deepspeedai / deepspeedai/DeepSpeed

not saving checkpoint by using multi-gpu

Open
#942 4 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, I'm using deepspeed to train bert and other pre-trained large models. When I tried the examples in bing_bert, I examine that I cannot save checkpoints by using multi-gpus, but can save them by using one gpu. It also happened when I try pipeline_parallelism/run.sh. What may gose wrong?

I think I installed the deepspeed environment correctly with cuda 10.1 pytorch 18.1+cu101. And the cpu_adm, fused_adam all installed with yes sign.

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 by reproducing the checkpoint-saving failure in the bing_bert examples and pipeline_parallelism/run.sh with multiple GPUs, then compare it with the reported single-GPU behavior. Inspect the checkpoint-saving path and capture the relevant error or logs. Done means identifying the multi-GPU cause and verifying that checkpoints save successfully in both reported examples.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
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.