GroUSE: A Benchmark to Evaluate Evaluators in Grounded Question Answering

TL;DR

This paper introduces GroUSE, a benchmark for evaluating the performance of judge models in grounded question answering, revealing current limitations and proposing improvements including fine-tuning models for better evaluation accuracy.

Contribution

The paper presents GroUSE, a comprehensive benchmark with unit tests for evaluating judge models, and demonstrates that fine-tuning Llama-3 enhances evaluation performance.

Findings

Existing evaluation frameworks often miss critical failure modes.

GPT-4's judgments are not fully indicative of practical judge performance.

Fine-tuning Llama-3 on GPT-4 traces improves evaluation accuracy.

Abstract

Retrieval-Augmented Generation (RAG) has emerged as a common paradigm to use Large Language Models (LLMs) alongside private and up-to-date knowledge bases. In this work, we address the challenges of using LLM-as-a-Judge when evaluating grounded answers generated by RAG systems. To assess the calibration and discrimination capabilities of judge models, we identify 7 generator failure modes and introduce GroUSE (Grounded QA Unitary Scoring of Evaluators), a meta-evaluation benchmark of 144 unit tests. This benchmark reveals that existing automated RAG evaluation frameworks often overlook important failure modes, even when using GPT-4 as a judge. To improve on the current design of automated RAG evaluation frameworks, we propose a novel pipeline and find that while closed models perform well on GroUSE, state-of-the-art open-source judges do not generalize to our proposed criteria, despite strong correlation with GPT-4's judgement. Our findings suggest that correlation with GPT-4 is an incomplete proxy for the practical performance of judge models and should be supplemented with evaluations on unit tests for precise failure mode detection. We further show that finetuning Llama-3 on GPT-4's reasoning traces significantly boosts its evaluation capabilities, improving upon both correlation with GPT-4's evaluations and calibration on reference situations.