young-geng/EasyLM
A JAX/Flax framework for pre-training, fine-tuning, evaluating, and serving large language models at scale across TPU/GPU accelerators.
Velocity · 7d
+1.9
★ / day
Trend
→steady
star history
EasyLM provides an easy-to-use and customizable codebase for training LLMs without the complexity of other frameworks. It leverages JAX’s pjit functionality to shard model weights and training data across hundreds of accelerators for multi-host training on Google Cloud TPU Pods. The project supports LLaMA, LLaMA 2, and LLaMA 3 architectures, building on HuggingFace transformers and datasets libraries.