vwxyzjn/hh-rlhf-trl-style

--- dataset_info: features: - name: info struct: - name: id dtype: string - name: post dtype: string - name: title dtype: string - name: subreddit dtype: string - name: site dtype: string - name: article dtype: string - name: summaries list: - name: text dtype: string - name: policy dtype: string - name: note dtype: string - name: choice dtype: int32 - name: worker dtype: string - name: batch dtype: string - name: split dtype: string - name: extra struct: - name: confidence dtype: int32 - name: prompt dtype: string - name: chosen list: - name: content dtype: string - name: role dtype: string - name: rejected list: - name: content dtype: string - name: role dtype: string splits: - name: train num_bytes: 315969 num_examples: 50 - name: validation num_bytes: 325197 num_examples: 50 download_size: 150469 dataset_size: 641166 configs: - config_name: default data_files: - split: train path: data/train-* - split: validation path: data/validation-* --- # TRL's Anthropic HH Dataset We preprocess the dataset using our standard `prompt, chosen, rejected` format. ## Reproduce this dataset 1. Download the `tldr_preference.py` from the https://huggingface.co/datasets/vwxyzjn/hh-rlhf-trl-style/tree/0.1.0. 2. Run `python examples/datasets/tldr_preference.py --debug --push_to_hub`

原始信息汇总

数据集概述

数据集结构

主要特征

• info

  • id: 字符串类型
  • post: 字符串类型
  • title: 字符串类型
  • subreddit: 字符串类型
  • site: 字符串类型
  • article: 字符串类型

• summaries

  • text: 字符串类型
  • policy: 字符串类型
  • note: 字符串类型

• choice: 整数类型（int32）

• worker: 字符串类型

• batch: 字符串类型

• split: 字符串类型

• extra

  • confidence: 整数类型（int32）

• prompt: 字符串类型

• chosen

  • content: 字符串类型
  • role: 字符串类型

• rejected

  • content: 字符串类型
  • role: 字符串类型

数据集分割

• train

  • 大小: 315969 字节
  • 示例数量: 50

• validation

  • 大小: 325197 字节
  • 示例数量: 50

数据集大小

• 下载大小: 150469 字节
• 数据集总大小: 641166 字节