search

layer6ai-labs / xpool

https://layer6ai-labs.github.io/xpool/
113 stars 8 forks source link

Can you provide multi-GPU version? #5

Open xmu-xiaoma666 opened 2 years ago

xmu-xiaoma666 commented 2 years ago

Can you provide multi-GPU version?

QinYang79 commented 1 year ago

Can you provide multi-GPU version?

I'm also experimenting with multiple GPUs. I found that using nn.DataParallel() and modifying some code as follows seem to achieve parallelism.

trainer.py text_embeds, video_features, video_embeds_pooled= self.model(data,True) video_embeds_pooled = self.model.module.pool_frames(text_embeds, video_features) output = sim_matrix_training(text_embeds, video_embeds_pooled, self.pooling_type)

Jiamian-Wang commented 1 year ago

Thanks for the nice and clear implementation! Is that possible to release a multi-GPU version?

X7J92 commented 10 months ago

您好呀，我想问一下，您在训练MSVD数据集，特定的迭代次数被系统自动杀死·的情况吗 @Jiamian-Wang

X7J92 commented 10 months ago

您好呀，我想问一下，您在训练MSVD数据集，特定的迭代次数被系统自动杀死·的情况吗？ @QinYang79