davanstrien/otb-cascade-phaseB-2tier

该数据集是一个由classify-and-augment工具生成的LLM标注数据集，主要用于分类任务。数据集包含200条输入行和200条输出行，使用了两级模型（LiquidAI/LFM2.5-1.2B-Instruct和Qwen/Qwen3-4B-Instruct-2507）进行标注。标签分为jim_crow和no_jim_crow两类，其中jim_crow标签有14条，no_jim_crow标签有186条。数据集还包含了级联审计的详细信息，如模型处理情况、熵值、解析失败百分比以及层级间的一致性等。

This dataset is an LLM-annotated dataset generated by the classify-and-augment tool, primarily used for classification tasks. The dataset contains 200 input rows and 200 output rows, annotated using a two-tier model (LiquidAI/LFM2.5-1.2B-Instruct and Qwen/Qwen3-4B-Instruct-2507). The labels are divided into jim_crow and no_jim_crow, with 14 jim_crow labels and 186 no_jim_crow labels. The dataset also includes detailed information on cascade audits, such as model processing, entropy, parse failure percentages, and inter-tier agreement.