Closed Yyb-XJTU closed 5 days ago
I have found this problem: I use "CUDA_VISIBLE_DEVICES=1, 2, 3 python runner.py" to perform multi-card training, and it seems that it is not supported. When I use "CUDA_VISIBLE_DEVICES=1 python runner.py", it can run normally.
When I try to train, it always gets stuck at this point, and the log shows the following figure: