loss.backward()无法进行反向传播

[feixuedudiao]

你好，loss不能做backward()，如何解决。

[huent189]

Hi, thank you for your interest. In your email, you said that you got the error RuntimeError: Trying to backward through the graph a second time (or directly access saved tensors after they have already been freed). when running code.
May I ask if you have done any modification to my code since this error often happens if you forget to detach the previous output from gradient computational graph in this line.

[feixuedudiao]

Thank you to answer my.I modify the loss.backward() for the loss after.

Is error to do this?

[huent189]

Do you use pytorch lightning for training? If yes, the loss.backward() will be handled automatically by the library. Putting this line into training code will cause error