained.yaml | 4 - .../base/ocr_kie_base_pretrained.yaml | 4 - .../PaddleOCR/base/ocr_rec_base.yaml | 2 - .../PaddleOCR/base/ocr_rec_base_distill.yaml | 2 - .../base/ocr_rec_base_pretrained.yaml | 4 - .../base/ocr_rec_base_pretrained_distill.yaml | 4 - .../PaddleOCR/base/ocr_sr_base.yaml | 2 - .../base/ocr_sr_base_pretrained.yaml | 4 - .../PaddleOCR/base/ocr_table_base.yaml | 2 - .../base/ocr_table_base_pretrained.yaml | 4 - .../PaddleOCR/diy_build/PaddleOCR_Build.py | 7 +- .../PaddleScience/tools/get_result.py | 1 - models_restruct/deepxde/tools/get_result.py | 1 - models_restruct/models_env/linux_env.sh | 40 +---- .../models_env/models_docker_build.sh | 3 +- models_restruct/models_env/windows_env.bat | 2 +- tools/linux_env_info.sh | 34 ---- 310 files changed, 66 insertions(+), 2546 deletions(-) delete mode 100644 inference/benchmark/jetson/LICENSE.md delete mode 100644 inference/benchmark/jetson/README.md delete mode 100644 inference/benchmark/jetson/benchmark.py delete mode 100644 inference/benchmark/jetson/benchmark_csv/nx-benchmarks.csv delete mode 100644 inference/benchmark/jetson/benchmark_csv/tx2-nano-benchmarks.csv delete mode 100644 inference/benchmark/jetson/benchmark_csv/xavier-benchmarks.csv delete mode 100644 inference/benchmark/jetson/install_requirements.sh delete mode 100644 inference/benchmark/jetson/utils/__init__.py delete mode 100644 inference/benchmark/jetson/utils/benchmark_argparser.py delete mode 100644 inference/benchmark/jetson/utils/download_models.py delete mode 100644 inference/benchmark/jetson/utils/load_store_engine.py delete mode 100644 inference/benchmark/jetson/utils/read_write_data.py delete mode 100644 inference/benchmark/jetson/utils/run_benchmark_models.py delete mode 100644 inference/benchmark/jetson/utils/run_xavier_maxn.sh delete mode 100644 inference/benchmark/jetson/utils/utilities.py diff --git a/.github/workflows/so_denpency_analyzer.yml b/.github/workflows/so_denpency_analyzer.yml index e16984577b..519d9236d8 100644 --- a/.github/workflows/so_denpency_analyzer.yml +++ b/.github/workflows/so_denpency_analyzer.yml @@ -36,7 +36,6 @@ jobs: - "TagBuild-Training-Linux-Gpu-Cuda11.8-Cudnn8.6-Mkl-Avx-Gcc8.2-SelfBuiltPypiUse" - "TagBuild-Training-Linux-Cpu-Mkl-Avx-Gcc82-SelfBuiltPypiUse" - "TagBuild-Training-Linux-Cpu-ARM-SelfBuiltPypiUse" - - "TagBuild-Training-Linux-Gpu-Cuda12.6-Cudnn9.5-Trt10.5-Mkl-Avx-Gcc11-SelfBuiltPypiUse" python_version: ["3.8", "3.10", "3.13"] is_cuda: ["True", "False"] exclude: @@ -44,8 +43,6 @@ jobs: is_cuda: "True" - ce_task_name: "TagBuild-Training-Linux-Cpu-ARM-SelfBuiltPypiUse" is_cuda: "True" - - ce_task_name: "TagBuild-Training-Linux-Gpu-Cuda12.6-Cudnn9.5-Trt10.5-Mkl-Avx-Gcc11-SelfBuiltPypiUse" - is_cuda: "False" - ce_task_name: "TagBuild-Training-Linux-Gpu-Cuda11.8-Cudnn8.6-Mkl-Avx-Gcc8.2-SelfBuiltPypiUse" is_cuda: "False" steps: diff --git a/distributed/CE_PDC/PaddleClas/test_tipc/configs/ResNet/CE_ResNet50_train_infer_python.txt b/distributed/CE_PDC/PaddleClas/test_tipc/configs/ResNet/CE_ResNet50_train_infer_python.txt index 95c2b7540b..e83bfd4025 100644 --- a/distributed/CE_PDC/PaddleClas/test_tipc/configs/ResNet/CE_ResNet50_train_infer_python.txt +++ b/distributed/CE_PDC/PaddleClas/test_tipc/configs/ResNet/CE_ResNet50_train_infer_python.txt @@ -42,7 +42,6 @@ inference:python/predict_cls.py -c configs/inference_cls.yaml -o Global.enable_mkldnn:False -o Global.cpu_num_threads:1 -o Global.batch_size:1 --o Global.use_tensorrt:False -o Global.use_fp16:False -o Global.inference_model_dir:../inference -o Global.infer_imgs:../dataset/ILSVRC2012/val/ILSVRC2012_val_00000001.JPEG diff --git a/distributed/CE_PDC/PaddleRec/test_tipc/configs/deep_walk/train_infer_python.txt b/distributed/CE_PDC/PaddleRec/test_tipc/configs/deep_walk/train_infer_python.txt index 9543636f30..b37ddf9d93 100755 --- a/distributed/CE_PDC/PaddleRec/test_tipc/configs/deep_walk/train_infer_python.txt +++ b/distributed/CE_PDC/PaddleRec/test_tipc/configs/deep_walk/train_infer_python.txt @@ -42,7 +42,6 @@ inference:null --enable_mkldnn:False --cpu_threads:1|6 --batchsize:10 ---enable_tensorRT:False --precision:fp32 --model_dir: --data_dir:test_tipc/data/infer diff --git a/distributed/CE_PDC/PaddleRec/test_tipc/configs/lightgcn/train_infer_python.txt b/distributed/CE_PDC/PaddleRec/test_tipc/configs/lightgcn/train_infer_python.txt index 95cae89836..eafe975672 100755 --- a/distributed/CE_PDC/PaddleRec/test_tipc/configs/lightgcn/train_infer_python.txt +++ b/distributed/CE_PDC/PaddleRec/test_tipc/configs/lightgcn/train_infer_python.txt @@ -42,7 +42,6 @@ inference:null --enable_mkldnn:False --cpu_threads:1|6 --batchsize:10 ---enable_tensorRT:False --precision:fp32 --model_dir: --data_dir:test_tipc/data/infer @@ -56,4 +55,4 @@ epoch:1 run_mode:PGLBOX fp_items:null device_num:N1C8|N2C16 -gpu_config:models/graph/lightgcn.yaml \ No newline at end of file +gpu_config:models/graph/lightgcn.yaml diff --git a/distributed/CE_PDC/PaddleRec/test_tipc/configs/transformer_conv/train_infer_python.txt b/distributed/CE_PDC/PaddleRec/test_tipc/configs/transformer_conv/train_infer_python.txt index 7fdd2c0e57..8fb97dd5db 100755 --- a/distributed/CE_PDC/PaddleRec/test_tipc/configs/transformer_conv/train_infer_python.txt +++ b/distributed/CE_PDC/PaddleRec/test_tipc/configs/transformer_conv/train_infer_python.txt @@ -42,7 +42,6 @@ inference:null --enable_mkldnn:False --cpu_threads:1|6 --batchsize:10 ---enable_tensorRT:False --precision:fp32 --model_dir: --data_dir:test_tipc/data/infer @@ -56,4 +55,4 @@ epoch:1 run_mode:PGLBOX fp_items:null device_num:N1C8|N2C16 -gpu_config:models/graph/transformer_conv.yaml \ No newline at end of file +gpu_config:models/graph/transformer_conv.yaml diff --git a/framework/e2e/PaddleLT_new/layercase/end2end/csp_darknet_det_backbone.py b/framework/e2e/PaddleLT_new/layercase/end2end/csp_darknet_det_backbone.py index fff75a04bf..067284dfca 100644 --- a/framework/e2e/PaddleLT_new/layercase/end2end/csp_darknet_det_backbone.py +++ b/framework/e2e/PaddleLT_new/layercase/end2end/csp_darknet_det_backbone.py @@ -282,7 +282,7 @@ class LayerCase(nn.Layer): return_idx (list): Index of stages whose feature maps are returned. """ - __shared__ = ['depth_mult', 'width_mult', 'act', 'trt'] + __shared__ = ['depth_mult', 'width_mult', 'act'] # in_channels, out_channels, num_blocks, add_shortcut, use_spp(use_sppf) # 'X' means setting used in YOLOX, 'P5/P6' means setting used in YOLOv5. @@ -302,7 +302,6 @@ def __init__(self, width_mult=1.0, depthwise=False, act='silu', - trt=False, return_idx=[2, 3, 4]): super(LayerCase, self).__init__() self.arch = arch diff --git a/framework/e2e/PaddleLT_new/layercase/end2end/cspresnet_det_backbone.py b/framework/e2e/PaddleLT_new/layercase/end2end/cspresnet_det_backbone.py index 4619cdfd42..6a38820b46 100644 --- a/framework/e2e/PaddleLT_new/layercase/end2end/cspresnet_det_backbone.py +++ b/framework/e2e/PaddleLT_new/layercase/end2end/cspresnet_det_backbone.py @@ -21,16 +21,10 @@ def silu(x): return F.silu(x) -def swish(x): - return x * F.sigmoid(x) - - -TRT_ACT_SPEC = {'swish': swish, 'silu': swish} - ACT_SPEC = {'mish': mish, 'silu': silu} -def get_act_fn(act=None, trt=False): +def get_act_fn(act=None): assert act is None or isinstance(act, ( str, dict)), 'name of activation should be str, dict or None' if not act: @@ -44,9 +38,7 @@ def get_act_fn(act=None, trt=False): name = act kwargs = dict() - if trt and name in TRT_ACT_SPEC: - fn = TRT_ACT_SPEC[name] - elif name in ACT_SPEC: + if name in ACT_SPEC: fn = ACT_SPEC[name] else: fn = getattr(F, name) @@ -251,7 +243,7 @@ def forward(self, x): class LayerCase(nn.Layer): - __shared__ = ['width_mult', 'depth_mult', 'trt'] + __shared__ = ['width_mult', 'depth_mult'] def __init__(self, layers=[3, 6, 6, 3], @@ -262,7 +254,6 @@ def __init__(self, use_large_stem=False, width_mult=1.0, depth_mult=1.0, - trt=False, use_checkpoint=False, use_alpha=False, **args): @@ -270,9 +261,8 @@ def __init__(self, self.use_checkpoint = use_checkpoint channels = [max(round(c * width_mult), 1) for c in channels] layers = [max(round(l * depth_mult), 1) for l in layers] - act = get_act_fn( - act, trt=trt) if act is None or isinstance(act, - (str, dict)) else act + act = get_act_fn(act) if act is None or isinstance(act, + (str, dict)) else act if use_large_stem: self.stem = nn.Sequential( diff --git a/framework/e2e/PaddleLT_new/layercase/end2end/mobileone_det_backbone.py b/framework/e2e/PaddleLT_new/layercase/end2end/mobileone_det_backbone.py index 5321fdae6a..2b369a712c 100644 --- a/framework/e2e/PaddleLT_new/layercase/end2end/mobileone_det_backbone.py +++ b/framework/e2e/PaddleLT_new/layercase/end2end/mobileone_det_backbone.py @@ -20,16 +20,10 @@ def silu(x): return F.silu(x) -def swish(x): - return x * F.sigmoid(x) - - -TRT_ACT_SPEC = {'swish': swish, 'silu': swish} - ACT_SPEC = {'mish': mish, 'silu': silu} -def get_act_fn(act=None, trt=False): +def get_act_fn(act=None): assert act is None or isinstance(act, ( str, dict)), 'name of activation should be str, dict or None' if not act: @@ -43,9 +37,7 @@ def get_act_fn(act=None, trt=False): name = act kwargs = dict() - if trt and name in TRT_ACT_SPEC: - fn = TRT_ACT_SPEC[name] - elif name in ACT_SPEC: + if name in ACT_SPEC: fn = ACT_SPEC[name] else: fn = getattr(F, name) diff --git a/framework/e2e/moduletrans/yaml/Det/modeling/backbones/csp_darknet.yml b/framework/e2e/moduletrans/yaml/Det/modeling/backbones/csp_darknet.yml index cf28efe865..8ddcc67563 100644 --- a/framework/e2e/moduletrans/yaml/Det/modeling/backbones/csp_darknet.yml +++ b/framework/e2e/moduletrans/yaml/Det/modeling/backbones/csp_darknet.yml @@ -403,7 +403,6 @@ csp_darknet_CSPDarkNet_0: width_mult: 1.0 depthwise: False act: 'silu' - trt: False return_idx: [2, 3, 4] DataGenerator: DataGenerator_name: "diy.data.struct_img_dataset.DictImageWithoutLabel" diff --git a/framework/e2e/moduletrans/yaml/Det/modeling/backbones/cspresnet.yml b/framework/e2e/moduletrans/yaml/Det/modeling/backbones/cspresnet.yml index 882e447efc..0de5a6d494 100644 --- a/framework/e2e/moduletrans/yaml/Det/modeling/backbones/cspresnet.yml +++ b/framework/e2e/moduletrans/yaml/Det/modeling/backbones/cspresnet.yml @@ -343,7 +343,6 @@ cspresnet_CSPResNet_0: use_large_stem: False width_mult: 1.0 depth_mult: 1.0 - trt: False use_checkpoint: False use_alpha: False DataGenerator: diff --git a/framework/e2e/moduletrans/yaml/Det/modeling/necks/custom_pan.yml b/framework/e2e/moduletrans/yaml/Det/modeling/necks/custom_pan.yml index 5720099da5..704d19ae95 100644 --- a/framework/e2e/moduletrans/yaml/Det/modeling/necks/custom_pan.yml +++ b/framework/e2e/moduletrans/yaml/Det/modeling/necks/custom_pan.yml @@ -131,7 +131,6 @@ custom_pan_CustomCSPPAN_0: data_format: 'NCHW' width_mult: 1.0 depth_mult: 1.0 - trt: False DataGenerator: DataGenerator_name: "diy.data.single_img_dataset.SingleImageWithoutLabel" data: @@ -210,7 +209,6 @@ custom_pan_CustomCSPPAN_1: data_format: 'NCHW' width_mult: 1.0 depth_mult: 1.0 - trt: False DataGenerator: DataGenerator_name: "diy.data.single_img_dataset.SingleImageWithoutLabel" data: diff --git a/framework/e2e/moduletrans/yaml/Det/modeling/necks/yolo_fpn.yml b/framework/e2e/moduletrans/yaml/Det/modeling/necks/yolo_fpn.yml index 40db662695..74f94d4c03 100644 --- a/framework/e2e/moduletrans/yaml/Det/modeling/necks/yolo_fpn.yml +++ b/framework/e2e/moduletrans/yaml/Det/modeling/necks/yolo_fpn.yml @@ -565,7 +565,6 @@ yolo_fpn_YOLOCSPPAN_0: depthwise: False data_format: 'NCHW' act: 'silu' - trt: False DataGenerator: DataGenerator_name: "diy.data.single_img_dataset.SingleImageWithoutLabel" data: diff --git a/framework/e2e/paddleLT/yaml/Det/modeling/backbones/csp_darknet.yml b/framework/e2e/paddleLT/yaml/Det/modeling/backbones/csp_darknet.yml index cf28efe865..8ddcc67563 100644 --- a/framework/e2e/paddleLT/yaml/Det/modeling/backbones/csp_darknet.yml +++ b/framework/e2e/paddleLT/yaml/Det/modeling/backbones/csp_darknet.yml @@ -403,7 +403,6 @@ csp_darknet_CSPDarkNet_0: width_mult: 1.0 depthwise: False act: 'silu' - trt: False return_idx: [2, 3, 4] DataGenerator: DataGenerator_name: "diy.data.struct_img_dataset.DictImageWithoutLabel" diff --git a/framework/e2e/paddleLT/yaml/Det/modeling/backbones/cspresnet.yml b/framework/e2e/paddleLT/yaml/Det/modeling/backbones/cspresnet.yml index 882e447efc..0de5a6d494 100644 --- a/framework/e2e/paddleLT/yaml/Det/modeling/backbones/cspresnet.yml +++ b/framework/e2e/paddleLT/yaml/Det/modeling/backbones/cspresnet.yml @@ -343,7 +343,6 @@ cspresnet_CSPResNet_0: use_large_stem: False width_mult: 1.0 depth_mult: 1.0 - trt: False use_checkpoint: False use_alpha: False DataGenerator: diff --git a/framework/e2e/paddleLT/yaml/Det/modeling/necks/custom_pan.yml b/framework/e2e/paddleLT/yaml/Det/modeling/necks/custom_pan.yml index 5720099da5..704d19ae95 100644 --- a/framework/e2e/paddleLT/yaml/Det/modeling/necks/custom_pan.yml +++ b/framework/e2e/paddleLT/yaml/Det/modeling/necks/custom_pan.yml @@ -131,7 +131,6 @@ custom_pan_CustomCSPPAN_0: data_format: 'NCHW' width_mult: 1.0 depth_mult: 1.0 - trt: False DataGenerator: DataGenerator_name: "diy.data.single_img_dataset.SingleImageWithoutLabel" data: @@ -210,7 +209,6 @@ custom_pan_CustomCSPPAN_1: data_format: 'NCHW' width_mult: 1.0 depth_mult: 1.0 - trt: False DataGenerator: DataGenerator_name: "diy.data.single_img_dataset.SingleImageWithoutLabel" data: diff --git a/framework/e2e/paddleLT/yaml/Det/modeling/necks/yolo_fpn.yml b/framework/e2e/paddleLT/yaml/Det/modeling/necks/yolo_fpn.yml index 40db662695..74f94d4c03 100644 --- a/framework/e2e/paddleLT/yaml/Det/modeling/necks/yolo_fpn.yml +++ b/framework/e2e/paddleLT/yaml/Det/modeling/necks/yolo_fpn.yml @@ -565,7 +565,6 @@ yolo_fpn_YOLOCSPPAN_0: depthwise: False data_format: 'NCHW' act: 'silu' - trt: False DataGenerator: DataGenerator_name: "diy.data.single_img_dataset.SingleImageWithoutLabel" data: diff --git a/inference/benchmark/jetson/LICENSE.md b/inference/benchmark/jetson/LICENSE.md deleted file mode 100644 index e980e71ba4..0000000000 --- a/inference/benchmark/jetson/LICENSE.md +++ /dev/null @@ -1,7 +0,0 @@ -Copyright (c) 2019-2020, NVIDIA CORPORATION. All rights reserved. - -Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. diff --git a/inference/benchmark/jetson/README.md b/inference/benchmark/jetson/README.md deleted file mode 100644 index 6e6a65fa2c..0000000000 --- a/inference/benchmark/jetson/README.md +++ /dev/null @@ -1,94 +0,0 @@ -# Benchmarks For Jetson - -## 参考链接： - -```shell -https://github.com/NVIDIA-AI-IOT/jetson_benchmarks -``` - -## 环境要求 -* JetPack 4.4 / 4.5 / 4.6 -* TensorRT 7 / 8 -* cuda -* cudnn - -## 前置条件 -```shell -mkdir models # 创建模型文件路径 -sudo sh install_requirements.sh #安装相关依赖库 -``` -## 下载模型 -```shell -# Jetson NX -python3 utils/download_models.py --all --csv_file_path ./benchmark_csv/nx-benchmarks.csv --save_dir ./models -# Jetson AGX -python3 utils/download_models.py --all --csv_file_path ./benchmark_csv/xavier-benchmarks.csv --save_dir ./models -# Jetson TX2 -python3 utils/download_models.py --all --csv_file_path ./benchmark_csv/tx2-nano-benchmarks.csv --save_dir ./models -``` - -## 运行单测(注意model 为绝对路径 否则会报错) -1.运行所有模型 -```shell -#Jetson NX -sudo python3 benchmark.py --all --csv_file_path ./benchmark_csv/nx-benchmarks.csv --model_dir model绝对路径 -#Jetson AGX -sudo python3 benchmark.py --all --csv_file_path ./benchmark_csv/xavier-benchmarks.csv \ - --model_dir model绝对路径 \ - --jetson_devkit xavier \ - --gpu_freq 1377000000 --dla_freq 1395200000 --power_mode 0 -#Jetson TX2 -sudo python3 benchmark.py --all --csv_file_path ./benchmark_csv/tx2-nano-benchmarks.csv \ - --model_dir 绝对路径 \ - --jetson_devkit tx2 \ - --gpu_freq 1122000000 --power_mode 3 --precision fp16 -``` -2.运行单个模型 -```shell -#Jetson NX -sudo python3 benchmark.py --model_name inception_v4 --csv_file_path ./benchmark_csv/nx-benchmarks.csv --model_dir model绝对路径 -#Jetson AGX -sudo python3 benchmark.py --model_name inception_v4 --csv_file_path ./benchmark_csv/xavier-benchmarks.csv \ - --model_dir model绝对路径 \ - --jetson_devkit xavier \ - --gpu_freq 1377000000 --dla_freq 1395200000 --power_mode 0 -#Jetson TX2 -sudo python3 benchmark.py --model_name inception_v4 --csv_file_path ./benchmark_csv/tx2-nano-benchmarks.csv \ - --model_dir 绝对路径 \ - --jetson_devkit tx2 \ - --gpu_freq 1122000000 --power_mode 3 --precision fp16 -``` -##注意事项 -* 由于 jetson tx2 和 nano 只有 GPU没有 DLA，结果获取的 FPS 即为 真实的GPU的 QPS -* 对于 jetson NX 和 AGX 存在 GPU和 DLA，结果获取的 FPS ： QPS （GPU）+ 2 ✖️ QPS（DLA ），因此需要修改单测文件只测试 GPU ： -```shell -# 1. 修改 jetson_benchmarks/utils/load_store_engine.py 12行 -self.num_devices = 1 # 3 if GPU+2DLA, 1 if GPU Only -# 2. 修改 jetson_benchmarks/utils/read_write_data.py 17行 -self.num_devices = 1 # data['Devices'][read_index] -``` -*还可以将 jetson_benchmarks/utils/load_store_engine.py 151-152 行 注释，从而在 jetson_benchmarks/models/inception_v4_b4_ws2048_gpu.txt 查看详细的输出耗时、性能分析数据。 -```shell -# 注释 jetson_benchmarks/utils/load_store_engine.py 151-152 行 -# if os.path.isfile(_txtout_path): -# os.remove(_txtout_path) -``` - -##Paddle GPU 测试步骤 -1.采用 E2E 测试方式 -```shell -time1 = time.time() -# core.nvprof_start() -# core.nvprof_enable_record_event() -for i in range(args.repeats): - # core.nvprof_nvtx_push("forward " + str(i)) - input_tensor.copy_from_cpu(img[0].copy()) - predictor.run() - output_names = predictor.get_output_names() - output_handle = predictor.get_output_handle(output_names[0]) - output_data = output_handle.copy_to_cpu() # numpy.ndarray类型 -# core.nvprof_nvtx_pop() -# core.nvprof_stop() -time2 = time.time() -total_inference_cost = (time2 - time1) * 1000 # total latency, ms -``` diff --git a/inference/benchmark/jetson/benchmark.py b/inference/benchmark/jetson/benchmark.py deleted file mode 100644 index 9fe9658982..0000000000 --- a/inference/benchmark/jetson/benchmark.py +++ /dev/null @@ -1,122 +0,0 @@ -#!/usr/bin/python -from utils import utilities, read_write_data, benchmark_argparser, run_benchmark_models -import sys -import os -import pandas as pd -import gc -import warnings -warnings.simplefilter("ignore") - -def main(): - # Set Parameters - arg_parser = benchmark_argparser() - args = arg_parser.make_args() - csv_file_path = args.csv_file_path - model_path = args.model_dir - precision = args.precision - - # System Check - system_check = utilities(jetson_devkit=args.jetson_devkit, gpu_freq=args.gpu_freq, dla_freq=args.dla_freq) - #system_check.close_all_apps() - if system_check.check_trt(): - sys.exit() - system_check.set_power_mode(args.power_mode, args.jetson_devkit) - system_check.clear_ram_space() - if args.jetson_clocks: - system_check.set_jetson_clocks() - else: - system_check.run_set_clocks_withDVFS() - system_check.set_jetson_fan(255) - - # Read CSV and Write Data - benchmark_data = read_write_data(csv_file_path=csv_file_path, model_path=model_path) - if args.all: - latency_each_model =[] - print("Running all benchmarks.. This will take at least 2 hours...") - for read_index in range (0,len(benchmark_data)): - gc.collect() - model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data) - download_err = model.execute(read_index=read_index) - if not download_err: - # Reading Results - latency_fps, error_log = model.report() - latency_each_model.append(latency_fps) - # Remove engine and txt files - if not error_log: - model.remove() - del gc.garbage[:] - system_check.clear_ram_space() - benchmark_table = pd.DataFrame(latency_each_model, columns=["GPU (ms)", "DLA0 (ms)", "DLA1 (ms)", "FPS", "Model Name"], dtype=float) - # Note: GPU, DLA latencies are measured in miliseconds, FPS = Frames per Second - print(benchmark_table[["Model Name", "FPS"]]) - if args.plot: - benchmark_data.plot_perf(latency_each_model) - - elif args.model_name == "inception_v4": - model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data) - download_err = model.execute(read_index=0) - if not download_err: - _, error_log = model.report() - if not error_log: - model.remove() - - elif args.model_name == "vgg19": - model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data) - download_err = model.execute(read_index=1) - if not download_err: - _, error_log = model.report() - if not error_log: - model.remove() - - elif args.model_name == "super_resolution": - model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data) - download_err = model.execute(read_index=2) - if not download_err: - _, error_log = model.report() - if not error_log: - model.remove() - - elif args.model_name == "unet": - model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data) - download_err = model.execute(read_index=3) - if not download_err: - _, error_log = model.report() - if not error_log: - model.remove() - - elif args.model_name == "pose_estimation": - model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data) - download_err = model.execute(read_index=4) - if not download_err: - _, error_log = model.report() - if not error_log: - model.remove() - - elif args.model_name == "tiny-yolov3": - model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data) - download_err = model.execute(read_index=5) - if not download_err: - _, error_log = model.report() - if not error_log: - model.remove() - - elif args.model_name == "resnet": - model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data) - download_err = model.execute(read_index=6) - if not download_err: - _, error_log = model.report() - if not error_log: - model.remove() - - elif args.model_name == "ssd-mobilenet-v1": - model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data) - download_err = model.execute(read_index=7) - if not download_err: - _, error_log = model.report() - if not error_log: - model.remove() - - system_check.clear_ram_space() - system_check.set_jetson_fan(0) -if __name__ == "__main__": - main() diff --git a/inference/benchmark/jetson/benchmark_csv/nx-benchmarks.csv b/inference/benchmark/jetson/benchmark_csv/nx-benchmarks.csv deleted file mode 100644 index 0d6331091e..0000000000 --- a/inference/benchmark/jetson/benchmark_csv/nx-benchmarks.csv +++ /dev/null @@ -1,9 +0,0 @@ -ModelName,FrameWork,Devices,BatchSizeGPU,BatchSizeDLA,WS_GPU,WS_DLA,input,output,URL -inception_v4,caffe,3,2,1,2048,1024,NA,prob,https://www.dropbox.com/s/b7masj8xdoycv2w/inception_v4.prototxt -vgg19_N2,caffe,1,1,0,2048,0,NA,prob,https://www.dropbox.com/s/t4qq079g5q4jibx/vgg19_N2.prototxt -super_resolution_bsd500,onnx,1,2,0,2048,0,NA,NA,https://www.dropbox.com/s/hdhxndo23cm9i5y/super_resolution_bsd500.zip -unet-segmentation,tensorrt,1,2,0,2048,None,"input_1,1,512,512",conv2d_19/Sigmoid,https://www.dropbox.com/s/85lttamnbjeig0e/unet-segmentation.uff -pose_estimation,caffe,1,2,0,2048,None,NA,Mconv7_stage2_L2,https://www.dropbox.com/s/hwa5i14v67u57ij/pose_estimation.prototxt -yolov3-tiny-416,onnx,1,8,0,2048,0,NA,NA,https://www.dropbox.com/s/ck9e40b57rd5o14/yolov3-tiny-416.zip -ResNet50_224x224,caffe,3,4,2,2048,1024,NA,prob,https://www.dropbox.com/s/9ohk387v0ki56wx/ResNet50_224x224.prototxt -ssd-mobilenet-v1,onnx,3,8,2,2048,1024,NA,NA,https://www.dropbox.com/s/gx5zayt76vszhpo/ssd-mobilenet-v1.zip \ No newline at end of file diff --git a/inference/benchmark/jetson/benchmark_csv/tx2-nano-benchmarks.csv b/inference/benchmark/jetson/benchmark_csv/tx2-nano-benchmarks.csv deleted file mode 100644 index b0e1b6f3fc..0000000000 --- a/inference/benchmark/jetson/benchmark_csv/tx2-nano-benchmarks.csv +++ /dev/null @@ -1,9 +0,0 @@ -ModelName,FrameWork,Devices,BatchSizeGPU,BatchSizeDLA,WS_GPU,WS_DLA,input,output,URL -inception_v4,caffe,1,1,0,1024,0,NA,prob,https://www.dropbox.com/s/b7masj8xdoycv2w/inception_v4.prototxt -vgg19_N2,caffe,1,1,0,512,0,NA,prob,https://www.dropbox.com/s/t4qq079g5q4jibx/vgg19_N2.prototxt -super_resolution_bsd500,onnx,1,1,0,1024,0,NA,NA,https://www.dropbox.com/s/hdhxndo23cm9i5y/super_resolution_bsd500.zip -unet-segmentation,tensorrt,1,1,0,1024,0,"input_1,1,512,512",conv2d_19/Sigmoid,https://www.dropbox.com/s/85lttamnbjeig0e/unet-segmentation.uff -pose_estimation,caffe,1,1,0,1024,0,NA,Mconv7_stage2_L2,https://www.dropbox.com/s/hwa5i14v67u57ij/pose_estimation.prototxt -yolov3-tiny-416,onnx,1,1,0,1024,0,NA,NA,https://www.dropbox.com/s/ck9e40b57rd5o14/yolov3-tiny-416.zip -ResNet50_224x224,caffe,1,1,0,1024,0,NA,prob,https://www.dropbox.com/s/9ohk387v0ki56wx/ResNet50_224x224.prototxt -ssd-mobilenet-v1,onnx,1,1,0,1024,0,NA,NA,https://www.dropbox.com/s/gx5zayt76vszhpo/ssd-mobilenet-v1.zip diff --git a/inference/benchmark/jetson/benchmark_csv/xavier-benchmarks.csv b/inference/benchmark/jetson/benchmark_csv/xavier-benchmarks.csv deleted file mode 100644 index da3f65134a..0000000000 --- a/inference/benchmark/jetson/benchmark_csv/xavier-benchmarks.csv +++ /dev/null @@ -1,9 +0,0 @@ -ModelName,FrameWork,Devices,BatchSizeGPU,BatchSizeDLA,WS_GPU,WS_DLA,input,output,URL -inception_v4,caffe,3,4,1,2048,2048,NA,prob,https://www.dropbox.com/s/b7masj8xdoycv2w/inception_v4.prototxt -vgg19_N2,caffe,1,4,0,2048,0,NA,prob,https://www.dropbox.com/s/t4qq079g5q4jibx/vgg19_N2.prototxt -super_resolution_bsd500,onnx,1,4,0,2048,2048,NA,NA,https://www.dropbox.com/s/hdhxndo23cm9i5y/super_resolution_bsd500.zip -unet-segmentation,tensorrt,1,2,0,2048,None,"input_1,1,512,512",conv2d_19/Sigmoid,https://www.dropbox.com/s/85lttamnbjeig0e/unet-segmentation.uff -pose_estimation,caffe,1,4,0,2048,None,NA,Mconv7_stage2_L2,https://www.dropbox.com/s/hwa5i14v67u57ij/pose_estimation.prototxt -yolov3-tiny-416,onnx,1,16,0,2048,2048,NA,NA,https://www.dropbox.com/s/ck9e40b57rd5o14/yolov3-tiny-416.zip -ResNet50_224x224,caffe,3,16,4,2048,2048,NA,prob,https://www.dropbox.com/s/9ohk387v0ki56wx/ResNet50_224x224.prototxt -ssd-mobilenet-v1,onnx,3,16,2,2048,2048,NA,NA,https://www.dropbox.com/s/gx5zayt76vszhpo/ssd-mobilenet-v1.zip diff --git a/inference/benchmark/jetson/install_requirements.sh b/inference/benchmark/jetson/install_requirements.sh deleted file mode 100644 index 2d461c6b06..0000000000 --- a/inference/benchmark/jetson/install_requirements.sh +++ /dev/null @@ -1,8 +0,0 @@ -#!/bin/bash -sudo apt-get update -sudo apt-get install -y python3-pip -sudo python3 -m pip install Cython -sudo python3 -m pip install numpy -sudo python3 -m pip install pandas -sudo apt-get install -y python3-matplotlib -sudo apt-get install -y python3-cairocffi diff --git a/inference/benchmark/jetson/utils/__init__.py b/inference/benchmark/jetson/utils/__init__.py deleted file mode 100644 index 6b0b873fff..0000000000 --- a/inference/benchmark/jetson/utils/__init__.py +++ /dev/null @@ -1,6 +0,0 @@ -from .download_models import download_models -from .load_store_engine import load_store_engine -from .read_write_data import read_write_data -from .utilities import utilities -from .benchmark_argparser import benchmark_argparser -from .run_benchmark_models import run_benchmark_models diff --git a/inference/benchmark/jetson/utils/benchmark_argparser.py b/inference/benchmark/jetson/utils/benchmark_argparser.py deleted file mode 100644 index b47a211d76..0000000000 --- a/inference/benchmark/jetson/utils/benchmark_argparser.py +++ /dev/null @@ -1,37 +0,0 @@ -import argparse - -class benchmark_argparser(): - def __init__(self): - self.parser = argparse.ArgumentParser(description="") - self.parser.add_argument("--csv_file_path", dest="csv_file_path", help="csv for model download and parameters", type=str) - self.parser.add_argument("--model_dir", dest="model_dir", help="path to downloaded path", type=str) - benchmark_group = self.parser.add_mutually_exclusive_group() - benchmark_group.add_argument("--model_name", dest="model_name", help="only specified models will be executed", type=str) - benchmark_group.add_argument("--all", dest="all", help="all models from DropBox will be downloaded", - action="store_true") - self.parser.add_argument("--jetson_devkit", dest="jetson_devkit", default="xavier-nx", help="Input Jetson Devkit name", type=str) - # For Jetson Xavier: set to "xavier" - # For Jetson TX2: set to "tx2" - # For Jetson Nano: set to "nano" - self.parser.add_argument("--power_mode", dest="power_mode", help="Jetson Power Mode", default=0, type=int) - # For Jetson Xavier: set to 0 (MAXN) - # For Jetson TX2: set to 3 (MAXP) - # For Jetson Nano: set to 0 (MAXN) - self.parser.add_argument("--precision", dest="precision", default="int8", - help="precision for model int8 or fp16", type=str) - # For Jetson Xavier: set to int8 - # For Jetson TX2: set to 3 fp16 - # For Jetson Nano: set to fp16 - self.parser.add_argument("--jetson_clocks", dest="jetson_clocks", help="Set Clock Frequency to Max (jetson_clocks)", - action="store_true") - self.parser.add_argument("--gpu_freq", dest="gpu_freq", default=1109250000,help="set GPU frequency", type=int) - # Default values are for Xavier-NX - # For Xavier set gpu_freq to 1377000000: Find using $sudo cat /sys/devices/17000000.gv11b/devfreq/17000000.gv11b/available_frequencies - # For TX2 set gpu freq to 1300500000: Find using $sudo cat /sys/devices/gpu.0/devfreq/17000000.gp10b/available_frequencies - # For Nano set gpu freq to 921600000: Find using $sudo cat /sys/devices/gpu.0/devfreq/57000000.gpu/available_frequencies - self.parser.add_argument("--dla_freq", dest="dla_freq", default=1100800000, help="set DLA frequency", type=int) - # Default values are for Xavier-NX - # For Xavier set dla_freq to 1395200000 : Find using $sudo cat /sys/kernel/debug/bpmp/debug/clk/nafll_dla/max_rate - self.parser.add_argument("--plot", dest="plot", help="Perf in Graph", action="store_true") - def make_args(self): - return self.parser.parse_args() diff --git a/inference/benchmark/jetson/utils/download_models.py b/inference/benchmark/jetson/utils/download_models.py deleted file mode 100644 index aaccfe1aef..0000000000 --- a/inference/benchmark/jetson/utils/download_models.py +++ /dev/null @@ -1,82 +0,0 @@ -#!