Checkpointing is not compatible with .grad() or when an inputs parameter is passed to .backward(). Please use .backward() and do not pass its inputs argument.

[yang-xidian]

hope this letter finds you well. I am one of the users of your [project/research], and I wanted to bring to your attention an issue I encountered while using your code.

In my application, I attempted to utilize PyTorch's checkpointing feature to reduce memory usage and optimize the training process of my model. However, when I tried to pass the inputs parameter to the .backward() method while performing backpropagation, I encountered a RuntimeError:

vbnet
Copy code
RuntimeError: Checkpointing is not compatible with .grad() or when an inputs parameter is passed to .backward(). Please use .backward() and do not pass its inputs argument.
I believe this issue may be related to the incompatibility of the checkpointing feature with the .grad() method or when passing the inputs parameter to the .backward() method simultaneously. While I did not encounter any problems when calling the .grad() method, passing the inputs parameter to the .backward() method resulted in this error.

I was wondering if you could provide some guidance or suggestions on how to address this issue. I am highly interested in your work, and I hope to fully leverage your code and apply it to my project.

Thank you very much for your time and assistance. I look forward to hearing from you.