T5 model returns tokens beginning a hash #. How can I get it to do it consistently?
I am actually trying to fine tune flan-t5-base to predict hashtags. After only one epoch I called the model to see if it was in working order. I am getting a range of outputs, I don’t understand why. These are some outputs for the same sample after multiple calls –
Transfomer model for implicitly learning text-based language style changes
I’ve been using Google’s t5-small, but I want to improve.