HomeToolsMCPHow It WorksStoriesPhilosophyArchitectureStar on GitHub
All Tools
L
Fine-tuningFreeOpen Source

LITGPT

Train, fine-tune, and serve open LLMs with minimal abstraction

Apache-2.0

ABOUT

Teams working with open LLMs often need one toolchain for inference, another for fine-tuning, and a third for scaling training across accelerators, which adds glue code and hides core model logic behind heavy abstractions. LitGPT provides transparent, high-performance implementations and recipes so developers can train, adapt, and deploy modern open models in one workflow.

INSTALL
pip install 'litgpt[extra]'

INTEGRATION GUIDE

1. Fine-tune open LLMs with LoRA, QLoRA, adapter, or full-tuning recipes 2. Pretrain or continue training language models across multi-GPU or TPU setups 3. Run quantized inference for open models with lower memory requirements 4. Experiment with transparent reference implementations instead of heavyweight frameworks

TAGS

pythonfine-tuningllmpretrainingquantizationinferencepytorchopen-source