EpistemeAI/reasoning-trace-gpt-oss-20b

该数据集包含了由OpenAI的gpt-oss-20b模型生成的细致推理轨迹。它包括指令、输入、推理、答案等特征，以及如生成的原始推理轨迹、模型ID和时间戳等元数据。数据集分为训练集，有特定的下载大小和总数据集大小。该数据集旨在用于大规模自回归变换器的调试、错误分析和解释性研究。数据集用英语描述，重点介绍其在探索开放权重模型的认识论方面的应用，并提供了Python的示例用法。还描述了实体类型，并提供了数据集的联系方式和引用信息。

This dataset contains fine-grained reasoning traces generated by OpenAIs gpt-oss-20b model. It includes features such as instruction, input, reasoning, answer, and metadata like generated raw reasoning trace, model ID, and timestamp. The dataset is split into a training set and has a specific download size and total dataset size. It is designed for debugging, error analysis, and interpretability research on large-scale autoregressive transformers. The dataset is described in English with a focus on its use for exploring the epistemology of open-weight models and includes example usage in Python. The entity types are described, and contact information along with citation information for the dataset is provided.