Facing NaN issue in train_second.py

[thenameisAnurag]

i have successfully trained the mode on train_first.py

this was the result in the train.logs

INFO:2025-09-25 17:50:53,200: Epoch [50/50], Step [240/243], Mel Loss: 0.28113, Gen Loss: 0.00000, Disc Loss: 0.00000, Mono Loss: 0.00000, S2S Loss: 0.00000, SLM Loss: 0.00000
INFO:2025-09-25 17:51:06,578: Validation loss: 0.275

now i have my config file with
epochs_1st: 0 epochs_2nd: 60 batch_size: 2 max_len: 500

and now i am facing this issue
`INFO:2025-09-30 15:24:19,919: Epoch [47/60], Step [20/486], Loss: nan, Disc Loss: nan, Dur Loss: nan, CE Loss: nan, Norm Loss: nan, F0 Loss: nan, LM Loss: nan, Gen Loss: nan, Sty Loss: nan, Diff Loss: nan, DiscLM Loss: 0.00000, GenLM Loss: 0.00000
INFO:2025-09-30 15:42:13,051: Epoch [47/60], Step [10/486], Loss: nan, Disc Loss: nan, Dur Loss: nan, CE Loss: nan, Norm Loss: nan, F0 Loss: nan, LM Loss: nan, Gen Loss: nan, Sty Loss: nan, Diff Loss: nan, DiscLM Loss: 0.00000, GenLM Loss: 0.00000

please suggest and help how can i solve this issue

i have
removed the
set_trace()

but still facing the issue what i need to do

[patriotyk]

You didn't train train_first.py properly because from your logs I see that GAN train(TMA_epoch parameter in your config seems to be greater or equals to epochs_1st) has newer started. Also it is possible to train to 0.24 validation loss before GAN train started. I recommend to increase epochs_1st to 200 and set TMA_epoch то 80. Also don't be confused when TMA_epoch start, mel loss should quickly go up then slowly begin decreasing.

[thenameisAnurag]

You didn't train train_first.py properly because from your logs I see that GAN train(TMA_epoch parameter in your config seems to be greater or equals to epochs_1st) has newer started. Also it is possible to train to 0.24 validation loss before GAN train started. I recommend to increase epochs_1st to 200 and set TMA_epoch то 80. Also don't be confused when TMA_epoch start, mel loss should quickly go up then slowly begin decreasing.

Thank you bro , i will implement and check the results