ABench-Physics
收藏资源简介:
ABench-Physics是一个旨在严格评估大型语言模型(LLM)物理推理和泛化能力的新型基准数据集。它由两部分组成:Phy A,一个包含400个静态的高难度研究生或奥林匹克级别的问题;Phy B,一个包含100个动态问题的子集,配备了一个自动变化引擎,以测试模型在不同条件下的鲁棒性。所有问题都需要精确的数值答案,具有严格的格式和公差限制。ABench-Physics提供了挑战性和诊断性的框架,用于推进LLM中的科学推理。
ABench-Physics is a novel benchmark dataset designed to rigorously evaluate the physical reasoning and generalization capabilities of large language models (LLMs). It consists of two components: Phy A, which contains 400 challenging static graduate-level or olympiad-level physics problems; and Phy B, a subset of 100 dynamic problems equipped with an automated variation engine to test the model's robustness under diverse conditions. All problems require precise numerical answers, with strict formatting and tolerance constraints. ABench-Physics provides a challenging and diagnostic framework for advancing scientific reasoning in LLMs.

- 1ABench-Physics: Benchmarking Physical Reasoning in LLMs via High-Difficulty and Dynamic Physics Problems浙江大学、蚂蚁集团 · 2025年



