How to use multiple GPU training？

[xdsonglinliu]

From the file REPRODUCE_RESULTS.md, We know the code to start training is python main.py -- train --pipeline_name unet_weighted

But how to use multiple GPU training？ I have two NVIDIA 1080 graphics cards, if I want use both of them at the same time in training, how should I set the parameters of the training command?

[jakubczakon]

@xdsonglinliu we are using DataParallel by default so all you need to do is go:

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py -- train --pipeline_name unet_weighted 

I hope this helps.