Model description

This is an mt5-base model, finetuned to generate questions using TyDi QA dataset. It was trained to take the context and answer as input to generate questions.

Overview

Language model: mT5-base
Language: Arabic, Bengali, English, Finnish, Indonesian, Korean, Russian, Swahili, Telugu
Task: Question Generation
Data: TyDi QA

Intented use and limitations

One can use this model to generate questions. Biases associated with pre-training of mT5 and TyDiQA dataset may be present.

Usage

One can use this model directly in the PrimeQA framework as in this example notebook.

Or

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tokenizer = AutoTokenizer.from_pretrained("PrimeQA/mt5-base-tydi-question-generator")
model = AutoModelForSeq2SeqLM.from_pretrained("PrimeQA/mt5-base-tydi-question-generator")

def get_question(answer, context, max_length=64):
  input_text = answer +" <<sep>> " + context
  features = tokenizer([input_text], return_tensors='pt')
  output = model.generate(input_ids=features['input_ids'], 
               attention_mask=features['attention_mask'],
               max_length=max_length)
  return tokenizer.decode(output[0])
context = "শচীন টেন্ডুলকারকে ক্রিকেট ইতিহাসের অন্যতম সেরা ব্যাটসম্যান হিসেবে গণ্য করা হয়।"
answer = "শচীন টেন্ডুলকার"
get_question(answer, context)
# output: ক্রিকেট ইতিহাসের অন্যতম সেরা ব্যাটসম্যান কে?

Citation

@inproceedings{xue2021mt5,
  title={mT5: A Massively Multilingual Pre-trained Text-to-Text Transformer},
  author={Xue, Linting and Constant, Noah and Roberts, Adam and
          Kale, Mihir and Al-Rfou, Rami and Siddhant, Aditya and
          Barua, Aditya and Raffel, Colin},
  booktitle={Proceedings of the 2021 Conference of the North American
             Chapter of the Association for Computational Linguistics:
             Human Language Technologies},
  pages={483--498},
  year={2021}
}