Usage Example

import requests
from PIL import Image
from transformers import MllamaForConditionalGeneration, AutoProcessor, BitsAndBytesConfig


def get_image_description(model, processor, image, initial_prompt='', max_new_tokens=70, *args, **kwargs):
    initial_prompt = initial_prompt if initial_prompt != '' else "How would you describe the contents of this photo?"
    messages = [
        {"role": "user", "content": [
            {"type": "image"},
            {"type": "text", "text": initial_prompt}
        ]}
    ]
    input_text = processor.apply_chat_template(
        messages, add_generation_prompt=True)
    inputs = processor(
        image,
        input_text,
        add_special_tokens=False,
        return_tensors="pt"
    ).to(model.device)
    output = model.generate(**inputs, max_new_tokens=max_new_tokens)
    return processor.decode(output[0])


def load_model(model_id="belkhir-nacim/l32vision_instruct"):
    bnb_config = BitsAndBytesConfig(
        load_in_4bit=True,  # Enable 4-bit quantization
    )
    model = MllamaForConditionalGeneration.from_pretrained(
        model_id, device_map="auto",quantization_config=bnb_config)
    processor = AutoProcessor.from_pretrained(model_id)
    return model, processor


model, processor = load_model()

url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/0052a70beed5bf71b92610a43a52df6d286cd5f3/diffusers/rabbit.jpg"
image = Image.open(requests.get(url, stream=True).raw)
result = get_image_description(
    model, processor, image, initial_prompt="Tell me what do you see in the image. use keywords to describe")
print(result)