GitHub / huggingface/candle issues and pull requests
#3022 - candle_datasets: Improve from_hub usability by exposing FileReader and cleaning up iterator logic
Issue -
State: open - Opened by xavierforge 15 days ago
#3017 - candle_nn: LayerNorm does not expose `remove_mean` and `eps` parameter
Issue -
State: open - Opened by nemosupremo 23 days ago
#3016 - Build fails on Maxwell GPU due to __dp4a undefined in quantized.cu
Issue -
State: open - Opened by fishonamos 24 days ago
#3015 - Unable to build candle with flash attention on iOS
Issue -
State: open - Opened by jpchen 25 days ago
#3014 - Short context length on Qwen quantized examples.
Issue -
State: open - Opened by AlpineVibrations 28 days ago
- 1 comment
#3013 - Word Timestamp for whisper
Issue -
State: open - Opened by bp7968h about 1 month ago
- 2 comments
#3012 - Fix LayerNorm gradient flow issue
Pull Request -
State: open - Opened by tymat about 1 month ago
- 2 comments
#3011 - LayerNorm Gradient Flow Issue in candle-nn
Issue -
State: open - Opened by tymat about 1 month ago
#3010 - Fix last_hidden_state slicing position in ModernBertForSequenceClassification
Pull Request -
State: open - Opened by whitebox2 about 1 month ago
#3009 - [ Gemma 3n ] - multi modal - small enough to work on a mobile
Issue -
State: closed - Opened by oddpxl about 1 month ago
- 3 comments
#3008 - CI: Fixed Apple Arm Error
Pull Request -
State: closed - Opened by greenrazer about 1 month ago
#3007 - Fixed Failing CI
Pull Request -
State: closed - Opened by greenrazer about 1 month ago
#3006 - fixed errors with hardswish merge
Pull Request -
State: closed - Opened by greenrazer about 1 month ago
#3005 - Qwen3: fix quality loss due to rope freq precision
Pull Request -
State: closed - Opened by zackangelo about 1 month ago
- 1 comment
#3004 - qwen3 q,k tensors not contiguous after rms norm
Issue -
State: closed - Opened by zackangelo about 1 month ago
- 1 comment
#3003 - Build for multiple arch?
Issue -
State: open - Opened by johnnynunez about 1 month ago
#3002 - Is there a roadmap or intention to support CUDA Graph?
Issue -
State: open - Opened by guoqingbao about 1 month ago
- 2 comments
#3001 - Unsupported ONNX operator: DequantizeLinear
Issue -
State: open - Opened by adamreichold about 1 month ago
#3000 - Issue: Inconsistent last_hidden_state slicing position compared to HuggingFace Transformers in ModernBertForSequenceClassification
Issue -
State: open - Opened by whitebox2 about 1 month ago
#2999 - Performance regression after upgrade to cudarc 0.14
Issue -
State: closed - Opened by zackangelo about 1 month ago
- 1 comment
#2998 - Add high-performance GLU activation variants (GLU, GeGLU, ReGLU) with comprehensive benchmarkingAr develop
Pull Request -
State: open - Opened by artem1984A about 1 month ago
#2997 - Implement Conv3D support for compatibility with Qwen-VL and similar models
Issue -
State: open - Opened by maximizemaxwell about 1 month ago
#2996 - Add Idefics3Model for ColSmol
Pull Request -
State: open - Opened by akshayballal95 about 1 month ago
#2995 - Add qwen2.5 vl
Pull Request -
State: open - Opened by maximizemaxwell about 2 months ago
- 9 comments
#2994 - [DRAFT] Add `candle-utils` helper crate for common patterns
Pull Request -
State: open - Opened by matthewhaynesonline about 2 months ago
- 1 comment
#2992 - Support new arch of GLM4 GGUF models
Pull Request -
State: open - Opened by guoqingbao about 2 months ago
#2991 - Support new arch of GLM4 models
Pull Request -
State: closed - Opened by guoqingbao about 2 months ago
- 5 comments
#2990 - question about VarBuilder:from_pth
Issue -
State: open - Opened by jingangdidi about 2 months ago
#2989 - fp8 support
Pull Request -
State: open - Opened by zackangelo about 2 months ago
- 7 comments
#2987 - Fix cuda memory error for Qwen3 non-quantized
Pull Request -
State: closed - Opened by akshayballal95 about 2 months ago
- 1 comment
#2986 - How to reset gradient before each batch
Issue -
