DSTI/traffic-accidents-reports-kd-smollm2-360M-7k

事故报告知识蒸馏数据集（单段式），用于从结构化事实生成单一自然段的交通事故/事件报告。该数据集结合了人类生成的目标数据和教师模型产生的软目标数据，以支持知识蒸馏。数据集包括训练和评估两部分，评估部分仅包含黄金标准数据，以进行无偏评估。该数据集基于zBotta/traffic-accidents-reports-5k数据集，并使用zBotta/smollm2-accident-reporter-360m模型生成教师目标。

Accident Reporting KD Dataset (One-Paragraph) for generating a single paragraph of accident/incident reports from structured facts. This dataset combines human-generated targets with soft targets produced by a teacher model to support knowledge distillation. It includes training and evaluation splits, with the evaluation split being gold-only for unbiased evaluation. The dataset is based on the zBotta/traffic-accidents-reports-5k dataset and uses the zBotta/smollm2-accident-reporter-360m model for generating teacher targets.