kxngluca/DL-Art-School
1
0
Fork 0
forked from mrq/DL-Art-School
Code Pull Requests Activity

Revert "Get rid of CUDA_VISIBLE_DEVICES"

Browse Source 
It is actually necessary for training in distributed mode. Only
do it then.
This commit is contained in:
James Betker 2020-12-31 10:31:40 -07:00
parent 8de5a02a48
commit 0eb1f4dd67
1 changed files with 7 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
codes/train.py
View File

@ -298,6 +298,12 @@ if __name__ == '__main__':
parser.add_argument('--local_rank', type=int, default=0) parser.add_argument('--local_rank', type=int, default=0)
args = parser.parse_args() args = parser.parse_args()
opt = option.parse(args.opt, is_train=True) opt = option.parse(args.opt, is_train=True)
if args.launcher != 'none':
# export CUDA_VISIBLE_DEVICES for running in distributed mode.
if 'gpu_ids' in opt.keys():
gpu_list = ','.join(str(x) for x in opt['gpu_ids'])
os.environ['CUDA_VISIBLE_DEVICES'] = gpu_list
print('export CUDA_VISIBLE_DEVICES=' + gpu_list)
trainer = Trainer() trainer = Trainer()
#### distributed training settings #### distributed training settings
@ -309,7 +315,7 @@ if __name__ == '__main__':
print('Disabled distributed training.') print('Disabled distributed training.')
else: else:
opt['dist'] = True opt['dist'] = True
init_dist('nccl') init_dist('nccl', opt)
trainer.world_size = torch.distributed.get_world_size() trainer.world_size = torch.distributed.get_world_size()
trainer.rank = torch.distributed.get_rank() trainer.rank = torch.distributed.get_rank()