ThinkBench: Dynamic Out-of-Distribution Evaluation for Robust LLM Reasoning

Evaluating large language models (LLMs) poses significant challenges, particularly due to issues of data contamination and the leakage of correct answers. To address these challenges, we introduce ThinkBench, a novel evaluation framework designed to robustly evaluate the reasoning capability of LLMs. ThinkBench proposes a dynamic data generation method for constructing out-of-distribution (OOD) datasets and offers an OOD dataset that contains 2, 912 samples drawn from reasoning tasks. ThinkBench unifies the evaluation of reasoning models and non-reasoning models. We evaluate 16 LLMs and 4 PRMs under identical experimental conditions and show that most of the LLMs' performance are far from robust and they face a certain level of data leakage. By dynamically generating OOD datasets, ThinkBench effectively provides a reliable evaluation of LLMs and reduces data contamination impact. Our data and codes are available at https: //github. com/huangshulin123/ThinkBench.

Authors

• Shulin Huang Shenzhen International Graduate School, Tsinghua University
• Linyi Yang
• Yan Song
• Shawn Chen
• Leyang Cui Tencent AI lab
• Ziyu Wan
• Qingcheng Zeng
• Ying Wen School of Communication and Electronic Engineering, East China Normal University, Shanghai, China.
• Kun Shao Huawei
• Weinan Zhang Shanghai Jiao Tong University
• Jun Wang
• Yue Zhang

Keywords

No keywords are indexed for this paper.