thu-ml / SageAttention
Quantized Attention that achieves speedups of 2.1-3.1x and 2.7-5.1x compared to FlashAttention2 and xformers, respectively, without lossing end-to-end metrics across various models.
See what the GitHub community is most excited about this month.
Quantized Attention that achieves speedups of 2.1-3.1x and 2.7-5.1x compared to FlashAttention2 and xformers, respectively, without lossing end-to-end metrics across various models.
[ICLR2025 Spotlight] SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models
Instant neural graphics primitives: lightning fast NeRF and more
NCCL Tests
FlashInfer: Kernel Library for LLM Serving
CUDA Library Samples
Causal depthwise conv1d in CUDA, with a PyTorch interface
[MICRO'23, MLSys'22] TorchSparse: Efficient Training and Inference Framework for Sparse Convolution on GPUs.
[ARCHIVED] Cooperative primitives for CUDA C++. See https://github.com/NVIDIA/cccl
Tile primitives for speedy kernels
cuGraph - RAPIDS Graph Analytics Library
CUDA Kernel Benchmarking Library
Lightning fast differentiable SSIM.
LLM training in simple, raw C/CUDA