Update app.py

app.py

@@ -11,239 +11,121 @@ logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
 try:
-    from langchain_community.document_loaders import PyPDFDirectoryLoader, PyPDFLoader
+    from langchain_community.document_loaders import PyPDFDirectoryLoader
     from langchain.text_splitter import RecursiveCharacterTextSplitter
     from langchain_community.embeddings import HuggingFaceEmbeddings
     from langchain_community.vectorstores import FAISS
     from langchain.prompts import PromptTemplate
     from langchain.chains import RetrievalQA
-    from langchain_community.llms import HuggingFaceEndpoint
+    from langchain.llms import HuggingFaceHub
     LANGCHAIN_AVAILABLE = True
 except ImportError as e:
     logger.error(f"LangChain import error: {e}")
     LANGCHAIN_AVAILABLE = False
 
-# Create PDFs folder if it doesn't exist
 PDF_FOLDER_PATH = "./pdfs"
 os.makedirs(PDF_FOLDER_PATH, exist_ok=True)
 
-# Global variables for the RAG system
 vectorstore = None
 retrieval_qa = None
 embedding_model = None
-
-# Check for pre-existing PDF folder
 PRELOADED_PDFS = os.path.exists(PDF_FOLDER_PATH) and len(os.listdir(PDF_FOLDER_PATH)) > 0
 
 def initialize_models():
-    """Initialize the embedding model and LLM"""
     global embedding_model
-    
     try:
-        # Initialize embedding model
         embedding_model = HuggingFaceEmbeddings(
             model_name="sentence-transformers/all-MiniLM-L6-v2",
             model_kwargs={'device': 'cpu'}
         )
-        
-        # Get HuggingFace token from environment
+
         hf_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
         if not hf_token:
-            return False, "❌ HuggingFace API token not found in environment variables"
-        
-        return True, "✅ Models initialized successfully"
-    
+            return False, "❌ HuggingFace API token not found"
+
+        return True, "✅ Models initialized"
     except Exception as e:
-        logger.error(f"Model initialization error: {e}")
-        return False, f"❌ Error initializing models: {str(e)}"
+        logger.error(f"Init error: {e}")
+        return False, str(e)
 
 def create_llm():
-    """Create and return the LLM instance"""
     hf_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
-    
-    llm = HuggingFaceEndpoint(
-        repo_id="google/flan-t5-base",
-        temperature=0.7,
-        max_new_tokens=512,
-        huggingfacehub_api_token=hf_token
-    )
-    
-    return llm
+    if not hf_token:
+        return create_fallback_llm()
+
+    models_to_try = [
+        "mistralai/Mistral-7B-Instruct-v0.2",
+        "google/flan-t5-base"
+    ]
+
+    for model_id in models_to_try:
+        try:
+            llm = HuggingFaceHub(
+                repo_id=model_id,
+                huggingfacehub_api_token=hf_token,
+                model_kwargs={
+                    "temperature": 0.7,
+                    "max_length": 512,
+                    "top_p": 0.9,
+                    "top_k": 50
+                }
+            )
+            return llm
+        except Exception as e:
+            logger.warning(f"Model {model_id} failed: {e}")
+    return create_fallback_llm()
+
+def create_fallback_llm():
+    class FallbackLLM:
+        def __call__(self, prompt):
+            return "Model is unavailable. Try again later."
+        def invoke(self, prompt):
+            return self.__call__(prompt)
+    return FallbackLLM()
 
 def load_preloaded_pdfs(chunk_size=1000, chunk_overlap=200):
-    """Load PDFs from the pre-existing folder"""
     global vectorstore, retrieval_qa, embedding_model
-    
+
     if not LANGCHAIN_AVAILABLE:
-        return "❌ LangChain is not available. Please check the installation."
-    
+        return "❌ LangChain not available"
+
     if not PRELOADED_PDFS:
-        return "❌ No pre-loaded PDFs found in ./pdfs folder."
-    
+        return "❌ No PDFs found"
+
     try:
-        # Initialize models if not already done
         if embedding_model is None:
-            success, message = initialize_models()
+            success, msg = initialize_models()
             if not success:
-                return message
-        
-        # Load documents from pre-existing folder
+                return msg
+
         loader = PyPDFDirectoryLoader(PDF_FOLDER_PATH)
         documents = loader.load()
-        
         if not documents:
-            return "❌ No documents were loaded from the PDFs folder."
-        
-        # Split documents into chunks
-        text_splitter = RecursiveCharacterTextSplitter(
-            chunk_size=int(chunk_size),
-            chunk_overlap=int(chunk_overlap)
-        )
-        chunks = text_splitter.split_documents(documents)
-        
-        # Create vector store
-        vectorstore = FAISS.from_documents(chunks, embedding_model)
-        retriever = vectorstore.as_retriever(search_kwargs={"k": 5})
-        
-        # Setup prompt template
-        prompt_template = """
-Use the following context to answer the question. If you cannot find the answer in the context, say "I don't have enough information to answer this question."
-
-Context:
-{context}
+            return "❌ No documents loaded"
 
