SLMEval: Entropy-Based Calibration for Human-Aligned Evaluation of Large Language Models
This addresses the need for more reliable and cost-effective evaluation of large language models in practical applications, though it is incremental as it builds on existing calibration techniques.
The paper tackles the problem that state-of-the-art calibrated evaluators for large language models often fail to align with human judgments in real-world, open-ended tasks, and proposes SLMEval, an entropy-based calibration method that achieves strong correlation with human evaluations, such as a Spearman correlation of 0.57 on one task, while reducing evaluation costs by 5-30x.
The LLM-as-a-Judge paradigm offers a scalable, reference-free approach for evaluating language models. Although several calibration techniques have been proposed to better align these evaluators with human judgment, prior studies focus primarily on narrow, well-structured benchmarks. As a result, it remains unclear whether such calibrations generalize to real-world, open-ended tasks. In this work, we show that SOTA calibrated evaluators often fail in these settings, exhibiting weak or even negative correlation with human judgments. To address this, we propose SLMEval, a novel and efficient calibration method based on entropy maximization over a small amount of human preference data. By estimating a latent distribution over model quality and reweighting evaluator scores accordingly, SLMEval achieves strong correlation with human evaluations across two real-world production use cases and the public benchmark. For example, on one such task, SLMEval achieves a Spearman correlation of 0.57 with human judgments, while G-Eval yields a negative correlation. In addition, SLMEval reduces evaluation costs by 5-30x compared to GPT-4-based calibrated evaluators such as G-eval.