fsdp

FSDP2 is a rewrite of PyTorch's Fully Sharded Data Parallel (FSDP) API, removing the FlatParameter abstraction for better composability and simpler implementation.

Skills
AI research
3

What it does and how to use it

A skill that teaches Claude Code to do ai research tasks well. It is packaged knowledge the model uses whenever it is relevant.

Install the skill (command above).
In Claude Code, describe what you need - the skill turns on by itself when the context fits.
No commands to memorize: you talk in plain language and the skill does the rest.

Install (Claude Code)

mkdir -p .claude/skills/fsdp && curl -fsSL https://doutorai.pt/data/lib/skills__fsdp.md -o .claude/skills/fsdp/SKILL.md

Paste this command into the terminal, in your project folder, and the component installs automatically.

Content

# FSDP2 in TorchTitan

## Why FSDP2?

FSDP2 is a rewrite of PyTorch's Fully Sharded Data Parallel (FSDP) API, removing the `FlatParameter` abstraction for better composability and simpler implementation.

### Key improvements over FSDP1

- **DTensor-based sharding**: Sharded parameters are `DTensor`s on dim-0, enabling easy manipulation and communication-free sharded state dicts
- **Better memory management**: Deterministic and lower GPU memory (7% reduction) by avoiding `recordStream`
- **Simplified API**: Fewer arguments, no wrapper class

### Performance

On Llama-7B with 8x H100s, FSDP2 ac

- 🔒 Full content on the Pro plan (€7.99/month) -

Need help with Claude Code?

Ask Dr AI - it explains things simply.

Need a tailor-made solution?

Not sure where to start with AI in your business? Get a free assessment and a tailored plan - no commitment.

Save hours every week with AI, no hassle

Doutor AI gives you 1,859 ready-to-copy AI prompts, by industry and in plain English - plus a complete technical library for those who want to go further.

See plans