-Question: {question}
-
-Helpful Answer:
-"""
-        prompt = PromptTemplate(
-            input_variables=["context", "question"], 
-            template=prompt_template
-        )
-        
-        # Initialize LLM using the new function
-        llm = create_llm()
-        
-        # Create RetrievalQA chain
-        retrieval_qa = RetrievalQA.from_chain_type(
-            llm=llm,
-            chain_type="stuff",
-            retriever=retriever,
-            return_source_documents=True,
-            chain_type_kwargs={"prompt": prompt}
+        splitter = RecursiveCharacterTextSplitter(
+            chunk_size=chunk_size, chunk_overlap=chunk_overlap
         )
-        
-        pdf_files = [f for f in os.listdir(PDF_FOLDER_PATH) if f.endswith('.pdf')]
-        return f"✅ Successfully processed {len(documents)} documents from {len(pdf_files)} PDF files into {len(chunks)} chunks. Ready for questions!"
-    
-    except Exception as e:
-        logger.error(f"Pre-loaded PDF processing error: {e}")
-        return f"❌ Error processing pre-loaded PDFs: {str(e)}"
-
-def extract_zip_to_pdfs(zip_file):
-    """Extract uploaded ZIP file to PDFs folder"""
-    if not zip_file:
-        return "❌ Please upload a ZIP file."
-    
-    try:
-        # Create PDFs directory if it doesn't exist
-        os.makedirs(PDF_FOLDER_PATH, exist_ok=True)
-        
-        # Extract ZIP file
-        with zipfile.ZipFile(zip_file.name, 'r') as zip_ref:
-            # Extract only PDF files
-            pdf_files = [f for f in zip_ref.namelist() if f.lower().endswith('.pdf')]
-            
-            if not pdf_files:
-                return "❌ No PDF files found in the ZIP archive."
-            
-            for pdf_file in pdf_files:
-                # Extract to PDFs folder
-                zip_ref.extract(pdf_file, PDF_FOLDER_PATH)
-                
-                # If file is in a subfolder, move it to the root of PDFs folder
-                extracted_path = os.path.join(PDF_FOLDER_PATH, pdf_file)
-                if os.path.dirname(pdf_file):  # File is in a subfolder
-                    new_path = os.path.join(PDF_FOLDER_PATH, os.path.basename(pdf_file))
-                    shutil.move(extracted_path, new_path)
-                    # Clean up empty directories
-                    try:
-                        os.rmdir(os.path.dirname(extracted_path))
-                    except:
-                        pass
-        
-        global PRELOADED_PDFS
-        PRELOADED_PDFS = True
-        
-        return f"✅ Successfully extracted {len(pdf_files)} PDF files. Now click 'Load Pre-existing PDFs' to process them."
-    
-    except Exception as e:
-        return f"❌ Error extracting ZIP file: {str(e)}"
-
-def process_pdfs(pdf_files, chunk_size, chunk_overlap):
-    """Process uploaded PDF files and create vector store"""
-    global vectorstore, retrieval_qa, embedding_model
-    
-    if not LANGCHAIN_AVAILABLE:
-        return "❌ LangChain is not available. Please check the installation."
-    
-    if not pdf_files:
-        return "❌ Please upload at least one PDF file or use pre-loaded PDFs."
-    
-    try:
-        # Initialize models if not already done
-        if embedding_model is None:
-            success, message = initialize_models()
-            if not success:
-                return message
-        
-        # Create temporary directory for PDFs
-        temp_dir = tempfile.mkdtemp()
-        
-        # Save uploaded files to temp directory
-        for pdf_file in pdf_files:
-            if pdf_file is not None:
-                temp_path = os.path.join(temp_dir, os.path.basename(pdf_file.name))
-                shutil.copy2(pdf_file.name, temp_path)
-        
-        # Load documents
-        loader = PyPDFDirectoryLoader(temp_dir)
-        documents = loader.load()
-        
-        if not documents:
-            return "❌ No documents were loaded. Please check your PDF files."
-        
-        # Split documents into chunks
-        text_splitter = RecursiveCharacterTextSplitter(
-            chunk_size=int(chunk_size),
-            chunk_overlap=int(chunk_overlap)
-        )
-        chunks = text_splitter.split_documents(documents)
-        
-        # Create vector store
+        chunks = splitter.split_documents(documents)
         vectorstore = FAISS.from_documents(chunks, embedding_model)
         retriever = vectorstore.as_retriever(search_kwargs={"k": 5})
-        
-        # Setup prompt template
+
         prompt_template = """
-Use the following context to answer the question. If you cannot find the answer in the context, say "I don't have enough information to answer this question."
+Use the following context to answer the question. If you cannot find the answer, say so.
 
