meituan-longcat/R-HORIZON-AMC23

R-HORIZON数据集旨在评估和训练大型推理模型在长推理任务上的性能。该数据集包括一个用于评估模型在复杂、多步骤推理场景中性能的基准，以及用于提高模型在这些任务中能力的训练数据集。README文件还描述了数据集的构建过程、数据集结构，并提供了基准评估和训练数据集的说明。该数据集可在Hugging Face Hub上获取，包括来自数学、代码生成和代理应用的数据集。

The R-HORIZON dataset is designed to evaluate and train large reasoning models on long-horizon reasoning tasks. It includes a benchmark for evaluating model performance on complex, multi-step reasoning scenarios and training datasets for improving model capabilities in such tasks. The README describes the dataset construction process, dataset structure, and provides instructions for benchmark evaluation and training with the dataset. The dataset is available on the Hugging Face Hub and includes datasets from mathematics, code generation, and agent applications.