marin-community/open-thoughts-4-30k-math-qwen3-32b-annotated

这个数据集是一个包含29,963个样本的子集，源自于marin-community/open-thoughts-4-math-qwen3-32b-annotated数据集，而后者又来源于mlfoundations-dev/hero_run_4_math数据集。数据集的主要特点是包含了由Qwen/Qwen3-32B模型生成的文本响应，这些响应是在温度为0.8和最大输出令牌为7500的条件下生成的。数据集的结构包括多个列，如instruction_seed（原始数学问题文本）、_source（原始数据集来源）、generated_text（由Qwen3-32B生成的包含思考链的响应）等。

This dataset is a 29,963 sample subset of marin-community/open-thoughts-4-math-qwen3-32b-annotated, originally derived from mlfoundations-dev/hero_run_4_math curated by the OpenThoughts4 team. The dataset features responses generated by the Qwen/Qwen3-32B model under the conditions of temperature = 0.8 and max output tokens = 7500. The dataset structure includes columns such as instruction_seed (original math problem text), _source (origin dataset), generated_text (response including chain-of-thought generated by Qwen3-32B), etc.