search

guoqincode / Open-AnimateAnyone

Unofficial Implementation of Animate Anyone
2.9k stars 233 forks source link

About multi-gpu training #22

Closed asdasdad738 closed 9 months ago

asdasdad738 commented 9 months ago

It seems that DDP isn't used in train.py.

ryancll commented 9 months ago

I tried to wrap the trainable modules with DDP and it works for stage2 training. But I meet error when training referencenet in stage1 (during backward). Could anyone provide some instructions? @guoqincode

guoqincode commented 9 months ago

I tried to wrap the trainable modules with DDP and it works for stage2 training. But I meet error when training referencenet in stage1 (during backward). Could anyone provide some instructions? @guoqincode

Maybe you can send your code to my email and I will check it when I have time.

asdasdad738 commented 9 months ago

I tried to wrap the trainable modules with DDP and it works for stage2 training. But I meet error when training referencenet in stage1 (during backward). Could anyone provide some instructions? @guoqincode

I just wrap the referencenet and pose guider with DDP directly, and it works.