question_generation

Why we included this project

This repo tackles question generation the practical way: given a paragraph, the model writes questions for it. Everything is built on transformer sequence-to-sequence models, mainly T5, and the author walks through the main variations, from answer-aware generation, where you hand the model the answer it should ask about, to end-to-end generation that needs no answer supervision at all, plus a multitask QA-QG setup. The scripts for data processing, training, and inference are kept deliberately simple, so the codebase works as a readable reference even if you plan to bring your own model, and the included Colab notebook gets you training on a GPU quickly. Anyone building quiz or study tools from plain text will find these recipes easy to adapt without pulling in a heavy pipeline.