deepspeedai / deepspeedai/DeepSpeed

When I load model and use deepspeed with ZERO stage 3, I have a problem with high cpu memory Usage

Open
#3,722 3 comments 0 reactions 1 assignee View on GitHub

@HeyangQin is already working on this.

Since Jun 16, 2023.

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

Description

deepspeed: 0.9.2
transformers: 4.30.0.dev0
torch: 1.12.1+cuda11.6
server info:
8*A100 80G
Memory: 500G

When load a model that has 15B parameters on single node(8*A100)，the memory usage exceeds 500 GB and the process was killed,
I have used the Zero stage 3.

how can I reduce the Memory Usage

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.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.