Errors encountered at the beginning of training. . . .（CUDA out of memory）

[Xiaoxiaosito]

The following are the details of error reporting. Before that, there was an error reporting like 'BrokenPipeError: [Errno 32] Broken pipe'. When I changed num_workers to 0, the problem was solved, but there is an new error reporting now. Who can help me? (Although he suggested that the memory was insufficient, it was obviously sufficient.)

[GoArsenal]

set batchsize to 16 or smaller