app.py

import os
from fastapi import Request
from starlette.responses import JSONResponse
import gradio as gr
from transformers import AutoProcessor, AutoModelForVision2Seq
from PIL import Image

# Get the secret key from Hugging Face secrets
API_KEY = os.environ.get("API_KEY")

# Load model and processor
processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct")
model = AutoModelForVision2Seq.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct", device_map="auto")

# Prompt for spine recognition
prompt = "What is the album title and artist on this CD spine?"

# Auth wrapper
def with_auth(fn):
    async def wrapper(image: Image.Image, request: Request):
        if request.headers.get("x-api-key") != API_KEY:
            return JSONResponse({"error": "unauthorized"}, status_code=401)
        return fn(image)
    return wrapper

# Inference function
@with_auth
def extract_text(image: Image.Image):
    inputs = processor(prompt=prompt, images=image, return_tensors="pt").to(model.device)
    generated_ids = model.generate(**inputs, max_new_tokens=128)
    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
    return {"text": generated_text}

# Gradio app
iface = gr.Interface(fn=extract_text, inputs="image", outputs="json")
iface.launch(
    server_name="0.0.0.0", server_port=7860, share=True, api_name="extract_text"
)