GeoThoughts数据集
收藏资源简介:
GeoThoughts数据集是一个包含两个子集的全面几何推理语料库:Geo-Thought-6K包含6,243个样本,其增强版本Geo-Thought-Augmented-10K包含10,834个样本。每个条目都包括视觉描述、逐步解决方案、明确的推理链、反思步骤和最终答案。该数据集用于训练GeoThought-MLLM,这是一个能够在解决问题时生成详细思维过程的数学推理多模态模型。该模型在几何任务中的表现优于现有的基准,证明了使用我们的Chain-of-Thought数据集训练可以改善几何推理能力。
The GeoThoughts dataset is a comprehensive geometric reasoning corpus consisting of two subsets: Geo-Thought-6K contains 6,243 samples, while its enhanced version Geo-Thought-Augmented-10K includes 10,834 samples. Each entry includes a visual description, step-by-step solution, explicit reasoning chain, reflective steps, and the final answer. This dataset is used to train GeoThought-MLLM, a mathematical reasoning multimodal model capable of generating detailed thought processes during problem-solving. This model outperforms existing benchmarks on geometric tasks, demonstrating that training with our Chain-of-Thought dataset can enhance geometric reasoning capabilities.




