AssertionError: Attempted unscale_ but _scale is None. This may indicate your script did not use scaler.scale(loss or outputs) earlier in the iteration

I was trying to run this notebook the speech emotion recognition one on a different dataset, giving me tons of errors. I tried resolving them but one such error related is to the automatic mixed precision. Can anybody please suggest how I can fix this issue? I tried training on the dataset given in the notebook but it’s giving a similar error. Here’s a link to my notebook
here