python

[leijdjcn]

running_loss += loss.item()
RuntimeError: CUDA error: unspecified launch failure
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
出现上面的错误怎么回事

[zhmou]

请尝试将模型和数据放在CPU上运行，观察是否会复现此错误。推测还是PyTorch问题，能否尝试重新创建整个环境。