deepspeedai / deepspeedai/DeepSpeedExamples
[DeepSpeed-Chat] Configuring parallelism
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 6.8k
- Forks
- 1.1k
- Avg merge
- 2d 16h
- Merged PRs (30d)
- 1
Description
Hi,
I'm trying to understand how the multi-GPU and multi-node deployments of DeepSpeed-Chat utilise resources in order to maximise parallelism. Specifically, I would like to know the dimensions (i.e., data, model, pipeline) of parallelism that are employed when running DeepSpeed-Chat on a multi-GPU, potentially, multi-node setting. This relates to all three steps of the RL-HF training pipeline but I'm also primarily interested in step#3 when the actual RL training takes place. Finally, is the PPO algorithm also running in a distributed manner?
Thank you very much,
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
No file, test, or entry point is mentioned. Start by reading the DeepSpeed-Chat RL-HF training flow, especially step #3, and trace how multi-GPU and multi-node resources are used. Done means documenting the data, model, and pipeline parallelism for all three steps and clarifying whether PPO runs distributed.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, machine-learning
- Issue type
- Documentation
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 30/100