asappresearch/gold

Official repository for "GOLD: Improving Out-of-Scope Detection in Dialogues using Data Augmentation"

gold's Issues

KeyError when using Roberta

If you want to use --model roberta, there will be a KeyError in the class BaseInstance in dataset,py, because the Roberta Tokenizer does not have token_type_ids.
Small workaround for single input sentences:
Replace self.segments = embed_data['token_type_ids'] with `self.segments = [0] * len(embed_data['input_ids'])``

A question about paper

Thanks for releasing your code. I'm interested in your paper. I have a question about your paper. When inference, you only consider a binary classifier to detect oos and in-domain intents or you also classify the in-domain intents to their classes.