Skip to content

Dataset · Rapidata

Text-to-image human preference dataset

Rapidata/text-to-image-human-feedback-benchmark

Largest open human preference dataset for text-to-image models with 7.7M votes.

What's inside

The dataset contains pairwise image comparisons from 57 models across 1,514 prompts, gathered into 890,698 pairs with a total of 7.7M votes in English. Created by Rapidata for Benchmark.ai using the Python SDK. Presented as pairwise preferences for RLHF, DPO, and reward models.

What you can do with it

  • Training reward models for text-to-image systems
  • Optimizing diffusion models using DPO and RLHF
  • Benchmarking and evaluating the quality of generative image models
Tasks
Text → image
Languages
English
Size
100K–1M rows
Downloads this month
11.5K
Likes
26

How to load it

from datasets import load_dataset

ds = load_dataset("Rapidata/text-to-image-human-feedback-benchmark")