Skip to content

Conversation

@ocss884
Copy link
Contributor

@ocss884 ocss884 commented Apr 15, 2025

Motivation

As we are integrating SGLang to training framework like veRL. Current unbalance constraint 10% is too thight and could cause the SGLang engine initialization fail. For example, the ray header would occupy more memory on one worker than others and makes the unbalance check fail.

Modifications

Checklist

@hnyls2002
Copy link
Collaborator

#5426 I think after this PR is merged, you can just diable the check in env.

@hnyls2002 hnyls2002 closed this Apr 18, 2025
@ocss884 ocss884 deleted the relax-mem-unbalance-check branch June 7, 2025 16:58
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants