app.py

import torch
from transformers import AutoModel, AutoTokenizer
from PIL import Image
import gradio as gr
import os

# Load the OCR model and tokenizer
tokenizer = AutoTokenizer.from_pretrained('ucaslcl/GOT-OCR2_0', trust_remote_code=True)
model = AutoModel.from_pretrained('ucaslcl/GOT-OCR2_0', trust_remote_code=True, 
                                  low_cpu_mem_usage=True, 
                                  pad_token_id=tokenizer.eos_token_id).eval()

# Ensure we are using CPU
device = torch.device('cpu')
model = model.to(device)

# Function to perform OCR on the image file
def perform_ocr(image_file_path):
    # Open the image using PIL
    image = Image.open(image_file_path)

    # Save the image temporarily
    temp_image_path = "temp_image.png"
    image.save(temp_image_path)

    # Use torch.no_grad() to avoid unnecessary memory usage
    with torch.no_grad():
        # Perform OCR using the model (pass the file path of the saved image)
        result = model.chat(tokenizer, temp_image_path, ocr_type='ocr')

    # Clean up the temporary image file
    os.remove(temp_image_path)

    # Return the extracted text
    return result

# Create the Gradio interface for file upload and OCR
iface = gr.Interface(fn=perform_ocr, inputs="file", outputs="text", 
                     title="OCR Application", description="Upload an image to extract text.")

# Launch the Gradio app
iface.launch()