{"full_name":"NVIDIA/cudnn-frontend","owner":"NVIDIA","name":"cudnn-frontend","shard":"1","url":"https://github.com/NVIDIA/cudnn-frontend","description":"cuDNN Frontend is NVIDIA's modern, open-source entry point to the cuDNN library and a growing collection of high-performance open-source kernels.","homepage":"https://docs.nvidia.com/deeplearning/cudnn/latest/","topics":["attention","blackwell","cuda-kernels","flash-attention","fp8","gemm","gpu","hopper","mixture-of-experts","moe","mxfp8","normalization","nvfp4","sdpa","transformer"],"tag_pages":["attention","blackwell","gpu","mixture-of-experts","moe","transformer"],"language":"Cuda","license":"Apache-2.0","avatar":"https://avatars.githubusercontent.com/u/1728152?v=4","created_at":"2021-01-22T16:16:29Z","pushed_at":"2026-11-15T18:57:55Z","metrics":{"stars":1000,"forks":338,"watchers":18,"issues":441,"prs":2173,"releases":111,"issues_open":139,"issues_closed":302,"prs_open":82,"prs_closed":206,"prs_merged":1885,"discussions":1},"growth":{"day":{"stars":0,"forks":1,"base_day":"2026-11-15"}},"history":{"dates":["2026-11-10","2026-11-11","2026-11-12","2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[996,997,997,998,999,1000,1000],"forks":[335,335,335,336,338,337,338],"watchers":[null,18,18,18,18,18,18],"issues_open":[null,129,131,131,135,135,139],"prs_open":[null,89,78,81,83,83,82]},"boards":{},"same_owner":[{"full_name":"NVIDIA/DALI","name":"DALI","shard":"1","boards":{}},{"full_name":"NVIDIA/Isaac-GR00T","name":"Isaac-GR00T","shard":"1","boards":{}},{"full_name":"NVIDIA/Megatron-LM","name":"Megatron-LM","shard":"1","boards":{}},{"full_name":"NVIDIA/Model-Optimizer","name":"Model-Optimizer","shard":"1","boards":{}},{"full_name":"NVIDIA/NemoClaw","name":"NemoClaw","shard":"1","boards":{}},{"full_name":"NVIDIA/OpenShell","name":"OpenShell","shard":"1","boards":{"monthly":107,"monthly_rate":96}},{"full_name":"NVIDIA/SkillEvaluator","name":"SkillEvaluator","shard":"1","boards":{"daily":18,"weekly":138,"daily_rate":4,"weekly_rate":3,"monthly_rate":21,"daily_forks":18,"daily_watchers":189}},{"full_name":"NVIDIA/SkillSpector","name":"SkillSpector","shard":"1","boards":{"daily":148,"weekly":170,"monthly":91,"monthly_rate":110}},{"full_name":"NVIDIA/TensorRT-LLM","name":"TensorRT-LLM","shard":"1","boards":{}},{"full_name":"NVIDIA/apex","name":"apex","shard":"1","boards":{}},{"full_name":"NVIDIA/cosmos","name":"cosmos","shard":"1","boards":{}},{"full_name":"NVIDIA/cuda-samples","name":"cuda-samples","shard":"1","boards":{}},{"full_name":"NVIDIA/cudf","name":"cudf","shard":"1","boards":{}},{"full_name":"NVIDIA/cutlass","name":"cutlass","shard":"1","boards":{}},{"full_name":"NVIDIA/garak","name":"garak","shard":"1","boards":{}},{"full_name":"NVIDIA/nvidia-docker","name":"nvidia-docker","shard":"1","boards":{}},{"full_name":"NVIDIA/open-gpu-kernel-modules","name":"open-gpu-kernel-modules","shard":"1","boards":{}}],"issues":[{"number":378,"title":"[RFE] 🦗 SDPA Support Matrix: Hopper (SM90) — open-source kernels wanted!","comments":14,"created_at":"2026-07-10T23:22:13Z"},{"number":431,"title":"Move License form MIT to Apache-2","comments":12,"created_at":"2026-07-23T18:17:14Z"},{"number":2284,"title":"norm L1: FP8 batchnorm backward never writes the DX AMAX output on SM90 (cuDNN 9.27)","comments":8,"created_at":"2026-11-05T19:09:06Z"},{"number":422,"title":"[Bug]: Segmentation fault in libcudnn_engines_precompiled.so.9.4.0 when executing fused INT8 convolution with vectCount=4","comments":8,"created_at":"2026-07-22T10:11:57Z"},{"number":377,"title":"[RFE] ⚡ SDPA Support Matrix: Ampere (SM80) — open-source kernels wanted!","comments":8,"created_at":"2026-07-10T23:22:12Z"},{"number":1439,"title":"sdpa fwd (frost): nonpaged THD split rule overcounts CTAs (no GQA packing, ragged max length, one-wave cap); 1.4-3x left unselected","comments":6,"created_at":"2026-10-07T07:14:00Z"},{"number":752,"title":"SDPA: CP-accurate numerics mode (chunk_size param, bitwise match to TE CP merge)","comments":6,"created_at":"2026-08-26T01:23:29Z"},{"number":737,"title":"frost(sdpa): THD ignores bound ragged offsets — TE-style padded THD is silently mis-addressed, and cannot be declined at plan time","comments":5,"created_at":"2026-08-25T18:20:13Z"}],"discussions":[{"number":239,"title":"Welcome to cudnn-frontend Discussions!","comments":0,"category":"Announcements","created_at":"2026-05-04T21:39:32Z"}],"latest_release":{"name":"v1.32.3 release","tag":"v1.32.3","published_at":"2026-11-10T18:11:41Z","body":"- DSA backward: Improve deterministic dKV relay performance on SM100 and add relay cluster-size selection ([#2072](https://github.com/NVIDIA/cudnn-frontend/pull/2072), [#2082](https://github.com/NVIDIA/cudnn-frontend/pull/2082)).\r\n- DSA backward: Add deterministic execution on SM90 with configurable relay cluster sizes ([#2209](https://github.com/NVIDIA/cudnn-frontend/pull/2209)).\r\n- DSA indexer: Add deterministic backward execution on SM90 and improve dW reduction for top-k values of 512 and above ([#2249](https://github.com/NVIDIA/cudnn-frontend/pull/2249)).\r\n- CSA compressor: Enable compression ratios 2 and 4 on SM90 and tune ratio-2 backward schedules ([#2247](https://github.com/NVIDIA/cudnn-frontend/pull/2247)).\r\n- PyTorch SDPA: Detect the cuDNN SDPA switch in `torch.nn.attention` to avoid duplicate provider registration ([#2368](https://github.com/NVIDIA/cudnn-frontend/pull/2368)).","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}