awkward training process loss

[mo-ezzat]

Hello, thank you for your work I'm a bit new in the machine learning so I'm a bit confused, I've tried the following:

• I trained a model using LJSpeech-1.1, the only parameter I edited at that training was changing max_mel_frames = 900 to 400
• I then replaced the audio files with my data, I changed max_mel_frames back to 900, and changed outputs_per_step to 2 now the avg loss is negative and so is the average as in the pic can anyone tell is it doing fine or not? another question, during the first training it used to produce wav files at logs-Tacotron-2/, it doesn't at the 2nd

any help?!

[SongoMen]

I have same situation as you have. My avg loss is also around -3. Did you fix it somehow or is it normal ?

[graham-eisele]

same problem