/usr/bin/python -import argparse -import pandas as pd -import subprocess -import shlex -import os - -def download_models(url, save_dir): - cmd = "wget --quiet --show-progress --progress=bar:force:noscroll --auth-no-challenge --no-check-certificate"+ " " + url+ " "+"-P"+ save_dir - args = shlex.split(cmd) - subprocess.call(args) - -def unzip_model_files(model_name,save_dir): - model_file_path = os.path.join(save_dir,model_name+".zip") - cmd = "unzip -qq"+" " +str(model_file_path)+" "+"-d"+" "+save_dir - args = shlex.split(cmd) - subprocess.call(args) - cmd_rm = "rm -rf"+" "+model_file_path - args_rm = shlex.split(cmd_rm) - subprocess.call(args_rm) - -def download_argparser(): - parser = argparse.ArgumentParser(description="Download Models from DropBox") - parser.add_argument("--save_dir", dest="save_dir", help="downloaded files will be stored here", type=str) - parser.add_argument("--csv_file_path", dest="csv_file_path", default="./nx-benchmarks.csv", help="csv contains url to download model", type=str) - downloader_group = parser.add_mutually_exclusive_group() - downloader_group.add_argument("--all", dest="all", help="all models from DropBox will be downloaded", action="store_true") - downloader_group.add_argument("--model_name", dest="model_name", help="only specified models will be downloaded", type=str) - args = parser.parse_args() - return args - -def main(): - downloader_args = download_argparser() - csv_file = downloader_args.csv_file_path - save_dir = downloader_args.save_dir - if downloader_args.all: - len_csv = len(pd.read_csv(csv_file)) - for read_index in range (0,len_csv): - url = pd.read_csv(csv_file)["URL"][read_index] - framework = pd.read_csv(csv_file)["FrameWork"][read_index] - if framework == "onnx": - model_name = pd.read_csv(csv_file)["ModelName"][read_index] - download_models(str(url), save_dir) - unzip_model_files(model_name=model_name, save_dir=save_dir) - else: - download_models(str(url), save_dir) - elif downloader_args.model_name == "inception_v4": - url = pd.read_csv(csv_file)["URL"][0] - download_models(url, save_dir) - elif downloader_args.model_name == "vgg19": - url = pd.read_csv(csv_file)["URL"][1] - download_models(url, save_dir) - elif downloader_args.model_name == "super_resolution": - url = pd.read_csv(csv_file)["URL"][2] - model_name = pd.read_csv(csv_file)["ModelName"][2] - download_models(str(url), save_dir) - unzip_model_files(model_name=model_name, save_dir=save_dir) - elif downloader_args.model_name == "unet": - url = pd.read_csv(csv_file)["URL"][3] - download_models(url, save_dir) - elif downloader_args.model_name == "pose_estimation": - url = pd.read_csv(csv_file)["URL"][4] - download_models(url, save_dir) - elif downloader_args.model_name == "tiny-yolov3": - url = pd.read_csv(csv_file)["URL"][5] - model_name = pd.read_csv(csv_file)["ModelName"][5] - download_models(str(url), save_dir) - unzip_model_files(model_name=model_name, save_dir=save_dir) - elif downloader_args.model_name == "resnet": - url = pd.read_csv(csv_file)["URL"][6] - download_models(url, save_dir) - elif downloader_args.model_name == "ssd-mobilenet-v1": - url = pd.read_csv(csv_file)["URL"][7] - model_name = pd.read_csv(csv_file)["ModelName"][7] - download_models(str(url), save_dir) - unzip_model_files(model_name=model_name, save_dir=save_dir) - -if __name__ == "__main__": - main() - - - diff --git a/inference/benchmark/jetson/utils/load_store_engine.py b/inference/benchmark/jetson/utils/load_store_engine.py deleted file mode 100644 index aeb2914c5d..0000000000 --- a/inference/benchmark/jetson/utils/load_store_engine.py +++ /dev/null @@ -1,150 +0,0 @@ -#!/usr/bin/python -import os -import subprocess -import threading -import time - -# Class for load, store, remove engine -class load_store_engine(): - def __init__(self, model_path, model_name, batch_size_gpu, batch_size_dla, num_devices, precision, ws_gpu, ws_dla, model_input, model_output ): - self.model_path = model_path # Directory - self.model_name = model_name # Model Name - self.num_devices = num_devices # 3 if GPU+2DLA, 1 if GPU Only - self.precision = precision # float16 or int8 - self.batch_size_gpu = batch_size_gpu # Batch Size for GPU - self.batch_size_dla = batch_size_dla # Batch Size for DLA - self.ws_gpu = ws_gpu # Workspace required for GPU - self.ws_dla =ws_dla # Workspace required for DLA - self. model_input = model_input # Input name of the model - self.model_output = model_output # Output name of the model - self.trt_process = [] - - def engine_gen(self): - cmd = [] - model = [] - self.framework = os.path.splitext(self.model_name)[1] - precision_cmd = str("--" + str(self.precision)) - for device_id in range(0, self.num_devices): - if device_id == 1 or device_id == 2: - self.device = "dla" - model_base_path = self._model2deploy() - dla_cmd = str("--useDLACore=" + str(device_id - 1)) - workspace_cmd = str("--workspace=" + str(self.ws_dla)) - _model = str(os.path.splitext(self.model_name)[0]) + "_b" + str(self.batch_size_dla) + "_ws" + str( - self.ws_dla) + "_" + str(self.device) + str(device_id) - engine_CMD = str( - "./trtexec" + " " + model_base_path + " " + precision_cmd + " " +"--allowGPUFallback" + " " + " " + dla_cmd + " " + - workspace_cmd) - else: - self.device = "gpu" - model_base_path = self._model2deploy() - workspace_cmd = str("--workspace=" + str(self.ws_gpu)) - _model = str(os.path.splitext(self.model_name)[0]) + "_b" + str(self.batch_size_gpu) + "_ws" + str( - self.ws_gpu) + "_" + str(self.device) - engine_CMD = str( - "./trtexec" + " " + model_base_path + " " + precision_cmd + " " + workspace_cmd) - cmd.append(engine_CMD) - model.append(_model) - return cmd, model - - def check_downloaded_models(self, model_name, framework): - model_files = [] - if framework == str("onnx"): - model_name_split = os.path.splitext(model_name)[0] - model_files.append(str(model_name_split + "-bs" + str(self.batch_size_gpu) + "." + framework)) - if self.num_devices > 1: - model_files.append(str(model_name_split + "-bs" + str(self.batch_size_dla) + "." + framework)) - else: - model_files.append(model_name) - - for e_id in range(0, len(model_files)): - model_file = os.path.join(self.model_path, model_files[e_id]) - if not os.path.isfile(model_file): - print("Could Not find model file {} in {}\nPlease Download all model files".format(model_files[e_id], self.model_path)) - return True - return False - - def _model2deploy(self): - if self.framework == str(".prototxt"): - _model_output = str("--output=" + str(self.model_output)) - _model_base = str("--deploy=" + str(os.path.join(self.model_path, self.model_name))) - if self.device=="gpu": - batch_cmd = str("--batch=" + str(self.batch_size_gpu)) - elif self.device == "dla": - batch_cmd = str("--batch=" + str(self.batch_size_dla)) - return str(_model_output + " " + _model_base+ " " + batch_cmd) - if self.framework == str(".onnx"): - batch_cmd = str("--explicitBatch") - model_name_split = os.path.splitext(self.model_name)[0] - if self.device == "gpu": - model_onnx = str(model_name_split+"-bs"+str(self.batch_size_gpu)+self.framework) - if self.device == "dla": - model_onnx = str(model_name_split+"-bs"+str(self.batch_size_dla)+self.framework) - return str("--onnx=" + str(os.path.join(self.model_path, model_onnx))+ " " + batch_cmd) - if self.framework == str(".uff"): - _model_input = str("--uffInput="+str(self.model_input)) - _model_output = str("--output="+str(self.model_output)) - _model_base = str("--uff=" + str(os.path.join(self.model_path, self.model_name))) - if self.device == "gpu": - batch_cmd = str("--batch=" + str(self.batch_size_gpu)) - elif self.device == "dla": - batch_cmd = str("--batch=" + str(self.batch_size_dla)) - return str(_model_input+" "+_model_output+" "+_model_base+ " " + batch_cmd) - - - def save_engine(self, _cmds, _models): - save_engine_path = str("--saveEngine=" + str(os.path.join(self.model_path, _models)) + ".engine") - cmd = str(_cmds)+" "+str(save_engine_path) - trt_process = subprocess.Popen([cmd], cwd="/usr/src/tensorrt/bin/", shell=True, stdout=subprocess.DEVNULL, - stderr=subprocess.STDOUT) - while trt_process.poll() == None: - trt_process.poll() - trt_process.kill() - - def save_all(self, commands, models): - for e_id in range(0, self.num_devices): - self.save_engine(commands[e_id], models[e_id]) - - def load_engine(self, _cmds, _models, load_output): - load_engine_path = str("--loadEngine=" + str(os.path.join(self.model_path, _models)) + ".engine") - avgruns_cmd = str("--avgRuns=100")+" "+"--duration=180" - cmd = str(_cmds)+" "+ avgruns_cmd + " " + str(load_engine_path) - _trt_process = subprocess.Popen([cmd], cwd="/usr/src/tensorrt/bin/", shell=True, stdout=load_output, - stderr=subprocess.STDOUT) - self.trt_process.append(_trt_process) - - def load_all(self, commands, models): - load_threads = [] - load_file_list = [] - for e_id in range(0, self.num_devices): - load_file = os.path.join(self.model_path, models[e_id] + ".txt") - load_output = open(load_file, "w") - _load_threads = threading.Thread(target=self.load_engine(commands[e_id], models[e_id], load_output)) - load_threads.append(_load_threads) - load_file_list.append(load_output) - time.sleep(10)# Load memory - # Start Threads - for lt in load_threads: - lt.start() - # Wait till threads are synchronize - for lt in load_threads: - lt.join() - # Kill the subprocessess once complete - for tp in self.trt_process: - while tp.poll() == None: - tp.poll() - tp.kill() - for flist in load_file_list: - flist.close() - - def remove_engine(self, models): - _engine_path = str(str(os.path.join(self.model_path, models)) + ".engine") - _txtout_path = str(str(os.path.join(self.model_path, models)) + ".txt") - if os.path.isfile(_engine_path): - os.remove(_engine_path) - if os.path.isfile(_txtout_path): - os.remove(_txtout_path) - - def remove_all(self, models): - for e_id in range(0, self.num_devices): - self.remove_engine(models[e_id]) diff --git a/inference/benchmark/jetson/utils/read_write_data.py b/inference/benchmark/jetson/utils/read_write_data.py deleted file mode 100644 index cf2b3645a6..0000000000 --- a/inference/benchmark/jetson/utils/read_write_data.py +++ /dev/null @@ -1,149 +0,0 @@ -import pandas as pd -import re -import os -import datetime -from datetime import timedelta -# Class for read csv and write to csv or panda or graph generate -class read_write_data(): - def __init__(self, csv_file_path, model_path): - self.csv_file_path =csv_file_path - self.model_path = model_path - self.start_valid_time = 0 - self.end_valid_time = 0 - def benchmark_csv(self, read_index): - data = pd.read_csv(self.csv_file_path) - model_name = data["ModelName"][read_index] - self.framework = data["FrameWork"][read_index] - self.num_devices = data["Devices"][read_index] - ws_gpu = data["WS_GPU"][read_index] - ws_dla = data["WS_DLA"][read_index] - model_input = data["input"][read_index] - model_output = data["output"][read_index] - batch_size_gpu = data["BatchSizeGPU"][read_index] - batch_size_dla = int(data["BatchSizeDLA"][read_index]) - return model_name, self.framework, self.num_devices, ws_gpu, ws_dla, model_input, model_output, batch_size_gpu, batch_size_dla - def __len__(self): - return len(pd.read_csv(self.csv_file_path)) - def framework2ext(self): - if self.framework == "caffe": - return str("prototxt") - if self.framework == "onnx": - return str("onnx") - if self.framework == "tensorrt": - return str("uff") - - def read_window_results(self, models): - self.time_value_window = [] - lpd = [0]*3 - thread_start_time = [datetime.datetime(1940, 12, 1, 23, 59, 59)]*3 - thread_end_time = [datetime.datetime(2040, 12, 1, 23, 59, 59)] * 3 - for e_id in range(0, self.num_devices): - read_file = os.path.join(self.model_path, str(models[e_id]) + ".txt") - thread_start_time[e_id], thread_end_time[e_id], thread_time_stamps, thread_latency = self.read_perf_time(read_file) - self.time_value_window.append([thread_time_stamps, thread_latency]) - try: - self.late_start(gpu_st=thread_start_time[0], dla0_st=thread_start_time[1], dla1_st=thread_start_time[2]) - self.earliest_end(gpu_et=thread_end_time[0], dla0_et=thread_end_time[1], dla1_et=thread_end_time[2]) - valid_window_frame = self.end_valid_time - self.start_valid_time - for e_id in range(0, len(self.time_value_window)): - lpd[e_id] = self.calculate_avg_latency(self.time_value_window[e_id]) - except IndexError: - pass - - return lpd[0], lpd[1], lpd[2] - - def earliest_end(self, gpu_et, dla0_et, dla1_et): - if (gpu_et < dla0_et) and (gpu_et < dla1_et): - self.end_valid_time = gpu_et - elif (dla0_et < gpu_et) and (dla0_et < dla1_et): - self.end_valid_time = dla0_et - else: - self.end_valid_time = dla1_et - - def late_start(self, gpu_st, dla0_st, dla1_st): - if (gpu_st > dla0_st) and (gpu_st > dla1_st): - self.start_valid_time = gpu_st - elif (dla0_st > gpu_st) and (dla0_st > dla1_st): - self.start_valid_time = dla0_st - else: - self.start_valid_time = dla1_st - - def read_perf_time(self,read_file): - time_stamps = [] - latencies = [] - add_time = 0 - start_time = datetime.datetime(1940, 12, 1, 23, 59, 59) - end_time = datetime.datetime(2040, 12, 1, 23, 59, 59) - with open(read_file, "r") as f: - for line in f: - if "Starting" in line: - match_start = re.search(r"\d{2}/\d{2}/\d{4}-\d{2}:\d{2}:\d{2}", line) - if match_start: - start_time = datetime.datetime.strptime(match_start.group(), "%m/%d/%Y-%H:%M:%S") - elif "Average on" in line: - matches = re.search(r"Average\s+on\s+(\d+)\s+runs.*?" - r"GPU\s+latency:\s+(\d+\.\d+)\s+.*?" - r"end\s+to\s+end\s+(\d+\.\d+)\s+ms", line) - if matches: - add_time += float(matches.group(1)) * float(matches.group(3)) / 1000 - time_thread = start_time + timedelta(seconds=add_time) - time_stamps.append(time_thread) - latencies.append(float(matches.group(2))) - else: - continue - if time_stamps: - end_time = time_stamps[len(time_stamps)-1] - return start_time, end_time, time_stamps, latencies - - def calculate_avg_latency(self, time_list): - _latency = 0 - count = 0 - for i in range(0, len(time_list[0])): - if self.start_valid_time < time_list[0][i] < self.end_valid_time: - _latency += time_list[1][i] - count += 1 - try: - return _latency / count - except ZeroDivisionError: - return 0 - - - def calculate_fps(self, models, batch_size_gpu, batch_size_dla): - latency_device = [0] * 5 - FPS = 0 - error_read = 0 - latency_device[0], latency_device[1], latency_device[2] = self.read_window_results(models) - for e_id in range(0, self.num_devices): - if latency_device[e_id] != 0: - if e_id ==0: - FPS += batch_size_gpu * (1000 / latency_device[e_id]) - elif e_id ==1 or e_id == 2: - FPS += batch_size_dla * (1000 / latency_device[e_id]) - else: - print("Error in Build, Please check the log in: {}".format(self.model_path)) - error_read = 1 - continue - if any(latency is 0 for latency in latency_device[0:self.num_devices]): - latency_device[len(latency_device) - 2] = 0 - print("We recommend to run benchmarking in headless mode") - else: - latency_device[len(latency_device)-2] = FPS - return latency_device, error_read - - def plot_perf(self, latency_each_model): - import matplotlib - matplotlib.use("Gtk3Agg") - import matplotlib.pyplot as plt - name = [] - fps = [] - for models in range(0,len(latency_each_model)): - fps.append(latency_each_model[models][len(latency_each_model[0]) - 2]) - name.append(latency_each_model[models][len(latency_each_model[0]) - 1]) - plt.bar(name, fps) - plt.figure(figsize=(20, 7)) - plt.bar(name, fps, color="Green") - plt.ylabel("FPS") - plt.title("Benchmark Analysis on Jetson") - plt.grid() - plt.savefig(str(os.path.join(self.model_path, str("perf_results.png")))) - print("Please find benchmark results in {}".format(self.model_path)) diff --git a/inference/benchmark/jetson/utils/run_benchmark_models.py b/inference/benchmark/jetson/utils/run_benchmark_models.py deleted file mode 100644 index 8e11b6df7e..0000000000 --- a/inference/benchmark/jetson/utils/run_benchmark_models.py +++ /dev/null @@ -1,47 +0,0 @@ -from utils import load_store_engine, read_write_data, utilities -import time -import subprocess - -class run_benchmark_models(): - def __init__(self, csv_file_path, model_path, precision, benchmark_data): - self.benchmark_data = benchmark_data - self.model_path = model_path - self.precision = precision - self.wall_time = 0 - self.download_error_flag = False - def execute(self, read_index): - wall_start_t0 = time.time() - self.model_name, _framework, num_devices, ws_gpu, ws_dla, model_input, model_output, self.batch_size_gpu, self.batch_size_dla = self.benchmark_data.benchmark_csv(read_index) - print("------------Executing {}------------\n".format(self.model_name)) - framework = self.benchmark_data.framework2ext() - # Save, Load and Delete Engine - model_ext = str(self.model_name) + "." + str(framework) - self.trt_engine = load_store_engine(model_path=self.model_path, model_name=model_ext, num_devices=num_devices, - batch_size_gpu=self.batch_size_gpu, batch_size_dla=self.batch_size_dla, - precision=self.precision, ws_gpu=ws_gpu, ws_dla=ws_dla, - model_input=model_input, model_output=model_output) - - self.download_error_flag = self.trt_engine.check_downloaded_models(model_name=model_ext, framework=framework) - - if not self.download_error_flag: - commands, self.models = self.trt_engine.engine_gen() - # Saving Engine - self.trt_engine.save_all(commands=commands, models=self.models) - # Loading Engine Concurrently - self.trt_engine.load_all(commands=commands, models=self.models) - wall_start_t1 = time.time() - self.wall_time = wall_start_t1 - wall_start_t0 - return self.download_error_flag - - def report(self): - latency_fps, error_log = self.benchmark_data.calculate_fps(models=self.models, batch_size_gpu=self.batch_size_gpu, batch_size_dla=self.batch_size_dla) - print("--------------------------\n") - print("Model Name: {} \nFPS:{:.2f} \n".format(self.model_name, latency_fps[3])) - print("--------------------------\n") - latency_fps[len(latency_fps) - 1] = self.model_name - return latency_fps, error_log - - def remove(self): - self.trt_engine.remove_all(models=self.models) - print("Wall Time for running model (secs): {}\n".format(self.wall_time)) - diff --git a/inference/benchmark/jetson/utils/run_xavier_maxn.sh b/inference/benchmark/jetson/utils/run_xavier_maxn.sh deleted file mode 100644 index b7228495ae..0000000000 --- a/inference/benchmark/jetson/utils/run_xavier_maxn.sh +++ /dev/null @@ -1,10 +0,0 @@ -#!/bin/sh -if [ "$#" -ne 1 ]; then - echo "Usgae ./run_xavier_maxn.sh " - echo " Example: ./run_xavier_maxn.sh "nvidia" " - exit -else - password=$1 - -echo $password | sudo -S nvpmodel -m0 -echo $password | sudo -S jetson_clocks diff --git a/inference/benchmark/jetson/utils/utilities.py b/inference/benchmark/jetson/utils/utilities.py deleted file mode 100644 index b7394c9faa..0000000000 --- a/inference/benchmark/jetson/utils/utilities.py +++ /dev/null @@ -1,107 +0,0 @@ -import os -import subprocess -import sys -import time -FNULL = open(os.devnull, "w") -# Class for Utilities (TRT check, Power mode switching) -# https://docs.nvidia.com/jetson/l4t/index.html#page/Tegra%2520Linux%2520Driver%2520Package%2520Development%2520Guide%2Fpower_management_jetson_xavier.html%23wwpID0E0KD0HA -class utilities(): - def __init__(self, jetson_devkit, gpu_freq, dla_freq): - self.jetson_devkit = jetson_devkit - self.gpu_freq = gpu_freq - self.dla_freq = dla_freq - def set_power_mode(self, power_mode, jetson_devkit): - power_cmd0 = "nvpmodel" - power_cmd1 = str("-m"+str(power_mode)) - subprocess.call("sudo {} {}".format(power_cmd0, power_cmd1), shell=True, - stdout=FNULL) - print("Setting Jetson {} in max performance mode".format(jetson_devkit)) - - def set_jetson_clocks(self): - clocks_cmd = "jetson_clocks" - subprocess.call("sudo {}".format(clocks_cmd), shell=True, - stdout=FNULL) - print("Jetson clocks are Set") - - def set_jetson_fan(self, switch_opt): - fan_cmd = "sh" + " " + "-c" + " " + ""echo" + " " + str( - switch_opt) + " " + ">" + " " + "/sys/devices/pwm-fan/target_pwm"" - subprocess.call("sudo {}".format(fan_cmd), shell=True, stdout=FNULL) - - def run_set_clocks_withDVFS(self): - if self.jetson_devkit == "tx2": - self.set_user_clock(device="gpu") - self.set_clocks_withDVFS(frequency=self.gpu_freq, device="gpu") - if self.jetson_devkit == "nano": - self.set_user_clock(device="gpu") - self.set_clocks_withDVFS(frequency=self.gpu_freq, device="gpu") - if self.jetson_devkit == "xavier" or self.jetson_devkit == "xavier-nx": - self.set_user_clock(device="gpu") - self.set_clocks_withDVFS(frequency=self.gpu_freq, device="gpu") - self.set_user_clock(device="dla") - self.set_clocks_withDVFS(frequency=self.dla_freq, device="dla") - - def set_user_clock(self, device): - if self.jetson_devkit == "tx2": - self.enable_register = "/sys/devices/gpu.0/aelpg_enable" - self.freq_register = "/sys/devices/gpu.0/devfreq/17000000.gp10b" - if self.jetson_devkit == "nano": - self.enable_register = "/sys/devices/gpu.0/aelpg_enable" - self.freq_register = "/sys/devices/gpu.0/devfreq/57000000.gpu" - if self.jetson_devkit == "xavier" or self.jetson_devkit == "xavier-nx": - if device == "gpu": - self.enable_register = "/sys/devices/gpu.0/aelpg_enable" - self.freq_register = "/sys/devices/gpu.0/devfreq/17000000.gv11b" - elif device == "dla": - base_register_dir = "/sys/kernel/debug/bpmp/debug/clk" - self.enable_register = base_register_dir + "/nafll_dla/mrq_rate_locked" - self.freq_register = base_register_dir + "/nafll_dla/rate" - - def set_clocks_withDVFS(self, frequency, device): - from_freq = self.read_internal_register(register=self.freq_register, device=device) - self.set_frequency(device=device, enable_register=self.enable_register, freq_register=self.freq_register, frequency=frequency, from_freq=from_freq) - time.sleep(1) - to_freq = self.read_internal_register(register=self.freq_register, device=device) - print("{} frequency is set from {} Hz --> to {} Hz".format(device, from_freq, to_freq)) - - def set_frequency(self, device, enable_register, freq_register, frequency, from_freq): - self.write_internal_register(enable_register, 1) - if device == "gpu": - max_freq_reg = freq_register+"/max_freq" - min_freq_reg = freq_register+"/min_freq" - if int(frequency) > int(from_freq): - self.write_internal_register(max_freq_reg, frequency) - self.write_internal_register(min_freq_reg, frequency) - else: - self.write_internal_register(min_freq_reg, frequency) - self.write_internal_register(max_freq_reg, frequency) - elif device =="dla": - self.write_internal_register(freq_register, frequency) - - def read_internal_register(self, register, device): - if device == "gpu": - register = register+"/cur_freq" - reg_read = open(register, "r") - reg_value = reg_read.read().rstrip("\n") - reg_read.close() - return reg_value - - def write_internal_register(self, register, value): - reg_write = open(register, "w") - reg_write.write("%s" % value) - reg_write.close() - - def clear_ram_space(self): - cmd_0 = str("sh" + " " + "-c") - cmd_1 = str(""echo") + " " + "2" + " " + " >" + " " + "/proc/sys/vm/drop_caches"" - cmd = cmd_0 + " " + cmd_1 - subprocess.call("sudo {}".format(cmd), shell=True) - - def close_all_apps(self): - input("Please close all other applications and Press Enter to continue...") - - def check_trt(self): - if not os.path.isfile("/usr/src/tensorrt/bin/trtexec"): # Check if TensorRT is installed - print("Exiting. Check if TensorRT is installed \n Use ``dpkg -l | grep nvinfer`` ") - return True - return False diff --git a/inference/benchmark/python/paddle/fast_rcnn.py b/inference/benchmark/python/paddle/fast_rcnn.py index f3ca268888..eaa1b76d8b 100644 --- a/inference/benchmark/python/paddle/fast_rcnn.py +++ b/inference/benchmark/python/paddle/fast_rcnn.py @@ -14,7 +14,6 @@ from paddle.inference import Config from paddle.inference import create_predictor -from paddle.inference import PrecisionType FORMAT = "%(asctime)s - %(name)s - %(levelname)s - %(message)s" @@ -43,20 +42,8 @@ def init_predictor(args): config = Config("./fast_rcnn/model.pdmodel", "./fast_rcnn/model.pdiparams") config.enable_memory_optim() - trt_precision_map = {"fp32": PrecisionType.Float32, "fp16": PrecisionType.Half, "int8": PrecisionType.Int8} if args.device == "gpu": config.enable_use_gpu(1000, 0) - if args.use_trt: - config.collect_shape_range_info("shape_range.pbtxt") - config.enable_tuned_tensorrt_dynamic_shape("shape_range.pbtxt", True) - config.enable_tensorrt_engine( - 1 << 30, # workspace_size - 10, # max_batch_size - 30, # min_subgraph_size - trt_precision_map[args.trt_precision], # precision - True, # use_static - False, # use_calib_mode - ) elif args.device == "cpu" and args.use_mkldnn: config.enable_mkldnn() @@ -97,11 +84,7 @@ def parse_args(): parser.add_argument("--warmup_times", type=int, default=10, help="warmup_times.") parser.add_argument("--repeats", type=int, default=1000, help="repeats.") parser.add_argument("--device", type=str, default="gpu", help="[gpu,cpu,xpu]") - parser.add_argument("--use_trt", type=bool, default=False, help="Whether use trt.") - parser.add_argument("--trt_precision", type=str, default="fp32", help="Whether use gpu.") - parser.add_argument( - "--use_mkldnn", type=int, default=False, help="trt precision, choice = ['fp32', 'fp16', 'int8']" - ) + parser.add_argument("--use_mkldnn", type=int, default=False, help="use mkldnn") return parser.parse_args() @@ -117,9 +100,6 @@ def summary_config(args, infer_time: float): logger.info("Batch size: {0}, Num of samples: {1}".format(args.batch_size, args.repeats)) logger.info("----------------------- Conf info -----------------------") logger.info("device: {0}".format(args.device)) - if args.use_trt: - logger.info("enable_tensorrt: {0}".format(args.use_trt)) - logger.info("trt_precision: {0}".format(args.trt_precision)) logger.info("----------------------- Perf info -----------------------") logger.info( "Average latency(ms): {0}, QPS: {1}".format( diff --git a/inference/benchmark/python/paddle/mobilenetv2.py b/inference/benchmark/python/paddle/mobilenetv2.py