 Context:
 {context}
 
 Question: {question}
 
-Helpful Answer:
+Answer:
 """
         prompt = PromptTemplate(
-            input_variables=["context", "question"], 
+            input_variables=["context", "question"],
             template=prompt_template
         )
-        
-        # Initialize LLM using the new function
+
         llm = create_llm()
-        
-        # Create RetrievalQA chain
         retrieval_qa = RetrievalQA.from_chain_type(
             llm=llm,
             chain_type="stuff",
@@ -251,412 +133,65 @@ Helpful Answer:
             return_source_documents=True,
             chain_type_kwargs={"prompt": prompt}
         )
-        
-        # Clean up temp directory
-        shutil.rmtree(temp_dir)
-        
-        return f"✅ Successfully processed {len(documents)} documents into {len(chunks)} chunks. Ready for questions!"
-    
+
+        return f"✅ {len(documents)} docs loaded, {len(chunks)} chunks"
     except Exception as e:
-        logger.error(f"PDF processing error: {e}")
-        return f"❌ Error processing PDFs: {str(e)}"
+        return f"❌ Error: {str(e)}"
 
 def answer_question(question):
-    """Answer a question using the RAG system"""
     global retrieval_qa
-    
     if not question.strip():
-        return "❌ Please enter a question.", ""
-    
+        return "❌ Enter a question", ""
     if retrieval_qa is None:
-        return "❌ Please upload and process PDF files first.", ""
-    
+        return "❌ Process documents first", ""
+
     try:
-        # Get answer from RAG system
         result = retrieval_qa({"query": question})
-        
-        answer = result["result"]
-        
-        # Format source documents
+        answer = result.get("result", "No answer")
+
         sources = []
         for i, doc in enumerate(result.get("source_documents", []), 1):
             source = doc.metadata.get("source", "Unknown")
             page = doc.metadata.get("page", "Unknown")
-            content_preview = doc.page_content[:200] + "..." if len(doc.page_content) > 200 else doc.page_content
-            
-            sources.append(f"**Source {i}:**\n- File: {Path(source).name}\n- Page: {page}\n- Preview: {content_preview}\n")
-        
-        sources_text = "\n".join(sources) if sources else "No sources found."
-        
-        return answer, sources_text
-    
+            preview = doc.page_content[:200] + "..." if len(doc.page_content) > 200 else doc.page_content
+            sources.append(f"**Source {i}:** {Path(source).name} (Page {page})\n{preview}")
+
+        return answer, "\n\n".join(sources)
     except Exception as e:
-        logger.error(f"Question answering error: {e}")
-        return f"❌ Error answering question: {str(e)}", ""
-
-def get_device_info():
-    """Simple function to detect if mobile (basic detection)"""
-    return """
-    <script>
-    function isMobile() {
-        return window.innerWidth <= 768;
-    }
-    
-    function adjustLayout() {
-        const isMob = isMobile();
-        const root = document.documentElement;
-        if (isMob) {
-            root.style.setProperty('--mobile-mode', '1');
-        } else {
-            root.style.setProperty('--mobile-mode', '0');
-        }
-    }
-    
-    window.addEventListener('resize', adjustLayout);
-    adjustLayout();
-    </script>
-    """
+        return f"❌ Error: {str(e)}", ""
 
