Finetuning SUPERB on custom dataset with 395 classes

[Sreeni1204]

Hello,

I am trying to fine tune the superb wav2vec2 base model for speaker identification task on a custom dataset based on voxceleb. What all changes do I need to perform in order to finetune the model.

All I did is to get my dataset in the VoxCeleb1 format with meta data file, and changed the num_classes in dataset.py file to 395.

Is there any more changes do I need to do?

If not I am facing an error stating: IndexError: Target -4511 is out of bounds. while calculating the cross entropy loss.

Please support