Skip to content

Dataset · PKU-Alignment

Multimodal model alignment dataset

PKU-Alignment/align-anything

Preference score data and text feedback for aligning models across text, images, audio, and tables.

What's inside

Contains any-to-any task examples with detailed preference annotations and language feedback. The dataset includes audio, images, tables, and text, with a size of 10k to 100k entries. Materials prepared by the PKU-Alignment team to advance multimodal foundational models.

What you can do with it

  • Aligning (RLHF / DPO) multimodal models with human preferences
  • SFT fine-tuning models on text and image instructions
  • Training reward models with text feedback generation
Tasks
Multimodal
Size
10K–100K rows
Downloads this month
7.1K
Likes
50

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("PKU-Alignment/align-anything")