Skip to content

Dataset · openbmb

Multimodal preferences for RLHF

openbmb/RLHF-V-Dataset

A small set of multimodal feedback for aligning vision-language models.

What's inside

RLHF-V-Dataset is a set of multimodal feedback for training vision-language models. The card specifies image/text modalities and English; volume is 1k to 10k examples. The data was used in MiniCPM-V 2.0; 7B and 12B models are available nearby.

What you can do with it

  • RLHF/RLAIF for vision-language models
  • Training reward models for VLM
  • Fine-tuning models like MiniCPM-V
Tasks
Text generation, Visual question answering
Languages
English
Size
1K–10K rows
Downloads this month
1.5K
Likes
87

How to load it

from datasets import load_dataset

ds = load_dataset("openbmb/RLHF-V-Dataset")