< Back to IRCAM Forum

Question: how long does it take for rave to finish training

Hey all, ML newbie here (please go easy on me haha).
Running an instance of RAVE in google colabs.

As many of you know the usage limits and timeout become a real burden, I am on day 2 of training my run. And currently on epoch 1228.

Due to limited documentation I am unsure of when to expect the training to finish, or even if a finite amount of epochs is even instantiated in the code. When can i expect training to be done.


Hi 50nny,

I’m not an advanced user, but my experience with colabs is that timeouts are a necessary step in the process. I trained for 1 week, got to 2 million steps (I find that steps, for me, are a better measure than epochs), and still needed some more training. It varies a lot from dataset to dataset.

I suggest viewing the discussions tab on the github page. This is the best current source of documentation, and they also discuss when to expect the training to be done.

1 Like