This was trained using 9:01 of dataset up to 190 epochs (5,130 steps) using RMVPE and the PABP 32k pretrain.
-