verl icon indicating copy to clipboard operation
verl copied to clipboard

Published 20 hours ago • verified publisher icon volcengine

[WIP] PRIME algorithm

Open ZefanW opened this issue 9 months ago • 2 comments

Refactor and merge PRIME algorithm into verl/main https://github.com/PRIME-RL/PRIME

ZefanW avatar Feb 24 '25 08:02 ZefanW

CLA assistant check
All committers have signed the CLA.

CLAassistant avatar Feb 26 '25 00:02 CLAassistant

@hiyouga have you seen the hf timeout issue before, in the geo3k test?

eric-haibin-lin avatar Mar 04 '25 22:03 eric-haibin-lin

Hi! I have a question about balance_batch. Why is it being made an optional feature in this PR? Are there any cases where balance_batch could have a negative impact? I find this a bit confusing and am unsure whether to enable it.

huiyeruzhou avatar Mar 13 '25 09:03 huiyeruzhou