# /// script
# requires-python = ">=3.12"
# dependencies = [
#     "transformers",
#     "torch",
# ]
# ///

try:
    # Use a pipeline as a high-level helper
    from transformers import pipeline
    
    pipe = pipeline("image-text-to-text", model="LiquidAI/LFM2-VL-1.6B", trust_remote_code=True)
    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)
    with open('LiquidAI_LFM2-VL-1.6B_0.txt', 'w') as f:
        f.write('Everything was good in LiquidAI_LFM2-VL-1.6B_0.txt')
except Exception as e:
    with open('LiquidAI_LFM2-VL-1.6B_0.txt', 'w') as f:
        import traceback
        traceback.print_exc(file=f)
finally:
    from huggingface_hub import upload_file
    upload_file(
        path_or_fileobj='LiquidAI_LFM2-VL-1.6B_0.txt',
        repo_id='model-metadata/custom_code_execution_files',
        path_in_repo='LiquidAI_LFM2-VL-1.6B_0.txt',
        repo_type='dataset',
    )