SLMEval: Entropy-Based Calibration for Human-Aligned Evaluation of Large Language Models

TL;DR

SLMEval introduces an entropy-based calibration method for large language model evaluators, improving alignment with human judgments in real-world tasks and significantly reducing evaluation costs.

Contribution

It presents a novel entropy maximization calibration technique that generalizes better to open-ended tasks compared to existing methods.

Findings

SLMEval achieves a Spearman correlation of 0.57 with human judgments.

SLMEval outperforms G-Eval, which has negative correlation.

Evaluation costs are reduced by 5-30 times.

Abstract

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.