index 158979937c..6a7f5db001 100644 --- a/inference/benchmark/python/paddle/mobilenetv2.py +++ b/inference/benchmark/python/paddle/mobilenetv2.py @@ -14,7 +14,6 @@ from paddle.inference import Config from paddle.inference import create_predictor -from paddle.inference import PrecisionType FORMAT = "%(asctime)s - %(name)s - %(levelname)s - %(message)s" @@ -40,25 +39,11 @@ def init_predictor(args): args : input args """ - use_calib_mode = False - if args.trt_precision == "int8": - use_calib_mode = True - config = Config("./MobileNetV2/inference.pdmodel", "./MobileNetV2/inference.pdiparams") config.enable_memory_optim() - trt_precision_map = {"fp32": PrecisionType.Float32, "fp16": PrecisionType.Half, "int8": PrecisionType.Int8} if args.device == "gpu": config.enable_use_gpu(1000, 0) - if args.use_trt: - config.enable_tensorrt_engine( - 1 << 30, # workspace_size - 10, # max_batch_size - 3, # min_subgraph_size - trt_precision_map[args.trt_precision], # precision - False, # use_static - use_calib_mode, # use_calib_mode - ) elif args.device == "cpu" and args.use_mkldnn: config.enable_mkldnn() @@ -99,11 +84,7 @@ def parse_args(): parser.add_argument("--warmup_times", type=int, default=10, help="warmup_times.") parser.add_argument("--repeats", type=int, default=1000, help="repeats.") parser.add_argument("--device", type=str, default="gpu", help="[gpu,cpu,xpu]") - parser.add_argument("--use_trt", type=bool, default=False, help="Whether use trt.") - parser.add_argument("--trt_precision", type=str, default="fp32", help="Whether use gpu.") - parser.add_argument( - "--use_mkldnn", type=int, default=False, help="trt precision, choice = ['fp32', 'fp16', 'int8']" - ) + parser.add_argument("--use_mkldnn", type=int, default=False, help="use mkldnn") return parser.parse_args() @@ -119,9 +100,6 @@ def summary_config(args, infer_time: float): logger.info("Batch size: {0}, Num of samples: {1}".format(args.batch_size, args.repeats)) logger.info("----------------------- Conf info -----------------------") logger.info("device: {0}".format(args.device)) - if args.use_trt: - logger.info("enable_tensorrt: {0}".format(args.use_trt)) - logger.info("trt_precision: {0}".format(args.trt_precision)) logger.info("----------------------- Perf info -----------------------") logger.info( "Average latency(ms): {0}, QPS: {1}".format( diff --git a/inference/benchmark/python/paddle/resnet101.py b/inference/benchmark/python/paddle/resnet101.py index cd5db4bf28..fc514ee6a4 100644 --- a/inference/benchmark/python/paddle/resnet101.py +++ b/inference/benchmark/python/paddle/resnet101.py @@ -14,7 +14,6 @@ from paddle.inference import Config from paddle.inference import create_predictor -from paddle.inference import PrecisionType FORMAT = "%(asctime)s - %(name)s - %(levelname)s - %(message)s" @@ -40,25 +39,11 @@ def init_predictor(args): args : input args """ - use_calib_mode = False - if args.trt_precision == "int8": - use_calib_mode = True - config = Config("./ResNet101/inference.pdmodel", "./ResNet101/inference.pdiparams") config.enable_memory_optim() - trt_precision_map = {"fp32": PrecisionType.Float32, "fp16": PrecisionType.Half, "int8": PrecisionType.Int8} if args.device == "gpu": config.enable_use_gpu(1000, 0) - if args.use_trt: - config.enable_tensorrt_engine( - 1 << 30, # workspace_size - 10, # max_batch_size - 3, # min_subgraph_size - trt_precision_map[args.trt_precision], # precision - False, # use_static - use_calib_mode, # use_calib_mode - ) elif args.device == "cpu" and args.use_mkldnn: config.enable_mkldnn() @@ -99,10 +84,6 @@ def parse_args(): parser.add_argument("--warmup_times", type=int, default=10, help="warmup_times.") parser.add_argument("--repeats", type=int, default=1000, help="repeats.") parser.add_argument("--device", type=str, default="gpu", help="[gpu,cpu,xpu]") - parser.add_argument("--use_trt", type=bool, default=False, help="Whether use trt.") - parser.add_argument( - "--trt_precision", type=str, default="fp32", help="trt precision, choice = ['fp32', 'fp16', 'int8']" - ) parser.add_argument("--use_mkldnn", type=int, default=False, help="use mkldnn") return parser.parse_args() @@ -119,9 +100,6 @@ def summary_config(args, infer_time: float): logger.info("Batch size: {0}, Num of samples: {1}".format(args.batch_size, args.repeats)) logger.info("----------------------- Conf info -----------------------") logger.info("device: {0}".format(args.device)) - if args.use_trt: - logger.info("enable_tensorrt: {0}".format(args.use_trt)) - logger.info("trt_precision: {0}".format(args.trt_precision)) logger.info("----------------------- Perf info -----------------------") logger.info( "Average latency(ms): {0}, QPS: {1}".format( diff --git a/inference/benchmark/python/paddle/squeezenet.py b/inference/benchmark/python/paddle/squeezenet.py index 64fe73b6b6..619d0f7292 100644 --- a/inference/benchmark/python/paddle/squeezenet.py +++ b/inference/benchmark/python/paddle/squeezenet.py @@ -14,7 +14,6 @@ from paddle.inference import Config from paddle.inference import create_predictor -from paddle.inference import PrecisionType FORMAT = "%(asctime)s - %(name)s - %(levelname)s - %(message)s" @@ -40,25 +39,11 @@ def init_predictor(args): args : input args """ - use_calib_mode = False - if args.trt_precision == "int8": - use_calib_mode = True - config = Config("./squeezenet/inference.pdmodel", "./squeezenet/inference.pdiparams") config.enable_memory_optim() - trt_precision_map = {"fp32": PrecisionType.Float32, "fp16": PrecisionType.Half, "int8": PrecisionType.Int8} if args.device == "gpu": config.enable_use_gpu(1000, 0) - if args.use_trt: - config.enable_tensorrt_engine( - 1 << 30, # workspace_size - 10, # max_batch_size - 3, # min_subgraph_size - trt_precision_map[args.trt_precision], # precision - False, # use_static - use_calib_mode, # use_calib_mode - ) elif args.device == "cpu" and args.use_mkldnn: config.enable_mkldnn() @@ -99,11 +84,7 @@ def parse_args(): parser.add_argument("--warmup_times", type=int, default=10, help="warmup_times.") parser.add_argument("--repeats", type=int, default=1000, help="repeats.") parser.add_argument("--device", type=str, default="gpu", help="[gpu,cpu,xpu]") - parser.add_argument("--use_trt", type=bool, default=False, help="Whether use trt.") - parser.add_argument("--trt_precision", type=str, default="fp32", help="Whether use gpu.") - parser.add_argument( - "--use_mkldnn", type=int, default=False, help="trt precision, choice = ['fp32', 'fp16', 'int8']" - ) + parser.add_argument("--use_mkldnn", type=int, default=False, help="use mkldnn") return parser.parse_args() @@ -119,9 +100,6 @@ def summary_config(args, infer_time: float): logger.info("Batch size: {0}, Num of samples: {1}".format(args.batch_size, args.repeats)) logger.info("----------------------- Conf info -----------------------") logger.info("device: {0}".format(args.device)) - if args.use_trt: - logger.info("enable_tensorrt: {0}".format(args.use_trt)) - logger.info("trt_precision: {0}".format(args.trt_precision)) logger.info("----------------------- Perf info -----------------------") logger.info( "Average latency(ms): {0}, QPS: {1}".format( diff --git a/inference/benchmark/python/paddle/vgg16.py b/inference/benchmark/python/paddle/vgg16.py index 7ee8650767..f806d1d69e 100644 --- a/inference/benchmark/python/paddle/vgg16.py +++ b/inference/benchmark/python/paddle/vgg16.py @@ -14,7 +14,6 @@ from paddle.inference import Config from paddle.inference import create_predictor -from paddle.inference import PrecisionType FORMAT = "%(asctime)s - %(name)s - %(levelname)s - %(message)s" @@ -40,25 +39,11 @@ def init_predictor(args): args : input args """ - use_calib_mode = False - if args.trt_precision == "int8": - use_calib_mode = True - config = Config("./VGG16/inference.pdmodel", "./VGG16/inference.pdiparams") config.enable_memory_optim() - trt_precision_map = {"fp32": PrecisionType.Float32, "fp16": PrecisionType.Half, "int8": PrecisionType.Int8} if args.device == "gpu": config.enable_use_gpu(1000, 0) - if args.use_trt: - config.enable_tensorrt_engine( - 1 << 30, # workspace_size - 10, # max_batch_size - 3, # min_subgraph_size - trt_precision_map[args.trt_precision], # precision - False, # use_static - use_calib_mode, # use_calib_mode - ) elif args.device == "cpu" and args.use_mkldnn: config.enable_mkldnn() @@ -99,11 +84,7 @@ def parse_args(): parser.add_argument("--warmup_times", type=int, default=10, help="warmup_times.") parser.add_argument("--repeats", type=int, default=1000, help="repeats.") parser.add_argument("--device", type=str, default="gpu", help="[gpu,cpu,xpu]") - parser.add_argument("--use_trt", type=bool, default=False, help="Whether use trt.") - parser.add_argument("--trt_precision", type=str, default="fp32", help="Whether use gpu.") - parser.add_argument( - "--use_mkldnn", type=int, default=False, help="trt precision, choice = ['fp32', 'fp16', 'int8']" - ) + parser.add_argument("--use_mkldnn", type=int, default=False, help="use mkldnn") return parser.parse_args() @@ -119,9 +100,6 @@ def summary_config(args, infer_time: float): logger.info("Batch size: {0}, Num of samples: {1}".format(args.batch_size, args.repeats)) logger.info("----------------------- Conf info -----------------------") logger.info("device: {0}".format(args.device)) - if args.use_trt: - logger.info("enable_tensorrt: {0}".format(args.use_trt)) - logger.info("trt_precision: {0}".format(args.trt_precision)) logger.info("----------------------- Perf info -----------------------") logger.info( "Average latency(ms): {0}, QPS: {1}".format( diff --git a/inference/benchmark/python/parse_log.py b/inference/benchmark/python/parse_log.py index 3b1e02c884..89f14857ad 100644 --- a/inference/benchmark/python/parse_log.py +++ b/inference/benchmark/python/parse_log.py @@ -65,8 +65,6 @@ def process_log(file_name: str, iden: str) -> list: output_dict["QPS"] = line_lists[-1].strip() if "cpu_math_library_num_threads:" in line_lists: output_dict["cpu_math_library_num_threads"] = line_lists[-1].strip() - if "trt_precision:" in line_lists: - output_dict["trt_precision"] = line_lists[-1].strip() except Exception: output_dict = {} output_list.append(output_dict) @@ -107,7 +105,7 @@ def main(): args = parse_args() # create empty DataFrame origin_df = pd.DataFrame( - columns=["frame_work", "model_name", "batch_size", "device", "trt_precision", "Average_latency(ms)", "QPS"] + columns=["frame_work", "model_name", "batch_size", "device", "Average_latency(ms)", "QPS"] ) iden = "----------------------- Model info ----------------------" @@ -117,7 +115,7 @@ def main(): if dict_log != {}: origin_df = origin_df.append(dict_log, ignore_index=True) - raw_df = origin_df.sort_values(by=["frame_work", "model_name", "batch_size", "device", "trt_precision"]) + raw_df = origin_df.sort_values(by=["frame_work", "model_name", "batch_size", "device"]) raw_df.to_excel(args.output_name) set_style(args.output_name) diff --git a/inference/benchmark/python/tensorflow/clas_keras_benchmark.py b/inference/benchmark/python/tensorflow/clas_keras_benchmark.py index 4343bb5373..771bca0e08 100644 --- a/inference/benchmark/python/tensorflow/clas_keras_benchmark.py +++ b/inference/benchmark/python/tensorflow/clas_keras_benchmark.py @@ -12,7 +12,6 @@ import numpy as np import tensorflow as tf # tf version should greater than 2.3.0 -from tensorflow.python.compiler.tensorrt import trt_convert as trt from tensorflow.python.saved_model import tag_constants FORMAT = "%(asctime)s - %(name)s - %(levelname)s - %(message)s" @@ -27,9 +26,6 @@ def parse_args(): """ parser = argparse.ArgumentParser() parser.add_argument("--model_name", type=str, help="model name") - parser.add_argument( - "--trt_precision", type=str, default="fp32", help="trt precision, choice = ['fp32', 'fp16', 'int8']" - ) parser.add_argument( "--image_shape", type=str, @@ -38,7 +34,6 @@ def parse_args(): ) parser.add_argument("--use_gpu", dest="use_gpu", action="store_true") - parser.add_argument("--use_trt", dest="use_trt", action="store_true") parser.add_argument("--use_xla", dest="use_xla", action="store_true") parser.add_argument("--batch_size", type=int, default=1, help="batch size") @@ -64,62 +59,7 @@ def prepare_model(args): else: sys.exit(0) - if args.use_trt and args.trt_precision == "fp32": - # convert model to trt fp32 - logger.info("Converting to TF-TRT FP32...") - conversion_params = trt.DEFAULT_TRT_CONVERSION_PARAMS._replace( - precision_mode=trt.TrtPrecisionMode.FP32, max_workspace_size_bytes=8000000000 - ) - - converter = trt.TrtGraphConverterV2( - input_saved_model_dir="{}_saved_model".format(args.model_name), conversion_params=conversion_params - ) - converter.convert() - converter.save(output_saved_model_dir="{}_saved_model_TFTRT_FP32".format(args.model_name)) - logger.info("Done Converting to TF-TRT FP32") - elif args.use_trt and args.trt_precision == "fp16": - logger.info("Converting to TF-TRT FP16...") - conversion_params = trt.DEFAULT_TRT_CONVERSION_PARAMS._replace( - precision_mode=trt.TrtPrecisionMode.FP16, max_workspace_size_bytes=8000000000 - ) - converter = trt.TrtGraphConverterV2( - input_saved_model_dir="{}_saved_model".format(args.model_name), conversion_params=conversion_params - ) - converter.convert() - converter.save(output_saved_model_dir="{}_saved_model_TFTRT_FP16".format(args.model_name)) - logger.info("Done Converting to TF-TRT FP16") - elif args.use_trt and args.trt_precision == "int8": - # convert model to trt int8 - logger.info("Converting to TF-TRT INT8...") - conversion_params = trt.DEFAULT_TRT_CONVERSION_PARAMS._replace( - precision_mode=trt.TrtPrecisionMode.INT8, max_workspace_size_bytes=8000000000, use_calibration=True - ) - converter = trt.TrtGraphConverterV2( - input_saved_model_dir="{}_saved_model".format(args.model_name), conversion_params=conversion_params - ) - - channels = int(args.image_shape.split(",")[0]) - height = int(args.image_shape.split(",")[1]) - width = int(args.image_shape.split(",")[2]) - logger.info("channels: {0}, height: {1}, width: {2}".format(channels, height, width)) - input_shape = (args.batch_size, height, width, channels) - - def calibration_input_fn(input_shape): - batched_input = tf.constant(np.ones(input_shape).astype("float")) - batched_input = tf.cast(batched_input, dtype="float") - yield (batched_input,) - - converter.convert(calibration_input_fn=calibration_input_fn(input_shape)) - converter.save(output_saved_model_dir="{}_saved_model_TFTRT_INT8".format(args.model_name)) - logger.info("Done Converting to TF-TRT INT8") - else: - logger.warn("No TensorRT precision was input, will not convert TensorRT graph to saved model") - - -def benchmark_tftrt(args, input_saved_model): - """ - trt inference - """ +def benchmark(args, input_saved_model): saved_model_loaded = tf.saved_model.load(input_saved_model, tags=[tag_constants.SERVING]) infer = saved_model_loaded.signatures["serving_default"] @@ -162,10 +102,6 @@ def summary_config(args, infer_time: float): logger.info("Batch size: {0}, Num of samples: {1}".format(args.batch_size, args.repeats)) logger.info("----------------------- Conf info -----------------------") logger.info("device: {0}".format("gpu" if args.use_gpu else "cpu")) - if args.use_gpu: - logger.info("enable_tensorrt: {0}".format(args.use_trt)) - if args.use_trt: - logger.info("trt_precision: {0}".format(args.trt_precision)) logger.info("enable_xla: {0}".format(args.use_xla)) logger.info("----------------------- Perf info -----------------------") logger.info( @@ -181,15 +117,7 @@ def run_demo(): """ args = parse_args() prepare_model(args) - if args.use_trt: - if args.trt_precision == "fp32": - total_time = benchmark_tftrt(args, "{}_saved_model_TFTRT_FP32".format(args.model_name)) - elif args.trt_precision == "fp16": - total_time = benchmark_tftrt(args, "{}_saved_model_TFTRT_FP16".format(args.model_name)) - elif args.trt_precision == "int8": - total_time = benchmark_tftrt(args, "{}_saved_model_TFTRT_INT8".format(args.model_name)) - else: - total_time = benchmark_tftrt(args, "{}_saved_model".format(args.model_name)) + total_time = benchmark(args, "{}_saved_model".format(args.model_name)) summary_config(args, total_time) diff --git a/inference/benchmark/python/torch/README.md b/inference/benchmark/python/torch/README.md index 03f5b7d514..e9786aa875 100644 --- a/inference/benchmark/python/torch/README.md +++ b/inference/benchmark/python/torch/README.md @@ -8,15 +8,10 @@ docker pull nvcr.io/nvidia/pytorch:22.01-py3 ## 相关依赖 ```shell python -m pip install opencv-python -git clone https://github.com/NVIDIA-AI-IOT/torch2trt -cd torch2trt -sudo python setup.py install --plugins ``` ## 执行方式 ```shell python clas_benchmark.py --model_name resnet101 --device cpu --batch_size 1 python clas_benchmark.py --model_name resnet101 --device gpu --batch_size 1 -python clas_benchmark.py --model_name resnet101 --device gpu --batch_size 1 --use_trt --trt_precision fp32 -python clas_benchmark.py --model_name resnet101 --device gpu --batch_size 1 --use_trt --trt_precision fp16 ``` diff --git a/inference/benchmark/python/torch/clas_benchmark.py b/inference/benchmark/python/torch/clas_benchmark.py index 0e841feb30..21529f6309 100644 --- a/inference/benchmark/python/torch/clas_benchmark.py +++ b/inference/benchmark/python/torch/clas_benchmark.py @@ -11,7 +11,6 @@ import cv2 import numpy as np import torch -from torch2trt import torch2trt import torchvision.models as models FORMAT = "%(asctime)s - %(name)s - %(levelname)s - %(message)s" @@ -77,9 +76,7 @@ def parse_args(): default="resnet50", choices=["resnet50", "resnet101", "alexnet", "vgg16", "squeezenet1_0", "inception_v3", "mobilenet_v2"], ) - parser.add_argument("--trt_precision", type=str, default="fp32", help="trt precision", choices=["fp32", "fp16"]) parser.add_argument("--device", default="gpu", type=str, choices=["gpu", "cpu"]) - parser.add_argument("--use_trt", dest="use_trt", action="store_true") parser.add_argument("--batch_size", type=int, default=1, help="batch size") parser.add_argument("--warmup_times", type=int, default=5, help="warmup") parser.add_argument("--repeats", type=int, default=1000, help="repeats") @@ -106,13 +103,6 @@ def forward_benchmark(args): # set running device on predictor = Predictor().to(device) # predictor = torch.jit.script(predictor).to(device) - if args.use_trt: - if args.trt_precision == "fp16": - image_tensor = image_tensor.half() - predictor = predictor.half() - predictor = torch2trt(predictor, [image_tensor], fp16_mode=True, max_batch_size=args.batch_size) - else: - predictor = torch2trt(predictor, [image_tensor], max_batch_size=args.batch_size) print(image_tensor.dtype) logger.info("input image tensor shape : {}".format(image_tensor.shape)) @@ -129,38 +119,6 @@ def forward_benchmark(args): return total_inference_cost, output -# def trt_benchmark(args): -# """ -# trt forward inference -# Args: -# args -# Returns: -# infernce trt benchmark time -# """ -# -# # Compile module -# predictor = Predictor() -# device = torch.device("cuda:0") -# image_tensor = torch.randn((1, 3, 224, 224)).to(device) -# # Trace the module with example data -# traced_model = torch.jit.trace(predictor.to(device), [image_tensor]).to(device) -# -# # Compile module -# compiled_trt_model = torch_tensorrt.compile( -# traced_model, -# inputs=[torch_tensorrt.Input(image_tensor.shape)], -# enabled_precisions={torch.float32}, # Run in FP32 -# ) -# for i in range(args.warmup_times): -# results = compiled_trt_model(image_tensor) -# time1 = time.time() -# for i in range(args.repeats): -# results = compiled_trt_model(image_tensor) -# time2 = time.time() -# total_inference_cost = (time2 - time1) * 1000 # total latency, ms -# return total_inference_cost, results - - def summary_config(args, infer_time: float): """ Args: @@ -173,9 +131,6 @@ def summary_config(args, infer_time: float): logger.info("Batch size: {0}, Num of samples: {1}".format(args.batch_size, args.repeats)) logger.info("----------------------- Conf info -----------------------") logger.info("device: {0}".format(args.device)) - if args.use_trt: - logger.info("enable_tensorrt: {0}".format(args.use_trt)) - logger.info("trt_precision: {0}".format(args.trt_precision)) logger.info("----------------------- Perf info -----------------------") logger.info( "Average latency(ms): {0}, QPS: {1}".format( diff --git a/inference/benchmark/python/torch/detection_benchmark.py b/inference/benchmark/python/torch/detection_benchmark.py index 71c552f0eb..63a0f59f7b 100644 --- a/inference/benchmark/python/torch/detection_benchmark.py +++ b/inference/benchmark/python/torch/detection_benchmark.py @@ -12,7 +12,6 @@ import wget import numpy as np import torch -from torch2trt import torch2trt import torchvision.models as models FORMAT = "%(asctime)s - %(name)s - %(levelname)s - %(message)s" @@ -66,9 +65,7 @@ def parse_args(): """ parser = argparse.ArgumentParser(description=__doc__) parser.add_argument("--model_name", type=str, default="yolov3", choices=["yolov3", "faster_rcnn"]) - parser.add_argument("--trt_precision", type=str, default="fp32", help="trt precision", choices=["fp32", "fp16"]) parser.add_argument("--device", default="gpu", type=str, choices=["gpu", "cpu"]) - parser.add_argument("--use_trt", dest="use_trt", action="store_true") parser.add_argument("--batch_size", type=int, default=1, help="batch size") parser.add_argument("--warmup_times", type=int, default=10, help="warmup") parser.add_argument("--repeats", type=int, default=1000, help="repeats") @@ -95,13 +92,6 @@ def forward_benchmark(args): # set running device on predictor = Predictor().to(device) # predictor = torch.jit.script(predictor).to(device) - if args.use_trt: - if args.trt_precision == "fp16": - image_tensor = image_tensor.half() - predictor = predictor.half() - predictor = torch2trt(predictor, [image_tensor], fp16_mode=True, max_batch_size=args.batch_size) - else: - predictor = torch2trt(predictor, [image_tensor], max_batch_size=args.batch_size) print(image_tensor.dtype) logger.info("input image tensor shape : {}".format(image_tensor.shape)) @@ -118,35 +108,6 @@ def forward_benchmark(args): return total_inference_cost, output -# def trt_benchmark(args): -# """ -# trt forward inference -# Args: -# args -# Returns: -# infernce trt benchmark time -# """ -# # Compile module -# predictor = Predictor() -# device = torch.device("cuda:0") -# image_tensor = torch.randn((1, 3, 224, 224)).to(device) -# # Trace the module with example data -# traced_model = torch.jit.trace(predictor.to(device), [image_tensor]).to(device) -# -# # Compile module -# compiled_trt_model = trtorch.compile( -# traced_model, {"input_shapes": [image_tensor.shape], "op_precision": torch.half} # Run in FP16 -# ) -# for i in range(args.warmup_times): -# results = compiled_trt_model(image_tensor.half()) -# time1 = time.time() -# for i in range(args.repeats): -# results = compiled_trt_model(image_tensor.half()) -# time2 = time.time() -# total_inference_cost = (time2 - time1) * 1000 # total latency, ms -# return total_inference_cost, results - - def summary_config(args, infer_time: float): """ Args: @@ -159,9 +120,6 @@ def summary_config(args, infer_time: float): logger.info("Batch size: {0}, Num of samples: {1}".format(args.batch_size, args.repeats)) logger.info("----------------------- Conf info -----------------------") logger.info("device: {0}".format(args.device)) - if args.use_trt: - logger.info("enable_tensorrt: {0}".format(args.use_trt)) - logger.info("trt_precision: {0}".format(args.trt_precision)) logger.info("----------------------- Perf info -----------------------") logger.info( "Average latency(ms): {0}, QPS: {1}".format( diff --git a/inference/benchmark/python/torch/seg_benchmark.py b/inference/benchmark/python/torch/seg_benchmark.py index 7066d07cb4..23dcd37ef2 100644 --- a/inference/benchmark/python/torch/seg_benchmark.py +++ b/inference/benchmark/python/torch/seg_benchmark.py @@ -10,7 +10,6 @@ import cv2 import torch -import trtorch import torchvision.models as models FORMAT = "%(asctime)s - %(name)s - %(levelname)s - %(message)s" @@ -61,11 +60,7 @@ def parse_args(): """ parser = argparse.ArgumentParser(description=__doc__) parser.add_argument("--model_name", type=str, default="deeplabv3_resnet50", choices=["deeplabv3_resnet50", "unet"]) - parser.add_argument( - "--trt_precision", type=str, default="fp32", help="trt precision, choice = ['fp32', 'fp16', 'int8']" - ) parser.add_argument("--device", default="gpu", type=str, choices=["gpu", "cpu"]) - parser.add_argument("--use_trt", dest="use_trt", action="store_true") parser.add_argument("--batch_size", type=int, default=1, help="batch size") parser.add_argument("--warmup_times", type=int, default=5, help="warmup") parser.add_argument("--repeats", type=int, default=1000, help="repeats") @@ -104,35 +99,6 @@ def forward_benchmark(args): return total_inference_cost, output -def trt_benchmark(args): - """ - trt forward inference - Args: - args - Returns: - infernce trt benchmark time - """ - # Compile module - predictor = Predictor() - device = torch.device("cuda:0") - image_tensor = torch.randn((1, 3, 224, 224)).to(device) - # Trace the module with example data - traced_model = torch.jit.trace(predictor.to(device), [image_tensor]).to(device) - - # Compile module - compiled_trt_model = trtorch.compile( - traced_model, {"input_shapes": [image_tensor.shape], "op_precision": torch.half} # Run in FP16 - ) - for i in range(args.warmup_times): - results = compiled_trt_model(image_tensor.half()) - time1 = time.time() - for i in range(args.repeats): - results = compiled_trt_model(image_tensor.half()) - time2 = time.time() - total_inference_cost = (time2 - time1) * 1000 # total latency, ms - return total_inference_cost, results - - def summary_config(args, infer_time: float): """ Args: @@ -145,9 +111,6 @@ def summary_config(args, infer_time: float): logger.info("Batch size: {0}, Num of samples: {1}".format(args.batch_size, args.repeats)) logger.info("----------------------- Conf info -----------------------") logger.info("device: {0}".format(args.device)) - if args.use_trt: - logger.info("enable_tensorrt: {0}".format(args.use_trt)) - logger.info("trt_precision: {0}".format(args.trt_precision)) logger.info("----------------------- Perf info -----------------------") logger.info( "Average latency(ms): {0}, QPS: {1}".format( @@ -161,10 +124,7 @@ def run_demo(): run_demo """ args = parse_args() - if args.use_trt: - total_time = trt_benchmark(args)[0] - else: - total_time = forward_benchmark(args)[0] + total_time = forward_benchmark(args)[0] summary_config(args, total_time) diff --git a/inference/python_api_test/test_int8_model/get_benchmark_info.py b/inference/python_api_test/test_int8_model/get_benchmark_info.py index dec62d0edf..5251e58e50 100644 --- a/inference/python_api_test/test_int8_model/get_benchmark_info.py +++ b/inference/python_api_test/test_int8_model/get_benchmark_info.py @@ -338,7 +338,6 @@ def res2db(env, benchmark_res, mode_list, metric_list): "model_name": model, "batch_size": info["batch_size"], "fp_mode": "int8", - "use_trt": False, "use_mkldnn": True, "jingdu": info["jingdu"]["value"], "jingdu_unit": info["jingdu"]["unit"], @@ -354,7 +353,6 @@ def res2db(env, benchmark_res, mode_list, metric_list): "python_version": env["python_version"], "cuda_version": env["cuda_version"], "cudnn_version": env["cudnn_version"], - "trt_version": env["trt_version"], "device": env["device"], "thread_num": 1, } @@ -376,11 +374,10 @@ def run(): python_version = sys.argv[6] cuda_version = sys.argv[7] cudnn_version = sys.argv[8] - trt_version = sys.argv[9] - device = sys.argv[10] - modes = sys.argv[11] - metrics = sys.argv[12] - save_file = sys.argv[13] + device = sys.argv[9] + modes = sys.argv[10] + metrics = sys.argv[11] + save_file = sys.argv[12] mode_list = modes.split(",") metric_list = metrics.split(",") @@ -395,7 +392,6 @@ def run(): "python_version": python_version, "cuda_version": cuda_version, "cudnn_version": cudnn_version, - "trt_version": trt_version, "device": device, "threshold": "时延/内存/显存 0.05，精度 0.01", } diff --git a/inference/python_api_test/test_int8_model/write_db.py b/inference/python_api_test/test_int8_model/write_db.py index 8a39890c6f..085f9f8c1e 100644 --- a/inference/python_api_test/test_int8_model/write_db.py +++ b/inference/python_api_test/test_int8_model/write_db.py @@ -48,13 +48,13 @@ def write(res): # cases sql_str = "insert into SlimResult \ (task_dt, \ - model_name, batch_size, fp_mode, use_trt, use_mkldnn, \ + model_name, batch_size, fp_mode, use_mkldnn, \ ips, ips_unit, cpu_mem, gpu_mem, \ frame, frame_branch, frame_commit, frame_version, \ - docker_image, python_version, cuda_version, cudnn_version, trt_version, \ + docker_image, python_version, cuda_version, cudnn_version, \ device_type, thread_num, jingdu, jingdu_unit) \ values (%s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, \ - %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s)" + %s, %s, %s, %s, %s, %s, %s, %s, %s, %s)" val = [] for item in res: val.append( @@ -63,7 +63,6 @@ def write(res): item["model_name"], item["batch_size"], item["fp_mode"], - item["use_trt"], item["use_mkldnn"], item["ips"], item["ips_unit"], @@ -77,7 +76,6 @@ def write(res): item["python_version"], item["cuda_version"], item["cudnn_version"], - item["trt_version"], item["device"], item["thread_num"], item["jingdu"], diff --git a/inference/python_api_test/test_int8_model/xly.sh b/inference/python_api_test/test_int8_model/xly.sh index c89dcd4865..45e69804ac 100644 --- a/inference/python_api_test/test_int8_model/xly.sh +++ b/inference/python_api_test/test_int8_model/xly.sh @@ -79,13 +79,12 @@ SAVE_FILE=${DT}_${FRAME}_${FRAME_BRANCH/\//-}_${PADDLE_COMMIT}_${DEVICE}.xlsx PYTHON_VERSION=${PYTHON_VERSION:-3.8} CUDA_VERSION=${CUDA_VERSION:-11.2} CUDNN_VERSION=${CUDNN_VERSION:-8.2} -TRT_VERSION=${TRT_VERSION:--} GPU=${DEVICE} CPU="-" bash run.sh -python get_benchmark_info.py ${FRAME} ${FRAME_BRANCH} ${PADDLE_COMMIT} ${FRAME_VERSION} ${DOCKER_IMAGE} ${PYTHON_VERSION} ${CUDA_VERSION} ${CUDNN_VERSION} ${TRT_VERSION} ${DEVICE} ${MODE} ${METRIC} ${SAVE_FILE} +python get_benchmark_info.py ${FRAME} ${FRAME_BRANCH} ${PADDLE_COMMIT} ${FRAME_VERSION} ${DOCKER_IMAGE} ${PYTHON_VERSION} ${CUDA_VERSION} ${CUDNN_VERSION} ${DEVICE} ${MODE} ${METRIC} ${SAVE_FILE} UPLOAD_FILE_PATH=`pwd`/${SAVE_FILE} diff --git a/inference/python_api_test/test_new_devices/benchmark.py b/inference/python_api_test/test_new_devices/benchmark.py index cfefd0eda2..98d0e1e7e6 100755 --- a/inference/python_api_test/test_new_devices/benchmark.py +++ b/inference/python_api_test/test_new_devices/benchmark.py @@ -184,7 +184,6 @@ def parse_args(): parser.add_argument("--enable_paddleort", type=str2bool, default=False) parser.add_argument("--enable_gpu", type=str2bool, default=False) parser.add_argument("--enable_pir", type=str2bool, default=False) - parser.add_argument("--enable_trt", type=str2bool, default=False) parser.add_argument("--enable_dynamic_shape", type=str2bool, default=True) parser.add_argument("--enable_tune", type=str2bool, default=False) parser.add_argument("--gen_calib", type=str2bool, default=False) @@ -194,7 +193,6 @@ def parse_args(): parser.add_argument("--return_result", type=str2bool, default=False) parser.add_argument("--enable_debug", type=str2bool, default=False) parser.add_argument("--enable_fd_paddle", type=str2bool, default=False) - parser.add_argument("--enable_fd_trt", type=str2bool, default=False) parser.add_argument("--enable_fd_ort", type=str2bool, default=False) parser.add_argument("--enable_fd_openvino", type=str2bool, default=False) @@ -470,7 +468,6 @@ def report(self, status=True): result["enable_mkldnn"] = self.conf.enable_mkldnn result["enable_gpu"] = self.conf.enable_gpu result["enable_pir"] = self.conf.enable_pir - result["enable_trt"] = self.conf.enable_trt result["input_shape"] = get_shape_str(self.conf.yaml_config["input_shape"], self.conf.test_num) print(result) with open("result.txt", "a+") as f: diff --git a/inference/python_api_test/test_new_devices/demo.sh b/inference/python_api_test/test_new_devices/demo.sh index e3b7a817e7..bb02303148 100755 --- a/inference/python_api_test/test_new_devices/demo.sh +++ b/inference/python_api_test/test_new_devices/demo.sh @@ -15,7 +15,6 @@ config_file=config.yaml gpu_id=12 enable_gpu=false enable_pir=false -enable_trt=false if [ $1 == "mask_rcnn_r50_fpn_1x_coco" ]; then subgraph_size_var=8 @@ -52,7 +51,7 @@ done #if [ $5 == "1" ];then # # auto tune -# python benchmark.py --model_dir=${model_dir} --config_file ${config_file} --precision ${precision} --enable_gpu=${enable_gpu} --enable_pir=${enable_pir} --gpu_id=${gpu_id} --enable_trt=${enable_trt} --backend_type=${backend_type} --batch_size=${batch_size} --paddle_model_file "$model_file" --paddle_params_file "$params_file" --enable_tune=true --return_result=true +# python benchmark.py --model_dir=${model_dir} --config_file ${config_file} --precision ${precision} --enable_gpu=${enable_gpu} --enable_pir=${enable_pir} --gpu_id=${gpu_id} --backend_type=${backend_type} --batch_size=${batch_size} --paddle_model_file "$model_file" --paddle_params_file "$params_file" --enable_tune=true --return_result=true #fi # infer -python benchmark.py --model_dir=${model_dir} --config_file ${config_file} --precision ${precision} --enable_gpu=${enable_gpu} --enable_pir=${enable_pir} --gpu_id=${gpu_id} --enable_trt=${enable_trt} --backend_type=${backend_type} --batch_size=${batch_size} --subgraph_size=${subgraph_size_var} --paddle_model_file "$model_file" --paddle_params_file "$params_file" --return_result=true +python benchmark.py --model_dir=${model_dir} --config_file ${config_file} --precision ${precision} --enable_gpu=${enable_gpu} --enable_pir=${enable_pir} --gpu_id=${gpu_id} --backend_type=${backend_type} --batch_size=${batch_size} --subgraph_size=${subgraph_size_var} --paddle_model_file "$model_file" --paddle_params_file "$params_file" --return_result=true diff --git a/inference/python_api_test/test_new_devices/inference_benchmark_new_devices.sh b/inference/python_api_test/test_new_devices/inference_benchmark_new_devices.sh index 5c25f09b21..33c8299228 100755 --- a/inference/python_api_test/test_new_devices/inference_benchmark_new_devices.sh +++ b/inference/python_api_test/test_new_devices/inference_benchmark_new_devices.sh @@ -5,12 +5,10 @@ export FLAGS_conv_workspace_size_limit=32 export FLAGS_initial_cpu_memory_in_mb=0 backend_type_list=(MLU) -enable_trt_list=(false) enable_gpu_list=(false) enable_mkldnn_list=(false) enable_gpu=false enable_pir=false -enable_trt=false precision=fp32 gpu_id=$1 batch_size_list=(1) @@ -58,9 +56,9 @@ run_benchmark(){ for batch_size in ${batch_size_var[@]};do for enable_gpu in ${enable_gpu_list[@]};do if [ ${enable_gpu} = "true" ]; then - python benchmark.py --model_dir=${model_dir} --config_file ${config_file} --precision ${precision} --enable_gpu=${enable_gpu} --enable_pir=${enable_pir} --gpu_id=${gpu_id} --enable_trt=${enable_trt} --backend_type=${backend_type} --batch_size=${batch_size} --subgraph_size=${subgraph_size_var} --paddle_model_file "$model_file" --paddle_params_file "$params_file" --return_result=true + python benchmark.py --model_dir=${model_dir} --config_file ${config_file} --precision ${precision} --enable_gpu=${enable_gpu} --enable_pir=${enable_pir} --gpu_id=${gpu_id} --backend_type=${backend_type} --batch_size=${batch_size} --subgraph_size=${subgraph_size_var} --paddle_model_file "$model_file" --paddle_params_file "$params_file" --return_result=true elif [ ${enable_gpu} = "false" ]; then - python benchmark.py --model_dir=${model_dir} --config_file ${config_file} --precision ${precision} --enable_gpu=${enable_gpu} --enable_pir=${enable_pir} --gpu_id=${gpu_id} --enable_trt=${enable_trt} --backend_type=${backend_type} --batch_size=${batch_size} --subgraph_size=${subgraph_size_var} --paddle_model_file "$model_file" --paddle_params_file "$params_file" --return_result=true + python benchmark.py --model_dir=${model_dir} --config_file ${config_file} --precision ${precision} --enable_gpu=${enable_gpu} --enable_pir=${enable_pir} --gpu_id=${gpu_id} --backend_type=${backend_type} --batch_size=${batch_size} --subgraph_size=${subgraph_size_var} --paddle_model_file "$model_file" --paddle_params_file "$params_file" --return_result=true fi done done diff --git a/inference/report/get_gsb.py b/inference/report/get_gsb.py index 01f06722eb..1f1a6c11eb 100644 --- a/inference/report/get_gsb.py +++ b/inference/report/get_gsb.py @@ -112,9 +112,7 @@ def select_compute(db_res, gsb, main_clas): # item is dict model_name = item["model_name"] mode = "" - if item["use_trt"] == 1: - mode = "trt" - elif item["use_mkldnn"] == 1: + if item["use_mkldnn"] == 1: mode = "mkldnn" else: mode = "native" diff --git a/inference/serving_api_test/paddle_serving_server/test_server.py b/inference/serving_api_test/paddle_serving_server/test_server.py index d41f8439e7..906e80b589 100644 --- a/inference/serving_api_test/paddle_serving_server/test_server.py +++ b/inference/serving_api_test/paddle_serving_server/test_server.py @@ -154,7 +154,6 @@ def test_prepare_engine_with_async_mode(self): assert model_engine_0.enable_batch_align == 1 assert model_engine_0.enable_memory_optimization is False assert model_engine_0.enable_ir_optimization is False - assert model_engine_0.use_trt is False assert model_engine_0.use_lite is False assert model_engine_0.use_xpu is False assert model_engine_0.use_gpu is True diff --git a/inference/serving_api_test/paddle_serving_server/util.py b/inference/serving_api_test/paddle_serving_server/util.py index cc0d4b2c6a..0d4511425b 100644 --- a/inference/serving_api_test/paddle_serving_server/util.py +++ b/inference/serving_api_test/paddle_serving_server/util.py @@ -67,7 +67,6 @@ def default_args(): args.max_body_size = 512 * 1024 * 1024 args.use_encryption_model = False args.use_multilang = False - args.use_trt = False args.use_lite = False args.use_xpu = False args.product_name = None diff --git a/models/Paddle2ONNX/Det2ONNX/infer_for_onnx.py b/models/Paddle2ONNX/Det2ONNX/infer_for_onnx.py index 181af8714d..a9988ddb61 100644 --- a/models/Paddle2ONNX/Det2ONNX/infer_for_onnx.py +++ b/models/Paddle2ONNX/Det2ONNX/infer_for_onnx.py @@ -58,13 +58,7 @@ class Detector(object): pred_config (object): config of model, defined by `Config(model_dir)` model_dir (str): root path of model.pdiparams, model.pdmodel and infer_cfg.yml device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU - run_mode (str): mode of running(paddle/trt_fp32/trt_fp16) batch_size (int): size of pre batch in inference - trt_min_shape (int): min shape for dynamic shape in trt - trt_max_shape (int): max shape for dynamic shape in trt - trt_opt_shape (int): opt shape for dynamic shape in trt - trt_calib_mode (bool): If the model is produced by TRT offline quantitative - calibration, trt_calib_mode need to set True cpu_threads (int): cpu threads enable_mkldnn (bool): whether to open MKLDNN """ @@ -74,12 +68,7 @@ def __init__( pred_config, model_dir, device="CPU", - run_mode="paddle", batch_size=1, - trt_min_shape=1, - trt_max_shape=1280, - trt_opt_shape=640, - trt_calib_mode=False, cpu_threads=1, enable_mkldnn=False, ): @@ -87,17 +76,10 @@ def __init__( default """ self.pred_config = pred_config + self.batch_size = batch_size self.predictor, self.config = load_predictor( model_dir, - run_mode=run_mode, - batch_size=batch_size, - min_subgraph_size=self.pred_config.min_subgraph_size, device=device, - use_dynamic_shape=self.pred_config.use_dynamic_shape, - trt_min_shape=trt_min_shape, - trt_max_shape=trt_max_shape, - trt_opt_shape=trt_opt_shape, - trt_calib_mode=trt_calib_mode, cpu_threads=cpu_threads, enable_mkldnn=enable_mkldnn, ) @@ -211,13 +193,7 @@ class DetectorSOLOv2(Detector): config (object): config of model, defined by `Config(model_dir)` model_dir (str): root path of model.pdiparams, model.pdmodel and infer_cfg.yml device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU - run_mode (str): mode of running(paddle/trt_fp32/trt_fp16) batch_size (int): size of pre batch in inference - trt_min_shape (int): min shape for dynamic shape in trt - trt_max_shape (int): max shape for dynamic shape in trt - trt_opt_shape (int): opt shape for dynamic shape in trt - trt_calib_mode (bool): If the model is produced by TRT offline quantitative - calibration, trt_calib_mode need to set True cpu_threads (int): cpu threads enable_mkldnn (bool): whether to open MKLDNN """ @@ -227,12 +203,7 @@ def __init__( pred_config, model_dir, device="CPU", - run_mode="paddle", batch_size=1, - trt_min_shape=1, - trt_max_shape=1280, - trt_opt_shape=640, - trt_calib_mode=False, cpu_threads=1, enable_mkldnn=False, ): @@ -242,15 +213,7 @@ def __init__( self.pred_config = pred_config self.predictor, self.config = load_predictor( model_dir, - run_mode=run_mode, - batch_size=batch_size, - min_subgraph_size=self.pred_config.min_subgraph_size, device=device, - use_dynamic_shape=self.pred_config.use_dynamic_shape, - trt_min_shape=trt_min_shape, - trt_max_shape=trt_max_shape, - trt_opt_shape=trt_opt_shape, - trt_calib_mode=trt_calib_mode, cpu_threads=cpu_threads, enable_mkldnn=enable_mkldnn, ) @@ -301,13 +264,7 @@ class DetectorPicoDet(Detector): config (object): config of model, defined by `Config(model_dir)` model_dir (str): root path of model.pdiparams, model.pdmodel and infer_cfg.yml device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU - run_mode (str): mode of running(paddle/trt_fp32/trt_fp16) batch_size (int): size of pre batch in inference - trt_min_shape (int): min shape for dynamic shape in trt - trt_max_shape (int): max shape for dynamic shape in trt - trt_opt_shape (int): opt shape for dynamic shape in trt - trt_calib_mode (bool): If the model is produced by TRT offline quantitative - calibration, trt_calib_mode need to set True cpu_threads (int): cpu threads enable_mkldnn (bool): whether to open MKLDNN """ @@ -317,12 +274,7 @@ def __init__( pred_config, model_dir, device="CPU", - run_mode="paddle", batch_size=1, - trt_min_shape=1, - trt_max_shape=1280, - trt_opt_shape=640, - trt_calib_mode=False, cpu_threads=1, enable_mkldnn=False, ): @@ -332,15 +284,7 @@ def __init__( self.pred_config = pred_config self.predictor, self.config = load_predictor( model_dir, - run_mode=run_mode, - batch_size=batch_size, - min_subgraph_size=self.pred_config.min_subgraph_size, device=device, - use_dynamic_shape=self.pred_config.use_dynamic_shape, - trt_min_shape=trt_min_shape, - trt_max_shape=trt_max_shape, - trt_opt_shape=trt_opt_shape, - trt_calib_mode=trt_calib_mode, cpu_threads=cpu_threads, enable_mkldnn=enable_mkldnn, ) @@ -461,10 +405,8 @@ def __init__(self, model_dir): self.check_model(yml_conf) self.arch = yml_conf["arch"] self.preprocess_infos = yml_conf["Preprocess"] - self.min_subgraph_size = yml_conf["min_subgraph_size"] self.labels = yml_conf["label_list"] self.mask = False - self.use_dynamic_shape = yml_conf["use_dynamic_shape"] if "mask" in yml_conf: self.mask = yml_conf["mask"] self.tracker = None @@ -500,15 +442,7 @@ def print_config(self): def load_predictor( model_dir, - run_mode="paddle", - batch_size=1, device="CPU", - min_subgraph_size=3, - use_dynamic_shape=False, - trt_min_shape=1, - trt_max_shape=1280, - trt_opt_shape=640, - trt_calib_mode=False, cpu_threads=1, enable_mkldnn=False, ): @@ -516,22 +450,9 @@ def load_predictor( Args: model_dir (str): root path of __model__ and __params__ device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU - run_mode (str): mode of running(paddle/trt_fp32/trt_fp16/trt_int8) - use_dynamic_shape (bool): use dynamic shape or not - trt_min_shape (int): min shape for dynamic shape in trt - trt_max_shape (int): max shape for dynamic shape in trt - trt_opt_shape (int): opt shape for dynamic shape in trt - trt_calib_mode (bool): If the model is produced by TRT offline quantitative - calibration, trt_calib_mode need to set True Returns: predictor (PaddlePredictor): AnalysisPredictor - Raises: - ValueError: predict by TensorRT need device == 'GPU'. """ - if device != "GPU" and run_mode != "paddle": - raise ValueError( - "Predict by TensorRT mode: {}, expect device=='GPU', but device == {}".format(run_mode, device) - ) config = Config(os.path.join(model_dir, "model.pdmodel"), os.path.join(model_dir, "model.pdiparams")) if device == "GPU": # initial GPU memory(M), device ID @@ -553,28 +474,6 @@ def load_predictor( print("The current environment does not support `mkldnn`, so disable mkldnn.") pass - precision_map = { - "trt_int8": Config.Precision.Int8, - "trt_fp32": Config.Precision.Float32, - "trt_fp16": Config.Precision.Half, - } - if run_mode in precision_map.keys(): - config.enable_tensorrt_engine( - workspace_size=1 << 25, - max_batch_size=batch_size, - min_subgraph_size=min_subgraph_size, - precision_mode=precision_map[run_mode], - use_static=False, - use_calib_mode=trt_calib_mode, - ) - - if use_dynamic_shape: - min_input_shape = {"image": [batch_size, 3, trt_min_shape, trt_min_shape]} - max_input_shape = {"image": [batch_size, 3, trt_max_shape, trt_max_shape]} - opt_input_shape = {"image": [batch_size, 3, trt_opt_shape, trt_opt_shape]} - config.set_trt_dynamic_shape_info(min_input_shape, max_input_shape, opt_input_shape) - print("trt set dynamic shape done!") - # disable print log when predict config.disable_glog_info() # enable shared memory @@ -738,12 +637,7 @@ def main(): pred_config, FLAGS.model_dir, device=FLAGS.device, - run_mode=FLAGS.run_mode, batch_size=FLAGS.batch_size, - trt_min_shape=FLAGS.trt_min_shape, - trt_max_shape=FLAGS.trt_max_shape, - trt_opt_shape=FLAGS.trt_opt_shape, - trt_calib_mode=FLAGS.trt_calib_mode, cpu_threads=FLAGS.cpu_threads, enable_mkldnn=FLAGS.enable_mkldnn, ) @@ -768,8 +662,7 @@ def main(): perf_info = detector.det_times.report(average=True) model_dir = FLAGS.model_dir - mode = FLAGS.run_mode - model_info = {"model_name": model_dir.strip("/").split("/")[-1], "precision": mode.split("_")[-1]} + model_info = {"model_name": model_dir.strip("/").split("/")[-1], "precision": "fp32"} data_info = {"batch_size": FLAGS.batch_size, "shape": "dynamic_shape", "data_num": perf_info["img_num"]} det_log = PaddleInferBenchmark(detector.config, model_info, data_info, perf_info, mems) det_log("Det") diff --git a/models/Paddle2ONNX/Det2ONNX/key_infer_for_onnx.py b/models/Paddle2ONNX/Det2ONNX/key_infer_for_onnx.py index 40df1614df..4e7b1455d8 100644 --- a/models/Paddle2ONNX/Det2ONNX/key_infer_for_onnx.py +++ b/models/Paddle2ONNX/Det2ONNX/key_infer_for_onnx.py @@ -50,13 +50,7 @@ class KeyPointDetector(Detector): Args: model_dir (str): root path of model.pdiparams, model.pdmodel and infer_cfg.yml device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU - run_mode (str): mode of running(paddle/trt_fp32/trt_fp16) batch_size (int): size of pre batch in inference - trt_min_shape (int): min shape for dynamic shape in trt - trt_max_shape (int): max shape for dynamic shape in trt - trt_opt_shape (int): opt shape for dynamic shape in trt - trt_calib_mode (bool): If the model is produced by TRT offline quantitative - calibration, trt_calib_mode need to set True cpu_threads (int): cpu threads enable_mkldnn (bool): whether to open MKLDNN use_dark(bool): whether to use postprocess in DarkPose @@ -66,12 +60,7 @@ def __init__( self, model_dir, device="CPU", - run_mode="paddle", batch_size=1, - trt_min_shape=1, - trt_max_shape=1280, - trt_opt_shape=640, - trt_calib_mode=False, cpu_threads=1, enable_mkldnn=False, output_dir="output", @@ -81,20 +70,17 @@ def __init__( """ default """ + pred_config = self.set_config(model_dir) super(KeyPointDetector, self).__init__( + pred_config=pred_config, model_dir=model_dir, device=device, - run_mode=run_mode, batch_size=batch_size, - trt_min_shape=trt_min_shape, - trt_max_shape=trt_max_shape, - trt_opt_shape=trt_opt_shape, - trt_calib_mode=trt_calib_mode, cpu_threads=cpu_threads, enable_mkldnn=enable_mkldnn, - output_dir=output_dir, - threshold=threshold, ) + self.output_dir = output_dir + self.threshold = threshold self.use_dark = use_dark def set_config(self, model_dir): @@ -316,10 +302,8 @@ def __init__(self, model_dir): self.arch = yml_conf["arch"] self.archcls = KEYPOINT_SUPPORT_MODELS[yml_conf["arch"]] self.preprocess_infos = yml_conf["Preprocess"] - self.min_subgraph_size = yml_conf["min_subgraph_size"] self.labels = yml_conf["label_list"] self.tagmap = False - self.use_dynamic_shape = yml_conf["use_dynamic_shape"] if self.archcls == "keypoint_bottomup": self.tagmap = True self.print_config() @@ -367,12 +351,7 @@ def main(): detector = KeyPointDetector( FLAGS.model_dir, device=FLAGS.device, - run_mode=FLAGS.run_mode, batch_size=FLAGS.batch_size, - trt_min_shape=FLAGS.trt_min_shape, - trt_max_shape=FLAGS.trt_max_shape, - trt_opt_shape=FLAGS.trt_opt_shape, - trt_calib_mode=FLAGS.trt_calib_mode, cpu_threads=FLAGS.cpu_threads, enable_mkldnn=FLAGS.enable_mkldnn, threshold=FLAGS.threshold, @@ -397,8 +376,7 @@ def main(): } perf_info = detector.det_times.report(average=True) model_dir = FLAGS.model_dir - mode = FLAGS.run_mode - model_info = {"model_name": model_dir.strip("/").split("/")[-1], "precision": mode.split("_")[-1]} + model_info = {"model_name": model_dir.strip("/").split("/")[-1], "precision": "fp32"} data_info = {"batch_size": 1, "shape": "dynamic_shape", "data_num": perf_info["img_num"]} det_log = PaddleInferBenchmark(detector.config, model_info, data_info, perf_info, mems) det_log("KeyPoint") diff --git a/models/Paddle2ONNX/Det2ONNX/utils_for_onnx.py b/models/Paddle2ONNX/Det2ONNX/utils_for_onnx.py index 8bbb00815a..0c5d0a65c9 100644 --- a/models/Paddle2ONNX/Det2ONNX/utils_for_onnx.py +++ b/models/Paddle2ONNX/Det2ONNX/utils_for_onnx.py @@ -59,9 +59,6 @@ def argsparser(): parser.add_argument( "--infer_output_np", type=str, default="infer_output_np", help="Directory of output np.array for onnx acc test." ) - parser.add_argument( - "--run_mode", type=str, default="paddle", help="mode of running(paddle/trt_fp32/trt_fp16/trt_int8)" - ) parser.add_argument( "--device", type=str, @@ -77,15 +74,6 @@ def argsparser(): ) parser.add_argument("--enable_mkldnn", type=ast.literal_eval, default=False, help="Whether use mkldnn with CPU.") parser.add_argument("--cpu_threads", type=int, default=1, help="Num of threads with CPU.") - parser.add_argument("--trt_min_shape", type=int, default=1, help="min_shape for TensorRT.") - parser.add_argument("--trt_max_shape", type=int, default=1280, help="max_shape for TensorRT.") - parser.add_argument("--trt_opt_shape", type=int, default=640, help="opt_shape for TensorRT.") - parser.add_argument( - "--trt_calib_mode", - type=bool, - default=False, - help="If the model is produced by TRT offline quantitative " "calibration, trt_calib_mode need to set True.", - ) parser.add_argument("--save_images", action="store_true", help="Save visualization image results.") parser.add_argument("--save_mot_txts", action="store_true", help="Save tracking results (txt).") parser.add_argument( diff --git a/models/Paddle2ONNX/Seg2ONNX/infer_for_onnx.py b/models/Paddle2ONNX/Seg2ONNX/infer_for_onnx.py index b5dbdba37a..4f5adf6a92 100644 --- a/models/Paddle2ONNX/Seg2ONNX/infer_for_onnx.py +++ b/models/Paddle2ONNX/Seg2ONNX/infer_for_onnx.py @@ -16,7 +16,7 @@ import yaml import numpy as np -from paddle.inference import create_predictor, PrecisionType +from paddle.inference import create_predictor from paddle.inference import Config as PredictConfig import paddleseg.transforms as T @@ -25,19 +25,6 @@ from paddleseg.utils.visualize import get_pseudo_color_map -def use_auto_tune(args): - """ - base - """ - return ( - hasattr(PredictConfig, "collect_shape_range_info") - and hasattr(PredictConfig, "enable_tuned_tensorrt_dynamic_shape") - and args.device == "gpu" - and args.use_trt - and args.enable_auto_tune - ) - - class DeployConfig: """ base @@ -87,55 +74,6 @@ def _load_transforms(self, t_list): return T.Compose(transforms) -def auto_tune(args, imgs, img_nums): - """ - Use images to auto tune the dynamic shape for trt sub graph. - The tuned shape saved in args.auto_tuned_shape_file. - Args: - args(dict): input args. - imgs(str, list[str]): the path for images. - img_nums(int): the nums of images used for auto tune. - Returns: - None - """ - logger.info("Auto tune the dynamic shape for GPU TRT.") - - assert use_auto_tune(args) - - if not isinstance(imgs, (list, tuple)): - imgs = [imgs] - num = min(len(imgs), img_nums) - - cfg = DeployConfig(args.cfg) - pred_cfg = PredictConfig(cfg.model, cfg.params) - pred_cfg.enable_use_gpu(100, 0) - if not args.print_detail: - pred_cfg.disable_glog_info() - pred_cfg.collect_shape_range_info(args.auto_tuned_shape_file) - - predictor = create_predictor(pred_cfg) - input_names = predictor.get_input_names() - input_handle = predictor.get_input_handle(input_names[0]) - - for i in range(0, num): - data = np.array([cfg.transforms(imgs[i])[0]]) - input_handle.reshape(data.shape) - input_handle.copy_from_cpu(data) - try: - predictor.run() - except: - logger.info( - "Auto tune fail. Usually, the error is out of GPU memory, " - "because the model and image is too large. \n" - ) - del predictor - if os.path.exists(args.auto_tuned_shape_file): - os.remove(args.auto_tuned_shape_file) - return - - logger.info("Auto tune success.