Any way to finetune?

[MCLYang]

This is such an interesting work. I hope I could finetune the model to generate particular type of genre and I want to listen to your suggestions. For example, I prepared a small set of jazz songs and attempt to train a little before I infer a new jazz song, where would you recommend to extend for the code?

Appreciate your help
Best,

[pietrobolcato]

Hello! :) did you manage to finetune the model? I am also finding difficulties.

My idea was to simply load the pre-trained model, and then train further on a small dataset with fit_one_cycle. Unfortunately, this doesn't get the wanted result and the training and validation losses increase rather than decrease. Any idea on how to do it?