deepspeedai / deepspeedai/DeepSpeed
[REQUEST] Disable zero stage2 all_gather‘s bucketing
Open
Nobody has claimed this yet.
enhancement
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
It seems that zero stage2 can never overlap all_gather with computation at the end of each step.
Bucketing the all_gather only makes it slower, and eats more GPU memory.
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 by locating the ZeRO stage 2 all_gather bucketing entry point and reviewing how it is scheduled at the end of each step. Done means stage 2 can disable this bucketing, avoids the reported extra GPU memory use, and its performance is verified against the current behavior.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- distributed-systems, machine-learning, performance
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100