Sorry for the inconvenience, but this would be an awesome way to speed up training. I changed the line 22 of file code/train.py to be gpus=-1 which is the value for pytorch lightning to use all available gpus. But when doing so the following error appears:
RuntimeError: Cowardly refusing to serialize non-leaf tensor which requires_grad, since autograd does not support crossing process boundaries. If you just want to transfer the data, call detach() on the tensor before serializing (e.g., putting it on the queue).
The traceback points to the trainer.fit() line but it does not give any additional information pointing to any other file of your repository. If you could fix it and try on a multi-gpu setting this could make the project more usable. Thanks in advance.
Sorry for the inconvenience, but this would be an awesome way to speed up training. I changed the line 22 of file code/train.py to be gpus=-1 which is the value for pytorch lightning to use all available gpus. But when doing so the following error appears:
RuntimeError: Cowardly refusing to serialize non-leaf tensor which requires_grad, since autograd does not support crossing process boundaries. If you just want to transfer the data, call detach() on the tensor before serializing (e.g., putting it on the queue).
The traceback points to the trainer.fit() line but it does not give any additional information pointing to any other file of your repository. If you could fix it and try on a multi-gpu setting this could make the project more usable. Thanks in advance.