deepspeedai / deepspeedai/DeepSpeed

[BUG] if use zero3+tp, use save_16bit_model to save weights, only save 1/tp weights

Open
#3,607 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

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

Description

only saved one file mp_rank_00_model_states.pt, only 1/tp weights in this file.

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

Start at the save_16bit_model entry point and reproduce the ZeRO-3 plus tensor-parallel save path described in the issue. Inspect how mp_rank_00_model_states.pt is produced and compare the saved contents with the expected full weights; done means the save contains all tensor-parallel weights rather than only 1/tp.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
distributed-systems, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.