deepspeedai / deepspeedai/DeepSpeedExamples
ppo zero stage 3 generate too slow
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 6.8k
- Forks
- 1.1k
- Avg merge
- 2d 16h
- Merged PRs (30d)
- 1
Description
Hi, when I ran ppo with bloomz-7b1-mt and bloom-560m (prompt_len = answer_len = 256) with zero stage 3 (8A100-40G), it seems the generation time is too slow (average about 72s). When I setting zero stage = 2, it consumes 0.8s with the same hyper-parameters for generation, but cause OOM after one step (I use totally 24A100-40G but still cause OOM). Currently, I consider the reason of generating slowly is using zero stage 3, but it consumes a lot of resource if I do not use this. Is there any solution for solving this problem? Thanks a lot!
Contributor guide
No contributing guide indexed for this repository
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 reproducing PPO generation with bloomz-7b1-mt and bloom-560m using the reported prompt and answer lengths, GPU setup, and ZeRO stage 3, then compare it with stage 2. The issue provides no files or tests; done would require identifying and validating a solution for the generation slowdown without causing the reported out-of-memory failure.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning, performance
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100