search

rohban-lab / Knowledge_Distillation_AD

82 stars 46 forks source link

Can you provide the accurate multi-GPU training version? #3

Open Pattaro opened 3 years ago

Pattaro commented 3 years ago

How to train model using multi-gpus? while using DataParall to train model ，I found that the memory of CUDA has been increasing. Is it because "with torch.no_grad()" was not used in the test？However the test needs to use “model.forward”, so can you provide the accurate multi-GPU training version? Thank you for your reading.

SMSD75 commented 3 years ago

Hi,

We will work on it to provide a multi-GPU version as well. However, if you are in hurry, you can convert the code to a multi-GPU setting yourself. I guess it wouldn't be hard.

Pattaro commented 3 years ago

Hi,

We will work on it to provide a multi-GPU version as well. However, if you are in hurry, you can convert the code to a multi-GPU setting yourself. I guess it wouldn't be hard.

yeah, I implement it, but the verification found that the AUC of MVTEC was reduced by 5%, so I still hope to get the official version