Zhe Wang
Results
1
comments of
Zhe Wang
> This is not a bug, generation with lora is supposed to be slow. A better approach to do lora RL is to merge weights back to main and perform...