Why DeepCopy automatically remove the token id before <endoftext> token?

XiangLi1999 / PrefixTuning

Prefix-Tuning: Optimizing Continuous Prompts for Generation

868 stars 158 forks source link

Why DeepCopy automatically remove the token id before <endoftext> token? #23

Closed 14H034160212 closed 2 years ago

14H034160212 commented 2 years ago

Hi,

I am curious about how deepcopy has been used in the language_model.py. Cause I got -100 for those input tokens before the first . Does anyone know how it is going? Thanks a lot.

14H034160212 commented 2 years ago

I got the answer.