GRPO Patch #22
Description
Hey guys,
Thanks for the great work as doing multi-gpu training is a huge need! I've saw multiple branches with dev/*grpo* and I was wondering does opensloth supports GRPO patch or not.
I looked into the code and it seems that the ddp_patch() only works for SFT right now. (I also tried running the GRPO training using the main branch patch regardless but it didn't work).
I'd appreciate if anyone or the team can give me some feedback. Thanks!