Dataset · nyu-visionx
Cambrian-10M for Multimodal Instruction Tuning
nyu-visionx/Cambrian-10M
A large English dataset of visual question-answering and instructions for training multimodal models.
What's inside
A large dataset for multimodal model instruction tuning: visual question-answering and image instructions. According to metadata, it contains between 1M and 10M English examples. It was created to address the shortage of high-quality multimodal instruction-tuning data and preserve the language capabilities of MLLMs.
What you can do with it
- Training VLMs or MLLMs to follow instructions
- Fine-tuning LoRA for visual question answering
- Testing multimodal chat
- Tasks
- Visual question answering, Question answering
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 54.5K
- Likes
- 132
Models trained on it
- JulietChoo/VisionSelector-Qwen2.5-VL-3BImage text → text 81
- JulietChoo/VisionSelector-Qwen2.5-VL-7BImage text → text 36
- junha1125/gaze-attention-qwen3-4b-videoVideo text → text 21
- junha1125/gaze-attention-qwen3-4b-384-onevisionImage text → text 17
- junha1125/gaze-attention-qwen3-4b-384Image text → text 15
- JulietChoo/VisionSelector-LLaVA-OV-1.5-8BImage text → text 13
- junha1125/gaze-attention-qwen3-4b-512Image text → text 13
- jeduardogruiz/Mixtral_etherReinforcement learning 0
How to load it
from datasets import load_dataset
ds = load_dataset("nyu-visionx/Cambrian-10M")