Error Pytorch backend training - Expected object of scalar type

[PiyushM1]

I used the pytorch backend to train 2 models with number of classes as 49 and 16. When I started training the third model with similar conditions, but with 174 classes, I received this error.

Attaching the screenshots. I believe there's nothing I can do because of the abstraction from using monk. Please suggest a fix.

[abhi-kumar]

Share the dataset details and your code. Please check if the training csv has the right labels.

[PiyushM1]

I had earlier trained the same model with a keras backend, it had worked that time, so I don't think the labels are incorrect. Moreover, the labels for the other two classifiers are also prepared in the same way.

Code: https://colab.research.google.com/drive/1Qt0UvKNbdUTwXbbh544bjJ6_7vhvol_f?usp=sharing

Check the Model Classifier part of the code (2nd classifier)

[abhi-kumar]

Your csv file for model classification contains a mix of strings and integers as classes, make sure every label is a string; and there are no white spaces in class names;

Suggestion 1- if your class name is an integer, say 30, change it to "cls_30"
Suggestion 2 - if you class name has white space, say "sonata hybrid", change it to "sonata_hybrid"

[PiyushM1]

Thank you. I replaced the spaces with underscores and it worked.