search

patil-suraj / longbart

A long version of BART model based on Longformer model
23 stars 5 forks source link

Longbart is broken #3

Closed michaelroyzen closed 4 years ago

michaelroyzen commented 4 years ago

I am trying to run your code as-is and am having tensor mismatch issues. Even after the conversion, it still seems to be asking for a max input of 1024.

I've attached a Colab notebook where I show my steps. Please take a look @patil-suraj and let me know how I can fix it. https://colab.research.google.com/drive/1cdYmC9J4RhAahUteAdl2aikrAW6UXOnk

Thanks!

michaelroyzen commented 4 years ago

Never mind, my issue was using BartForConditionalGeneration, not LongBartForConditionalGeneration.