Cohesive Long-form Answers from Passages in Natural Questions for Retrieval Augmented Generation (RAG) Systems

CLAPNQ is a benchmark dataset for evaluating the full Retrieval Augmented Generation (RAG) pipeline, including retrieval, generation, and the combined RAG system. The dataset provides long, concise, complete, and cohesive answers that are grounded in a single gold passage, as well as unanswerable questions.