flashattention
A C++23 library for local LLMs at the metal — inference and training, built from explicit neural-network components you can read and understand.
Cross-platform FlashAttention-2 Triton implementation for Turing+ GPUs with custom configuration mode
