ahmad21omar/SLR-Bench-French

SLR-Bench-French 是一个用于评估和训练大型语言模型（LLM）在逻辑推理方面的法语数据集。它包含具有不同复杂性的自然语言提示、可执行验证程序和潜在的真实规则。数据集的结构包括 20 个复杂度级别，分为四个层级（基本、简单、中等、困难）。数据集还提供了自动生成具有可控复杂性和新逻辑规则的归纳推理任务的功能，并支持通过课程学习评估和训练模型。SLR-Bench-French is a dataset designed for evaluating and training large language models (LLMs) in logical reasoning using a curriculum of tasks with varying complexity. It includes natural language prompts, executable validation programs, and latent ground-truth rules. The datasets structure consists of 20 complexity levels grouped into four tiers (basic, easy, medium, hard). SLR-Bench-French also features the ability to automatically generate inductive reasoning tasks with controllable complexity and novel logic rules, supporting evaluation and training of models via curriculum learning.