Download
Megatron is a large, powerful transformer developed by the Applied Deep Learning Research team at NVIDIA. This repository is for ongoing research on training large transformer language models at scale. We developed efficient, model-parallel (tensor, sequence, and pipeline), and multi-node pre-training of transformer based models such as GPT, BERT, and T5 using mixed precision.
Megatron is also used in NeMo Megatron, a framework to help enterprises overcome the challenges of building and training sophisticated natural language processing models with billions and trillions of parameters.
Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
Project Samples
Follow Megatron
Other Useful Business Software
Build AI Apps with Gemini 3 on Vertex AI
Vertex AI gives developers access to Gemini 3—Google’s most advanced reasoning and coding model—plus 200+ foundation models including Claude, Llama, and Gemma. Build generative AI apps with Vertex AI Studio, customize with fine-tuning, and deploy to production with enterprise-grade MLOps. New customers get $300 in free credits.
Rate This Project
Login To Rate This Project
User Reviews
Be the first to post a review of Megatron!
