Dataset · wyu1
Multimodal SFT Dataset with Multi-Image and Text
wyu1/Leopard-Instruct
925K English instructions emphasizing multi-image and dense text tasks.
What's inside
Contains 925,000 image-text pairs for English SFT training, of which 739,000 are specifically designed for multi-image and heavy text scenarios. Authors trained Leopard-LLaVA and Leopard-Idefics2 models on these data.
What you can do with it
- Fine-tune multimodal models (VLM) for document, screenshot, and infographic analysis
- Train models to reason across multiple images in a single context
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 35.5K
- Likes
- 64
How to load it
from datasets import load_dataset
ds = load_dataset("wyu1/Leopard-Instruct")