class DecoderModel(ModelBase) using padding token for prediction

bigscience-workshop / t-zero

Reproduce results and replicate training fo T0 (Multitask Prompted Training Enables Zero-Shot Task Generalization)

Apache License 2.0

457 stars 53 forks source link

class DecoderModel(ModelBase) using padding token for prediction #27

Closed richardbaihe closed 2 years ago

richardbaihe commented 2 years ago

Hi There,

In the t0/model.py's DecoderModel class, I noticed that there are padding tokens between the input_ids and labels. It seems padding tokens are used when predicting the labels. Is this an issue?

VictorSanh commented 2 years ago

fixed with #32 thanks to @thomasw21