Hey guys, first time trying to train my own model, I setup Applio with rmvpe, 8 batch size, disabled pretrained and started training, It's been here for around 11h, it's on epoch 1400 and I still don't seem to be close to the point where the model starts overtraining, I see most models don't even go above 1000 epoch, specially on a small dataset like mine (17 minutes of nonstop high quality speaking at 44.1khz), is this normal? Should I just let it train more? I tested the latest pth file on infer and it still sounds a bit garbage :/
#Point of Overtraining
1 messages · Page 1 of 1 (latest)
when it comes to your graph, it still needs more actually it is still going down, but when it comes to the sound itself, it could be overcooked