kssmmm

The rough order of magnitude for the loss during pretraining is how many?

3

I pretrained the model with librispeech960h and get the loss of 0.2. However, when I used the checkpoint to finetune with the librispeech100h, I got a dev-wer about 100. Did...