Dataset · BAAI
Tens of millions of multimodal instructions
BAAI/Infinity-MM
A large multimodal instruction dataset from BAAI: images with QA pairs in English and Chinese.
What's inside
Infinity-MM is a large-scale multimodal instruction dataset from BAAI containing tens of millions of examples, filtered for quality and deduplicated. Synthetic data was generated using open models and a labeling system with detailed image captions and diverse questions. Languages are English and Chinese.
What you can do with it
- Train VLMs on images and instructions
- Fine-tune image-to-text models
- Evaluate multimodal models on en/zh
- Tasks
- Image → text
- Languages
- English, Chinese
- Size
- 10M–100M rows
- Downloads this month
- 13.5K
- Likes
- 120
- Access
- on request on Hugging Face
Models trained on it
- NovaSearch/jasper_en_vision_language_v1 715
- BAAI/Aquila-VL-2B-llava-qwenVisual question answering 233
- vincentkaufmann/gpt-oss-20b-vision-previewImage text → text 115
- LiAutoAD/Ristretto-3BImage text → text 74
- TULLUS/Aquila-VL-2B-llava-qwenVisual question answering 26
- FlashVL/FlashVL-2B-StaticImage text → text 18
- FlashVL/FlashVL-2B-DynamicImage text → text 17
- alexlimh/jasper_en_vision_language_v1 16
How to load it
from datasets import load_dataset
ds = load_dataset("BAAI/Infinity-MM")