Can I not use torch.distributed?

Open yeRuiHuan opened this issue 6 years ago • 4 comments

After read the train.py and evaluate.py, I don't find a option that turn torch.distributed off.(I try to run your code in one GPU)

Dec 06 '19 04:12 yeRuiHuan

Actually, It's a bug of package Inplace-abn. It needs to run with the context of torch.distributed.

Dec 14 '19 05:12 speedinghzl

我也只有一张gpu，一张gpu是否可以直接用nn.BatchNorm2d

Jun 27 '20 13:06 swjtulinxi

你好，请问单张gpu是不是可以直接用nn.BatchNorm2d,然后DataParallelModel(deeplab)，，DataParallelCriterion(criterion)这两句都可以不要对吧

Jun 28 '20 01:06 swjtulinxi

还有你这个代码当中的模型中输出两个值下，x和dsn，dsn用于辅助损失函数，像pspnet那样，但计算loss时并没有把dsn考虑进去

Jun 28 '20 02:06 swjtulinxi