LLM training in simple, raw C/CUDA

A massively parallel, optimal functional runtime in Rust

DeepEP: an efficient expert-parallel communication library

Tile primitives for speedy kernels

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 …

UNet diffusion model in pure CUDA

cuVS - a library for vector search and clustering on the GPU