aniketr/hpdv2-test-no-embeds

---
dataset_info:
features:
- name: user_id
dtype: int64
- name: prompt
dtype: string
- name: image_left_id
dtype: string
- name: image_right_id
dtype: string
- name: pref
dtype: int64
- name: ranking_id
dtype: string
splits:
- name: train
num_bytes: 20770830
num_examples: 153000
download_size: 1465804
dataset_size: 20770830
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
task_categories:
- text-to-image
language:
- en
size_categories:
- 100K<n<1M
---
Constructing a preference dataset from the test split of [HPDv2](https://huggingface.co/datasets/ymhao/HPDv2).

Each row in the dataset is a unique ranking of preference on a single image pair and contains
1. `user_id`: unique identifier of the annotator
2. `prompt`: text prompt used to generate image candidates
3. `image_left_id`: id of left image in preference pair
4. `image_right_id`: id of right image in preference pair
5. `pref`: image preferred by annotator. 1 for left image, -1 for right image
6. `ranking_id`: unique identifier for the ranking pair

原始信息汇总

数据集概述

数据集信息

• 特征列表:

  • user_id: 用户ID，数据类型为int64
  • prompt: 文本提示，数据类型为string
  • image_left_id: 左侧图像ID，数据类型为string
  • image_right_id: 右侧图像ID，数据类型为string
  • pref: 偏好选择，数据类型为int64
  • ranking_id: 排名ID，数据类型为string

• 数据分割:

  • train: 训练集，包含20770830字节，153000个样本

• 下载大小: 1465804字节

• 数据集大小: 20770830字节

配置信息

• 默认配置:
  • 数据文件路径: data/train-*

任务类别

• text-to-image

语言

• en

数据集大小类别

• 100K<n<1M