thu-ml / SageAttention
[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.
See what the GitHub community is most excited about today.
[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.
FlashInfer: Kernel Library for LLM Serving
NCCL Tests
CUDA accelerated rasterization of gaussian splatting
DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling
LLM training in simple, raw C/CUDA
Fast CUDA matrix multiplication from scratch
DeepEP: an efficient expert-parallel communication library
Instant neural graphics primitives: lightning fast NeRF and more
cuVS - a library for vector search and clustering on the GPU
CUDA Kernel Benchmarking Library
Causal depthwise conv1d in CUDA, with a PyTorch interface
Graphics Processing Units Molecular Dynamics
[ARCHIVED] Cooperative primitives for CUDA C++. See https://github.com/NVIDIA/cccl
GPU accelerated decision optimization
CUDA Library Samples
RAFT contains fundamental widely-used algorithms and primitives for machine learning and information retrieval. The algorithms are CUDA-accelerated and form building blocks for more easily writing high performance applications.