How to use one GPU for training?

microsoft / transductive-vos.pytorch

a transductive approach for video object segmentation

155 stars 32 forks source link

How to use one GPU for training? #10

Closed kelisiya closed 4 years ago

mofang2016 commented 4 years ago

Maybe you should make sure --nproc_per_node=1. In addition, the devices can be set via CUDA_VISIBLE_DEVICES. CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.launch --master_port 12347 --nproc_per_node=1 main.py

eastOffice commented 4 years ago

See #7