LeakPro/synthetic_tab

Synthetic Text Anonymization Benchmark (TAB)数据集是一个合成的文本匿名化基准数据集，包含了1014个由欧洲人权法院（ECHR）生成的案例。该数据集是为了微调QLoRA（Quantized Low-Rank Adapter）模型而创建的，这些模型是基于mistralai/Mistral-7B-Instruct-v0.3。每个案例都是一个JSON文档，包含合成案例文本、生成器是否完成案例序列的标记以及文本的令牌长度。The Synthetic Text Anonymization Benchmark (TAB) dataset is a collection of 1,014 synthetic cases generated from the European Court of Human Rights (ECHR). This dataset was created to fine-tune QLoRA (Quantized Low-Rank Adapter) models based on mistralai/Mistral-7B-Instruct-v0.3. Each case is a JSON document that includes the synthetic case text, a flag indicating whether the generator finished the case sequence, and the token length of the text.