Dataset · m-a-p
Massive PIN-200M multimodal corpus
m-a-p/PIN-200M
About 200 million English and Chinese examples for training multimodal models.
What's inside
A reduced version of the research-oriented PIN dataset, combining paired and interleaved multimodal documents in English and Chinese. It contains about 200 million samples in a specialized PIN format with quality evaluation signals and occupies roughly 312 TB.
What you can do with it
- Pre-training large multimodal language models (VLM)
- Filtering and selecting high-quality data using built-in signals
- Research in processing interleaved text and images
- Languages
- English, Chinese
- Size
- 10K–100K rows
- Downloads this month
- 57.1K
- Likes
- 34
How to load it
from datasets import load_dataset
ds = load_dataset("m-a-p/PIN-200M")