\n") - - class Predictor: """ base @@ -188,32 +126,6 @@ def _init_gpu_config(self): """ logger.info("Use GPU") self.pred_cfg.enable_use_gpu(100, 0) - precision_map = {"fp16": PrecisionType.Half, "fp32": PrecisionType.Float32, "int8": PrecisionType.Int8} - precision_mode = precision_map[self.args.precision] - - if self.args.use_trt: - logger.info("Use TRT") - self.pred_cfg.enable_tensorrt_engine( - workspace_size=1 << 30, - max_batch_size=1, - min_subgraph_size=50, - precision_mode=precision_mode, - use_static=False, - use_calib_mode=False, - ) - - if use_auto_tune(self.args) and os.path.exists(self.args.auto_tuned_shape_file): - logger.info("Use auto tuned dynamic shape") - allow_build_at_runtime = True - self.pred_cfg.enable_tuned_tensorrt_dynamic_shape( - self.args.auto_tuned_shape_file, allow_build_at_runtime - ) - else: - logger.info("Use manual set dynamic shape") - min_input_shape = {"x": [1, 3, 100, 100]} - max_input_shape = {"x": [1, 3, 2000, 3000]} - opt_input_shape = {"x": [1, 3, 512, 1024]} - self.pred_cfg.set_trt_dynamic_shape_info(min_input_shape, max_input_shape, opt_input_shape) def run(self, imgs): """ @@ -323,31 +235,6 @@ def parse_args(): "--device", choices=["cpu", "gpu"], default="gpu", help="Select which device to inference, defaults to gpu." ) - parser.add_argument( - "--use_trt", - default=False, - type=eval, - choices=[True, False], - help="Whether to use Nvidia TensorRT to accelerate prediction.", - ) - parser.add_argument( - "--precision", default="fp32", type=str, choices=["fp32", "fp16", "int8"], help="The tensorrt precision." - ) - parser.add_argument( - "--enable_auto_tune", - default=False, - type=eval, - choices=[True, False], - help="Whether to enable tuned dynamic shape. We uses some images to collect " - "the dynamic shape for trt sub graph, which avoids setting dynamic shape manually.", - ) - parser.add_argument( - "--auto_tuned_shape_file", - type=str, - default="auto_tune_tmp.pbtxt", - help="The temp file to save tuned dynamic shape.", - ) - parser.add_argument("--cpu_threads", default=10, type=int, help="Number of threads to predict when using cpu.") parser.add_argument( "--enable_mkldnn", @@ -382,16 +269,9 @@ def main(args): """ imgs_list, _ = get_image_list(args.image_path) - if use_auto_tune(args): - tune_img_nums = 10 - auto_tune(args, imgs_list, tune_img_nums) - predictor = Predictor(args) predictor.run(imgs_list) - if use_auto_tune(args) and os.path.exists(args.auto_tuned_shape_file): - os.remove(args.auto_tuned_shape_file) - if args.benchmark: predictor.autolog.report() diff --git a/models/PaddleClas/Full_Chain/tipc.sh b/models/PaddleClas/Full_Chain/tipc.sh index e90e3e60b0..540b380bcd 100644 --- a/models/PaddleClas/Full_Chain/tipc.sh +++ b/models/PaddleClas/Full_Chain/tipc.sh @@ -8,8 +8,6 @@ REPO=$1 # 参数作为配置文件传入 # DOCKER_IMAGE=registry.baidubce.com/paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82 # DOCKER_NAME=paddle_whole_chain_test_clas -# # COMPILE_PATH=https://paddle-qa.bj.bcebos.com/paddle-pipeline/Master_GpuAll_LinuxUbuntu_Gcc82_Cuda10.1_Trton_Py37_Compile_H_DISTRIBUTE/latest/paddlepaddle_gpu-0.0.0-cp37-cp37m-linux_x86_64.whl -# COMPILE_PATH=https://paddle-qa.bj.bcebos.com/paddle-pipeline/Master_GpuAll_LinuxUbuntu_Gcc82_Cuda10.1_Trton_Py37_Compile_H_DISTRIBUTE/latest/paddlepaddle_gpu-0.0.0-cp37-cp37m-linux_x86_64.whl # define version compare function function version_lt() { test "$(echo "$@" | tr " " "\n" | sort -rV | head -n 1)" != "$1"; } diff --git a/models/PaddleDetection/module_test/heads/test_FCOSHead.py b/models/PaddleDetection/module_test/heads/test_FCOSHead.py index f6b37b22d4..836f5022e8 100644 --- a/models/PaddleDetection/module_test/heads/test_FCOSHead.py +++ b/models/PaddleDetection/module_test/heads/test_FCOSHead.py @@ -34,7 +34,6 @@ def __init__(self): sqrt_score=False, fcos_loss="FCOSLoss", nms="MultiClassNMS", - trt=False, ) self.net.eval() feat1 = paddle.rand(shape=[4, 256, 32, 32]) diff --git a/models/PaddleDetection/module_test/heads/test_FCOSRHead.py b/models/PaddleDetection/module_test/heads/test_FCOSRHead.py index 28248536fc..1ee668ea83 100644 --- a/models/PaddleDetection/module_test/heads/test_FCOSRHead.py +++ b/models/PaddleDetection/module_test/heads/test_FCOSRHead.py @@ -28,7 +28,6 @@ def __init__(self): stacked_convs=3, act="relu", fpn_strides=[4, 8, 16], - trt=False, loss_weight={"class": 1.0, "probiou": 1.0}, norm_cfg={"name": "gn", "num_groups": 1}, assigner="FCOSRAssigner", diff --git a/models/PaddleDetection/module_test/heads/test_PPYOLOERHead.py b/models/PaddleDetection/module_test/heads/test_PPYOLOERHead.py index 6acd463ad1..735374e24c 100644 --- a/models/PaddleDetection/module_test/heads/test_PPYOLOERHead.py +++ b/models/PaddleDetection/module_test/heads/test_PPYOLOERHead.py @@ -34,7 +34,6 @@ def __init__(self): angle_max=90, use_varifocal_loss=True, static_assigner_epoch=-1, - trt=False, export_onnx=False, static_assigner=fcosrassigner, assigner=rotatedassigner, diff --git a/models/PaddleDetection/module_test/heads/test_YOLOFHead.py b/models/PaddleDetection/module_test/heads/test_YOLOFHead.py index b86166e2b6..e7f7980d3a 100644 --- a/models/PaddleDetection/module_test/heads/test_YOLOFHead.py +++ b/models/PaddleDetection/module_test/heads/test_YOLOFHead.py @@ -35,7 +35,6 @@ def __init__(self): prior_prob=0.01, nms_pre=1000, use_inside_anchor=False, - trt=False, exclude_nms=False, ) self.net.eval() diff --git a/models/PaddleGAN/Full_Chain/tipc.sh b/models/PaddleGAN/Full_Chain/tipc.sh index 4c25014996..2be8b5a04a 100644 --- a/models/PaddleGAN/Full_Chain/tipc.sh +++ b/models/PaddleGAN/Full_Chain/tipc.sh @@ -8,8 +8,6 @@ REPO=$1 # 参数作为配置文件传入 # DOCKER_IMAGE=registry.baidubce.com/paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82 # DOCKER_NAME=paddle_whole_chain_test_gan -# # COMPILE_PATH=https://paddle-qa.bj.bcebos.com/paddle-pipeline/Master_GpuAll_LinuxUbuntu_Gcc82_Cuda10.1_Trton_Py37_Compile_H_DISTRIBUTE/latest/paddlepaddle_gpu-0.0.0-cp37-cp37m-linux_x86_64.whl -# COMPILE_PATH=https://paddle-qa.bj.bcebos.com/paddle-pipeline/Master_GpuAll_LinuxUbuntu_Gcc82_Cuda10.1_Trton_Py37_Compile_H_DISTRIBUTE/890bd6266c1ba638ded7487e189fcf658e0579a1/paddlepaddle_gpu-0.0.0-cp37-cp37m-linux_x86_64.whl # #1207 COMPILE_PATH # define version compare function diff --git a/models/PaddleMIX/CE/ppdiffusers/deploy/script/gather_img_video_to_one_file.sh b/models/PaddleMIX/CE/ppdiffusers/deploy/script/gather_img_video_to_one_file.sh index 66f1c26a61..6261590a45 100644 --- a/models/PaddleMIX/CE/ppdiffusers/deploy/script/gather_img_video_to_one_file.sh +++ b/models/PaddleMIX/CE/ppdiffusers/deploy/script/gather_img_video_to_one_file.sh @@ -20,7 +20,7 @@ exit_code=0 cd ${work_path} # 遍历所有子目录 -find . -type d \( -name "results-paddle" -o -name "results-paddle-fp16" -o -name "results-paddle_tensorrt" -o -name "results-paddle_tensorrt-fp16" \) | while read dir; do +find . -type d \( -name "results-paddle" -o -name "results-paddle-fp16" \) | while read dir; do # 提取父目录路径作为子目录名 echo "Processing: $dir"; PARENT_DIR=$(basename "$(dirname "$dir")") @@ -39,7 +39,7 @@ done set -x cd ${work_path}/ipadapter/ set +x -find . -type d \( -name "results-paddle" -o -name "results-paddle-fp16" -o -name "results-paddle_tensorrt" -o -name "results-paddle_tensorrt-fp16" \) | while read dir; do +find . -type d \( -name "results-paddle" -o -name "results-paddle-fp16" \) | while read dir; do # 提取父目录路径作为子目录名 echo "Processing: $dir"; PARENT_DIR=$(basename "$(dirname "$dir")") diff --git a/models/PaddleNLP/CI/ci_case.sh b/models/PaddleNLP/CI/ci_case.sh index 226a53f992..79b0e63ea6 100644 --- a/models/PaddleNLP/CI/ci_case.sh +++ b/models/PaddleNLP/CI/ci_case.sh @@ -250,7 +250,6 @@ print_info $? gpt_p_depoly # # FT # cd ${nlp_dir}/ # export PYTHONPATH=$PWD/PaddleNLP/:$PYTHONPATH -# wget -q https://paddle-inference-lib.bj.bcebos.com/2.4.0/cxx_c/Linux/GPU/x86-64_gcc8.2_avx_mkl_cuda10.2_cudnn8.1.1_trt7.2.3.4/paddle_inference.tgz # tar -zxf paddle_inference.tgz # cd ${nlp_dir}/paddlenlp/ops # export CC=/usr/local/gcc-8.2/bin/gcc @@ -727,7 +726,6 @@ print_info $? transformer_infer # # FT # cd ${nlp_dir}/ # export PYTHONPATH=$PWD/PaddleNLP/:$PYTHONPATH -# wget -q https://paddle-inference-lib.bj.bcebos.com/2.4.0/cxx_c/Linux/GPU/x86-64_gcc8.2_avx_mkl_cuda10.2_cudnn8.1.1_trt7.2.3.4/paddle_inference.tgz # tar -zxf paddle_inference.tgz # export CC=/usr/local/gcc-8.2/bin/gcc # export CXX=/usr/local/gcc-8.2/bin/g++ diff --git a/models/PaddleSeg/CE/whole_function/run_PaddleSeg.sh b/models/PaddleSeg/CE/whole_function/run_PaddleSeg.sh index c42b815854..2da5291a4e 100644 --- a/models/PaddleSeg/CE/whole_function/run_PaddleSeg.sh +++ b/models/PaddleSeg/CE/whole_function/run_PaddleSeg.sh @@ -41,7 +41,6 @@ wget ${paddle_inference} tar xvf paddle_inference.tgz WITH_MKL=ON WITH_GPU=ON -USE_TENSORRT=OFF DEMO_NAME=test_seg work_path=$(dirname $(readlink -f $0)) LIB_DIR="${work_path}/paddle_inference" @@ -52,7 +51,6 @@ cmake .. \ -DDEMO_NAME=${DEMO_NAME} \ -DWITH_MKL=${WITH_MKL} \ -DWITH_GPU=${WITH_GPU} \ - -DUSE_TENSORRT=${USE_TENSORRT} \ -DWITH_STATIC_LIB=OFF \ -DPADDLE_LIB=${LIB_DIR} make -j diff --git a/models/PaddleVideo/CI/test_video.sh b/models/PaddleVideo/CI/test_video.sh index 366ab6b851..e2eadd3f07 100644 --- a/models/PaddleVideo/CI/test_video.sh +++ b/models/PaddleVideo/CI/test_video.sh @@ -79,8 +79,7 @@ INFER(){ --config ${config} \ --model_file inference/${model}/${model}.pdmodel \ --params_file inference/${model}/${model}.pdiparams \ - --use_gpu=True \ - --use_tensorrt=False >log/${model}/${model}_infer.log 2>&1 + --use_gpu=True >log/${model}/${model}_infer.log 2>&1 print_result } model_list='TSM ppTSN' diff --git a/models/ROCM/RocmTestFramework.py b/models/ROCM/RocmTestFramework.py index d2ebfce7e9..1acf3d2e03 100644 --- a/models/ROCM/RocmTestFramework.py +++ b/models/ROCM/RocmTestFramework.py @@ -321,13 +321,13 @@ class predict cmd_gpu = ( "cd PaddleClas; cd deploy; python python/predict_cls.py -c configs/inference_cls.yaml \ -o Global.inference_model_dir=../inference/%s -o Global.batch_size=1 -o Global.use_gpu=True \ - -o Global.use_tensorrt=False -o Global.enable_mkldnn=False" + -o Global.enable_mkldnn=False" % self.model ) cmd_cpu = ( "cd PaddleClas; cd deploy; python python/predict_cls.py -c configs/inference_cls.yaml \ -o Global.inference_model_dir=../inference/%s -o Global.batch_size=1 -o Global.use_gpu=False \ - -o Global.use_tensorrt=False -o Global.enable_mkldnn=False" + -o Global.enable_mkldnn=False" % self.model ) for cmd in [cmd_gpu, cmd_cpu]: diff --git a/models_restruct/Paddle3D/diy_build/Paddle3D_Build.py b/models_restruct/Paddle3D/diy_build/Paddle3D_Build.py index f6acea7e3d..f5b776de44 100644 --- a/models_restruct/Paddle3D/diy_build/Paddle3D_Build.py +++ b/models_restruct/Paddle3D/diy_build/Paddle3D_Build.py @@ -168,7 +168,6 @@ def compile_c_predict_demo(self): LIB_DIR = os.environ.get("paddle_inference_LIB_DIR") CUDA_LIB_DIR = os.environ.get("CUDA_LIB_DI") CUDNN_LIB_DIR = os.environ.get("CUDNN_LIB_DIR") - TENSORRT_DIR = os.environ.get("TENSORRT_DIR") os.chdir("Paddle3D/deploy/smoke/cpp") # paddle_inference @@ -186,9 +185,9 @@ def compile_c_predict_demo(self): cmd = ( "export OpenCV_DIR=%s; cmake .. -DPADDLE_LIB=%s -DWITH_MKL=ON -DDEMO_NAME=infer -DWITH_GPU=OFF \ - -DWITH_STATIC_LIB=OFF -DUSE_TENSORRT=OFF -DWITH_ROCM=OFF -DROCM_LIB=/opt/rocm/lib \ - -DCUDNN_LIB=%s -DCUDA_LIB=%s -DTENSORRT_ROOT=%s" - % (OPENCV_DIR, LIB_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR, TENSORRT_DIR) + -DWITH_STATIC_LIB=OFF -DWITH_ROCM=OFF -DROCM_LIB=/opt/rocm/lib \ + -DCUDNN_LIB=%s -DCUDA_LIB=%s" + % (OPENCV_DIR, LIB_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR) ) print(cmd) repo_result = subprocess.getstatusoutput(cmd) @@ -208,11 +207,11 @@ def compile_c_predict_demo(self): cmd = ( "cmake .. -DPADDLE_LIB=%s -DWITH_MKL=ON -DDEMO_NAME=main -DWITH_GPU=OFF \ - -DWITH_STATIC_LIB=OFF -DUSE_TENSORRT=OFF -DWITH_ROCM=OFF \ - -DROCM_LIB=/opt/rocm/lib -DCUDNN_LIB=%s -DCUDA_LIB=%s -DTENSORRT_ROOT=%s \ + -DWITH_STATIC_LIB=OFF -DWITH_ROCM=OFF \ + -DROCM_LIB=/opt/rocm/lib -DCUDNN_LIB=%s -DCUDA_LIB=%s \ -DCUSTOM_OPERATOR_FILES='custom_ops/iou3d_cpu.cpp;custom_ops/\ iou3d_nms_api.cpp;custom_ops/iou3d_nms.cpp;custom_ops/iou3d_nms_kernel.cu'" - % (LIB_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR, TENSORRT_DIR) + % (LIB_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR) ) print(cmd) repo_result = subprocess.getstatusoutput(cmd) @@ -233,10 +232,10 @@ def compile_c_predict_demo(self): cmd = ( "cmake .. -DPADDLE_LIB=%s -DWITH_MKL=ON -DDEMO_NAME=main -DWITH_GPU=OFF -DWITH_STATIC_LIB=OFF \ - -DUSE_TENSORRT=OFF -DWITH_ROCM=OFF -DROCM_LIB=/opt/rocm/lib -DCUDNN_LIB=%s -DCUDA_LIB=%s -DTENSORRT_ROOT=%s \ + -DWITH_ROCM=OFF -DROCM_LIB=/opt/rocm/lib -DCUDNN_LIB=%s -DCUDA_LIB=%s \ -DCUSTOM_OPERATOR_FILES='custom_ops/voxelize_op.cu;custom_ops/voxelize_op.cc;\ custom_ops/iou3d_nms_kernel.cu;custom_ops/postprocess.cc;custom_ops/postprocess.cu'" - % (LIB_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR, TENSORRT_DIR) + % (LIB_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR) ) print(cmd) repo_result = subprocess.getstatusoutput(cmd) @@ -256,10 +255,10 @@ def compile_c_predict_demo(self): cmd = ( "cmake .. -DOPENCV_DIR=%s -DPADDLE_LIB=%s -DWITH_MKL=ON -DDEMO_NAME=main \ - -DWITH_GPU=OFF -DWITH_STATIC_LIB=OFF -DUSE_TENSORRT=OFF \ + -DWITH_GPU=OFF -DWITH_STATIC_LIB=OFF \ -DWITH_ROCM=OFF -DROCM_LIB=/opt/rocm/lib -DCUDNN_LIB=%s \ - -DCUDA_LIB=%s -DTENSORRT_ROOT=%s -DCUSTOM_OPERATOR_FILES=''" - % (OPENCV_DIR, LIB_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR, TENSORRT_DIR) + -DCUDA_LIB=%s -DCUSTOM_OPERATOR_FILES=''" + % (OPENCV_DIR, LIB_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR) ) print(cmd) repo_result = subprocess.getstatusoutput(cmd) @@ -279,10 +278,10 @@ def compile_c_predict_demo(self): cmd = ( "cmake .. -DOPENCV_DIR=%s -DPADDLE_LIB=%s -DWITH_MKL=ON \ - -DDEMO_NAME=main -DWITH_GPU=OFF -DWITH_STATIC_LIB=OFF -DUSE_TENSORRT=OFF \ - -DWITH_ROCM=OFF -DROCM_LIB=/opt/rocm/lib -DCUDNN_LIB=%s -DCUDA_LIB=%s -DTENSORRT_ROOT=%s \ + -DDEMO_NAME=main -DWITH_GPU=OFF -DWITH_STATIC_LIB=OFF \ + -DWITH_ROCM=OFF -DROCM_LIB=/opt/rocm/lib -DCUDNN_LIB=%s -DCUDA_LIB=%s \ -DCUSTOM_OPERATOR_FILES='custom_ops/iou3d_nms.cpp;custom_ops/iou3d_nms_api.cpp;custom_ops/iou3d_nms_kernel.cu'" - % (OPENCV_DIR, LIB_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR, TENSORRT_DIR) + % (OPENCV_DIR, LIB_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR) ) print(cmd) repo_result = subprocess.getstatusoutput(cmd) diff --git a/models_restruct/PaddleClas/base/slim_base.yaml b/models_restruct/PaddleClas/base/slim_base.yaml index ed45e7605e..ba60b127e0 100644 --- a/models_restruct/PaddleClas/base/slim_base.yaml +++ b/models_restruct/PaddleClas/base/slim_base.yaml @@ -1,4 +1,3 @@ -#230207 修改slim为默认开启trt模式 skipped train: - name: function diff --git a/models_restruct/PaddleClas/tools/get_result.py b/models_restruct/PaddleClas/tools/get_result.py index b8e8c09020..92a5a55dee 100644 --- a/models_restruct/PaddleClas/tools/get_result.py +++ b/models_restruct/PaddleClas/tools/get_result.py @@ -327,7 +327,6 @@ def update_kpi(self): and ( tag_value["name"] == "trained" or tag_value["name"] == "trained_mkldnn" - or tag_value["name"] == "trained_trt" ) ): try: # 增加尝试方式报错，定死指标为class_ids 变成退出码 exit_code diff --git a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_fpn_1x_coco.yml index 56e9d6399a..1538830d12 100644 --- a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_fpn_1x_coco.yml @@ -37,10 +37,6 @@ case: name: python # - # name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_fpn_1x_coco.yml index 298c744d41..288632ae34 100644 --- a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_fpn_1x_coco.yml @@ -48,10 +48,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_vd_fpn_ssld_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_vd_fpn_ssld_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_vd_fpn_ssld_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_vd_fpn_ssld_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_vd_fpn_ssld_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_vd_fpn_ssld_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_vd_fpn_ssld_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^cascade_rcnn^cascade_rcnn_r50_vd_fpn_ssld_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_dla34_140e_coco.yml b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_dla34_140e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_dla34_140e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_dla34_140e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv1_140e_coco.yml b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv1_140e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv1_140e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv1_140e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv3_large_140e_coco.yml b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv3_large_140e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv3_large_140e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv3_large_140e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv3_small_140e_coco.yml b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv3_small_140e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv3_small_140e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_mbv3_small_140e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_r50_140e_coco.yml b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_r50_140e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_r50_140e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_r50_140e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_shufflenetv2_140e_coco.yml b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_shufflenetv2_140e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^centernet^centernet_shufflenetv2_140e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^centernet^centernet_shufflenetv2_140e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^cascade_rcnn_dcn_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^cascade_rcnn_dcn_r50_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^cascade_rcnn_dcn_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^cascade_rcnn_dcn_r50_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^cascade_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^cascade_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^cascade_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^cascade_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r101_vd_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r101_vd_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r101_vd_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r101_vd_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_vd_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_vd_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_vd_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_vd_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_vd_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_vd_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_vd_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_r50_vd_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^faster_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r101_vd_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r101_vd_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r101_vd_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r101_vd_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r50_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r50_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r50_vd_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r50_vd_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r50_vd_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_r50_vd_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^dcn^mask_rcnn_dcn_x101_vd_64x4d_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^deformable_detr^deformable_detr_r50_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^deformable_detr^deformable_detr_r50_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^deformable_detr^deformable_detr_r50_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^deformable_detr^deformable_detr_r50_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^detr^detr_r50_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^detr^detr_r50_1x_coco.yml index 163990fea8..657cf47b04 100644 --- a/models_restruct/PaddleDetection/cases/configs^detr^detr_r50_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^detr^detr_r50_1x_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^face_detection^blazeface_1000e.yml b/models_restruct/PaddleDetection/cases/configs^face_detection^blazeface_1000e.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^face_detection^blazeface_1000e.yml +++ b/models_restruct/PaddleDetection/cases/configs^face_detection^blazeface_1000e.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^face_detection^blazeface_fpn_ssh_1000e.yml b/models_restruct/PaddleDetection/cases/configs^face_detection^blazeface_fpn_ssh_1000e.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^face_detection^blazeface_fpn_ssh_1000e.yml +++ b/models_restruct/PaddleDetection/cases/configs^face_detection^blazeface_fpn_ssh_1000e.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_vd_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_vd_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_vd_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_vd_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_vd_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_vd_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_vd_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r101_vd_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r34_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r34_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r34_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r34_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r34_vd_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r34_vd_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r34_vd_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r34_vd_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_fpn_1x_coco.yml index 4b999deded..456b29b453 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_fpn_1x_coco.yml @@ -48,10 +48,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_ssld_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_ssld_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_ssld_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_ssld_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_ssld_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_ssld_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_ssld_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_r50_vd_fpn_ssld_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_1x_coco.yml index f46a8668b1..264889814e 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_1x_coco.yml @@ -37,10 +37,6 @@ case: name: python # - # name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_2x_coco.yml index aa010c4ce9..e3831656f3 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_x101_vd_64x4d_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_x101_vd_64x4d_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_x101_vd_64x4d_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_x101_vd_64x4d_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_x101_vd_64x4d_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_x101_vd_64x4d_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_x101_vd_64x4d_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^faster_rcnn^faster_rcnn_x101_vd_64x4d_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^fcos^fcos_dcn_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^fcos^fcos_dcn_r50_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^fcos^fcos_dcn_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^fcos^fcos_dcn_r50_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_1x_coco.yml index 9bb737ed2a..5c7b62771c 100644 --- a/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_1x_coco.yml @@ -47,10 +47,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_iou_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_iou_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_iou_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_iou_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_iou_multiscale_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_iou_multiscale_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_iou_multiscale_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_iou_multiscale_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_multiscale_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_multiscale_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_multiscale_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^fcos^fcos_r50_fpn_multiscale_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r101vd_fpn_mstrain_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r101vd_fpn_mstrain_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r101vd_fpn_mstrain_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r101vd_fpn_mstrain_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r18vd_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r18vd_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r18vd_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r18vd_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r34vd_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r34vd_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r34vd_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r34vd_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r50_fpn_1x_coco.yml index e794be3d3c..97e3cccf78 100644 --- a/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^gfl^gfl_r50_fpn_1x_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^gfl^gflv2_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^gfl^gflv2_r50_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^gfl^gflv2_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^gfl^gflv2_r50_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^gn^cascade_mask_rcnn_r50_fpn_gn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^gn^cascade_mask_rcnn_r50_fpn_gn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^gn^cascade_mask_rcnn_r50_fpn_gn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^gn^cascade_mask_rcnn_r50_fpn_gn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^gn^cascade_rcnn_r50_fpn_gn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^gn^cascade_rcnn_r50_fpn_gn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^gn^cascade_rcnn_r50_fpn_gn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^gn^cascade_rcnn_r50_fpn_gn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^gn^faster_rcnn_r50_fpn_gn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^gn^faster_rcnn_r50_fpn_gn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^gn^faster_rcnn_r50_fpn_gn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^gn^faster_rcnn_r50_fpn_gn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^gn^mask_rcnn_r50_fpn_gn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^gn^mask_rcnn_r50_fpn_gn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^gn^mask_rcnn_r50_fpn_gn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^gn^mask_rcnn_r50_fpn_gn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^hrnet^faster_rcnn_hrnetv2p_w18_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^hrnet^faster_rcnn_hrnetv2p_w18_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^hrnet^faster_rcnn_hrnetv2p_w18_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^hrnet^faster_rcnn_hrnetv2p_w18_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^hrnet^faster_rcnn_hrnetv2p_w18_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^hrnet^faster_rcnn_hrnetv2p_w18_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^hrnet^faster_rcnn_hrnetv2p_w18_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^hrnet^faster_rcnn_hrnetv2p_w18_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_512.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_512.yml index 66fa184166..1972188709 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_512.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_512.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_512_swahr.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_512_swahr.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_512_swahr.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_512_swahr.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_640.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_640.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_640.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^higherhrnet^higherhrnet_hrnet_w32_640.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^dark_hrnet_w32_256x192.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^dark_hrnet_w32_256x192.yml index d8e92beeb8..98d8a5e242 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^dark_hrnet_w32_256x192.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^dark_hrnet_w32_256x192.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^dark_hrnet_w32_384x288.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^dark_hrnet_w32_384x288.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^dark_hrnet_w32_384x288.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^dark_hrnet_w32_384x288.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^hrnet_w32_256x192.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^hrnet_w32_256x192.yml index db99dbe859..1fb8a39ba1 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^hrnet_w32_256x192.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^hrnet_w32_256x192.yml @@ -47,10 +47,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^hrnet_w32_384x288.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^hrnet_w32_384x288.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^hrnet_w32_384x288.