axolotl-ai-co/omni_med_vqa_mini

from datasets import load_dataset
raw_ds = load_dataset("simwit/omni-med-vqa-mini")
system_message = (
    "You are a medical Vision Language Model specialized in analyzing medical images and providing clinical insights. "
    "Provide concise, clinically relevant answers based on the image and question."
)
def format_medical_sample(sample):
    return {
        "messages": [
            {"role": "system", "content": [{"type": "text", "text": system_message}]},
            {
                "role": "user",
                "content": [
                    {"type": "image", "text": None},
                    {"type": "text", "text": sample["question"]},
                ],
            },
            {"role": "assistant", "content": [{"type": "text", "text": sample["gt_answer"]}]},
        ],
        "image": sample["image"],
    }
train_dataset = raw_ds.map(
    format_medical_sample,
    batched=False,
)
train_dataset.push_to_hub("axolotl-ai-co/omni_med_vqa_mini")