movic
难道不是nn.Dropout的问题吗？
 As the pivot column moving from left to right, suppose the union of `s1`, `pivot` and `s2` is equal to the total set of anchor scores, the individual loss...
Since the code can run on your local machine, I guess the reason is that the version of PyTorch you used are different. But from the Traceback info, you could...
Well then, try to change the input back to float, and convert the target label to float as well, I guess the label is double type.
Sorry, I've deleted the checkpoint file to save storage. But it takes only 2~3 hours to train the model, you can try to retrain by yourself.
> Yeah it would be nice to make them serializable! It would be nice if it was defined on the original object though (vs a separate `fcl2.py` + inheritance), what...
> Yeah it would be nice to make them serializable! It would be nice if it was defined on the original object though (vs a separate `fcl2.py` + inheritance), what...