Dataset · lmms-lab
LLaVA-OneVision multimodal data
lmms-lab/LLaVA-OneVision-Data
Images and text in English and Chinese for training and evaluating multimodal models.
What's inside
LLaVA-OneVision-Data is a multimodal dataset with images and text in English and Chinese. The card mentions subsets VisualWebInstruct (filtered), ureader_kg, and ureader_qa, uploaded in Hugging Face format with JSON and image archives. Size ranges from 1 million to 10 million examples.
What you can do with it
- Training LLaVA-like vision-language models
- LoRA fine-tuning for VQA
- Evaluating multimodal assistants
- Languages
- English, Chinese
- Size
- 1M–10M rows
- Downloads this month
- 32.4K
- Likes
- 246
Models trained on it
- HuggingFaceTB/SmolVLM2-500M-Video-InstructImage text → text 1.7M
- HuggingFaceTB/SmolVLM2-2.2B-InstructImage text → text 194.6K
- llava-hf/llava-onevision-qwen2-0.5b-ov-hfImage text → text 188K
- llava-hf/llava-onevision-qwen2-7b-ov-hfImage text → text 49.8K
- HuggingFaceTB/SmolVLM2-256M-Video-InstructImage text → text 44.7K
- lmms-lab/llava-onevision-qwen2-7b-ovText generation 29K
- lmms-lab/LLaVA-Video-7B-Qwen2Video text → text 12.3K
- DAMO-NLP-SG/VideoLLaMA3-7BVideo text → text 5.5K
How to load it
from datasets import load_dataset
ds = load_dataset("lmms-lab/LLaVA-OneVision-Data")