Update app.py

app.py

@@ -9,17 +9,23 @@ import torch
 import torchvision
 import subprocess
 
+# Run the commands from setup.sh to install poppler-utils
 def install_poppler():
     try:
         subprocess.run(["pdfinfo"], check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
     except FileNotFoundError:
         print("Poppler not found. Installing...")
+        # Run the setup commands
         subprocess.run("apt-get update", shell=True)
         subprocess.run("apt-get install -y poppler-utils", shell=True)
 
+# Call the Poppler installation check
 install_poppler()
+
+# Install flash-attn if not already installed
 subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
 
+# Load the RAG Model and the Qwen2-VL-2B-Instruct model
 RAG = RAGMultiModalModel.from_pretrained("vidore/colpali")
 model = Qwen2VLForConditionalGeneration.from_pretrained("Qwen/Qwen2-VL-2B-Instruct",
                                                         trust_remote_code=True, torch_dtype=torch.bfloat16).cuda().eval()
@@ -27,17 +33,24 @@ processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct", trust_rem
 
 @spaces.GPU()
 def process_pdf_and_query(pdf_file, user_query):
-    images = convert_from_path(pdf_file.name)
+    # Convert the PDF to images
+    images = convert_from_path(pdf_file.name)  # pdf_file.name gives the file path
     num_images = len(images)
+
+    # Indexing the PDF in RAG
     RAG.index(
         input_path=pdf_file.name,
-        index_name="image_index",
+        index_name="image_index",  # index will be saved at index_root/index_name/
         store_collection_with_index=False,
         overwrite=True
     )
+
+    # Search the query in the RAG model
     results = RAG.search(user_query, k=1)
     if not results:
         return "No results found.", num_images
+
+    # Retrieve the page number and process image
     image_index = results[0]["page_num"] - 1
     messages = [
         {
@@ -51,6 +64,8 @@ def process_pdf_and_query(pdf_file, user_query):
             ],
         }
     ]
+
+    # Generate text with the Qwen model
     text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
     image_inputs, video_inputs = process_vision_info(messages)
     inputs = processor(
@@ -61,6 +76,8 @@ def process_pdf_and_query(pdf_file, user_query):
         return_tensors="pt",
     )
     inputs = inputs.to("cuda")
+
+    # Generate the output response
     generated_ids = model.generate(**inputs, max_new_tokens=50)
     generated_ids_trimmed = [
         out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
@@ -68,45 +85,39 @@ def process_pdf_and_query(pdf_file, user_query):
     output_text = processor.batch_decode(
         generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
     )
+
     return output_text[0], num_images
 
-footer = """
-<div style="text-align: center; margin-top: 20px;">
-    <a href="https://www.linkedin.com/in/pejman-ebrahimi-4a60151a7/" target="_blank">LinkedIn</a> |
-    <a href="https://github.com/arad1367" target="_blank">GitHub</a> |
-    <a href="https://arad1367.pythonanywhere.com/" target="_blank">Live demo of my PhD defense</a> |
-    <a href="https://huggingface.co/Qwen/Qwen2-VL-2B-Instruct" target="_blank">Qwen/Qwen2-VL-2B-Instruct</a> |
-    <a href="https://github.com/AnswerDotAI/byaldi" target="_blank">Byaldi</a> |
-    <a href="https://github.com/illuin-tech/colpali" target="_blank">ColPali</a>
-    <br>
-    Made with 💖 by Pejman Ebrahimi
-</div>
-"""
-
-pdf_input = gr.File(label="Upload PDF")
-query_input = gr.Textbox(label="Enter your query", placeholder="Ask a question about the PDF")
-output_text = gr.Textbox(label="Model Answer")
-output_images = gr.Textbox(label="Number of Images in PDF")
-duplicate_button = gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button", color="green")
-
-explanation_text = """
-<div style='text-align: center; font-size: 16px;'>
-    <h2 style='font-weight: bold;'>Multimodal RAG Overview</h2>
-    <p>This application utilizes a Multimodal RAG (Retrieve-and-Generate) approach, enabling users to query information from PDF documents 
-    by extracting relevant text and images. The ColPali model serves as a multimodal retriever, while the Byaldi library simplifies 
-    the integration of ColPali. The Qwen/Qwen2-VL-2B-Instruct LLM enhances the generation of responses based on the retrieved content.</p>
-</div>
-"""
-
-demo = gr.Interface(
-    fn=process_pdf_and_query, 
-    inputs=[pdf_input, query_input], 
-    outputs=[output_text, output_images], 
-    title="<div style='text-align: center; font-size: 24px; font-weight: bold;'>Multimodal RAG with Image Query</div>",
-    description=explanation_text,
-    theme='freddyaboulton/dracula_revamped'
-)
-
-demo.launch(debug=True)
-demo.append(duplicate_button)
-demo.append(gr.HTML(footer))
+
+with gr.Blocks(theme='freddyaboulton/dracula_revamped') as demo:
+    gr.HTML("<h1 style='text-align: center; font-size: 30px;'><a href='https://github.com/arad1367'>Multimodal RAG with Image Query - By Pejman Ebrahimi</a></h1>")
+    gr.Markdown("Multimodal RAG is a technique that combines both textual and visual data to provide more accurate and comprehensive results. In this application, we use ColPali, a multimodal retriever, and Byaldi, a new library by answer.ai to easily use ColPali. We also use Qwen/Qwen2-VL-2B-Instruct LLM.")
+
+    pdf_input = gr.File(label="Upload PDF")  
+    query_input = gr.Textbox(label="Enter your query", placeholder="Ask a question about the PDF")  
+    output_text = gr.Textbox(label="Model Answer")  
+    output_images = gr.Textbox(label="Number of Images in PDF")  
+
+    submit_btn = gr.Button("Submit", variant="primary")
+    submit_btn.style(full_width=True)
+
+    duplicate_btn = gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button")
+    duplicate_btn.style(full_width=True)
+
+    submit_btn.click(fn=process_pdf_and_query, inputs=[pdf_input, query_input], outputs=[output_text, output_images])
+
+    footer = """
+    <div style="text-align: center; margin-top: 20px;">
+        <a href="https://www.linkedin.com/in/pejman-ebrahimi-4a60151a7/" target="_blank">LinkedIn</a> |
+        <a href="https://github.com/arad1367" target="_blank">GitHub</a> |
+        <a href="https://arad1367.pythonanywhere.com/" target="_blank">Live demo of my PhD defense</a> |
+        <a href="https://huggingface.co/Qwen/Qwen2-VL-2B-Instruct" target="_blank">Qwen/Qwen2-VL-2B-Instruct</a> |
+        <a href="https://github.com/AnswerDotAI/byaldi" target="_blank">Byaldi</a> |
+        <a href="https://github.com/illuin-tech/colpali" target="_blank">ColPali</a>
+        <br>
+        Made with 💖 by Pejman Ebrahimi
+    </div>
+    """
+    gr.HTML(footer)
+
+demo.launch(debug=True)