What it does
Megatron-LM is NVIDIA's toolkit for training transformer models, the design behind most modern AI language models, at very large scale. Training such models needs many graphics cards working together, and this software splits the work between them efficiently.
The repository has two parts. Megatron Core is a library of optimised building blocks for people making their own training systems. Megatron-LM adds ready-made training scripts for researchers and for learning how large-scale training works.
A companion tool converts models between Megatron and the popular Hugging Face format.
Who it's for
Research teams and engineers training large AI models on powerful GPU clusters.
Can I use it without coding?
Developers only. Needs programming skills and access to NVIDIA graphics cards, usually many of them. New to this? Start here.
Can I run it?
- Cost
- FreeNothing to pay or sign up for.
- Runs on
- Server
- Computer
- Made for servers or data centres
- API key
- Not needed
Try it
Using Claude Code, Cursor or another AI coding assistant? Paste the prompt and it will walk you through the setup.New to this?
Before you install
- Only download from the official GitHub page. Copies with similar names exist for many popular projects.
- No standard license listed. Check the GitHub page before using it for business.
- More on staying safe
Where people are talking about it
- Scalable Training of Mixture-of-Experts Models with Megatron CoreResearch paper · Mar 2026
- CUDA 11.0Hacker News · Jul 2020
- CUDA Toolkit Release NotesHacker News · Nov 2019
- Megatron-LM: Training Multi-Billion Parameter Language Models Using Model ParallelismResearch paper · Sept 2019