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^hrnet^hrnet_w32_384x288.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_18_256x192_coco.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_18_256x192_coco.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_18_256x192_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_18_256x192_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_18_384x288_coco.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_18_384x288_coco.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_18_384x288_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_18_384x288_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_30_256x192_coco.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_30_256x192_coco.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_30_256x192_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_30_256x192_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_30_384x288_coco.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_30_384x288_coco.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_30_384x288_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^lite_hrnet_30_384x288_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^wider_naive_hrnet_18_256x192_coco.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^wider_naive_hrnet_18_256x192_coco.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^wider_naive_hrnet_18_256x192_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^lite_hrnet^wider_naive_hrnet_18_256x192_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^tiny_pose^tinypose_128x96.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^tiny_pose^tinypose_128x96.yml index 8c5ca7f9b4..e4d56db9b5 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^tiny_pose^tinypose_128x96.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^tiny_pose^tinypose_128x96.yml @@ -41,10 +41,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^keypoint^tiny_pose^tinypose_256x192.yml b/models_restruct/PaddleDetection/cases/configs^keypoint^tiny_pose^tinypose_256x192.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/configs^keypoint^tiny_pose^tinypose_256x192.yml +++ b/models_restruct/PaddleDetection/cases/configs^keypoint^tiny_pose^tinypose_256x192.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r101_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r101_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r101_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r101_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r101_vd_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r101_vd_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r101_vd_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r101_vd_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_1x_coco.yml index 94d548ffe5..0189a9e64d 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_1x_coco.yml @@ -47,10 +47,6 @@ case: name: python # - # name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_fpn_1x_coco.yml index 56e9d6399a..1538830d12 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_fpn_1x_coco.yml @@ -37,10 +37,6 @@ case: name: python # - # name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_x101_vd_64x4d_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_x101_vd_64x4d_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_x101_vd_64x4d_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_x101_vd_64x4d_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_x101_vd_64x4d_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_x101_vd_64x4d_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_x101_vd_64x4d_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^mask_rcnn^mask_rcnn_x101_vd_64x4d_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608.yml b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608.yml index 6265d4609b..bf861adb61 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608.yml @@ -48,10 +48,6 @@ case: - name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608_airplane.yml b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608_airplane.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608_airplane.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608_airplane.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608_bytetracker.yml b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608_bytetracker.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608_bytetracker.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_1088x608_bytetracker.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_576x320.yml b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_576x320.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_576x320.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_576x320.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_864x480.yml b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_864x480.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_864x480.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_dla34_30e_864x480.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_enhance_dla34_60e_1088x608.yml b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_enhance_dla34_60e_1088x608.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_enhance_dla34_60e_1088x608.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_enhance_dla34_60e_1088x608.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_enhance_hardnet85_30e_1088x608.yml b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_enhance_hardnet85_30e_1088x608.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_enhance_hardnet85_30e_1088x608.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_enhance_hardnet85_30e_1088x608.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_1088x608.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_576x320.yml b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_576x320.yml index 47d70b36e8..abe36887f9 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_576x320.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_576x320.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_864x480.yml b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_864x480.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_864x480.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^fairmot^fairmot_hrnetv2_w18_dlafpn_30e_864x480.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^headtracking21^fairmot_dla34_30e_1088x608_headtracking21.yml b/models_restruct/PaddleDetection/cases/configs^mot^headtracking21^fairmot_dla34_30e_1088x608_headtracking21.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^headtracking21^fairmot_dla34_30e_1088x608_headtracking21.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^headtracking21^fairmot_dla34_30e_1088x608_headtracking21.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_1088x608.yml b/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_1088x608.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_1088x608.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_1088x608.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_576x320.yml b/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_576x320.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_576x320.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_576x320.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_864x480.yml b/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_864x480.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_864x480.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^jde^jde_darknet53_30e_864x480.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_dla34_30e_1088x608_visdrone.yml b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_dla34_30e_1088x608_visdrone.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_dla34_30e_1088x608_visdrone.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_dla34_30e_1088x608_visdrone.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_dla34_30e_1088x608_visdrone_vehicle_bytetracker.yml b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_dla34_30e_1088x608_visdrone_vehicle_bytetracker.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_dla34_30e_1088x608_visdrone_vehicle_bytetracker.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_dla34_30e_1088x608_visdrone_vehicle_bytetracker.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone.yml b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone_vehicle_bytetracker.yml b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone_vehicle_bytetracker.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone_vehicle_bytetracker.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone_vehicle_bytetracker.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_576x320_bdd100k_mcmot.yml b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_576x320_bdd100k_mcmot.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_576x320_bdd100k_mcmot.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_576x320_bdd100k_mcmot.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone.yml b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_864x480_visdrone.yml b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_864x480_visdrone.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_864x480_visdrone.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^mcfairmot^mcfairmot_hrnetv2_w18_dlafpn_30e_864x480_visdrone.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_dla34_30e_1088x608_pathtrack.yml b/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_dla34_30e_1088x608_pathtrack.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_dla34_30e_1088x608_pathtrack.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_dla34_30e_1088x608_pathtrack.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_dla34_30e_1088x608_visdrone_pedestrian.yml b/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_dla34_30e_1088x608_visdrone_pedestrian.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_dla34_30e_1088x608_visdrone_pedestrian.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_dla34_30e_1088x608_visdrone_pedestrian.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone_pedestrian.yml b/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone_pedestrian.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone_pedestrian.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_1088x608_visdrone_pedestrian.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone_pedestrian.yml b/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone_pedestrian.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone_pedestrian.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone_pedestrian.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_864x480_visdrone_pedestrian.yml b/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_864x480_visdrone_pedestrian.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_864x480_visdrone_pedestrian.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^pedestrian^fairmot_hrnetv2_w18_dlafpn_30e_864x480_visdrone_pedestrian.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_bdd100kmot_vehicle.yml b/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_bdd100kmot_vehicle.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_bdd100kmot_vehicle.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_bdd100kmot_vehicle.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_kitti_vehicle.yml b/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_kitti_vehicle.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_kitti_vehicle.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_kitti_vehicle.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_visdrone_vehicle.yml b/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_visdrone_vehicle.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_visdrone_vehicle.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_dla34_30e_1088x608_visdrone_vehicle.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_hrnetv2_w18_dlafpn_30e_576x320_bdd100kmot_vehicle.yml b/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_hrnetv2_w18_dlafpn_30e_576x320_bdd100kmot_vehicle.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_hrnetv2_w18_dlafpn_30e_576x320_bdd100kmot_vehicle.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_hrnetv2_w18_dlafpn_30e_576x320_bdd100kmot_vehicle.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone_vehicle.yml b/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone_vehicle.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone_vehicle.yml +++ b/models_restruct/PaddleDetection/cases/configs^mot^vehicle^fairmot_hrnetv2_w18_dlafpn_30e_576x320_visdrone_vehicle.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_lcnet_1_5x_640_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_lcnet_1_5x_640_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_lcnet_1_5x_640_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_lcnet_1_5x_640_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_mobilenetv3_large_1x_416_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_mobilenetv3_large_1x_416_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_mobilenetv3_large_1x_416_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_mobilenetv3_large_1x_416_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_r18_640_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_r18_640_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_r18_640_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_r18_640_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_shufflenetv2_1x_416_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_shufflenetv2_1x_416_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_shufflenetv2_1x_416_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^more_config^picodet_shufflenetv2_1x_416_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_320_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_320_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_320_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_320_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_416_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_416_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_416_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_416_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_640_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_640_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_640_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_l_640_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_m_320_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_m_320_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_m_320_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_m_320_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_m_416_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_m_416_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_m_416_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_m_416_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_s_416_coco.yml b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_s_416_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_s_416_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^legacy_model^picodet_s_416_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_320_coco_lcnet.yml b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_320_coco_lcnet.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_320_coco_lcnet.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_320_coco_lcnet.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_416_coco_lcnet.yml b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_416_coco_lcnet.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_416_coco_lcnet.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_416_coco_lcnet.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_640_coco_lcnet.yml b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_640_coco_lcnet.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_640_coco_lcnet.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_l_640_coco_lcnet.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_m_320_coco_lcnet.yml b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_m_320_coco_lcnet.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_m_320_coco_lcnet.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_m_320_coco_lcnet.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_m_416_coco_lcnet.yml b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_m_416_coco_lcnet.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_m_416_coco_lcnet.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_m_416_coco_lcnet.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_416_coco_lcnet.yml b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_416_coco_lcnet.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_416_coco_lcnet.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_416_coco_lcnet.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_416_coco_npu.yml b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_416_coco_npu.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_416_coco_npu.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_s_416_coco_npu.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_xs_320_coco_lcnet.yml b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_xs_320_coco_lcnet.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_xs_320_coco_lcnet.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_xs_320_coco_lcnet.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_xs_416_coco_lcnet.yml b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_xs_416_coco_lcnet.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^picodet^picodet_xs_416_coco_lcnet.yml +++ b/models_restruct/PaddleDetection/cases/configs^picodet^picodet_xs_416_coco_lcnet.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_mbv3_small_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_mbv3_small_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_mbv3_small_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_mbv3_small_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r18vd_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r18vd_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r18vd_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r18vd_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_voc.yml b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_voc.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_voc.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolo_r50vd_dcn_voc.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolov2_r101vd_dcn_365e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolov2_r101vd_dcn_365e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolov2_r101vd_dcn_365e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyolo^ppyolov2_r101vd_dcn_365e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_l_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_l_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_l_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_l_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_m_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_m_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_m_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_m_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_s_400e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_s_400e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_s_400e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_s_400e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_x_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_x_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_x_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_crn_x_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_l_80e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_l_80e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_l_80e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_l_80e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_m_80e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_m_80e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_m_80e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_m_80e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_x_80e_coco.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_x_80e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_x_80e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^ppyoloe_plus_crn_x_80e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^voc^ppyoloe_plus_crn_l_30e_voc.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^voc^ppyoloe_plus_crn_l_30e_voc.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^voc^ppyoloe_plus_crn_l_30e_voc.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^voc^ppyoloe_plus_crn_l_30e_voc.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ppyoloe^voc^ppyoloe_plus_crn_s_30e_voc.yml b/models_restruct/PaddleDetection/cases/configs^ppyoloe^voc^ppyoloe_plus_crn_s_30e_voc.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ppyoloe^voc^ppyoloe_plus_crn_s_30e_voc.yml +++ b/models_restruct/PaddleDetection/cases/configs^ppyoloe^voc^ppyoloe_plus_crn_s_30e_voc.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rcnn_enhance^faster_rcnn_enhance_3x_coco.yml b/models_restruct/PaddleDetection/cases/configs^rcnn_enhance^faster_rcnn_enhance_3x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rcnn_enhance^faster_rcnn_enhance_3x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^rcnn_enhance^faster_rcnn_enhance_3x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^res2net^faster_rcnn_res2net50_vb_26w_4s_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^res2net^faster_rcnn_res2net50_vb_26w_4s_fpn_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^res2net^faster_rcnn_res2net50_vb_26w_4s_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^res2net^faster_rcnn_res2net50_vb_26w_4s_fpn_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^res2net^mask_rcnn_res2net50_vb_26w_4s_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^res2net^mask_rcnn_res2net50_vb_26w_4s_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^res2net^mask_rcnn_res2net50_vb_26w_4s_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^res2net^mask_rcnn_res2net50_vb_26w_4s_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^res2net^mask_rcnn_res2net50_vd_26w_4s_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^res2net^mask_rcnn_res2net50_vd_26w_4s_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^res2net^mask_rcnn_res2net50_vd_26w_4s_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^res2net^mask_rcnn_res2net50_vd_26w_4s_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r101_distill_r50_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r101_distill_r50_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r101_distill_r50_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r101_distill_r50_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r101_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r101_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r101_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r101_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r50_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r50_fpn_2x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r50_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^retinanet^retinanet_r50_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^fcosr^fcosr_x50_3x_dota.yml b/models_restruct/PaddleDetection/cases/configs^rotate^fcosr^fcosr_x50_3x_dota.yml index 163990fea8..657cf47b04 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^fcosr^fcosr_x50_3x_dota.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^fcosr^fcosr_x50_3x_dota.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_l_3x_dota.yml b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_l_3x_dota.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_l_3x_dota.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_l_3x_dota.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_l_3x_dota_ms.yml b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_l_3x_dota_ms.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_l_3x_dota_ms.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_l_3x_dota_ms.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_m_3x_dota.yml b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_m_3x_dota.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_m_3x_dota.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_m_3x_dota.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_m_3x_dota_ms.yml b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_m_3x_dota_ms.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_m_3x_dota_ms.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_m_3x_dota_ms.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_s_3x_dota.yml b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_s_3x_dota.yml index 163990fea8..657cf47b04 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_s_3x_dota.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_s_3x_dota.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_s_3x_dota_ms.yml b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_s_3x_dota_ms.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_s_3x_dota_ms.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_s_3x_dota_ms.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_x_3x_dota.yml b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_x_3x_dota.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_x_3x_dota.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_x_3x_dota.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_x_3x_dota_ms.yml b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_x_3x_dota_ms.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_x_3x_dota_ms.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^ppyoloe_r^ppyoloe_r_crn_x_3x_dota_ms.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_1x_spine.yml b/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_1x_spine.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_1x_spine.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_1x_spine.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_alignconv_2x_dota.yml b/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_alignconv_2x_dota.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_alignconv_2x_dota.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_alignconv_2x_dota.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_conv_2x_dota.yml b/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_conv_2x_dota.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_conv_2x_dota.yml +++ b/models_restruct/PaddleDetection/cases/configs^rotate^s2anet^s2anet_conv_2x_dota.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^faster_rcnn_r50_fpn_2x_coco_sup010.yml b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^faster_rcnn_r50_fpn_2x_coco_sup010.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^faster_rcnn_r50_fpn_2x_coco_sup010.yml +++ b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^faster_rcnn_r50_fpn_2x_coco_sup010.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^fcos_r50_fpn_2x_coco_sup005.yml b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^fcos_r50_fpn_2x_coco_sup005.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^fcos_r50_fpn_2x_coco_sup005.yml +++ b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^fcos_r50_fpn_2x_coco_sup005.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^fcos_r50_fpn_2x_coco_sup010.yml b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^fcos_r50_fpn_2x_coco_sup010.