Multiple training codes are stopped without error

cuda 11.4 
pytorch 1.10.2

It works normally when running up to two training codes.
However, when I add one more, the added code fail to run 1 epoch and stops without error.

The code still hold GPU memory but not running.

I tried to apply num_workers from 0 to 4 but nothing changed.
How can I fix it?
스크린샷 2022-02-16 13.05.52