deepseek-ai / FlashMLA

FlashMLA: Efficient Multi-head Latent Attention Kernels

microsoft / BitNet

Official inference framework for 1-bit LLMs