Dataset · openbmb
RLAIF-V data for trustworthy VLMs
openbmb/RLAIF-V-Dataset
English images and text for training and validating trustworthy multimodal models via RLAIF.
What's inside
An English image-text dataset containing 10k to 100k examples. It accompanies the RLAIF-V paper on open AI-feedback to enhance the reliability of GPT-4V level models and was used in MiniCPM-V 4.5. The exact storage format is not specified in the card.
What you can do with it
- Fine-tune LoRA for image-text-to-text and visual-question-answering.
- Evaluate the trustworthiness of a multimodal model.
- Train or test VLMs on VQA tasks.
- Tasks
- Image text → text, Visual question answering, Multimodal
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 2.6K
- Likes
- 227
Models trained on it
- openbmb/MiniCPM-V-4_5Image text → text 68.3K
- openbmb/MiniCPM-o-2_6Multimodal 38.8K
- openbmb/MiniCPM-V-4_5-ggufImage text → text 34.1K
- openbmb/MiniCPM-V-4Image text → text 30.4K
- openbmb/MiniCPM-V-2_6Image text → text 16.9K
- openbmb/MiniCPM-V-4_5-AWQImage text → text 13.5K
- openbmb/MiniCPM-Llama3-V-2_5Image text → text 11K
- openbmb/MiniCPM-V-4-ggufImage text → text 3.9K
How to load it
from datasets import load_dataset
ds = load_dataset("openbmb/RLAIF-V-Dataset")