Dataset · mvp-lab
LLaVA-OneVision-2 Multimodal and Video Data
mvp-lab/LLaVA-OneVision-2-Data
Large collection of video clips, captions, and spatial reasoning tasks for training multimodal models.
What's inside
The official English training corpus for the LLaVA-OneVision-2 model family, comprising text, images, and video. The first repository section contains 60-second video snippets, text captions for clips of varying durations, and spatial reasoning data used during the mid-training stage. Due to its large scale, the data is split across two Hugging Face repositories.
What you can do with it
- Training and fine-tuning multimodal models (VLM) for video and image understanding
- Developing spatial reasoning and visual question answering (VQA) capabilities
- Tasks
- Video text → text, Visual question answering, Image text → text
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 59.3K
- Likes
- 51
How to load it
from datasets import load_dataset
ds = load_dataset("mvp-lab/LLaVA-OneVision-2-Data")