Instructions to use benbayibaurba/cardea-v0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use benbayibaurba/cardea-v0 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="benbayibaurba/cardea-v0")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("benbayibaurba/cardea-v0")
model = AutoModelForMultimodalLM.from_pretrained("benbayibaurba/cardea-v0", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use benbayibaurba/cardea-v0 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "benbayibaurba/cardea-v0"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "benbayibaurba/cardea-v0",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/benbayibaurba/cardea-v0

SGLang

How to use benbayibaurba/cardea-v0 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 "benbayibaurba/cardea-v0" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "benbayibaurba/cardea-v0",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "benbayibaurba/cardea-v0" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "benbayibaurba/cardea-v0",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use benbayibaurba/cardea-v0 with Docker Model Runner:
```
docker model run hf.co/benbayibaurba/cardea-v0
```

CARDEA

CARDEA is a single vision-language model that interprets coronary angiography end-to-end, from raw multi-view sequences through keyframe selection to study-level diagnosis. It grounds its conclusions in explicit spatial evidence through Chain-of-Box (CoB) reasoning.

CARDEA is built on Qwen3-VL-30B-A3B-Thinking.

Pipeline

CARDEA runs a two-pass pipeline. The first pass selects diagnostic keyframes and classifies each view as LCA, RCA, or OTHER. The second pass reasons over the curated LCA and RCA keyframes to produce:

coronary dominance classification;
SYNTAX complexity assessment; and
zero-shot structured report generation over LM, LAD, LCX, and RCA.

Usage

Use CARDEA through the official inference pipeline and demo in the CARDEA GitHub repository, which provides deployment instructions and task-level usage.

Training

CARDEA was trained exclusively on closed-ended tasks using only public datasets:

Single-view: ARCADE, CADICA, and selected non-CAG examples from PubMedVision.
Study-level: CoronaryDominance and CardioSyntax.

Training proceeded in three stages:

Visual feature alignment
Self-distilled CoB cold start
Reinforcement learning with verifiable rewards

Open-ended tasks were excluded from training; report generation was evaluated zero-shot on AngioCAD.

Evaluation

Pass 1

Keyframe selection: mean frame distance Dₖ 0.60; lower is better.
View classification (LCA/RCA/OTHER): accuracy 0.99; Macro-F1 0.99.

Pass 2

Dominance classification (domain shift): accuracy 0.91; Macro-F1 0.89.
Complexity assessment: accuracy 0.90; Macro-F1 0.80.
Report generation (zero-shot): two-class Vessel Severity Macro-F1 0.686 (LLM-judged with google/medgemma-27b-it), compared with 0.513 for the untuned base model and 0.312 for the always-normal baseline; 0.716 on the supplementary valid-views subset, in which both LCA and RCA views were identified.

The full methodology, baselines, results, and analysis are available in the paper.

Limitations

CARDEA is a research model and has not been validated for clinical use. See the paper for a full discussion of its limitations.

License

CARDEA is available for non-commercial use under the PolyForm Noncommercial License 1.0.0.

Citation

Jia-Jen Lee, Shih-Yen Hou, Kee Koon Ng, Wei-Chun Wang, and Shih-Sheng Chang. CARDEA: Auditable Reasoning Grounded in Spatial Evidence for End-to-End Coronary Angiography Interpretation. arXiv:2609.06931 [cs.CV], 2026.

@misc{lee2026cardea,
  title={CARDEA: Auditable Reasoning Grounded in Spatial Evidence for End-to-End Coronary Angiography Interpretation},
  author={Jia-Jen Lee and Shih-Yen Hou and Kee Koon Ng and Wei-Chun Wang and Shih-Sheng Chang},
  year={2026},
  eprint={2609.06931},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2609.06931}
}