[Question] Problematic multi-agent implementation for GRPO
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 8.5k
- Forks
- 1.3k
- Avg merge
- 5h 36m
- Merged PRs (30d)
- 22
Description
Looking at this section of code below in the context of multi-agent, the number of samples returned from a multi-agent generation must match --n-samples-per-prompt, otherwise GRPO is disabled since the groups are uneven?
So in your multi-agent example, could this constraint be properly documented by saying something like "Requirement: Number of samples returned from generate function must always match --n-samples-per-prompt". For example, if you want to implement n-shot with self-verification and you have --n-samples-per-prompt 8, you would want to set n=7 and have a single sample for the verification, meaning you return 8 samples for the multi-agent workflow.
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 multi-agent generation section in slime/ray/rollout.py at lines 187-205, then locate the documentation or example covering that workflow. Document that the number of samples returned by generate must match --n-samples-per-prompt, including the described self-verification example, and confirm the constraint is clear to users.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Documentation
- Difficulty
- 2/5
- Estimated time
- 1-3 hours
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 52/100