deepspeedai / deepspeedai/DeepSpeed
[BUG] OOM when train 70B models using deepspeed 0.16.4
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
We found that using OpenRLHF + DeepSpeed 0.15.0, SFT + Adam Offload can train a 70B model with 8 A100 70G + ZeRO3, whereas DeepSpeed 0.16.4 results in OOM. You can try the script https://github.com/OpenRLHF/OpenRLHF/blob/main/examples/scripts/train_sft_llama.sh and use the 70B model + Adam Offload to reproduce the issue.
It looks like this is a serious bug that deepspeed 16.4 can't train 70b models.
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 running examples/scripts/train_sft_llama.sh with a 70B model, Adam Offload, ZeRO3, and the reported 8 A100 70G setup against DeepSpeed 0.15.0 and 0.16.4. Compare memory use and identify the regression; done means restoring the ability to train the 70B model without OOM under 0.16.4.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, machine-learning, performance
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 25/100