app.py

import torch
from transformers import AutoModelForCausalLM, AutoProcessor
from PIL import Image
import gradio as gr

# Load the MAGMA model and processor
model_id = "microsoft/Magma-8B"
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, trust_remote_code=True)
model.to("cuda" if torch.cuda.is_available() else "cpu")

def magma_inference(image: Image.Image, prompt: str) -> str:
    # Prepare conversation
    convs = [
        {"role": "system", "content": "You are an agent that can see, talk, and act."},
        {"role": "user", "content": prompt}
    ]
    # Generate prompt
    text_prompt = processor.tokenizer.apply_chat_template(convs, tokenize=False, add_generation_prompt=True)
    # Process inputs
    inputs = processor(images=[image], texts=text_prompt, return_tensors="pt").to(model.device)
    # Generate output
    with torch.inference_mode():
        generate_ids = model.generate(**inputs, max_new_tokens=50)
        generate_ids = generate_ids[:, inputs["input_ids"].shape[-1]:]
        response = processor.decode(generate_ids[0], skip_special_tokens=True).strip()
    return response

# Gradio interface
interface = gr.Interface(
    fn=magma_inference,
    inputs=[
        gr.Image(type="pil", label="Input Image"),
        gr.Textbox(label="Prompt")
    ],
    outputs=gr.Textbox(label="MAGMA Output"),
    title="MAGMA Image + Text to Text API",
    description="Upload an image and enter a prompt. Returns MAGMA's textual response."
)

app = gr.mount_gradio_app(app=interface, path="/")