Lightning-AI/litgpt

Health v2 · maintenance only

Adoption not part of the score

13629 stars · 1488 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

LitGPT is a Python library providing from-scratch, hackable implementations of 20+ open-source large language models with recipes for pretraining, finetuning (LoRA, QLoRA, adapters), and deployment at scale. It supports distributed training on 1-1000+ GPUs/TPUs with features like Flash attention, FSDP, and fp4/8/16/32 quantization.

Use cases

• finetune an open-source LLM with LoRA or QLoRA
• pretrain a large language model from scratch on multiple GPUs
• run fast LLM inference locally
• deploy a finetuned LLM as an API
• reduce GPU memory usage with quantization
• read clean from-scratch transformer implementations for learning
• scale LLM training across hundreds of GPUs

When to choose

• you want full control and readable, abstraction-free LLM code
• you need to pretrain or finetune LLMs at scale with FSDP
• you want Apache-2.0 licensed enterprise-ready LLM tooling
• you want to experiment with many open LLM architectures in one library

When to avoid

• you just want a plug-and-play inference server with the widest model coverage
• you need a high-level API with minimal configuration
• you are not working with PyTorch or GPUs

Facets

library · maturity active

llm-training llm-inference machine-learning deep-learning cli large-language-models deep-learning machine-learning artificial-intelligence python cli cross-platform pytorch finetuning pretraining lora flash-attention fsdp model-deployment open-source-llms gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="Lightning-AI/litgpt")