Dataset · FreedomIntelligence
Visual instructions and QA in English
FreedomIntelligence/ALLaVA-4V
Images with captions and complex question-answers in English for training multimodal models.
What's inside
Image and text modality dataset in English, sized between 100k and 1M examples. Assembled via the ALLaVA-4V pipeline: GPT-4V generated captions and complex reasoning QA for LAION and Vision-FLAN, while Wizard answers were regenerated by GPT-4-Turbo. The card describes the initial sources and naming structure, but the file format is not specified.
What you can do with it
- Fine-tuning vision-language model on image-text pairs
- SFT of multimodal LoRA for visual questions
- Model evaluation on complex visual QA
- Tasks
- Question answering, Text generation
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 3.4K
- Likes
- 99
Models trained on it
- Geraldxm/Qwen2.5-VL-72B-Instruct-eagle3-sgl 159
- FreedomIntelligence/ALLaVA-3BText generation 79
- Rayzl/qwen2.5-vl-7b-eagle3-sgl-en 64
- FreedomIntelligence/ALLaVA-Phi3-mini-128kText generation 49
- Geraldxm/Qwen2.5-VL-7B-Instruct-eagle3-sgl 46
- FreedomIntelligence/ALLaVA-StableLM2-1_6BText generation 45
- FreedomIntelligence/ALLaVA-Phi2-2_7BText generation 32
- FreedomIntelligence/ALLaVA-3B-LongerText generation 27
How to load it
from datasets import load_dataset
ds = load_dataset("FreedomIntelligence/ALLaVA-4V")