stalaei/sdft-science-distil

该数据集是一个用于问答和多项选择任务的NLP数据集，主要涉及化学领域。数据集名为SDFT Science (SciKnowEval Chemistry L-3) — distil format，是用于自我蒸馏微调（SDFT）论文的管道就绪版本。数据集包含训练集和评估集，分别有2674和507行数据。数据集的模式包括prompt、teacher_prompt和chat_template_kwargs三个列。数据集的来源是SciKnowEval Chemistry L-3，教师提示中的演示来自SDFT论文的train_data，由GPT-4o专家编写。评估标签来自eval_data。

This dataset is an NLP dataset for question-answering and multiple-choice tasks, primarily in the field of chemistry. The dataset is named SDFT Science (SciKnowEval Chemistry L-3) — distil format and is a pipeline-ready version used in the Self-Distillation Fine-Tuning (SDFT) paper. The dataset includes training and evaluation sets with 2674 and 507 rows of data, respectively. The schema of the dataset consists of three columns: prompt, teacher_prompt, and chat_template_kwargs. The dataset originates from SciKnowEval Chemistry L-3, with demonstrations in the teacher_prompt coming from the SDFT papers train_data, expert-written by GPT-4o. The evaluation labels are from eval_data.