app.py

import gradio as gr
from PIL import Image
import io
import base64
from huggingface_hub import InferenceClient

# Initialize the Hugging Face Inference Client
client = InferenceClient("microsoft/llava-med-7b-delta")

# Custom Field for Base64 Encoded Image
class Base64ImageField(gr.Field):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)

    def preprocess(self, image):
        buffered = io.BytesIO()
        image.save(buffered, format="PNG")
        img_str = base64.b64encode(buffered.getvalue()).decode('utf-8')
        return img_str

# Function to interact with LLAVA model
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
    image=None
):
    messages = [{"role": "system", "content": system_message}]

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

    messages.append({"role": "user", "content": message})

    if image:
        # Convert image(s) to base64 using the custom field
        if isinstance(image, Image.Image):
            image_b64 = Base64ImageField().preprocess(image)
            messages.append({"role": "user", "content": "Image uploaded", "image": image_b64})
        else:
            for img in image:
                image_b64 = Base64ImageField().preprocess(img)
                messages.append({"role": "user", "content": "Image uploaded", "image": image_b64})

    # Call Hugging Face model for response
    try:
        responses = []
        generated_image = None

        for response in client.chat_completion(
            messages,
            max_tokens=max_tokens,
            stream=True,
            temperature=temperature,
            top_p=top_p,
        ):
            token = response.choices[0].delta.content
            responses.append(token)

            # Check if the response contains an image to be displayed
            if response.choices[0].delta.image:
                image_b64 = response.choices[0].delta.image
                image_data = base64.b64decode(image_b64)
                generated_image = Image.open(io.BytesIO(image_data))
                # Optionally convert to RGB if needed
                # generated_image = generated_image.convert("RGB")

        yield responses, generated_image

    except Exception as e:
        yield [str(e)], None

# Debugging print statements
print("Starting Gradio interface setup...")
try:
    # Create a Gradio interface
    demo = gr.Interface(
        fn=respond,
        inputs=[
            gr.Textbox(label="Message"),
            gr.Image(label="Upload Medical Image (Optional)", type="pil")
        ],
        outputs=[
            gr.Textbox(label="Response", placeholder="Model response will appear here..."),
            gr.Image(label="Generated Image", type="pil", output=True)
        ],
        title="LLAVA Model - Medical Image and Question",
        description="Upload a medical image and ask a specific question about the image for a medical description.",
        additional_inputs=[
            gr.Textbox(label="System message", value="You are a friendly Chatbot."),
            gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
            gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
            gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)")
        ]
    )

    # Launch the Gradio interface
    if __name__ == "__main__":
        print("Launching Gradio interface...")
        demo.launch()

except Exception as e:
    print(f"Error during Gradio setup: {str(e)}")