thu-ml / SageAttention
Quantized Attention achieves speedup of 2-5x and 3-11x compared to FlashAttention and xformers, without lossing end-to-end metrics across language, image, and video models.
See what the GitHub community is most excited about this month.
Quantized Attention achieves speedup of 2-5x and 3-11x compared to FlashAttention and xformers, without lossing end-to-end metrics across language, image, and video models.
NCCL Tests
Causal depthwise conv1d in CUDA, with a PyTorch interface
Instant neural graphics primitives: lightning fast NeRF and more
CUDA accelerated rasterization of gaussian splatting
LLM training in simple, raw C/CUDA
NVIDIA cuOpt is an open-source GPU-accelerated optimization engine delivering near real-time solutions for complex decision-making challenges.
DeepEP: an efficient expert-parallel communication library
CUDA Library Samples
cuVS - a library for vector search and clustering on the GPU
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.
Lightning fast differentiable SSIM.