Preview · under training
Dhad Vision v1
Describes images in Arabic and answers simple questions about them. This model is still under training. This is the latest trained version, not the final one.
Try it on an image1.2B
Parameters
v1
Latest version
Preview
Status
What it does
One job, done well.
Describing an image
A general description of what is in the picture, in Arabic.
Simple questions
Colours, objects and counts.
A preview
It can get details wrong, and it should not be trusted to read text inside images.
Versions
What exists so far.
v1
Preview · Latest
v2
Upcoming
Specs
What is inside.
- Name
- dhad-1-vision-1.2b
- Parameters
- 1.2B
- Image encoder
- 86M (frozen)
- Connector
- 7M
- Image tokens
- Up to 144