Dataset · huaweilin
ViRA training data for visual agents
huaweilin/ViRA-Train
A training dataset for the ViRA model designed for ultra-high-resolution tasks.
What's inside
The dataset contains materials for training visual agents based on policy optimization methods. The data is associated with ViRA-Bench benchmarks for evaluating model accuracy across various pixel budgets and at maximum resolutions.
What you can do with it
- Training multimodal visual agents such as ViRA-9B
- Optimizing cross-input relative policies for VQA
- Tasks
- Visual question answering
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 1.3K
- Likes
- 2
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("huaweilin/ViRA-Train")