Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
cuDNN Frontend is NVIDIA's modern, open-source entry point to the cuDNN library and a growing collection of high-performance open-source kernels.
| Date | Stars |
|---|---|
| 2026-07-24 | 887 |
| 2026-07-25 | 888 |
| 2026-07-28 | 888 |
| 2026-07-30 | 888 |
| 2026-08-06 | 888 |
| 2026-08-07 | 900 |
| 2026-08-15 | 905 |
| 2026-08-18 | 907 |
| 2026-08-19 | 910 |
| 2026-08-20 | 910 |
| 2026-08-21 | 911 |
| 2026-08-22 | 914 |
| 2026-08-23 | 914 |
| 2026-08-24 | 915 |
| 2026-08-25 | 915 |
| 2026-08-26 | 916 |
| 2026-08-27 | 916 |
| 2026-08-28 | 916 |
| 2026-08-29 | 918 |
| 2026-08-30 | 918 |
| 2026-08-31 | 920 |
| 2026-09-01 | 921 |
| 2026-09-02 | 923 |
| 2026-09-03 | 926 |
| 2026-09-04 | 926 |
| 2026-09-05 | 929 |
| 2026-09-06 | 929 |
| 2026-09-07 | 931 |
| 2026-09-08 | 932 |
| 2026-09-09 | 932 |
| 2026-09-10 | 933 |
| 2026-09-11 | 932 |
| 2026-09-12 | 933 |
| 2026-09-13 | 933 |
| 2026-09-14 | 937 |
| 2026-09-15 | 938 |
| 2026-09-16 | 939 |
| 2026-09-17 | 939 |
| 2026-09-18 | 940 |
| 2026-09-19 | 942 |
| 2026-09-20 | 942 |
Today
— stars today
This week
+9 stars this week
This month
+31 stars this month
Momentum
44.0
growth rate 0.97%/day
# cuDNN Frontend (FE) [](https://pypi.org/project/nvidia-cudnn-frontend/) [](https://pypi.org/project/nvidia-cudnn-frontend/) [](https://pypi.org/project/nvidia-cudnn-frontend/) [](LICENSE.txt) [](https://nvidia.github.io/cudnn-frontend/) **cuDNN Frontend** is NVIDIA's modern, open-source entry point to the cuDNN library and a growing collection of high-performance open-source kernels — scaled dot-product attention (**SDPA / Flash Attention**), grouped GEMM fusions for **Mixture-of-Experts (MoE)** training, fused normalization + activation, and more. It provides a **header-only C++ API** and a **Python interface** (with native PyTorch integration) to the cuDNN Graph API, targeting NVIDIA **Hopper** (H100/H200) and **Blackwell** (B200/GB200/GB300) GPUs across FP16, BF16, FP8, and **MXFP8** precision. **Links:** [Documentation](https://docs.nvidia.com/deeplearning/cudnn/frontend/latest/) · [Blog & Deep Dives](https://nvidia.github.io/cudnn-frontend/) · [PyPI](https://pypi.org/project/nvidia-cudnn-frontend/) · [Release Notes](https://github.com/NVIDIA/cudnn-frontend/releases) · [Samples](samples/) ## 🚀 Latest news: We will begin open-sourcing kernels based on customer needs, with the goal to educate developers and enable them to customize as needed. We are now shipping **OSS kernels**, allowing you to inspect, modify, and contribute to the core logic. Check out our latest implementations: * **[FROST GEMM engine](https://github.com/NVIDIA/cudnn-frontend/tree/main/python/cudnn/gemm/frost):** JIT-compiled Blackwell GEMM engine reachable through the ordinary `cudnn.pygraph` API — matmul, grouped (MoE) matmul, block-scaled FP4/FP8, and chained pointwise epilogues are fused into one kernel from the graph you already built. Opt in with `CUDNN_FRONTEND_ENABLE_FROST_ENGINES=1`; it is then a candidate for every matmul graph it can serve, ranked against the backend's own plans. * **[GEMM + Amax](https://github.com/NVIDIA/cudnn-frontend/tree/main/python/cudnn/gemm/cutedsl/dense/amax):** Optimized FP8 matrix multiplication with absolute maximum calculation. * **[GEMM + SwiGLU](https://github.com/NVIDIA/cudnn-frontend/tree/main/python/cudnn/gemm/cutedsl/dense/swiglu):** High-performance implementation of the SwiGLU activation fused with GEMM. * **[GEMM + sReLU](https://github.com/NVIDIA/cudnn-frontend/tree/main/python/cudnn/gemm/cutedsl/dense/srelu):** High-performance implementation of squared-ReLU fused with GEMM. * **[GEMM + dsReLU](https://github.com/NVIDIA/cudnn-frontend/tree/main/python/cudnn/gemm/cutedsl/dense/dsrelu):** High-performance implementation of dsquared-ReLU fused with GEMM. * **[Grouped GEMM (BF16)](https://github.com/NVIDIA/cudnn-frontend/tree/main/python/cudnn/gemm/cutedsl/grouped/unfused):** Unfused BF16 grouped GEMM with dense and discrete MoE weight layouts. * **[Grouped GEMM + GLU](https://github.com/NVIDIA/cudnn-frontend/tree/main/python/cudnn/gemm/cutedsl/grouped/glu):** Unified BF16 and legacy block-scaled grouped GEMM GLU API supporting dense and discrete MoE weight layouts. * **[Grouped GEMM + GLU + Hadamard](https://github.com/NVIDIA/cudnn-frontend/tree/main/python/cudnn/gemm/cutedsl/grouped/glu_hadamard):** Dense grouped GEMM GLU forward fusion with a fused Hadamard transform and per-expert AMAX reduction. * **[Grouped GEMM + dGLU](https://github.com/NVIDIA/cudnn-frontend/tree/main/python/cudnn/gemm/cutedsl/grouped/dglu):** Unified BF16 and legacy block-scaled grouped GEMM dGLU backward API supporting dense and discrete MoE weight layouts. * **[Grouped GEMM + SwiGLU](https://github.com/NVIDIA/cudnn-frontend/tre
Excerpt of 15,238 characters
Read on GitHub
Anerudhan Gopal
134
17
11
9
9
8
7
6
5
4
4
3
2
2
2
2
2
2
2
2
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:5748f7c0f834aea8, topic:cuda-kernels, topic:flash-attention, readme:fp8
matched fp:5748f7c0f834aea8, topic:gpu, topic:cuda
matched fp:5748f7c0f834aea8, topic:deep-learning
matched fp:5748f7c0f834aea8, topic:transformer, topic:moe, readme:mixture of experts