 def create_interface():
-    """Create the fully responsive Gradio interface"""
-    
-    # Custom CSS for better responsiveness
-    custom_css = """
-    /* Base responsive styles */
-    .gradio-container {
-        max-width: 100% !important;
-        margin: 0 auto;
-        padding: 10px;
-    }
-    
-    /* Mobile-first responsive design */
-    @media (max-width: 768px) {
-        .gradio-container {
-            padding: 5px;
-        }
-        
-        /* Stack elements vertically on mobile */
-        .gr-row {
-            flex-direction: column !important;
-            gap: 10px !important;
-        }
-        
-        /* Full width on mobile */
-        .gr-column {
-            width: 100% !important;
-            min-width: 100% !important;
-        }
-        
-        /* Adjust component spacing */
-        .gr-form > * {
-            margin-bottom: 8px !important;
-        }
-        
-        /* Better button sizing */
-        .gr-button {
-            width: 100% !important;
-            min-height: 44px !important;
-            font-size: 14px !important;
-        }
-        
-        /* Text input improvements */
-        .gr-textbox textarea {
-            min-height: 60px !important;
-            font-size: 16px !important; /* Prevents zoom on iOS */
-        }
-        
-        /* File upload improvements */
-        .gr-file {
-            min-height: 100px !important;
-        }
-        
-        /* Slider improvements */
-        .gr-slider {
-            margin: 10px 0 !important;
-        }
-        
-        /* Tab improvements */
-        .gr-tab-nav {
-            flex-wrap: wrap !important;
-        }
-        
-        .gr-tab-nav > button {
-            flex: 1 1 auto !important;
-            min-width: 80px !important;
-            font-size: 12px !important;
-        }
-    }
-    
-    /* Tablet styles */
-    @media (min-width: 769px) and (max-width: 1024px) {
-        .gradio-container {
-            padding: 15px;
-        }
-        
-        .gr-button {
-            min-height: 40px !important;
-        }
-    }
-    
-    /* Desktop styles */
-    @media (min-width: 1025px) {
-        .gradio-container {
-            max-width: 1400px;
-            padding: 20px;
-        }
-    }
-    
-    /* Improve readability */
-    .gr-markdown h1 {
-        font-size: clamp(1.5rem, 4vw, 2.5rem) !important;
-        line-height: 1.2 !important;
-        margin-bottom: 1rem !important;
-    }
-    
-    .gr-markdown h3 {
-        font-size: clamp(1.1rem, 3vw, 1.4rem) !important;
-        margin: 1rem 0 0.5rem 0 !important;
-    }
-    
-    .gr-markdown p, .gr-markdown li {
-        font-size: clamp(0.9rem, 2.5vw, 1rem) !important;
-        line-height: 1.5 !important;
-    }
-    
-    /* Status text improvements */
-    .gr-textbox[data-testid="textbox"] {
-        font-family: monospace !important;
-        font-size: clamp(0.8rem, 2vw, 0.9rem) !important;
-    }
-    
-    /* Accessibility improvements */
-    .gr-button:focus,
-    .gr-textbox:focus,
-    .gr-file:focus {
-        outline: 2px solid #2563eb !important;
-        outline-offset: 2px !important;
-    }
-    
-    /* Dark mode considerations */
-    @media (prefers-color-scheme: dark) {
-        .gr-button {
-            border: 1px solid #374151 !important;
-        }
-    }
-    """
-    
-    with gr.Blocks(
-        title="PDF RAG System", 
-        theme=gr.themes.Soft(),
-        css=custom_css
-    ) as demo:
-        
-        # Add device detection script
-        gr.HTML(get_device_info())
-        
-        gr.Markdown("""
-        # 📚 PDF Question Answering System
-        
-        Upload your PDF documents and ask questions about their content!
-        
-        **Quick Start:**
-        1. Upload PDFs or use pre-loaded ones
-        2. Click Process to prepare your documents  
-        3. Ask questions about the content
-        """)
-        
-        # Check for pre-loaded PDFs
-        if PRELOADED_PDFS:
-            gr.Markdown("""
-            <div style="background: linear-gradient(90deg, #10b981, #059669); 
-                        color: white; padding: 12px; border-radius: 8px; margin: 10px 0;">
-            🎉 <strong>Pre-loaded PDFs detected!</strong> Use the 'Load Pre-existing PDFs' button to get started quickly.
-            </div>
-            """)
-        
-        # Main layout - responsive columns
+    with gr.Blocks(title="RAG PDF QA") as demo:
+        gr.Markdown("## PDF QA with LangChain + HuggingFaceHub")
+
         with gr.Row():
-            # Left column - Upload & Settings (collapses to full width on mobile)
-            with gr.Column(scale=1, min_width=300):
-                gr.Markdown("### 📄 Document Management")
-                
-                with gr.Tabs():
-                    with gr.TabItem("📁 Upload PDFs"):
-                        pdf_files = gr.File(
-                            label="Select PDF Files",
-                            file_count="multiple",
-                            file_types=[".pdf"],
