Skip to content

Pull requests: huggingface/candle

New pull request New
273 Open 2,291 Closed
273 Open 2,291 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

doc: yolo26-rs external
#3677 opened Jun 28, 2026 by wzh19960613 Loading…
docs: yolo26-wasm demo
#3676 opened Jun 28, 2026 by wzh19960613 Loading…
Support ONNX DynamicQuantizeLinear
#3675 opened Jun 27, 2026 by jesco-absolut Loading…
1
Support ONNX Mod
#3674 opened Jun 27, 2026 by jesco-absolut Loading…
1
Support ONNX MaxPool padding
#3673 opened Jun 27, 2026 by jesco-absolut Loading…
2
Support BoolStorage in pickle loader
#3672 opened Jun 27, 2026 by jesco-absolut Loading…
1
Support linear Resize in candle-onnx
#3671 opened Jun 27, 2026 by jesco-absolut Loading…
1
Support llama GGUF tokenizers without explicit merges
#3670 opened Jun 27, 2026 by jesco-absolut Loading…
1
Add CUDA Graph capture for decode + FlashInfer-style decode-attention backend (CUDA/Metal/CPU) (#3651)
#3669 opened Jun 27, 2026 by astorise Loading…
1
cpu-optimization: wire f16kv changes to Qwen3 (and remove feature toggles for PR 3664-3667 27% gain in prefill, 14% gain decode)
#3668 opened Jun 27, 2026 by DrJesseGlass Contributor Loading…
cpu-optimiziation: aarch64 CPU prefill lane=row Q4_K kernel + a packed Q6_K dtype (20% prefill gain at 6 threads and 512 tokens)
#3667 opened Jun 27, 2026 by DrJesseGlass Contributor Loading…
Fused CPU neox RoPE for decode
#3666 opened Jun 27, 2026 by DrJesseGlass Contributor Loading…
cpu-optimize: add f16 KV cache to the CPU flash-attention path (14% decode gain) vec approx expf (2% decode gain)
#3665 opened Jun 27, 2026 by DrJesseGlass Contributor Loading…
cpu-optimize: Parallelize contiguous f32 elementwise ops over the barrier pool (prefill 8% gain)
#3664 opened Jun 27, 2026 by DrJesseGlass Contributor Loading…
cuda: reject launch configs above u32::MAX
#3663 opened Jun 27, 2026 by ogarciarevett Loading…
1
Add block-wise FP8 quantized linear layer support (#3650)
#3662 opened Jun 27, 2026 by astorise Loading…
5 tasks done
1
Add AWQ quantized linear layer support and unify with GPTQ (#3650)
#3661 opened Jun 27, 2026 by astorise Loading…
6 tasks done
Add GPTQ quantized linear layer support for Qwen2 (#3650)
#3660 opened Jun 27, 2026 by astorise Loading…
6 tasks done
1
Add device-agnostic (CPU/Metal) PagedAttention to complement the CUDA kernels (#3655)
#3657 opened Jun 25, 2026 by astorise Loading…
1
1
candle-onnx: add pipeline-parallel evaluation via simple_eval_with_placement
#3648 opened Jun 25, 2026 by astorise Loading…
Upstream onnx device candle-onnx: propagate device through simple_eval instead of hard-coding CPU
#3647 opened Jun 25, 2026 by astorise Loading…
1
Fix Qwen3 causal mask for batch_size > 1
#3646 opened Jun 24, 2026 by Olcmyk Loading…
1
Fix Metal device creation panic with bounds check (fixes #3566)
#3645 opened Jun 24, 2026 by Olcmyk Loading…
1
Load added tokens from GGUF metadata (fixes missing <think>/</think> on Qwen)
#3641 opened Jun 23, 2026 by jaweed3 Loading…
1
Metal apple silicon fixes
#3640 opened Jun 22, 2026 by tyt2y3 Loading…
ProTip! Mix and match filters to narrow down what you’re looking for.