HazyResearch / ThunderKittens
Tile primitives for speedy kernels
See what the GitHub community is most excited about this week.
Tile primitives for speedy kernels
cuGraph - RAPIDS Graph Analytics Library
CUDA Kernel Benchmarking Library
CUDA Library Samples
Instant neural graphics primitives: lightning fast NeRF and more
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.
Sample codes for my CUDA programming book
Causal depthwise conv1d in CUDA, with a PyTorch interface
LLM training in simple, raw C/CUDA
[ARCHIVED] Cooperative primitives for CUDA C++. See https://github.com/NVIDIA/cccl
NCCL Tests
FlashInfer: Kernel Library for LLM Serving
This is a series of GPU optimization topics. Here we will introduce how to optimize the CUDA kernel in detail. I will introduce several basic kernel optimizations, including: elementwise, reduce, sgemv, sgemm, etc. The performance of these kernels is basically at or near the theoretical limit.
[ICLR2025 Spotlight] SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models