Jianshu001/arabic-daily-batch01-smoke20-rewriter

该数据集名为Batch 01 — Smoke 20 (Gemma-as-rewriter)，包含20条记录，用于验证Gemma-as-rewriter清理架构。采用了一种特定的重写方法，通过Gemma作为重写器来清理思考字段。数据集的架构包括用户消息字段（turn, role, text）和助手消息字段（turn, role, thinking, text）。数据集的语言为阿拉伯语，许可证为MIT，标签包括阿拉伯语、cascade-rewrite、gemma-4和gemma-as-rewriter，规模小于1K。

The dataset is named Batch 01 — Smoke 20 (Gemma-as-rewriter), containing 20 records for validating the Gemma-as-rewriter cleanup architecture. It employs a specific rewriting approach using Gemma as a rewriter to clean up the thinking field. The dataset schema includes user message fields (turn, role, text) and assistant message fields (turn, role, thinking, text). The dataset is in Arabic, licensed under MIT, tagged with Arabic, cascade-rewrite, gemma-4, and gemma-as-rewriter, and has a size category of less than 1K.