Skip to content

Actions: eulogik/TinyDoc-VLM

Actions

CI

Actions

Loading...
Loading

Show workflow options

Create status badge

Loading
ci.yml
148 workflow runs
148 workflow runs

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
fix: use Path(output_dir) before 'out' is defined in train() CI #148: Commit 0bdee57 pushed by gautamkishore
3m 0s main
main
3m 0s
fix: resume_ckpt NameError in train() - use output_dir/latest instead CI #147: Commit 4cb34c9 pushed by gautamkishore
2m 43s main
main
2m 43s
fix: save/load optimizer+scheduler+scaler on checkpoint resume; fix -… CI #146: Commit 4b1c707 pushed by gautamkishore
3m 58s main
main
3m 58s
eval kernel: force T4 via machine_shape=NvidiaTeslaT4 (P100 default o… CI #145: Commit b53f8ba pushed by gautamkishore
3m 0s main
main
3m 0s
eval_768: make repo root importable for tinydoc_vlm (fixes Kaggle run) CI #144: Commit 9ad461d pushed by gautamkishore
3m 11s main
main
3m 11s
kaggle notebook: single-GPU by default (DDP SIGSEGVs on T4 x2; kaggle… CI #143: Commit a9271ad pushed by gautamkishore
3m 44s main
main
3m 44s
kaggle: retry HF_TOKEN secrets fetch with backoff; fix token injectio… CI #142: Commit fe6ccff pushed by gautamkishore
2m 44s main
main
2m 44s
run_kaggle.sh: BATCH default 8 -> 2 (matches notebook's memory-safe d… CI #141: Commit fc5297c pushed by gautamkishore
2m 49s main
main
2m 49s
train on T4 x2 via DDP; fresh-start from empty hub checkpoint repo CI #140: Commit 34b4bc6 pushed by gautamkishore
3m 3s main
main
3m 3s
CRITICAL FIX: collate_fn only injected 1 <image> token — model never … CI #139: Commit b23008d pushed by gautamkishore
2m 57s main
main
2m 57s
fix: suppress HTTP logs, fix opt.zero_grad bug, single T4, always re-… CI #138: Commit 8e37b96 pushed by gautamkishore
3m 17s main
main
3m 17s
kaggle: fix torch-pin install (cu124 index dropped cudnn 9.1.0.70) CI #137: Commit 65ab990 pushed by gautamkishore
2m 51s main
main
2m 51s
kaggle: auto-relaunch on signal-death (VM kills trainer every ~40 min) CI #136: Commit b202182 pushed by gautamkishore
3m 2s main
main
3m 2s
kaggle: pin torch 2.6.0+cu124 on T4 (4x SIGSEGV on image torch 2.10) CI #135: Commit eb6ae36 pushed by gautamkishore
3m 1s main
main
3m 1s
kaggle: single-GPU first; DDP opt-in (3x SIGSEGV today) CI #134: Commit 938678b pushed by gautamkishore
2m 45s main
main
2m 45s
kaggle: add single-GPU last-resort fallback after DDP SIGSEGV CI #133: Commit 821507f pushed by gautamkishore
3m 7s main
main
3m 7s
kaggle: remove fp16-only dtype guard (would discard new fp32 checkpoi… CI #132: Commit 9a33567 pushed by gautamkishore
3m 1s main
main
3m 1s
kaggle: self-healing memory plan - grad-ckpt first, batch-1 fallback … CI #131: Commit 989a511 pushed by gautamkishore
2m 45s main
main
2m 45s
kaggle: enable grad-checkpointing (fp32 master weights OOM the 14.5Gi… CI #130: Commit 3824017 pushed by gautamkishore
2m 48s main
main
2m 48s
fix: NaN cascade from half-precision AdamW state (root cause of every… CI #129: Commit 03de742 pushed by gautamkishore
3m 8s main
main
3m 8s
fix: args -> num_workers inside train() function CI #128: Commit 2151a66 pushed by gautamkishore
2m 51s main
main
2m 51s
amp: add --num-workers arg (default 0 for MPS, 2 for CUDA); kaggle us… CI #127: Commit 1316ce4 pushed by gautamkishore
3m 5s main
main
3m 5s
kaggle: disable grad-checkpoint (SIGSEGV trigger on T4) and use num_w… CI #126: Commit 49bd0c4 pushed by gautamkishore
2m 44s main
main
2m 44s
amp: split autocast vs scaler — always autocast on CUDA for numerical… CI #125: Commit 17ad9d9 pushed by gautamkishore
2m 50s main
main
2m 50s
amp: only use GradScaler with fp32 params (fp16/bf16 checkpoints -> f… CI #124: Commit a2682de pushed by gautamkishore
3m 12s main
main
3m 12s