Modifying learning rate for fine tuning

[stevenjnovotny]

I am attempting to fine tune a ljspeech model using 26hrs of voice data. I want to tune slowly, thus I modified the learning rate in the lightning.py file. I also added a LearningRateMonitor as a callback for the trainer. In addition, I set auto_lr_find to False.

What happens:

• the version yaml files shows the modified value for lr correctly
• the lr monitor functions correctly
• no matter what I set as the lr, the monitor shows 1.25e-4 as the initial learning rate

Am I setting the lr incorrectly? Is there a better way to modify the learning rate for slow fine tuning?