Fine tuning with 0.4.1

Deep Speech
#1

I am trying to fine tune with 0.4.1. A couple of things that I noted are, it has an ‘–early_stop’ and ‘–earlystop_nsteps’ flag. Does this mean I don’t need to observe the loss of each epoch manually and it will automatically stop when loss starts to increase or in other words when the learning stops? Also, it is stated in ‘Help’ for ‘–early_stop’ that

Make sure that dev FLAG is enabled for this to work

But, I couldn’t find a dev FLAG. So, what does this mean?

#2

The --dev_files etc. flags?

#3

It just says dev flag. That’s why the confusion. Because how can we even train without specifying —dev-files