Xun Wu

https://yushuiwx.github.io/

Microsoft Research

Results 5 comments of


                                            Xun Wu

内存利用率随着save_freq增长是为什么呀

同样的问题，最后内存炸了:(

内存利用率随着save_freq增长是为什么呀

> 同样的问题，最后内存炸了:( 是将vllm升级到0.7.3，verl的code更新到最新版的时候出现的，关闭了VLLM_ATTENTION_BACKEND=XFORMERS的设置

Ray OOM causes the process to be killed

same issue, ![Image](https://github.com/user-attachments/assets/5ba70195-e1a2-4c56-a348-6af6b619cf8f)

ray.exceptions.ActorDiedError: The actor died unexpectedly before finishing this task.

same error

[mcore] verl+megatron development tracking

mark👀