yl4579
Train first stage on LJSpeech， train gen_loss is slower increase？？？Is this correct？？
The generator loss should at one point stabilize and not go any higher. Here are my current graphs on training first stage.
Are you training single language or multi? Also, is this for English or other language?
The generator loss should at one point stabilize and not go any higher. Here are my current graphs on training first stage.
Are you training single language or multi? Also, is this for English or other language?
Thanks for your reply, I trained the first stage on single language (LJSpeech Dataset).
Interesting - I've only ever trained on custom datasets, so I've not seen graphs from LJSpeech training. Did the loss stabilize at the end?
Interesting - I've only ever trained on custom datasets, so I've not seen graphs from LJSpeech training. Did the loss stabilize at the end? no，gen_loss didn't stabilize at the end. gen_loss was slower increasing at the end. I trained first stage on single H20 GPU with LJSpeech dataset.
@renwotao , did you finished your training? I am facing the same issue on a custom dataset, training on a rtx5090.
