Open alanngnet opened 6 months ago
Verified that simply taking a pretrained model's checkpoint files as a starting to start a training run on a different dataset (sized with the same size CQT arrays) works, with clearly better results than training from scratch on the second dataset.
Unclear how much this is a software vs documentation task. What is the most efficient way to take a trained model and teach it some new tunes?