app.py

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

# =========================
# Model Setup & Patch
# =========================
model_id = 'microsoft/Florence-2-large'
device = "cuda" if torch.cuda.is_available() else "cpu"

# PATCH: Explicitly handle the Florence2 configuration bug
config = AutoConfig.from_pretrained(model_id, trust_remote_code=True)
if not hasattr(config, 'forced_bos_token_id'):
    config.forced_bos_token_id = None

# Load model and processor
model = AutoModelForCausalLM.from_pretrained(
    model_id, 
    config=config,
    trust_remote_code=True
).to(device).eval()

processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)

def run_ocr(image):
    if image is None:
        return "⚠️ Please upload an image."

    # Using <DETAILED_CAPTION> or <OCR> task for better text flow
    # Florence-2 works best with these specific task tags
    prompt = "<OCR>"
    
    inputs = processor(text=prompt, images=image, return_tensors="pt").to(device)

    with torch.no_grad():
        generated_ids = model.generate(
            input_ids=inputs["input_ids"],
            pixel_values=inputs["pixel_values"],
            max_new_tokens=1024,
            do_sample=False,
            num_beams=3
        )

    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
    
    # Clean up the output
    parsed_answer = processor.post_process_generation(
        generated_text, 
        task=prompt, 
        image_size=(image.width, image.height)
    )

    return parsed_answer[prompt]

# =========================
# Gradio UI
# =========================
with gr.Blocks() as demo:
    gr.Markdown("## 🖋️ Handwritten Note to Text (Florence-2)")
    
    with gr.Row():
        input_img = gr.Image(type="pil")
        output_text = gr.Textbox(label="Extracted Text", lines=10)
    
    btn = gr.Button("Convert to Text", variant="primary")
    btn.click(fn=run_ocr, inputs=input_img, outputs=output_text)

if __name__ == "__main__":
    demo.launch()