deepspeedai / deepspeedai/DeepSpeed
[Question] Deploying Mixture of Expert Model with expert-only parallelism
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
I'm new to DeepSpeed and I'm trying to deploy mixture of expert model based on the tutorial. In the tutorial it says that DeepSpeed support expert only parallelism (the "E" setting in the table) that can parallelize expert layer only but not the rest of the model. However, the group initialize method indicates that it only support 4 scenarios: D, E+D, M and E+D+M. Just wondering if there is anyway I can force Deepspeed to parallelize the expert layer only. Thanks!
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 mixture-of-experts tutorial and deepspeed/utils/groups.py, focusing on the documented E setting and the group initialization scenarios. Determine whether expert-only parallelism is supported by the current implementation; done means clarifying the mismatch or identifying the required change.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, machine-learning
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100