Dataset · Open-Bee
Bee multimodal image-text corpus
Open-Bee/Honey-Data-15M
15M English image-text pairs for training and evaluating open multimodal models.
What's inside
A multimodal corpus for image-text-to-text tasks containing English images and texts, around 15M examples. Collected in the Bee project via the HoneyPipe pipeline; the dataset card notes a newer version, Honey-Data-V2, which fixes community-reported defects. Data is available on Hugging Face.
What you can do with it
- SFT of a multimodal model for image-text-to-text
- Training a LoRA for an open MLLM
- Evaluating answer quality on images
- Tasks
- Image text → text
- Languages
- English
- Size
- 10M–100M rows
- Downloads this month
- 15K
- Likes
- 121
Models trained on it
- Open-Bee/Bee-8B-RLImage text → text 8.7K
- Open-Bee/Bee-8B-SFTImage text → text 2.1K
- Open-Bee/Bee-8B-Stage3Image text → text 29
- Open-Bee/Bee-8B-Stage2Image text → text 28
- axs2k/anubisReinforcement learning 0
- joanxprat/test_qualify_imagesDepth estimation 0
- Thando1221/DAunwell 0
- moreiraj93/jennyImage text → text 0
How to load it
from datasets import load_dataset
ds = load_dataset("Open-Bee/Honey-Data-15M")