deepspeedai / deepspeedai/DeepSpeed
[QUESTION] About Pipeline Parallelism save_checkpoint
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
Thx project!
I used Pipeline Parallelism mode,and My model loss has convergenced.
But when I used model_engine.save_checkpoint(). This saved model is not has trained!(acc1 is 0.1% acc5 is 0.5%, It is random parameter)
So,I want to confirm the Pipeline Parallelism's save_checkpoint can correct the Pipeline Parallelism model?
Thx again
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the Pipeline Parallelism workflow and the model_engine.save_checkpoint() entry point described in the report. Reproduce the save and restore process, then verify that the restored model retains the trained accuracy rather than random-parameter results.
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