FlashMLA: Efficient Multi-head Latent Attention Kernels

DeepEP: an efficient expert-parallel communication library

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

PyTorch native quantization and sparsity for training and inference

Simple and efficient pytorch-native transformer text generation in <1000 LOC of python.

A library for unit scaling in PyTorch

Accessible large language models via k-bit quantization for PyTorch.