Skip to content

Dataset · lmqg

Synthetic question-answer pairs for SQuADShifts

lmqg/qa_squadshifts_synthetic

Between 1 and 10 million synthetic English question-answer pairs for evaluating question generation models.

What's inside

The corpus contains between 1 and 10 million English text examples generated by fine-tuned question generation models on top of the lmqg/qa_squadshifts dataset. The test split represents the original validation split of the source set. The data is intended for evaluating question generation quality through question-answering (QAE) task solving.

What you can do with it

  • Evaluating the quality of question generation (QG) models using the QAE method
  • Supervised fine-tuning models for question answering under domain shift conditions
Tasks
Question answering
Languages
English
Size
1M–10M rows
Downloads this month
47.8K
Likes
4

How to load it

from datasets import load_dataset

ds = load_dataset("lmqg/qa_squadshifts_synthetic")