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^fcos_r50_fpn_2x_coco_sup010.yml +++ b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^fcos_r50_fpn_2x_coco_sup010.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^ppyoloe_plus_crn_s_80e_coco_sup005.yml b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^ppyoloe_plus_crn_s_80e_coco_sup005.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^ppyoloe_plus_crn_s_80e_coco_sup005.yml +++ b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^ppyoloe_plus_crn_s_80e_coco_sup005.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^ppyoloe_plus_crn_s_80e_coco_sup010.yml b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^ppyoloe_plus_crn_s_80e_coco_sup010.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^ppyoloe_plus_crn_s_80e_coco_sup010.yml +++ b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^ppyoloe_plus_crn_s_80e_coco_sup010.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^retinanet_r50_fpn_2x_coco_sup010.yml b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^retinanet_r50_fpn_2x_coco_sup010.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^retinanet_r50_fpn_2x_coco_sup010.yml +++ b/models_restruct/PaddleDetection/cases/configs^semi_det^baseline^retinanet_r50_fpn_2x_coco_sup010.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^sniper^faster_rcnn_r50_fpn_1x_sniper_visdrone.yml b/models_restruct/PaddleDetection/cases/configs^sniper^faster_rcnn_r50_fpn_1x_sniper_visdrone.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^sniper^faster_rcnn_r50_fpn_1x_sniper_visdrone.yml +++ b/models_restruct/PaddleDetection/cases/configs^sniper^faster_rcnn_r50_fpn_1x_sniper_visdrone.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^sniper^faster_rcnn_r50_fpn_1x_visdrone.yml b/models_restruct/PaddleDetection/cases/configs^sniper^faster_rcnn_r50_fpn_1x_visdrone.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^sniper^faster_rcnn_r50_fpn_1x_visdrone.yml +++ b/models_restruct/PaddleDetection/cases/configs^sniper^faster_rcnn_r50_fpn_1x_visdrone.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r101_vd_fpn_3x_coco.yml b/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r101_vd_fpn_3x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r101_vd_fpn_3x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r101_vd_fpn_3x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_enhance_coco.yml b/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_enhance_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_enhance_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_enhance_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_fpn_1x_coco.yml index 31ab63cfa4..840d454367 100644 --- a/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_fpn_1x_coco.yml @@ -48,10 +48,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_fpn_3x_coco.yml b/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_fpn_3x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_fpn_3x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^solov2^solov2_r50_fpn_3x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ssd^ssd_mobilenet_v1_300_120e_voc.yml b/models_restruct/PaddleDetection/cases/configs^ssd^ssd_mobilenet_v1_300_120e_voc.yml index 5abca8ec77..8459079004 100644 --- a/models_restruct/PaddleDetection/cases/configs^ssd^ssd_mobilenet_v1_300_120e_voc.yml +++ b/models_restruct/PaddleDetection/cases/configs^ssd^ssd_mobilenet_v1_300_120e_voc.yml @@ -48,10 +48,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ssd^ssd_vgg16_300_240e_voc.yml b/models_restruct/PaddleDetection/cases/configs^ssd^ssd_vgg16_300_240e_voc.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ssd^ssd_vgg16_300_240e_voc.yml +++ b/models_restruct/PaddleDetection/cases/configs^ssd^ssd_vgg16_300_240e_voc.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_ghostnet_320_coco.yml b/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_ghostnet_320_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_ghostnet_320_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_ghostnet_320_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v1_300_coco.yml b/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v1_300_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v1_300_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v1_300_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v3_large_320_coco.yml b/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v3_large_320_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v3_large_320_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v3_large_320_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v3_small_320_coco.yml b/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v3_small_320_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v3_small_320_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ssd^ssdlite_mobilenet_v3_small_320_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^tood^tood_r50_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^tood^tood_r50_fpn_1x_coco.yml index 163990fea8..657cf47b04 100644 --- a/models_restruct/PaddleDetection/cases/configs^tood^tood_r50_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^tood^tood_r50_fpn_1x_coco.yml @@ -37,10 +37,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/configs^ttfnet^pafnet_10x_coco.yml b/models_restruct/PaddleDetection/cases/configs^ttfnet^pafnet_10x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ttfnet^pafnet_10x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ttfnet^pafnet_10x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ttfnet^pafnet_lite_mobilenet_v3_20x_coco.yml b/models_restruct/PaddleDetection/cases/configs^ttfnet^pafnet_lite_mobilenet_v3_20x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^ttfnet^pafnet_lite_mobilenet_v3_20x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ttfnet^pafnet_lite_mobilenet_v3_20x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^ttfnet^ttfnet_darknet53_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^ttfnet^ttfnet_darknet53_1x_coco.yml index 1f11e480f8..d17ab61aa0 100644 --- a/models_restruct/PaddleDetection/cases/configs^ttfnet^ttfnet_darknet53_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^ttfnet^ttfnet_darknet53_1x_coco.yml @@ -48,10 +48,6 @@ case: - name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolof^yolof_r50_c5_1x_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolof^yolof_r50_c5_1x_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolof^yolof_r50_c5_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolof^yolof_r50_c5_1x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_darknet53_270e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_darknet53_270e_coco.yml index 03552bc963..3cfe540153 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_darknet53_270e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_darknet53_270e_coco.yml @@ -48,10 +48,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_270e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_270e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_270e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_270e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_270e_voc.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_270e_voc.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_270e_voc.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_270e_voc.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_roadsign.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_roadsign.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_roadsign.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_roadsign.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_ssld_270e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_ssld_270e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_ssld_270e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_ssld_270e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_ssld_270e_voc.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_ssld_270e_voc.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_ssld_270e_voc.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v1_ssld_270e_voc.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_270e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_270e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_270e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_270e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_270e_voc.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_270e_voc.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_270e_voc.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_270e_voc.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_ssld_270e_voc.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_ssld_270e_voc.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_ssld_270e_voc.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_mobilenet_v3_large_ssld_270e_voc.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_r34_270e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_r34_270e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_r34_270e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_r34_270e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_r50vd_dcn_270e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_r50vd_dcn_270e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_r50vd_dcn_270e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolov3^yolov3_r50vd_dcn_270e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_cdn_tiny_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_cdn_tiny_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_cdn_tiny_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_cdn_tiny_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_crn_s_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_crn_s_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_crn_s_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_crn_s_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_l_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_l_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_l_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_l_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_m_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_m_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_m_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_m_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_nano_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_nano_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_nano_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_nano_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_s_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_s_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_s_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_s_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_tiny_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_tiny_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_tiny_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_tiny_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_x_300e_coco.yml b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_x_300e_coco.yml index 8386b53ddf..443b8dd5c2 100644 --- a/models_restruct/PaddleDetection/cases/configs^yolox^yolox_x_300e_coco.yml +++ b/models_restruct/PaddleDetection/cases/configs^yolox^yolox_x_300e_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/test_configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_1x_coco.yml b/models_restruct/PaddleDetection/cases/test_configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_1x_coco.yml index 7140574c9e..48b16568f6 100644 --- a/models_restruct/PaddleDetection/cases/test_configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_1x_coco.yml +++ b/models_restruct/PaddleDetection/cases/test_configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_1x_coco.yml @@ -37,10 +37,6 @@ case: name: python # - # name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 # - name: paddle2onnx # - diff --git a/models_restruct/PaddleDetection/cases/test_configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_2x_coco.yml b/models_restruct/PaddleDetection/cases/test_configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_2x_coco.yml index aa010c4ce9..e3831656f3 100644 --- a/models_restruct/PaddleDetection/cases/test_configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_2x_coco.yml +++ b/models_restruct/PaddleDetection/cases/test_configs^faster_rcnn^faster_rcnn_swin_tiny_fpn_2x_coco.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/test_configs^keypoint^tiny_pose^tinypose_128x96.yml b/models_restruct/PaddleDetection/cases/test_configs^keypoint^tiny_pose^tinypose_128x96.yml index 8c5ca7f9b4..e4d56db9b5 100644 --- a/models_restruct/PaddleDetection/cases/test_configs^keypoint^tiny_pose^tinypose_128x96.yml +++ b/models_restruct/PaddleDetection/cases/test_configs^keypoint^tiny_pose^tinypose_128x96.yml @@ -41,10 +41,6 @@ case: name: python - name: mkldnn - # - - # name: trt_fp32 - # - - # name: trt_fp16 - name: paddle2onnx - diff --git a/models_restruct/PaddleDetection/cases/test_keypoint.yml b/models_restruct/PaddleDetection/cases/test_keypoint.yml index ceb19e5f57..3b82c0e8bf 100644 --- a/models_restruct/PaddleDetection/cases/test_keypoint.yml +++ b/models_restruct/PaddleDetection/cases/test_keypoint.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleDetection/cases/test_mot.yml b/models_restruct/PaddleDetection/cases/test_mot.yml index 67dc1a4955..38e65e2404 100644 --- a/models_restruct/PaddleDetection/cases/test_mot.yml +++ b/models_restruct/PaddleDetection/cases/test_mot.yml @@ -38,10 +38,6 @@ case: # - # name: mkldnn # - - # name: trt_fp32 - # - - # name: trt_fp16 - # - # name: paddle2onnx # - # name: onnx_infer diff --git a/models_restruct/PaddleLLM/tools/run.sh b/models_restruct/PaddleLLM/tools/run.sh index 7e4bcbb38d..94f17dd847 100644 --- a/models_restruct/PaddleLLM/tools/run.sh +++ b/models_restruct/PaddleLLM/tools/run.sh @@ -22,7 +22,6 @@ mv -v PaddleNLP ./TestFrameWork/PaddleLLM unset http_proxy && unset https_proxy # python -m pip install -r TestFrameWork/requirements.txt -# python -m pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release-TagBuild-Training-Linux-Gpu-Cuda12.8-Cudnn9.7-Trt10.5-Mkl-Avx-Gcc11-SelfBuiltPypiUse/3b5fe1f4e5b4bd71f1c0b8e33d459f2f4caff554/paddlepaddle_gpu-3.0.0.dev20250423-cp310-cp310-linux_x86_64.whl --force-reinstall --no-dependencies #### for cuda12.8 pdc image ##### export LD_LIBRARY_PATH=/usr/local/lib/python3.10/dist-packages/nvidia/cusparse/lib/:${LD_LIBRARY_PATH} diff --git a/models_restruct/PaddleLLM/tools/run_build.sh b/models_restruct/PaddleLLM/tools/run_build.sh index 084c7a9cde..c620649181 100644 --- a/models_restruct/PaddleLLM/tools/run_build.sh +++ b/models_restruct/PaddleLLM/tools/run_build.sh @@ -10,7 +10,6 @@ rm -rf PaddleNLP && tar xf PaddleNLP.tar && rm -rf PaddleNLP.tar sed -i '/from transformer_engine import transformer_engine_paddle as tex/,/^ }/d' PaddleNLP/paddlenlp/quantization/qat_utils.py unset http_proxy && unset https_proxy -# python -m pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release-TagBuild-Training-Linux-Gpu-Cuda12.8-Cudnn9.7-Trt10.5-Mkl-Avx-Gcc11-SelfBuiltPypiUse/3b5fe1f4e5b4bd71f1c0b8e33d459f2f4caff554/paddlepaddle_gpu-3.0.0.dev20250423-cp310-cp310-linux_x86_64.whl --force-reinstall --no-dependencies #### for cuda12.8 pdc image ##### export LD_LIBRARY_PATH=/usr/local/lib/python3.10/dist-packages/nvidia/cusparse/lib/:${LD_LIBRARY_PATH} diff --git a/models_restruct/PaddleOCR/base/ocr_cls_base.yaml b/models_restruct/PaddleOCR/base/ocr_cls_base.yaml index f0a33d01a9..f12e0c5d07 100755 --- a/models_restruct/PaddleOCR/base/ocr_cls_base.yaml +++ b/models_restruct/PaddleOCR/base/ocr_cls_base.yaml @@ -152,7 +152,6 @@ predict: - --image_dir="doc/imgs_words_en/word_10.png" - --cls_model_dir="./models_inference/"${qa_yaml_name} - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -161,7 +160,6 @@ predict: - --image_dir="doc/imgs_words_en/word_10.png" - --cls_model_dir="./models_inference/"${qa_yaml_name} - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True api: name: trained_C_plus_plus_GPU diff --git a/models_restruct/PaddleOCR/base/ocr_cls_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_cls_base_pretrained.yaml index 09ece540aa..e8cd3e83d6 100755 --- a/models_restruct/PaddleOCR/base/ocr_cls_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_cls_base_pretrained.yaml @@ -152,7 +152,6 @@ predict: - --image_dir="doc/imgs_words_en/word_10.png" - --cls_model_dir="./models_inference/"${qa_yaml_name} - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -161,7 +160,6 @@ predict: - --image_dir="doc/imgs_words_en/word_10.png" - --cls_model_dir="./models_inference/"${qa_yaml_name} - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - name: pretrained @@ -170,7 +168,6 @@ predict: - --image_dir="doc/imgs_words_en/word_10.png" - --cls_model_dir="./models_inference/"${model} - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: pretrained_mkldnn @@ -179,7 +176,6 @@ predict: - --image_dir="doc/imgs_words_en/word_10.png" - --cls_model_dir="./models_inference/"${model} - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True api: name: trained_C_plus_plus_GPU diff --git a/models_restruct/PaddleOCR/base/ocr_det_base.yaml b/models_restruct/PaddleOCR/base/ocr_det_base.yaml index aa200916dd..b88b5c1cf1 100755 --- a/models_restruct/PaddleOCR/base/ocr_det_base.yaml +++ b/models_restruct/PaddleOCR/base/ocr_det_base.yaml @@ -116,7 +116,6 @@ predict: - --image_dir=doc/imgs_en/img_10.jpg - --det_algorithm=${algorithm} - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - --det_model_dir=./models_inference/${qa_yaml_name} - @@ -126,7 +125,6 @@ predict: - --image_dir=doc/imgs_en/img_10.jpg - --det_algorithm=${algorithm} - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - --det_model_dir=./models_inference/${qa_yaml_name} api: diff --git a/models_restruct/PaddleOCR/base/ocr_det_base_distill.yaml b/models_restruct/PaddleOCR/base/ocr_det_base_distill.yaml index f310787200..edc1b21ac5 100755 --- a/models_restruct/PaddleOCR/base/ocr_det_base_distill.yaml +++ b/models_restruct/PaddleOCR/base/ocr_det_base_distill.yaml @@ -116,7 +116,6 @@ predict: - --image_dir=doc/imgs_en/img_10.jpg - --det_algorithm=${algorithm} - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - --det_model_dir=./models_inference/${qa_yaml_name}/Student - @@ -126,7 +125,6 @@ predict: - --image_dir=doc/imgs_en/img_10.jpg - --det_algorithm=${algorithm} - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - --det_model_dir=./models_inference/${qa_yaml_name}/Student api: diff --git a/models_restruct/PaddleOCR/base/ocr_det_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_det_base_pretrained.yaml index 7ab9c5f0cc..6a96e71148 100755 --- a/models_restruct/PaddleOCR/base/ocr_det_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_det_base_pretrained.yaml @@ -146,7 +146,6 @@ predict: - --image_dir=doc/imgs_en/img_10.jpg - --det_algorithm=${algorithm} - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - --det_model_dir=./models_inference/${qa_yaml_name} - @@ -156,7 +155,6 @@ predict: - --image_dir=doc/imgs_en/img_10.jpg - --det_algorithm=${algorithm} - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - --det_model_dir=./models_inference/${qa_yaml_name} - @@ -166,7 +164,6 @@ predict: - --image_dir=doc/imgs_en/img_10.jpg - --det_algorithm=${algorithm} - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - --det_model_dir=./models_inference/${model} - @@ -176,7 +173,6 @@ predict: - --image_dir=doc/imgs_en/img_10.jpg - --det_algorithm=${algorithm} - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - --det_model_dir=./models_inference/${model} api: diff --git a/models_restruct/PaddleOCR/base/ocr_e2e_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_e2e_base_pretrained.yaml index 8bca67d507..f9f8ced4a2 100755 --- a/models_restruct/PaddleOCR/base/ocr_e2e_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_e2e_base_pretrained.yaml @@ -173,7 +173,6 @@ predict: - --e2e_model_dir="./models_inference/"${qa_yaml_name} - --e2e_algorithm=PGNet - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -183,7 +182,6 @@ predict: - --e2e_model_dir="./models_inference/"${qa_yaml_name} - --e2e_algorithm=PGNet - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - name: pretrained @@ -193,7 +191,6 @@ predict: - --e2e_model_dir="./models_inference/"${model} - --e2e_algorithm=PGNet - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: pretrained_mkldnn @@ -203,5 +200,4 @@ predict: - --e2e_model_dir="./models_inference/"${model} - --e2e_algorithm=PGNet - --use_gpu=$False - - --use_tensorrt=False - --enable_mkldnn=True diff --git a/models_restruct/PaddleOCR/base/ocr_kie_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_kie_base_pretrained.yaml index ed5e2cc8dc..a2791cbdce 100755 --- a/models_restruct/PaddleOCR/base/ocr_kie_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_kie_base_pretrained.yaml @@ -147,7 +147,6 @@ predict: - --vis_font_path=./doc/fonts/simfang.ttf - --ocr_order_method="tb-yx" - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -160,7 +159,6 @@ predict: - --vis_font_path=./doc/fonts/simfang.ttf - --ocr_order_method="tb-yx" - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - name: pretrained @@ -173,7 +171,6 @@ predict: - --vis_font_path=./doc/fonts/simfang.ttf - --ocr_order_method=tb-yx - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: pretrained_mkldnn @@ -186,5 +183,4 @@ predict: - --vis_font_path=./doc/fonts/simfang.ttf - --ocr_order_method=tb-yx - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True diff --git a/models_restruct/PaddleOCR/base/ocr_rec_base.yaml b/models_restruct/PaddleOCR/base/ocr_rec_base.yaml index dd4ce19374..c25ad43e6f 100755 --- a/models_restruct/PaddleOCR/base/ocr_rec_base.yaml +++ b/models_restruct/PaddleOCR/base/ocr_rec_base.yaml @@ -137,7 +137,6 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -150,7 +149,6 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True api: name: trained_C_plus_plus_GPU diff --git a/models_restruct/PaddleOCR/base/ocr_rec_base_distill.yaml b/models_restruct/PaddleOCR/base/ocr_rec_base_distill.yaml index 7283daa515..f885b6c06d 100755 --- a/models_restruct/PaddleOCR/base/ocr_rec_base_distill.yaml +++ b/models_restruct/PaddleOCR/base/ocr_rec_base_distill.yaml @@ -143,7 +143,6 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -156,5 +155,4 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True diff --git a/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained.yaml index 3da219f535..5d1241e4a8 100755 --- a/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained.yaml @@ -166,7 +166,6 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -179,7 +178,6 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - name: pretrained @@ -192,7 +190,6 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: pretrained_mkldnn @@ -205,7 +202,6 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True api: name: pretrained_C_plus_plus_GPU diff --git a/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained_distill.yaml b/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained_distill.yaml index 3480b2254d..ec0becd1cf 100755 --- a/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained_distill.yaml +++ b/models_restruct/PaddleOCR/base/ocr_rec_base_pretrained_distill.yaml @@ -172,7 +172,6 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -185,7 +184,6 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - name: pretrained @@ -198,7 +196,6 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: pretrained_mkldnn @@ -211,5 +208,4 @@ predict: - --rec_char_dict_path=${rec_dict} - --use_space_char=False - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True diff --git a/models_restruct/PaddleOCR/base/ocr_sr_base.yaml b/models_restruct/PaddleOCR/base/ocr_sr_base.yaml index da86fa2aca..9a0e7ffa7c 100755 --- a/models_restruct/PaddleOCR/base/ocr_sr_base.yaml +++ b/models_restruct/PaddleOCR/base/ocr_sr_base.yaml @@ -87,7 +87,6 @@ predict: - --sr_model_dir="./models_inference/"${qa_yaml_name} - --sr_image_shape=${image_shape} - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -97,5 +96,4 @@ predict: - --sr_model_dir="./models_inference/"${qa_yaml_name} - --sr_image_shape=${image_shape} - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True diff --git a/models_restruct/PaddleOCR/base/ocr_sr_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_sr_base_pretrained.yaml index 958a887f3e..73df3201c1 100755 --- a/models_restruct/PaddleOCR/base/ocr_sr_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_sr_base_pretrained.yaml @@ -113,7 +113,6 @@ predict: - --sr_model_dir="./models_inference/"${qa_yaml_name} - --sr_image_shape=${image_shape} - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -123,7 +122,6 @@ predict: - --sr_model_dir="./models_inference/"${qa_yaml_name} - --sr_image_shape=${image_shape} - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - name: pretrained @@ -133,7 +131,6 @@ predict: - --sr_model_dir="./models_inference/"${model} - --sr_image_shape=${image_shape} - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: pretrained_mkldnn @@ -143,5 +140,4 @@ predict: - --sr_model_dir="./models_inference/"${model} - --sr_image_shape=${image_shape} - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True diff --git a/models_restruct/PaddleOCR/base/ocr_table_base.yaml b/models_restruct/PaddleOCR/base/ocr_table_base.yaml index ac2e3838ad..161ee918e8 100755 --- a/models_restruct/PaddleOCR/base/ocr_table_base.yaml +++ b/models_restruct/PaddleOCR/base/ocr_table_base.yaml @@ -143,7 +143,6 @@ predict: - --table_max_len=488 - --vis_font_path=./doc/fonts/simfang.ttf - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -155,5 +154,4 @@ predict: - --table_max_len=488 - --vis_font_path=./doc/fonts/simfang.ttf - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True diff --git a/models_restruct/PaddleOCR/base/ocr_table_base_pretrained.yaml b/models_restruct/PaddleOCR/base/ocr_table_base_pretrained.yaml index 088f866ea7..5a9a62cce6 100755 --- a/models_restruct/PaddleOCR/base/ocr_table_base_pretrained.yaml +++ b/models_restruct/PaddleOCR/base/ocr_table_base_pretrained.yaml @@ -172,7 +172,6 @@ predict: - --table_max_len=488 - --vis_font_path=./doc/fonts/simfang.ttf - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: pretrained_mkldnn @@ -184,7 +183,6 @@ predict: - --table_max_len=488 - --vis_font_path=./doc/fonts/simfang.ttf - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - name: trained @@ -196,7 +194,6 @@ predict: - --table_max_len=488 - --vis_font_path=./doc/fonts/simfang.ttf - --use_gpu=${use_gpu} - - --use_tensorrt=False - --enable_mkldnn=False - name: trained_mkldnn @@ -208,7 +205,6 @@ predict: - --table_max_len=488 - --vis_font_path=./doc/fonts/simfang.ttf - --use_gpu=False - - --use_tensorrt=False - --enable_mkldnn=True - name: pretrained_C_plus_plus_GPU diff --git a/models_restruct/PaddleOCR/diy_build/PaddleOCR_Build.py b/models_restruct/PaddleOCR/diy_build/PaddleOCR_Build.py index d5e9aec40f..b3d2c13c47 100755 --- a/models_restruct/PaddleOCR/diy_build/PaddleOCR_Build.py +++ b/models_restruct/PaddleOCR/diy_build/PaddleOCR_Build.py @@ -200,7 +200,6 @@ def compile_c_predict_demo(self): LIB_DIR = os.environ.get("paddle_inference_LIB_DIR") CUDA_LIB_DIR = os.environ.get("CUDA_LIB_DI") CUDNN_LIB_DIR = os.environ.get("CUDNN_LIB_DIR") - TENSORRT_DIR = os.environ.get("TENSORRT_DIR") if os.path.exists("build"): shutil.rmtree("build") @@ -208,9 +207,9 @@ def compile_c_predict_demo(self): os.chdir("build") print(os.getcwd()) cmd = ( - "cmake .. -DPADDLE_LIB=%s -DWITH_MKL=ON -DWITH_GPU=OFF -DWITH_STATIC_LIB=OFF -DWITH_TENSORRT=OFF \ - -DOPENCV_DIR=%s -DCUDNN_LIB=%s -DCUDA_LIB=%s -DTENSORRT_DIR=%s" - % (LIB_DIR, OPENCV_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR, TENSORRT_DIR) + "cmake .. -DPADDLE_LIB=%s -DWITH_MKL=ON -DWITH_GPU=OFF -DWITH_STATIC_LIB=OFF \ + -DOPENCV_DIR=%s -DCUDNN_LIB=%s -DCUDA_LIB=%s" + % (LIB_DIR, OPENCV_DIR, CUDNN_LIB_DIR, CUDA_LIB_DIR) ) print(cmd) repo_result = subprocess.getstatusoutput(cmd) diff --git a/models_restruct/PaddleScience/tools/get_result.py b/models_restruct/PaddleScience/tools/get_result.py index 758e985c40..d7e90ce36b 100644 --- a/models_restruct/PaddleScience/tools/get_result.py +++ b/models_restruct/PaddleScience/tools/get_result.py @@ -327,7 +327,6 @@ def update_kpi(self): and ( tag_value["name"] == "trained" or tag_value["name"] == "trained_mkldnn" - or tag_value["name"] == "trained_trt" ) ): try: # 增加尝试方式报错，定死指标为class_ids 变成退出码 exit_code diff --git a/models_restruct/deepxde/tools/get_result.py b/models_restruct/deepxde/tools/get_result.py index 758e985c40..d7e90ce36b 100644 --- a/models_restruct/deepxde/tools/get_result.py +++ b/models_restruct/deepxde/tools/get_result.py @@ -327,7 +327,6 @@ def update_kpi(self): and ( tag_value["name"] == "trained" or tag_value["name"] == "trained_mkldnn" - or tag_value["name"] == "trained_trt" ) ): try: # 增加尝试方式报错，定死指标为class_ids 变成退出码 exit_code diff --git a/models_restruct/models_env/linux_env.sh b/models_restruct/models_env/linux_env.sh index 6dcc389efb..9d1c2ae2ab 100644 --- a/models_restruct/models_env/linux_env.sh +++ b/models_restruct/models_env/linux_env.sh @@ -105,11 +105,9 @@ if [[ ${AGILE_PIPELINE_NAME} =~ "Cuda102" ]] && [[ ${AGILE_PIPELINE_NAME} =~ "Py if [[ ${AGILE_PIPELINE_NAME} =~ "Develop" ]];then export paddle_whl=${paddle_whl:-"https://paddle-wheel.bj.bcebos.com/develop/linux/linux-gpu-cuda10.2-cudnn7-mkl-gcc8.2-avx/paddlepaddle_gpu-0.0.0.post102-cp36-cp36m-linux_x86_64.whl"} export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-GpuAll-Centos-Gcc82-Cuda102-Cudnn81-Trt7234-Py38-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-7.1.3.4"} else export paddle_whl=${paddle_whl:-"https://paddle-wheel.bj.bcebos.com/develop/linux/linux-gpu-cuda10.2-cudnn7-mkl-gcc8.2-avx/paddlepaddle_gpu-0.0.0.post102-cp36-cp36m-linux_x86_64.whl"} export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release-GpuAll-Centos-Gcc82-Cuda102-Cudnn81-Trt7234-Py38-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-7.1.3.4"} fi elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda102" ]] && [[ ${AGILE_PIPELINE_NAME} =~ "Python37" ]];then @@ -117,7 +115,6 @@ elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda102" ]] && [[ ${AGILE_PIPELINE_NAME} =~ " linux_env_info_main get_wheel_url Cuda102 Python37 Develop # cudnn7用的是cudnn8的预测库的包 export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-GpuAll-Centos-Gcc82-Cuda102-Cudnn81-Trt7234-Py38-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-7.0.0.11"} #230223 stride test # export paddle_whl=${paddle_whl:-"https://paddle-qa.bj.bcebos.com/xieyunshen/TempPRBuild/50444/paddlepaddle_gpu-0.0.0-cp37-cp37m-linux_x86_64.whl"} @@ -125,86 +122,55 @@ elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda102" ]] && [[ ${AGILE_PIPELINE_NAME} =~ " linux_env_info_main get_wheel_url Cuda102 Python37 Release # cudnn7用的是cudnn8的预测库的包 export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release-GpuAll-Centos-Gcc82-Cuda102-Cudnn81-Trt7234-Py38-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-7.0.0.11"} fi elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda112" ]] && [[ ${AGILE_PIPELINE_NAME} =~ "Python38" ]];then if [[ ${AGILE_PIPELINE_NAME} =~ "Develop" ]];then linux_env_info_main get_wheel_url Cuda112 Python38 Develop ON - # export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-GpuAll-Centos-Gcc82-Cuda112-Cudnn82-Trt8034-Py38-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.0.3.4"} else linux_env_info_main get_wheel_url Cuda112 Python38 Release ON - # export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release-GpuAll-Centos-Gcc82-Cuda112-Cudnn82-Trt8034-Py38-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.0.3.4"} fi elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda112" ]] && [[ ${AGILE_PIPELINE_NAME} =~ "Python311" ]];then if [[ ${AGILE_PIPELINE_NAME} =~ "Develop" ]];then linux_env_info_main get_wheel_url Cuda112 Python311 Develop ON - # export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-GpuAll-Centos-Gcc82-Cuda112-Cudnn82-Trt8034-Py38-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.0.3.4"} else linux_env_info_main get_wheel_url Cuda112 Python311 Release ON - # export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release-GpuAll-Centos-Gcc82-Cuda112-Cudnn82-Trt8034-Py38-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.0.3.4"} fi elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda112" ]] && [[ ${AGILE_PIPELINE_NAME} =~ "Python312" ]];then if [[ ${AGILE_PIPELINE_NAME} =~ "Develop" ]];then linux_env_info_main get_wheel_url Cuda112 Python312 Develop ON - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.0.3.4"} else linux_env_info_main get_wheel_url Cuda112 Python312 Release ON - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.0.3.4"} fi elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda116" ]] && [[ ${AGILE_PIPELINE_NAME} =~ "Python39" ]];then if [[ ${AGILE_PIPELINE_NAME} =~ "Develop" ]];then linux_env_info_main get_wheel_url Cuda116 Python39 Develop ON - # export paddle_inference=${paddle_inference:-"https://paddle-inference-lib.bj.bcebos.com/develop/cxx_c/Linux/GPU/x86-64_gcc8.2_avx_mkl_cuda11.6_cudnn8.4.0-trt8.4.0.6/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.4.0.6"} else linux_env_info_main get_wheel_url Cuda116 Python39 Release ON - # export paddle_inference=${paddle_inference:-"https://paddle-inference-lib.bj.bcebos.com/release/2.5/cxx_c/Linux/GPU/x86-64_gcc8.2_avx_mkl_cuda11.6_cudnn8.4.0-trt8.4.0.6/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.4.0.6"} fi elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda116" ]] && [[ ${AGILE_PIPELINE_NAME} =~ "Python310" ]];then if [[ ${AGILE_PIPELINE_NAME} =~ "Develop" ]];then linux_env_info_main get_wheel_url_covall Cuda116 Python310 Develop ON - # export paddle_inference=${paddle_inference:-"https://paddle-inference-lib.bj.bcebos.com/develop/cxx_c/Linux/GPU/x86-64_gcc8.2_avx_mkl_cuda11.6_cudnn8.4.0-trt8.4.0.6/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.4.0.6"} else linux_env_info_main get_wheel_url_covall Cuda116 Python310 Release ON - # export paddle_inference=${paddle_inference:-"https://paddle-inference-lib.bj.bcebos.com/release/2.5/cxx_c/Linux/GPU/x86-64_gcc8.2_avx_mkl_cuda11.6_cudnn8.4.0-trt8.4.0.6/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.4.0.6"} fi elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda117" ]] && [[ ${AGILE_PIPELINE_NAME} =~ "Python310" ]];then if [[ ${AGILE_PIPELINE_NAME} =~ "Develop" ]];then linux_env_info_main get_wheel_url Cuda117 Python310 Develop ON - # export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-GpuAll-LinuxCentos-Gcc82-Cuda117-Cudnn84-Trt84-Py39-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.4.2.4"} else linux_env_info_main get_wheel_url Cuda117 Python310 Release ON - # export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release-GpuAll-LinuxCentos-Gcc82-Cuda117-Cudnn84-Trt84-Py39-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.4.2.4"} fi elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda118" ]] && [[ ${AGILE_PIPELINE_NAME} =~ "Python310" ]];then if [[ ${AGILE_PIPELINE_NAME} =~ "Develop" ]];then linux_env_info_main get_wheel_url Cuda118 Python310 Develop ON - # export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-GpuAll-LinuxCentos-Gcc82-Cuda117-Cudnn84-Trt84-Py39-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.5.1.7"} else linux_env_info_main get_wheel_url Cuda118 Python310 Release ON - # export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release-GpuAll-LinuxCentos-Gcc82-Cuda117-Cudnn84-Trt84-Py39-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.5.1.7"} fi elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda120" ]] && [[ ${AGILE_PIPELINE_NAME} =~ "Python311" ]];then if [[ ${AGILE_PIPELINE_NAME} =~ "Develop" ]];then linux_env_info_main get_wheel_url Cuda120 Python311 Develop ON - # export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-GpuAll-LinuxCentos-Gcc82-Cuda117-Cudnn84-Trt84-Py39-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.6.1.6"} else linux_env_info_main get_wheel_url Cuda120 Python311 Release ON - # export paddle_inference=${paddle_inference:-"https://paddle-qa.bj.bcebos.com/paddle-pipeline/Release-GpuAll-LinuxCentos-Gcc82-Cuda117-Cudnn84-Trt84-Py39-Compile/latest/paddle_inference.tgz"} - export TENSORRT_DIR=${TENSORRT_DIR:-"/usr/local/TensorRT-8.6.1.6"} fi else if [[ ${paddle_whl} ]];then @@ -312,7 +278,6 @@ echo "@@@use_data_cfs: ${use_data_cfs}" echo "@@@plot: ${plot}" echo "@@@c_plus_plus_predict: ${c_plus_plus_predict}" echo "@@@paddle_inference: ${paddle_inference}" -echo "@@@TENSORRT_DIR: ${TENSORRT_DIR}" echo "@@@FLAGS_use_cinn: ${FLAGS_use_cinn}" echo "@@@FLAGS_prim_all: ${FLAGS_prim_all}" @@ -455,7 +420,6 @@ if [[ "${docker_flag}" == "" ]]; then -e get_repo=${get_repo} \ -e paddle_whl=${paddle_whl} \ -e paddle_inference=${paddle_inference} \ - -e TENSORRT_DIR=${TENSORRT_DIR} \ -e dataset_org=${dataset_org} \ -e dataset_target=${dataset_target} \ -e set_cuda=${set_cuda} \ @@ -602,7 +566,7 @@ if [[ "${docker_flag}" == "" ]]; then python -m pip install --ignore-installed six -i https://pypi.tuna.tsinghua.edu.cn/simple python -m pip uninstall -y opencv-python python -m pip config list - python main.py --models_list=${models_list:-None} --models_file=${models_file:-None} --system=${system:-linux} --step=${step:-train} --reponame=${reponame:-PaddleClas} --mode=${mode:-function} --use_build=${use_build:-yes} --branch=${branch:-develop} --get_repo=${get_repo:-wget} --paddle_whl=${paddle_whl:-None} --dataset_org=${dataset_org:-None} --dataset_target=${dataset_target:-None} --set_cuda=${set_cuda:-0,1} --timeout=${timeout:-3600} --binary_search_flag=${binary_search_flag:-False} --use_data_cfs=${use_data_cfs:-False} --plot=${plot:-False} --c_plus_plus_predict=${c_plus_plus_predict:-False} --paddle_inference=${paddle_inference:-None} --TENSORRT_DIR=${TENSORRT_DIR:-None} --PaddleX=${PaddleX:-None} + python main.py --models_list=${models_list:-None} --models_file=${models_file:-None} --system=${system:-linux} --step=${step:-train} --reponame=${reponame:-PaddleClas} --mode=${mode:-function} --use_build=${use_build:-yes} --branch=${branch:-develop} --get_repo=${get_repo:-wget} --paddle_whl=${paddle_whl:-None} --dataset_org=${dataset_org:-None} --dataset_target=${dataset_target:-None} --set_cuda=${set_cuda:-0,1} --timeout=${timeout:-3600} --binary_search_flag=${binary_search_flag:-False} --use_data_cfs=${use_data_cfs:-False} --plot=${plot:-False} --c_plus_plus_predict=${c_plus_plus_predict:-False} --paddle_inference=${paddle_inference:-None} --PaddleX=${PaddleX:-None} ' & wait $! docker start ce_${AGILE_PIPELINE_NAME}_${AGILE_JOB_BUILD_ID} @@ -730,5 +694,5 @@ else python -m pip install --ignore-installed six -i https://mirror.baidu.com/pypi/simple python -m pip uninstall -y opencv-python python -m pip config list - python main.py --models_list=${models_list:-None} --models_file=${models_file:-None} --system=${system:-linux} --step=${step:-train} --reponame=${reponame:-PaddleClas} --mode=${mode:-function} --use_build=${use_build:-yes} --branch=${branch:-develop} --get_repo=${get_repo:-wget} --paddle_whl=${paddle_whl:-None} --dataset_org=${dataset_org:-None} --dataset_target=${dataset_target:-None} --set_cuda=${set_cuda:-0,1} --timeout=${timeout:-3600} --binary_search_flag=${binary_search_flag:-False} --use_data_cfs=${use_data_cfs:-False} --plot=${plot:-False} --c_plus_plus_predict=${c_plus_plus_predict:-False} --paddle_inference=${paddle_inference:-None} --TENSORRT_DIR=${TENSORRT_DIR:-None} --PaddleX=${PaddleX:-None} + python main.py --models_list=${models_list:-None} --models_file=${models_file:-None} --system=${system:-linux} --step=${step:-train} --reponame=${reponame:-PaddleClas} --mode=${mode:-function} --use_build=${use_build:-yes} --branch=${branch:-develop} --get_repo=${get_repo:-wget} --paddle_whl=${paddle_whl:-None} --dataset_org=${dataset_org:-None} --dataset_target=${dataset_target:-None} --set_cuda=${set_cuda:-0,1} --timeout=${timeout:-3600} --binary_search_flag=${binary_search_flag:-False} --use_data_cfs=${use_data_cfs:-False} --plot=${plot:-False} --c_plus_plus_predict=${c_plus_plus_predict:-False} --paddle_inference=${paddle_inference:-None} --PaddleX=${PaddleX:-None} fi diff --git a/models_restruct/models_env/models_docker_build.sh b/models_restruct/models_env/models_docker_build.sh index 51a863b410..a34838f705 100644 --- a/models_restruct/models_env/models_docker_build.sh +++ b/models_restruct/models_env/models_docker_build.sh @@ -12,7 +12,6 @@ if [[ ${AGILE_PIPELINE_NAME} =~ "Cuda102" ]];then linux_env_info_main get_docker_images Centos Cuda102 else linux_env_info_main get_docker_images Ubuntu Cuda102 - #230320 change registry.baidubce.com/paddlepaddle/paddle:latest-gpu-cuda10.2-cudnn7-dev for add trt fi elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda112" ]];then if [[ ${AGILE_PIPELINE_NAME} =~ "Centos" ]];then @@ -45,7 +44,7 @@ elif [[ ${AGILE_PIPELINE_NAME} =~ "Cuda120" ]];then linux_env_info_main get_docker_images Ubuntu Cuda120 fi else - Image_version="registry.baidubce.com/paddlepaddle/paddleqa:latest-dev-cuda10.2-cudnn7.6-trt7.0-gcc8.2" + Image_version="registry.baidubce.com/paddlepaddle/paddle:latest-gpu-cuda10.2-cudnn7-dev" fi echo "Image_version: ${Image_version}" diff --git a/models_restruct/models_env/windows_env.bat b/models_restruct/models_env/windows_env.bat index 960784d291..30236d8985 100644 --- a/models_restruct/models_env/windows_env.bat +++ b/models_restruct/models_env/windows_env.bat @@ -97,7 +97,7 @@ if %errorlevel% equ 0 ( ) rem set path -set "PATH=C:\Program Files\Git\bin;C:\Program Files\Git\cmd;C:\Windows\System32;C:\Windows\SysWOW64;C:\zip_unzip;D:\TensorRT-8.4.1.5\lib;%PATH%" +set "PATH=C:\Program Files\Git\bin;C:\Program Files\Git\cmd;C:\Windows\System32;C:\Windows\SysWOW64;C:\zip_unzip;%PATH%" rem cuda_version echo %AGILE_PIPELINE_NAME% | findstr "Cuda102" >nul diff --git a/tools/linux_env_info.sh b/tools/linux_env_info.sh index a4a936f60b..15f00768b5 100644 --- a/tools/linux_env_info.sh +++ b/tools/linux_env_info.sh @@ -36,91 +36,78 @@ function DockerImages () { export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda10.2-cudnn7.6-trt7.0-gcc8.2" export env_cuda_version="10.2" export env_cudnn_version="7.6.5" - export env_trt_version="7.0.0.11" ;; "Cuda112") echo "Selected Centos: Cuda112" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda11.2-cudnn8.2-trt8.0-gcc82" export env_cuda_version="11.2" export env_cudnn_version="8.2.1" - export env_trt_version="8.0.3.4" ;; "Cuda116") echo "Selected Centos: Cuda116" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda11.6-cudnn8.4-trt8.4-gcc8.2" export env_cuda_version="11.6" export env_cudnn_version="8.4.0" - export env_trt_version="8.4.0.6" ;; "Cuda117") echo "Selected Centos: Cuda117" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda11.7-cudnn8.4-trt8.4-gcc8.2" export env_cuda_version="11.7" export env_cudnn_version="8.4.1" - export env_trt_version="8.4.2.4" ;; "Cuda118") echo "Selected Almalinux: Cuda118" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda11.8-cudnn8.9-trt8.6-gcc11" export env_cuda_version="11.8" export env_cudnn_version="8.9.7" - export env_trt_version="8.6.1.6" ;; "Cuda120") echo "Selected Centos: Cuda120" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda12.0-cudnn8.9-trt8.6-gcc12.2" export env_cuda_version="12.0" export env_cudnn_version="8.9.1" - export env_trt_version="8.6.1.6" ;; "Cuda123") echo "Selected Centos: Cuda123" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda12.3-cudnn9.1-trt10.5-gcc11" export env_cuda_version="12.3" export env_cudnn_version="9.1.0" - export env_trt_version="10.5.0.18" ;; "Cuda124") echo "Selected Almalinux: Cuda124" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda12.4-cudnn9.1-trt10.5-gcc11" export env_cuda_version="12.4" export env_cudnn_version="9.1.1" - export env_trt_version="10.5.0.18" ;; "Cuda126") echo "Selected Almalinux: Cuda126" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda12.6-cudnn9.5-trt10.5-gcc11" export env_cuda_version="12.6" export env_cudnn_version="9.5.1" - export env_trt_version="10.5.0.18" ;; "Cuda128") echo "Selected Almalinux: Cuda128" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda12.8-cudnn9.7-trt10.5-gcc11" export env_cuda_version="12.8" export env_cudnn_version="9.7.0" - export env_trt_version="10.5.0.18" ;; "Cuda129") echo "Selected Almalinux: Cuda129" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda12.9-cudnn9.9-trt10.5-gcc11" export env_cuda_version="12.9" export env_cudnn_version="9.9.0" - export env_trt_version="10.5.0.18" ;; "Cuda130") echo "Selected Almalinux: Cuda130" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda13.0-cudnn9.13-trt10.13-gcc11" export env_cuda_version="13.0" export env_cudnn_version="9.13.0" - export env_trt_version="10.13.3.9" ;; "Cuda132") echo "Selected Almalinux: Cuda132" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle_manylinux_devel:cuda13.2-cudnn9.20-trt10.16-gcc11" export env_cuda_version="13.2" export env_cudnn_version="9.20.0" - export env_trt_version="10.16.1.14" ;; *) DOCKER_EXIT_CODE=101 @@ -128,40 +115,29 @@ function DockerImages () { esac elif [[ "${docker_type}" == "UbuntuTiny" ]];then case ${cuda_version} in - "Cuda118TRT") - echo "Selected Ubuntu: Cuda118 With TensorRT" - export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:cuda118-dev-trt8.6" - export env_cuda_version="11.8" - export env_cudnn_version="8.9.6" - export env_trt_version="8.6.1.6" - ;; "Cuda118") echo "Selected Ubuntu: Cuda118" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:cuda118-dev" export env_cuda_version="11.8" export env_cudnn_version="8.9.6" - export env_trt_version="" ;; "Cuda126") echo "Selected Ubuntu: Cuda126" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:cuda126-dev" export env_cuda_version="12.6" export env_cudnn_version="9.3.0" - export env_trt_version="" ;; "Cuda129") echo "Selected Ubuntu: Cuda129" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:cuda129-dev" export env_cuda_version="12.9" export env_cudnn_version="9.9.0" - export env_trt_version="" ;; "Cuda130") echo "Selected Ubuntu: Cuda130" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:cuda130-dev" export env_cuda_version="13.0" export env_cudnn_version="9.13.0" - export env_trt_version="" ;; *) DOCKER_EXIT_CODE=101 @@ -175,70 +151,60 @@ function DockerImages () { export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:latest-dev-cuda10.2-cudnn7.6-trt7.0-gcc8.2" export env_cuda_version="10.2" export env_cudnn_version="7.6.5" - export env_trt_version="7.0.0.11" ;; "Cuda112") echo "Selected Ubuntu: Cuda112" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:latest-dev-cuda11.2-cudnn8.2-trt8.0-gcc82" export env_cuda_version="11.2" export env_cudnn_version="8.2.1" - export env_trt_version="8.0.3.4" ;; "Cuda116") echo "Selected Ubuntu: Cuda116" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:latest-dev-cuda11.6-cudnn8.4-trt8.4-gcc82" export env_cuda_version="11.6" export env_cudnn_version="8.4.0" - export env_trt_version="8.4.0.6" ;; "Cuda117") echo "Selected Ubuntu: Cuda117" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:latest-dev-cuda11.7-cudnn8.4-trt8.4-gcc82" export env_cuda_version="11.7" export env_cudnn_version="8.4.1" - export env_trt_version="8.4.2.4" ;; "Cuda118") echo "Selected Ubuntu: Cuda118" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:latest-dev-cuda11.8-cudnn8.6-trt8.5-gcc82" export env_cuda_version="11.8" export env_cudnn_version="8.6.0" - export env_trt_version="8.5.3.1" ;; "Cuda120") echo "Selected Ubuntu: Cuda120" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:latest-dev-cuda12.0-cudnn8.9-trt8.6-gcc12.2" export env_cuda_version="12.0" export env_cudnn_version="8.9.1" - export env_trt_version="8.6.1.6" ;; "Cuda123") echo "Selected Ubuntu: Cuda123" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:latest-dev-cuda12.3-cudnn9.0-trt8.6-gcc12.2" export env_cuda_version="12.3" export env_cudnn_version="9.0.0" - export env_trt_version="8.6.1.6" ;; "Cuda126") echo "Selected Ubuntu: Cuda126" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/latest-dev-cuda12.6-cudnn9.5-trt10.5.0.18-ubuntu24:latest" export env_cuda_version="12.6" export env_cudnn_version="9.5.1" - export env_trt_version="10.5.0.18" ;; "Cuda128") echo "Selected Ubuntu: Cuda128" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/latest-dev-cuda12.8-cudnn9.7-trt10.5-gcc12:latest" export env_cuda_version="12.8" export env_cudnn_version="9.7.0" - export env_trt_version="10.5.0.18" ;; "Cuda129") echo "Selected Ubuntu: Cuda129" export Image_version="ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:latest-dev-cuda12.9-cudnn9.9-trt10.5-gcc13.3" export env_cuda_version="12.9" export env_cudnn_version="9.9.0" - export env_trt_version="10.5.0.18" ;; *) DOCKER_EXIT_CODE=101 From 068e7ff97b2e704a85c91759c3d98f9ce71b1155 Mon Sep 17 00:00:00 2001 From: gouzi <530971494@qq.com> Date: Sun, 19 Jul 2026 15:06:51 +0800 Subject: [PATCH 3/5] validate int8 deploy backend choices --- inference/python_api_test/test_int8_model/test_bert_infer.py | 1 + .../test_int8_model/test_image_classification_infer.py | 1 + inference/python_api_test/test_int8_model/test_nlp_infer.py | 1 + inference/python_api_test/test_int8_model/test_ppyoloe_infer.py | 1 + .../python_api_test/test_int8_model/test_segmentation_infer.py | 1 + .../python_api_test/test_int8_model/test_yolo_series_infer.py | 1 + 6 files changed, 6 insertions(+) diff --git a/inference/python_api_test/test_int8_model/test_bert_infer.py b/inference/python_api_test/test_int8_model/test_bert_infer.py index 5a1d4d4fde..50ce28d8b3 100644 --- a/inference/python_api_test/test_int8_model/test_bert_infer.py +++ b/inference/python_api_test/test_int8_model/test_bert_infer.py @@ -122,6 +122,7 @@ def argsparser(): "--deploy_backend", type=str, default="paddle_inference", + choices=["paddle_inference", "onnxruntime"], help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) parser.add_argument("--model_name", type=str, default="", help="model_name for benchmark") diff --git a/inference/python_api_test/test_int8_model/test_image_classification_infer.py b/inference/python_api_test/test_int8_model/test_image_classification_infer.py index bcd58cde01..894fc6bb3d 100644 --- a/inference/python_api_test/test_int8_model/test_image_classification_infer.py +++ b/inference/python_api_test/test_int8_model/test_image_classification_infer.py @@ -59,6 +59,7 @@ def argsparser(): "--deploy_backend", type=str, default="paddle_inference", + choices=["paddle_inference", "onnxruntime"], help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) parser.add_argument( diff --git a/inference/python_api_test/test_int8_model/test_nlp_infer.py b/inference/python_api_test/test_int8_model/test_nlp_infer.py index 1a339af846..64d93c7ff7 100644 --- a/inference/python_api_test/test_int8_model/test_nlp_infer.py +++ b/inference/python_api_test/test_int8_model/test_nlp_infer.py @@ -114,6 +114,7 @@ def argsparser(): "--deploy_backend", type=str, default="paddle_inference", + choices=["paddle_inference", "onnxruntime"], help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) parser.add_argument("--model_name", type=str, default="", help="model_name for benchmark") diff --git a/inference/python_api_test/test_int8_model/test_ppyoloe_infer.py b/inference/python_api_test/test_int8_model/test_ppyoloe_infer.py index 0b49ed3869..3c1f243b36 100644 --- a/inference/python_api_test/test_int8_model/test_ppyoloe_infer.py +++ b/inference/python_api_test/test_int8_model/test_ppyoloe_infer.py @@ -43,6 +43,7 @@ def argsparser(): "--deploy_backend", type=str, default="paddle_inference", + choices=["paddle_inference", "onnxruntime"], help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) parser.add_argument( diff --git a/inference/python_api_test/test_int8_model/test_segmentation_infer.py b/inference/python_api_test/test_int8_model/test_segmentation_infer.py index 9d7486f24d..c9f8f38347 100644 --- a/inference/python_api_test/test_int8_model/test_segmentation_infer.py +++ b/inference/python_api_test/test_int8_model/test_segmentation_infer.py @@ -47,6 +47,7 @@ def argsparser(): "--deploy_backend", type=str, default="paddle_inference", + choices=["paddle_inference", "onnxruntime"], help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) parser.add_argument("--dataset_config", type=str, default=None, help="path of dataset config.") diff --git a/inference/python_api_test/test_int8_model/test_yolo_series_infer.py b/inference/python_api_test/test_int8_model/test_yolo_series_infer.py index c3dbcb6247..8a5eb8e443 100644 --- a/inference/python_api_test/test_int8_model/test_yolo_series_infer.py +++ b/inference/python_api_test/test_int8_model/test_yolo_series_infer.py @@ -44,6 +44,7 @@ def argsparser(): "--deploy_backend", type=str, default="paddle_inference", + choices=["paddle_inference", "onnxruntime"], help="deploy backend, it can be: `paddle_inference`, `onnxruntime`", ) parser.add_argument("--use_l3", type=bool, default=False, help="Whether use L3_cache or not.") From 6e2a8f183ee7afc6ee80b9d47049f1908ea0c29d Mon Sep 17 00:00:00 2001 From: gouzi <530971494@qq.com> Date: Sun, 19 Jul 2026 22:24:40 +0800 Subject: [PATCH 4/5] restore OCR GPU predictor coverage --- models/AutomaticTestSystem/test_ocr_acc.py | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+) diff --git a/models/AutomaticTestSystem/test_ocr_acc.py b/models/AutomaticTestSystem/test_ocr_acc.py index 4b29a7833c..ecfd17bca8 100644 --- a/models/AutomaticTestSystem/test_ocr_acc.py +++ b/models/AutomaticTestSystem/test_ocr_acc.py @@ -177,6 +177,26 @@ def test_ocr_accuracy_predict_mkl(yml_name, enable_mkldnn): model.test_ocr_rec_predict(False, enable_mkldnn) +@allure.story("predict") +@pytest.mark.parametrize("yml_name", get_model_list()) +def test_ocr_accuracy_predict_gpu(yml_name): + """ + test_ocr_accuracy_predict_gpu + """ + model_name = os.path.splitext(os.path.basename(yml_name))[0] + allure.dynamic.title(model_name + "_GPU_predict") + allure.dynamic.description("预测库预测") + + if model_name == "re_vi_layoutxlm_xfund_zh": + pytest.skip("not supported") + + r = re.search("/(.*)/", yml_name) + category = r.group(1) + print(category) + model = TestOcrModelFunction(model=model_name, yml=yml_name, category=category) + model.test_ocr_rec_predict(True, False) + + def test_ocr_accuracy_predict_recovery(): """ test_ocr_accuracy_predict_recovery From 00d233c5d532314866b2ab1291fb75cb822469ed Mon Sep 17 00:00:00 2001 From: gouzi <530971494@qq.com> Date: Sun, 19 Jul 2026 22:49:31 +0800 Subject: [PATCH 5/5] restore Serving use_trt default --- inference/serving_api_test/paddle_serving_server/util.py | 1 + 1 file changed, 1 insertion(+) diff --git a/inference/serving_api_test/paddle_serving_server/util.py b/inference/serving_api_test/paddle_serving_server/util.py index 0d4511425b..cc0d4b2c6a 100644 --- a/inference/serving_api_test/paddle_serving_server/util.py +++ b/inference/serving_api_test/paddle_serving_server/util.py @@ -67,6 +67,7 @@ def default_args(): args.max_body_size = 512 * 1024 * 1024 args.use_encryption_model = False args.use_multilang = False + args.use_trt = False args.use_lite = False args.use_xpu = False args.product_name = None