Dataset · liuhaotian
Multimodal instructions on 150k examples
liuhaotian/LLaVA-Instruct-150K
150k image-instruction pairs from GPT-4 for visual instruction tuning and VQA.
What's inside
A set of multimodal instructions for training models to follow visual commands; data generated via the GPT-4-0314 API in April 2023. Contains about 150k examples in English, size ranging from 100k to 1M records. Designed for visual instruction tuning and developing capabilities of large multimodal models in the spirit of GPT-4 vision/language.
What you can do with it
- Fine-tune a LLaVA-like VLM
- Perform SFT for visual question answering
- Evaluate following visual instructions
- Tasks
- Visual question answering, Question answering
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 12.7K
- Likes
- 650
Models trained on it
- llava-hf/llava-1.5-7b-hfImage text → text 632.8K
- openbmb/MiniCPM-V-2Visual question answering 10K
- TheBloke/CausalLM-7B-GGUFText generation 6.6K
- TheBloke/CausalLM-14B-GGUFText generation 4.4K
- mlx-community/llava-1.5-7b-4bitImage text → text 2.9K
- bczhou/tiny-llava-v1-hfImage text → text 2.6K
- tastypear/CausalLM-7B-DPO-alpha-GGUFText generation 2.4K
- xtuner/llava-internlm2-20bImage text → text 2.2K
How to load it
from datasets import load_dataset
ds = load_dataset("liuhaotian/LLaVA-Instruct-150K")