为什么单卡不超显存，多卡会超显存

[eightmusic]

[qingyuanxingsi]

+1,8卡会OOM,求指导

[zhou-linpeng]

you can use deepspeed to reduce gpu memory

[zhangvia]

you can use deepspeed to reduce gpu memory

i did, but the process will stuck at backward in 5th step. i was confused

[jiliu112]

+1， 请问这个问题解决了吗

[SystemErrorWang]

you can use deepspeed to reduce gpu memory

i did, but the process will stuck at backward in 5th step. i was confused

same problem，training with deepspeed got stuck after 4 steps. have you got any solution for this? Thank you!