flagos-ai/FlagGems

FlagScale is a large model toolkit based on open-sourced projects.

A kernel library written in tilelang

~1300 line, minimal, extensible LLM inference engine built from scratch.

Go ahead and axolotl questions

Scale LLM Engine public repository

A simple, performant, and scalable Jax LLM!

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

Efficient Triton Kernels for LLM Training

Gemma open-weight LLM library, from Google DeepMind