Enrique Manjavacas

Hey. Nice you got it wrapped up. I have two comments on this. - I'd really try to avoid duplicating classes/scripts just because of the optimization (in this case it's...

Actually better to use "devices" because it wouldn't have to be a cuda device after all. On Sat, Jun 27, 2020 at 5:44 PM Thibault Clérice wrote: > *@PonteIneptique* commented...

Hey! This went missing... sorry! In the current setup, if you use the LM loss, you are effectively leveraging contextual embeddings (even before they were cool). Of course you are...

Yes, you need input word embeddings if you use context word.