OpenRLHF icon indicating copy to clipboard operation
OpenRLHF copied to clipboard

Published 20 hours ago • verified publisher icon OpenLLMAI

PPO采用zero 3 stage后产生time out error

Open victorShawFan opened this issue 1 month ago • 1 comments

image image image 脚本修改如下，ckpt换为Qwen： image

victorShawFan avatar May 14 '24 02:05 victorShawFan

有可能是deepspeed的问题 可以尝试降低一下版本号，比如 v0.13.5 see https://github.com/microsoft/DeepSpeed/issues/5520

hijkzzz avatar May 14 '24 04:05 hijkzzz