Astris/toxic-dpo-v0.2-embedded

--- dataset_info: features: - name: prompt dtype: string - name: chosen dtype: string - name: rejected dtype: string - name: id dtype: string - name: chosen_embed sequence: float16 - name: rejected_embed sequence: float16 splits: - name: train num_bytes: 2625194 num_examples: 541 download_size: 2133832 dataset_size: 2625194 configs: - config_name: default data_files: - split: train path: data/train-* --- unalignment's Toxic-DPO-V0.2 dataset, coupled with embeddings of the 'chosen' and 'rejected' columns. The model used for embeddings was 'jinaai/jina-embeddings-v2-base-es'. To ensure that the responses were contextualized, each response was run through the model with its accompanying prompt. Then, to pool the embeddings, only embeddings of the response were included.

原始信息汇总

数据集概述

数据集特征

• prompt: 数据类型为字符串。
• chosen: 数据类型为字符串。
• rejected: 数据类型为字符串。
• id: 数据类型为字符串。
• chosen_embed: 数据类型为序列，具体为float16。
• rejected_embed: 数据类型为序列，具体为float16。

数据集分割

• 训练集（train）:
  • 数据量：2625194字节
  • 示例数量：541个

数据集大小

• 下载大小：2133832字节
• 数据集总大小：2625194字节

配置信息

• 配置名称: default
• 数据文件:
  • 分割类型: 训练集（train）
  • 路径: data/train-*