Multi-state Bar Examination (MBE) questions
收藏资源简介:
本研究使用了由阿尔伯特·路德维希斯·弗莱堡大学和ELLIS学院共同收集的1,514条关于多州律师资格考试(MBE)的问题数据集。该数据集由网上学习资料整理而来,包含了问题正文、四个可选答案以及正确答案的解释。为了提高模型对法律问题的推理能力,研究团队采用了监督微调(SFT)的方法对LLama 2 7B和LLama 3 8B模型进行训练。数据集经过精心整理,部分数据被转化为IRAC(问题、规则、应用、结论)格式,以指导模型进行结构化的推理过程。该数据集旨在帮助小型语言模型在法律问题回答方面达到更接近人类的性能水平。
This study employs a dataset of 1,514 Multistate Bar Examination (MBE) questions jointly collected by the University of Freiburg (Albert-Ludwigs-Universität Freiburg) and the ELLIS Institute. Sourced from curated online learning materials, the dataset includes the full question text, four multiple-choice options, and explanatory notes for the correct answers. To enhance the model's legal reasoning capabilities, the research team adopted Supervised Fine-Tuning (SFT) to train the LLaMA 2 7B and LLaMA 3 8B models. The dataset was meticulously curated further, with a subset of the data converted into the IRAC (Issue, Rule, Application, Conclusion) format to guide the model through structured reasoning processes. The core objective of this dataset is to assist small language models in achieving performance that closely matches that of human experts in legal question answering tasks.

- 1A Llama walks into the 'Bar': Efficient Supervised Fine-Tuning for Legal Reasoning in the Multi-state Bar Exam阿尔伯特·路德维希斯·弗莱堡大学,德国;ELLIS学院,图宾根,德国 · 2025年



