Training error caused by inplace operations

Open LeoDarcy opened this issue 1 year ago • 1 comments

https://github.com/nv-tlabs/XCube/blob/d0ee1c26ebd67a9a8ff803835333b088155fb051/xcube/modules/autoencoding/base_encoder.py#L63

It leads to "RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation" during training.

Sep 24 '24 15:09 LeoDarcy

After making modifications, I still encounter similar errors. May I ask if there are any other code that need to be modified? Thank you very much.

Sep 25 '24 04:09 LeoDarcy