app.py

import gradio as gr
import time
import spaces
from PIL import Image
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
from qwen_vl_utils import process_vision_info
import torch
import uuid
import os
import numpy as np

# Load model and processor
model_name = "oddadmix/Qaari-0.1-Urdu-OCR-2B"
model = Qwen2VLForConditionalGeneration.from_pretrained(
                model_name,
                torch_dtype="auto",
                device_map="cuda"
            )
processor = AutoProcessor.from_pretrained(model_name)
max_tokens = 2000


@spaces.GPU
def perform_ocr(image):
    inputArray = np.any(image)
    if inputArray == False:
        return "Error Processing"
    """Process image and extract text using OCR model"""
    image = Image.fromarray(image)
    src = str(uuid.uuid4()) + ".png"
    prompt = "Below is the image of one page of a document, as well as some raw textual content that was previously extracted for it. Just return the plain text representation of this document as if you were reading it naturally. Do not hallucinate."
    image.save(src)
    
    messages = [
        {
            "role": "user",
            "content": [
                {"type": "image", "image": f"file://{src}"},
                {"type": "text", "text": prompt},
            ],
        }
    ]
    
    # Process inputs
    text = processor.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )
    image_inputs, video_inputs = process_vision_info(messages)
    inputs = processor(
        text=[text],
        images=image_inputs,
        videos=video_inputs,
        padding=True,
        return_tensors="pt",
    )
    inputs = inputs.to("cuda")
    
    # Generate text
    generated_ids = model.generate(**inputs, max_new_tokens=max_tokens, use_cache=True)
    generated_ids_trimmed = [
        out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
    ]
    output_text = processor.batch_decode(
        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
    )[0]
    
    # Cleanup
    os.remove(src)
    return output_text

# Create Gradio interface
with gr.Blocks(title="Qaari-0.1-Urdu-OCR-2B Urdu OCR") as demo:
    gr.Markdown("# Qaari-0.1-Urdu-OCR-2B Urdu OCR")
    gr.Markdown("Upload an image to extract Urdu text in real-time. This model is specialized for Urdu document OCR.")
    
    with gr.Row():
        with gr.Column(scale=1):
            # Input image
            image_input = gr.Image(type="numpy", label="Upload Image")
            
            # Example gallery
            gr.Examples(
                examples=[
                    ["1.jpg"],
                    ["2.jpg"]
                ],
                inputs=image_input,
                label="Example Images",
                examples_per_page=4
            )
            
            # Submit button
            submit_btn = gr.Button("Extract Text")
        
        with gr.Column(scale=1):
            # Output text
            output = gr.Textbox(label="Extracted Text", lines=20, show_copy_button=True, rtl=True)
            
            # Model details
            with gr.Accordion("Model Information", open=False):
                gr.Markdown("""
                **Model:** Qaari-0.1-Urdu-OCR-2B
                **Description:** Urdu OCR model based on Qwen2-VL architecture
                **Size:** 2B parameters
                **Context window:** Supports up to 2000 output tokens
                """)
    
    # Set up processing flow
    submit_btn.click(fn=perform_ocr, inputs=image_input, outputs=output)
    image_input.change(fn=perform_ocr, inputs=image_input, outputs=output)

demo.launch()