from transformers import AutoProcessor, AutoModelForImageTextToText

# Requires transformers>=5.3.0
model_id = "OpenMed/Qwen3.5-2B-MedVL"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(model_id, torch_dtype="auto", device_map="auto")

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://example.com/xray.jpg"},
            {"type": "text", "text": "What are the key findings in this chest X-ray?"},
        ],
    }
]

inputs = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=512)
print(processor.decode(output[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))

vLLM

python

from vllm import LLM, SamplingParams

llm = LLM(model="OpenMed/Qwen3.5-2B-MedVL", max_model_len=4096, limit_mm_per_prompt={"image": 1})

messages = [{"role": "user", "content": [
    {"type": "image_url", "image_url": {"url": "https://example.com/xray.jpg"}},
    {"type": "text", "text": "What are the key findings in this chest X-ray?"},
]}]

output = llm.chat(messages, SamplingParams(temperature=0, max_tokens=512))
print(output[0].outputs[0].text)

SGLang

bash

# Launch server
python -m sglang.launch_server --model-path OpenMed/Qwen3.5-2B-MedVL --port 8000

python

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
response = client.chat.completions.create(
    model="OpenMed/Qwen3.5-2B-MedVL",
    messages=[{"role": "user", "content": [
        {"type": "image_url", "image_url": {"url": "https://example.com/xray.jpg"}},
        {"type": "text", "text": "What are the key findings in this chest X-ray?"},
    ]}],
    max_tokens=512,
)
print(response.choices[0].message.content)

Training Details

Base model: Qwen/Qwen3.5-2B
Data: ~200K medical VQA records from the SynthVision pipeline
Method: LoRA (rank=64, alpha=64)
Target modules: q_proj, v_proj, k_proj, o_proj, gate_proj, up_proj, down_proj
Learning rate: 1e-4, cosine schedule
Steps: 350
Weight decay: 0.01
Hardware: 4x NVIDIA A100 80GB (48 vCPU, 568 GB RAM) via Hugging Face Jobs
Training time: ~1h

flyingfishinwater/Qwen3.5-2B-MedVL-MLX-4bit

Qwen3.5-2B-MedVL

Benchmark Results (Exact Match)

Usage

Transformers

vLLM

SGLang

Training Details

Links

flyingfishinwater/
Qwen3.5-2B-MedVL-MLX-4bit