Update app.py

app.py

@@ -1,70 +1,146 @@
 import gradio as gr
 import torch
-from PIL import Image
-import time
-from threading import Thread
-from transformers import (
-    Qwen2VLForConditionalGeneration,
-    AutoProcessor,
-    TextIteratorStreamer,
-)
+from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
+from qwen_vl_utils import process_vision_info
+import re
 
-# Load model and processor - CPU version
-MODEL_ID = "prithivMLmods/Qwen2-VL-OCR-2B-Instruct"
-processor = AutoProcessor.from_pretrained(MODEL_ID, trust_remote_code=True)
-model = Qwen2VLForConditionalGeneration.from_pretrained(
-    MODEL_ID,
-    trust_remote_code=True,
-    torch_dtype=torch.float32  # Using float32 for CPU compatibility
-).to("cpu").eval()
+# 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
 
-def extract_medicines(image):
-    """Extract medicine names from prescription images."""
+# 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 a prescription image."
+        return "Please upload an image."
     
-    # Process the image
-    text = "Extract ONLY the names of medications/medicines from this prescription image. Format the output as a numbered list of medicine names only, without dosages or instructions."
+    model, processor = get_model_and_processor()
     
-    messages = [{
-        "role": "user",
-        "content": [
-            {"type": "image", "image": Image.open(image)},
-            {"type": "text", "text": text},
-        ],
-    }]
+    # 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."},
+            ],
+        }
+    ]
     
-    prompt_full = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    # 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=[prompt_full],
-        images=[Image.open(image)],
-        return_tensors="pt",
+        text=[text],
+        images=image_inputs,
+        videos=video_inputs,
         padding=True,
-    ).to("cpu")
-    
-    # Generate response
-    with torch.no_grad():
-        output = model.generate(**inputs, max_new_tokens=512)
+        return_tensors="pt",
+    )
     
-    # Decode and return response
-    response = processor.decode(output[0], skip_special_tokens=True)
+    # 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]
     
-    # Clean up the response to get just the model's answer
-    if "<|assistant|>" in response:
-        response = response.split("<|assistant|>")[1].strip()
+    # Remove <|im_end|> and any other special tokens that might appear in the output
+    output_text = output_text.replace("<|im_end|>", "").strip()
     
-    return response
+    return output_text
 
-# Create a simple Gradio interface
-demo = gr.Interface(
-    fn=extract_medicines,
-    inputs=gr.Image(type="filepath", label="Upload Prescription Image"),
-    outputs=gr.Textbox(label="Extracted Medicine Names"),
-    title="Medicine Name Extractor",
-    description="Upload prescription images to extract medicine names",
-    examples=[["examples/prescription1.jpg"]], # Update with your actual example paths or remove if not available
-    cache_examples=True,
-)
+# 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__":
-    demo.launch(debug=True)
+    app.launch()