---
title: 'Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism'
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
license_note: 'paper is arXiv non-exclusive distribution license (verified on arXiv abs page 2026-11-12); this pointer note is our own words and is CC BY 4.0; we link to the paper rather than reproduce it'
source: https://iinventedlight.com/library/notes/1909-08053-megatron-lm.md
source_license: 'arXiv non-exclusive distribution license'
attribution: 'Pointer by iinventedlight (AI-assisted). Source: Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper, Bryan Catanzaro, Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism (arXiv non-exclusive distribution license). This note links to the paper rather than reproducing it.'
author: 'iinventedlight (AI-assisted)'
date: 2026-11-12
status: published
approved_by: "Cole Withington, 2026-11-12 14:44 PT, Librarian chat widget t51s316: \"Yes, publish all 100 second-list arXiv notes\""
checked_on: 2026-11-12
as_of: 2026-11-12
confidence: medium
untested: 'Pointer note only. Paper not reproduced. Abs page and license checked 2026-11-12.'
category: arxiv
topic_area: 'ai-research: efficiency'
tags: arxiv, efficiency, model-parallelism, training
arxiv_id: 1909.08053
alias: 1909.08053-megatron-lm
type: pointer
format: pointer
sources:
  - id: S1
    title: 'Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism'
    publisher: arXiv
    authors: 'Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper, Bryan Catanzaro'
    url: https://arxiv.org/abs/1909.08053
    version_read: 'v4 (13 Mar 2020)'
    accessed: 2026-11-12
    license: 'arXiv non-exclusive distribution license'
---
# Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism

## Summary
[Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) (2019) described an intra-layer model-parallel method for training multi-billion-parameter transformer language models, addressing the memory limits of putting an entire model on one device. It became a foundational training system, later known as Megatron-LM.

What it introduced: a simple, efficient split of transformer layers across devices so that very large models can be trained with model parallelism, complementing ordinary data parallelism rather than replacing it.

What it led to: the Megatron family of training libraries and later combinations with ZeRO-style data-parallel sharding. See the ZeRO pointer in this batch. Live note: [2305-13245-gqa](https://iinventedlight.com/library/notes/2305-13245-gqa/), which later reduced inference-time memory in a different way.

Use it when you want the original intra-layer model-parallel training paper. Where to read it: https://arxiv.org/abs/1909.08053 (v4 (13 Mar 2020)).

The paper is posted on arXiv under the arXiv non-exclusive distribution license. This note links to it rather than reproducing it. The note itself is CC BY 4.0.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Pointer by iinventedlight (AI-assisted). Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it.

---

Agents: [/agents.md](/agents.md) | [/.well-known/agent-library.json](/.well-known/agent-library.json).
