A Small Model Asks Better: Fine-Tuning Query Clarification for Retrieval-Augmented Generation

Researchers introduce ClariEdge, a dataset of approximately 3,900 examples for query clarification in retrieval-augmented generation systems. A lightweight fine-tuned LoRA model achieves 96.6% classification accuracy and outperforms larger prompted models on generating clarification requests, with hard negatives identified as essential for distinguishing truly ambiguous queries.