Dataset · HuggingFaceM4
Giant Collection of Images and Dialogues for VLMs
HuggingFaceM4/FineVisionMax
24.3 million interleaved image-text examples in English and Chinese for training multimodal models.
What's inside
It contains 17.3 million images, 24.3 million examples, and 88.9 million turns in English and Chinese (totaling 9.5 billion response tokens). This is a unified and interleaved version of the original FineVision dataset by the HuggingFaceM4 team, prepared specifically for convenient streaming loading. The data is intended for the image-text-to-text task.
What you can do with it
- Pre-training and fine-tuning open Vision-Language Models (VLMs)
- Streaming training of multimodal neural networks without downloading the entire corpus to disk
- Tasks
- Image text → text
- Languages
- English, Chinese
- Size
- 10M–100M rows
- Downloads this month
- 291.9K
- Likes
- 48
How to load it
from datasets import load_dataset
ds = load_dataset("HuggingFaceM4/FineVisionMax")