app.py

import gradio as gr
import torch
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
from qwen_vl_utils import process_vision_info
import re

# Load the model on CPU
def load_model():
    model = Qwen2VLForConditionalGeneration.from_pretrained(
        "prithivMLmods/Qwen2-VL-OCR-2B-Instruct", 
        torch_dtype=torch.float32,
        device_map="cpu"
    )
    processor = AutoProcessor.from_pretrained("prithivMLmods/Qwen2-VL-OCR-2B-Instruct")
    return model, processor

# Function to extract medicine names
def extract_medicine_names(image):
    model, processor = load_model()
    
    # Prepare the message with the specific prompt for medicine extraction
    messages = [
        {
            "role": "user",
            "content": [
                {
                    "type": "image",
                    "image": image,
                },
                {"type": "text", "text": "Extract and list ONLY the names of medicines/drugs from this prescription image. Output the medicine names as a numbered list without any additional information or descriptions."},
            ],
        }
    ]
    
    # Prepare for inference
    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",
    )
    
    # Generate output
    generated_ids = model.generate(**inputs, max_new_tokens=256)
    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]
    
    # Remove <|im_end|> and any other special tokens that might appear in the output
    output_text = output_text.replace("<|im_end|>", "").strip()
    
    return output_text

# Create a singleton model and processor to avoid reloading for each request
model_instance = None
processor_instance = None

def get_model_and_processor():
    global model_instance, processor_instance
    if model_instance is None or processor_instance is None:
        model_instance, processor_instance = load_model()
    return model_instance, processor_instance

# Optimized extraction function that uses the singleton model
def extract_medicine_names_optimized(image):
    if image is None:
        return "Please upload an image."
    
    model, processor = get_model_and_processor()
    
    # Prepare the message with the specific prompt for medicine extraction
    messages = [
        {
            "role": "user",
            "content": [
                {
                    "type": "image",
                    "image": image,
                },
                {"type": "text", "text": "Extract and list ONLY the names of medicines/drugs from this prescription image. Output the medicine names as a numbered list without any additional information or descriptions."},
            ],
        }
    ]
    
    # Prepare for inference
    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",
    )
    
    # Generate output
    generated_ids = model.generate(**inputs, max_new_tokens=256)
    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]
    
    # Remove <|im_end|> and any other special tokens that might appear in the output
    output_text = output_text.replace("<|im_end|>", "").strip()
    
    return output_text

# Create Gradio interface
with gr.Blocks(title="Medicine Name Extractor") as app:
    gr.Markdown("# Medicine Name Extractor")
    gr.Markdown("Upload a medical prescription image to extract the names of medicines.")
    
    with gr.Row():
        with gr.Column():
            input_image = gr.Image(type="pil", label="Upload Prescription Image")
            extract_btn = gr.Button("Extract Medicine Names", variant="primary")
        
        with gr.Column():
            output_text = gr.Textbox(label="Extracted Medicine Names", lines=10)
    
    extract_btn.click(
        fn=extract_medicine_names_optimized,
        inputs=input_image,
        outputs=output_text
    )
    
    gr.Markdown("### Notes")
    gr.Markdown("- This tool uses the Qwen2-VL-OCR model to extract text from prescription images")
    gr.Markdown("- For best results, ensure the prescription image is clear and readable")
    gr.Markdown("- Processing may take some time as the model runs on CPU")

# Launch the app
if __name__ == "__main__":
    app.launch()