metadata

language:
  - en
tags:
  - text2text-generation
datasets:
  - domenicrosati/QA2D
widget:
  - text: Where in the world is Carmen Sandiego. She is in Abruzzo
    example_title: Where is Carmen Sandiego?
  - text: Halifax is a city in which province. Nova Scotia
    example_title: A Halifact

Question-Answer to Statement Converter

A question answer pair to statement converter from https://github.com/jifan-chen/QA-Verification-Via-NLI

See:

@article{chen2021can,
  title={Can NLI Models Verify QA Systems' Predictions?},
  author={Chen, Jifan and Choi, Eunsol and Durrett, Greg},
  journal={EMNLP Findings},
  year={2021}
}

Note: I am not the maintainer or orginal author just keeping it here to use huggingface APIs to produce statements from question answer pair for downstream applications.

TL;DR:

We fine-tune a seq2seq model, T5-3B (Raffel et al., 2020), using the $(a, q, d)$ pairs annotated by Demszky et al. (2018).

Where a is answer, q is question, and d is declerative sentence (i.e. a statement).

See Appendex B.2 of Chen et al. for more.

Usage

The prompt should be {question} {seperator} {answer} where the seperator is </s>.

from transformers import AutoTokenizer,  AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained('domenicrosati/question_converter-3b')
model = AutoModelForSeq2SeqLM.from_pretrained('domenicrosati/question_converter-3b')

question = "Where in the world is Carmen Sandiego?"
answer = "She is in Abruzzo"

prompt = f'{question} </s> {answer}'
input_ids = tokenizer(prompt, return_tensors='pt').input_ids
output_ids = model.generate(input_ids)
responses = tokenizer.batch_decode(output_ids, skip_special_tokens=True)

['Carmen Sandiego is in Abruzzo.']