State: open - Opened by lokxii about 2 months ago
#2985 - candle_onnx compatible object detection models
Issue -
State: open - Opened by magi-2 about 2 months ago
- 1 comment
#2983 - Add Qwen2.x instruct model including wasm demo
Pull Request -
State: open - Opened by ITHwang about 2 months ago
#2982 - candle-onnx: Implement SoftmaxCrossEntopyLoss operator
Pull Request -
State: open - Opened by Michall00 about 2 months ago
#2981 - candle-onnx: Implement Eye Like operator
Pull Request -
State: open - Opened by Michall00 about 2 months ago
#2980 - candle-onnx: Implement Hard Swish operator
Pull Request -
State: closed - Opened by Michall00 about 2 months ago
- 2 comments
#2979 - candle-onnx: Implement One Hot operator
Pull Request -
State: open - Opened by Michall00 about 2 months ago
#2978 - candle-onnx: Implement Selu operator
Pull Request -
State: open - Opened by Michall00 about 2 months ago
#2977 - No backward pass for `LayerNorm` if tensor is contiguous and remove_mean is true
Issue -
State: open - Opened by toolness 2 months ago
#2976 - Hang in quantized_phi::ModelWeights::forward() with Phi-2 GGUF on CPU (Candle main branch)
Issue -
State: open - Opened by EarthSports 2 months ago
#2975 - candle_onnx: Unimplimented op_types for grounding-dino-tiny
Issue -
State: open - Opened by magi-2 2 months ago
#2974 - Any good first issues a newcomer could tackle?
Issue -
State: open - Opened by Heidar-An 2 months ago
- 3 comments
#2973 - Fused CPU attention kernels (~4x performance increase)
Pull Request -
State: open - Opened by EricLBuehler 2 months ago
- 6 comments
#2972 - feat: enhance linear algebra operations
Pull Request -
State: closed - Opened by ssfdust 2 months ago
- 3 comments
#2971 - Enhance the usability of the tensor struct
Issue -
State: closed - Opened by ssfdust 2 months ago
- 1 comment
#2970 - (hotfix) fix the doc test for indexer#i(&self, (a b):(A, B))
Pull Request -
State: closed - Opened by klion26 2 months ago
- 2 comments
#2969 - Add fine-tuned text classifier to xlm roberta example
Pull Request -
State: closed - Opened by jpe90 2 months ago
- 1 comment
#2968 - Use a tanh activation in the xlm-roberta classification head.
Pull Request -
State: closed - Opened by LaurentMazare 2 months ago
#2967 - Logit Discrepancy Between Candle and PyTorch When Using XLM-RoBERTa Model
Issue -
State: closed - Opened by jpe90 2 months ago
- 2 comments
#2966 - Add support for sm_120
Issue -
State: open - Opened by Wolframko 2 months ago
#2965 - Are there any support for complex number?
Issue -
State: open - Opened by hndrbrm 2 months ago
#2964 - candle-onnx: Implement RNN operator
Pull Request -
State: closed - Opened by BrunoSienkiewicz 2 months ago
- 1 comment
#2963 - allow hooking whisper model cross-attention
Pull Request -
State: closed - Opened by jnnnnn 2 months ago
#2962 - Surprising lack of gradient clipping
Issue -
State: open - Opened by grimtin10 2 months ago
- 1 comment
#2961 - Shape Mismatch in MatMul During Forward Pass of ModernBertForSequenceClassification
Issue -
State: closed - Opened by whitebox2 2 months ago
- 2 comments
#2960 - Proper support for phi-4
Pull Request -
State: closed - Opened by LaurentMazare 2 months ago
#2959 - phi4 example does not work (cli args are swapped, model is not supported)
Issue -
State: open - Opened by mcharytoniuk 2 months ago
- 3 comments
#2958 - Fix typos
Pull Request -
State: closed - Opened by omahs 2 months ago
- 1 comment
#2957 - Use HF Papers
Pull Request -
State: open - Opened by qgallouedec 3 months ago
#2956 - Fixes for clippy 1.87.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2955 - Fix docs quantized qwen3
Pull Request -
State: closed - Opened by maximizemaxwell 3 months ago
- 1 comment
#2954 - Olmo 2 model
Pull Request -
State: closed - Opened by janimo 3 months ago
- 1 comment
#2953 - Optimize Quantized Qwen3 KvCache
Pull Request -
State: closed - Opened by b0r3k 3 months ago
- 1 comment
#2952 - candle-onnx: Implement Trilu and ScatterND ops
