deepspeedai / deepspeedai/DeepSpeed

How to save DS Zero model weights in meta tensor format for DS Inference

Open
#3,835 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
43.1k
Forks
5k
Avg merge
4d 15h
Merged PRs (30d)
112

Description

Hi there, I'm wondering if there's a way to save the weights of a model that's been trained using DS Zero directly into the meta tensor format that is used to speed up model loading times for DS Inference.

Currently, I've been saving the model back into a single .bin file and then loading that into DS Inference & then saving the meta tensor out, but wondering if there is a quicker way to do this.

Any help would be greatly appreciated!

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

The issue concerns saving DeepSpeed ZeRO-trained weights directly in the meta tensor format used by DS Inference. Start by tracing the existing ZeRO save path and DS Inference loading path, then determine whether a direct conversion path exists; done means the supported workflow is implemented or documented and avoids the intermediate single .bin step.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.