AlignBench: Benchmarking Chinese Alignment of Large Language Models

TL;DR

AlignBench is a comprehensive benchmark designed to evaluate the alignment of Chinese large language models across multiple dimensions, utilizing human-verified data and an LLM-based evaluation approach for high reliability.

Contribution

This paper introduces AlignBench, the first multi-dimensional Chinese LLM alignment benchmark with a human-in-the-loop data curation pipeline and an LLM-as-Judge evaluation method.

Findings

AlignBench covers 683 real-scenario queries with verified references.

It employs a rule-calibrated LLM-as-Judge for reliable evaluation.

AlignBench has been adopted by top Chinese LLMs for alignment assessment.

Abstract

Alignment has become a critical step for instruction-tuned Large Language Models (LLMs) to become helpful assistants. However, the effective evaluation of alignment for emerging Chinese LLMs is still largely unexplored. To fill in this gap, we introduce AlignBench, a comprehensive multi-dimensional benchmark for evaluating LLMs' alignment in Chinese. We design a human-in-the-loop data curation pipeline, containing eight main categories, 683 real-scenario rooted queries and corresponding human verified references. To ensure the correctness of references, each knowledge-intensive query is accompanied with evidences collected from reliable web sources (including URLs and quotations) by our annotators. For automatic evaluation, our benchmark employs a rule-calibrated multi-dimensional LLM-as-Judge~\cite{zheng2023judging} approach with Chain-of-Thought to generate explanations and final ratings, ensuring high reliability and interpretability. All evaluation code, data, and LLM generations are available at \url{https://github.com/THUDM/AlignBench}. Since its release, AlignBench has been adopted by top (Chinese) LLMs for evaluating their alignment capabilities in Chinese, including ChatGLM, Qwen, DeepSeek, Yi, Baichuan, and Abab.