JamesXZ/tts-knowledge

该数据集包含用于研究推理模型在知识密集型任务中测试时缩放效果（如低推理努力、增加思考预算）的模型输出、推理轨迹和元数据。数据集基于SimpleQA和FRAMES两个基准，包含问题的原始文本、参考答案、模型生成的最终答案、推理轨迹（如果有）、评价标签、生成的令牌数、使用的模型以及测试时缩放设置。

This dataset contains model outputs, reasoning traces, and metadata for studying the effect of test-time scaling in reasoning models on knowledge-intensive tasks (e.g., low reasoning effort, increased thinking budget). It is based on the SimpleQA and FRAMES benchmarks and includes the original text of the questions, reference answers, the models final answers, reasoning traces (if available), evaluation labels, number of tokens generated, the model used, and the test-time scaling setting.