Summary

  • Vectara, in conjunction with the University of Waterloo, has launched an open-source framework that assesses and benchmarks retrieval-augmented generation (RAG) systems.
  • RAG systems are increasingly popular in the enterprise sector as multi-step agentic systems, but they are difficult to evaluate objectively.
  • The Open RAG Eval framework uses automatic grading of “nuggets” of information — basic factual elements of responses — to allow organisations to test any RAG pipeline.
  • Four metrics are used to test the systems: hallucination detection, citation, auto nugget and UMBRELA ( Unifiied Method for Benchmarking Retrieval Evaluation with LLM Assessment)
  • TheOpen RAG Eval framework is available to download at https://github.

By Sean Michael Kerner

Original Article