Skills
.Work. You
Rest

megatron-fsdp

Official

Scale large Megatron models with FSDP.

Software Engineering#gpu#distributed-training#fsdp#megatron#megatron-bridge#training-optimization
AuthorNVIDIA
Version1.0.0
Installs0

System Documentation

What problem does it solve?

Megatron-FSDP enables Fully Sharded Data Parallel in Megatron-Bridge to optimize memory usage and training scalability for very large Megatron models.

Core Features & Use Cases

  • Enable Megatron's FSDP by toggling configuration flags in Bridge: dist.use_megatron_fsdp and ddp.use_megatron_fsdp.
  • Provide code anchors, validation steps, pitfalls guidance, and a runtime path choice between Megatron FSDP and Torch FSDP2 for robust experimentation.
  • Suitable for large-scale pretraining and fine-tuning workflows that demand distributed memory management and performance gains.

Quick Start

Set cfg.dist.use_megatron_fsdp and cfg.ddp.use_megatron_fsdp to True, then run the Megatron-Bridge performance harness to verify the setup.

Dependency Matrix

Required Modules

None required

Components

Standard package

💻 Claude Code Installation

Recommended: Let Claude install automatically. Simply copy and paste the text below to Claude Code.

Please help me install this Skill:
Name: megatron-fsdp
Download link: https://github.com/NVIDIA/skills/archive/main.zip#megatron-fsdp

Please download this .zip file, extract it, and install it in the .claude/skills/ directory.
View Source Repository

Agent Skills Search Helper

Install a tiny helper to your Agent, search and equip skill from 471,000+ vetted skills library on demand.