Qwen25VLMEDVQAGI

Fine-tuned on Kvasir-VQA-x1 for the ImageCLEFmed MEDVQA-GI 2026 challenge.

Model Details

• Base model: Qwen/Qwen2.5-VL-7B-Instruct
• Fine-tuning: QLoRA (rank=16, alpha=32, 4-bit NF4)
• Training stage: stage1_subtask1
• Task: GI endoscopy VQA — Subtask 1 + Subtask 2

Usage

import torch
from swift.llm import PtEngine, RequestConfig, InferRequest
from transformers import BitsAndBytesConfig

engine = PtEngine(
    adapters=["sageofai/Qwen25VLMEDVQAGI"],
    model_id_or_path="Qwen/Qwen2.5-VL-7B-Instruct",
    quantization_config=BitsAndBytesConfig(
        load_in_4bit=True, bnb_4bit_quant_type='nf4',
        bnb_4bit_use_double_quant=True,
        bnb_4bit_compute_dtype=torch.float16,
    ),
    attn_impl='sdpa', use_hf=True,
)
req_cfg = RequestConfig(max_tokens=512, temperature=0.3)
resp = engine.infer([InferRequest(messages=[{
    "role": "user",
    "content": [
        {"type": "image", "image": "path/to/gi_image.jpg"},
        {"type": "text",  "text": "Is there a polyp?"},
    ]
}])], req_cfg)
print(resp[0].choices[0].message.content)