asahi417/lm-question-generation

Supports three distinct QAG architectures—Pipeline, Multitask, and End2end—that trade off between generation volume and inference speed, all built on transformer language models (T5) and hosted on HuggingFace. Beyond QA pair generation, enables conditional question generation given answer spans, model fine-tuning on custom datasets, and REST API deployment. Covers nine languages with unified APIs and includes an online service (AutoQG) for zero-setup inference.

372 stars. No commits in the last 6 months.