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...