Image-Text-to-Text
Transformers
Safetensors
Arabic
English
arabic
vision
image-captioning
image-understanding
siglip2
jais
Instructions to use dhadlab/dhad-vision with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use dhadlab/dhad-vision with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="dhadlab/dhad-vision")# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("dhadlab/dhad-vision", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use dhadlab/dhad-vision with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "dhadlab/dhad-vision" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "dhadlab/dhad-vision", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/dhadlab/dhad-vision
- SGLang
How to use dhadlab/dhad-vision with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "dhadlab/dhad-vision" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "dhadlab/dhad-vision", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "dhadlab/dhad-vision" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "dhadlab/dhad-vision", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use dhadlab/dhad-vision with Docker Model Runner:
docker model run hf.co/dhadlab/dhad-vision
File size: 1,833 Bytes
e3f7919 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 | # NOTICE — licences and third-party components
This repository ships **only** Dhad's own connector weights (`connector.pt`), its
architecture/inference code (`code/`), and the unfinished stage-2 snapshot. It does
**not** re-upload the base models, the frozen encoder, or any training data.
## Shipped here (Dhad) — Apache License 2.0
| File | Copyright | Licence |
|---|---|---|
| `connector.pt` (connector + markers) | Dhad / dhadlab | Apache-2.0 (see `LICENSE`) |
| `code/connector.py`, `code/graft.py`, `code/infer.py`, `code/encoder.py` | Dhad / dhadlab | Apache-2.0 |
| `stage2-step675/*` (unfinished) | Dhad / dhadlab | Apache-2.0 |
## Referenced, not shipped
| Component | Use | Licence |
|---|---|---|
| `inception42/Jais-3-0.9B` | frozen language trunk | Apache-2.0 (declared in its model card; the repo has no LICENSE/NOTICE file) |
| `google/siglip2-base-patch16-naflex` | frozen vision encoder | Apache-2.0; cite [arXiv:2502.14786](https://huggingface.co/papers/2502.14786) |
| `FreedomIntelligence/ALLaVA-4V` | training images (not shipped) | CC-BY-NC-4.0 (card); images are LAION web crawl, rights not owned by the uploaders, per-image licence unknown |
| `FreedomIntelligence/ALLaVA-4V-Arabic` | training captions (not shipped) | Apache-2.0 (card); machine translation of GPT-4V captions |
## Notes
- We redistribute only a trained connector. No images, captions, or SigLIP2 feature
files are included.
- The upstream training-data terms are stated above for provenance. If you
redistribute this model or build on it, review those upstream terms yourself.
- The unfinished stage-2 LoRA targets the Jais trunk; it is published only as an
explicitly-labelled unfinished preview.
The project's full licence audit (`docs/LICENSES_AUDIT.md`) is not published here.
Nothing in this notice is legal advice.
|