search

uber-research / UPSNet

UPSNet: A Unified Panoptic Segmentation Network
Other
649 stars 120 forks source link

GPU allocation #89

Closed discretecoder closed 5 years ago

discretecoder commented 5 years ago

I have 8 GPUs available and I wish to train on the GPUs 3,4,5,6, when i change the gpus to '3,4,5,6' and also the same for CUDA devices, I get an error from train_model = eval(config.symbol)().cuda() in the training script. it looks like it still uses gpus 0,1,2,3 for this step. what would be a solution? Thank you

YuwenXiong commented 5 years ago

Just set CUDA_VISIBLE_DEVICES=3,4,5,6 and leave the config files unchanged.