A general-purpose evaluation dataset for scientific paper question-answering systems. Contains question-answer pairs generated from scientific papers across multiple domains.
Originally built for evaluating Open Paper, but designed to be useful for any research paper QA or RAG system. The dataset-generation pipeline, benchmark harness, citation matcher, and LLM judge are open-source at github.com/khoj-ai/openpaper/tree/master/server/evals.