Pull Request -
State: closed - Opened by greenrazer 3 months ago
#2951 - Fixed Quantized Qwen3 Model
Pull Request -
State: closed - Opened by nosnakeob 3 months ago
- 11 comments
#2950 - quantized_qwen3 CUDA_ERROR_OUT_OF_MEMORY with Irrespective of Size
Issue -
State: closed - Opened by nosnakeob 3 months ago
- 12 comments
#2949 - Using the `--model` parameter with the quantized-qwen3 example fails
Issue -
State: open - Opened by azw413 3 months ago
- 2 comments
#2948 - Optimized `cumsum` operation
Issue -
State: open - Opened by EricLBuehler 3 months ago
- 1 comment
#2947 - it run in colab cpu very good
Issue -
State: open - Opened by werruww 3 months ago
- 1 comment
#2946 - Add Resize to onnx ops
Pull Request -
State: closed - Opened by greenrazer 3 months ago
#2945 - Operating steps from scratch for beginners?
Issue -
State: open - Opened by Qarqor5555555 3 months ago
#2944 - finetuning yolo 8 candle model
Issue -
State: open - Opened by flutter-painter 3 months ago
#2943 - chore: edited the badge on Discord
Pull Request -
State: open - Opened by sashaphmn 3 months ago
#2942 - Bump cudarc to 0.16.3.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2940 - Indexing with max-value results in zero/no-op.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2939 - Qwen3 quantized implementation
Pull Request -
State: closed - Opened by ljt019 3 months ago
- 16 comments
#2938 - Plans to support AMD
Issue -
State: open - Opened by radudiaconu0 3 months ago
- 3 comments
#2937 - fix kv cache issue with quantized_phi3 implementation
Pull Request -
State: open - Opened by ljt019 3 months ago
#2936 - Add a scattered kv cache.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2935 - Bump the candle version to 0.9.1.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2934 - Add Qwen3 MoE
Pull Request -
State: closed - Opened by greenrazer 3 months ago
- 3 comments
#2933 - SmolDocling model support
Issue -
State: open - Opened by ljt019 3 months ago
#2932 - Add support for Helium-v1.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2931 - Is qwen3 being worked on for candle-transformers?
Issue -
State: closed - Opened by ljt019 3 months ago
- 1 comment
#2930 - Updating `Add qwen3` (PR 2903) to use HF weights
Pull Request -
State: closed - Opened by greenrazer 3 months ago
- 5 comments
#2929 - Switch Tensor::full to return a contiguous tensor.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2928 - Fix the gumbel softmax by casting to f32.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2927 - Optimize Tensor::new when called on nested Vec<..>.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2926 - Support for "unbatched" rope.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2925 - Remove redundant mlx gemm dtype check
Pull Request -
State: closed - Opened by ivarflakstad 3 months ago
- 1 comment
#2924 - Bump the crate version to 0.9.0.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2923 - Add the scatter in place ops.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2922 - Added tracing page to the candle book.
Pull Request -
State: closed - Opened by greenrazer 3 months ago
- 1 comment
#2921 - Add the scatter op.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago
#2920 - candle-onnx: Implement LRN operator
Pull Request -
State: open - Opened by BrunoSienkiewicz 3 months ago
#2919 - candle-onnx: Implement layer normalization operator
Pull Request -
State: open - Opened by BrunoSienkiewicz 3 months ago
- 3 comments
#2918 - Fixed Quantized Gemma3 Model and example
Pull Request -
State: closed - Opened by greenrazer 3 months ago
- 2 comments
#2917 - Fixed Gemma3 model and example
Pull Request -
State: closed - Opened by greenrazer 3 months ago
- 2 comments
#2916 - how to save and load the model
Issue -
State: closed - Opened by liguheng 3 months ago
- 3 comments
#2915 - Update to cudarc 0.16.1.
Pull Request -
State: closed - Opened by LaurentMazare 3 months ago