-                            height=120
-                        )
-                        process_btn = gr.Button(
-                            "🔄 Process PDFs", 
-                            variant="primary",
-                            size="lg"
-                        )
-                    
-                    with gr.TabItem("🗂️ ZIP Upload"):
-                        zip_file = gr.File(
-                            label="Upload ZIP (with PDFs)",
-                            file_count="single",
-                            file_types=[".zip"],
-                            height=80
-                        )
-                        extract_btn = gr.Button(
-                            "📦 Extract ZIP", 
-                            variant="secondary",
-                            size="lg"
-                        )
-                        extract_output = gr.Textbox(
-                            label="Extraction Status", 
-                            lines=2,
-                            max_lines=3
-                        )
-                    
-                    with gr.TabItem("💾 Pre-loaded"):
-                        if PRELOADED_PDFS:
-                            pdf_list = [f for f in os.listdir(PDF_FOLDER_PATH) if f.endswith('.pdf')]
-                            gr.Markdown(f"**Found {len(pdf_list)} PDF files**")
-                            
-                            # Show files in a more mobile-friendly way
-                            if len(pdf_list) <= 5:
-                                for pdf in pdf_list:
-                                    gr.Markdown(f"📄 {pdf}")
-                            else:
-                                for pdf in pdf_list[:3]:
-                                    gr.Markdown(f"📄 {pdf}")
-                                gr.Markdown(f"*... and {len(pdf_list) - 3} more files*")
-                        else:
-                            gr.Markdown("No pre-loaded PDFs found.")
-                        
-                        preload_btn = gr.Button(
-                            "📚 Load Pre-existing PDFs", 
-                            variant="primary",
-                            size="lg",
-                            interactive=PRELOADED_PDFS
-                        )
-                
-                # Settings section - collapsible on mobile
-                with gr.Accordion("⚙️ Advanced Settings", open=False):
-                    chunk_size = gr.Slider(
-                        minimum=200,
-                        maximum=2000,
-                        value=1000,
-                        step=100,
-                        label="Chunk Size",
-                        info="Larger chunks = more context, smaller = more precise"
-                    )
-                    
-                    chunk_overlap = gr.Slider(
-                        minimum=0,
-                        maximum=500,
-                        value=200,
-                        step=50,
-                        label="Chunk Overlap",
-                        info="Overlap between text chunks"
-                    )
-                
-                # Status display
-                process_output = gr.Textbox(
-                    label="📊 Processing Status", 
-                    lines=3,
-                    max_lines=5,
-                    placeholder="Status updates will appear here..."
-                )
-            
-            # Right column - Q&A Section (collapses to full width on mobile)
-            with gr.Column(scale=2, min_width=400):
-                gr.Markdown("### ❓ Ask Questions")
-                
-                question_input = gr.Textbox(
-                    label="Your Question",
-                    placeholder="What would you like to know about your documents?",
-                    lines=2,
-                    max_lines=4
-                )
-                
-                ask_btn = gr.Button(
-                    "🤔 Ask Question", 
-                    variant="secondary",
-                    size="lg"
-                )
-                
-                # Results section - stack vertically on mobile
-                with gr.Row():
-                    answer_output = gr.Textbox(
-                        label="💡 Answer",
-                        lines=6,
-                        max_lines=12,
-                        placeholder="Your answer will appear here..."
-                    )
-                    
-                    sources_output = gr.Textbox(
-                        label="📚 Sources",
-                        lines=6,
-                        max_lines=12,
-                        placeholder="Source references will appear here..."
-                    )
-        
-        # Event handlers (unchanged)
+            with gr.Column():
+                pdf_files = gr.File(file_types=[".pdf"], file_count="multiple", label="Upload PDFs")
+                chunk_size = gr.Slider(200, 2000, value=1000, label="Chunk Size")
+                chunk_overlap = gr.Slider(0, 500, value=200, label="Chunk Overlap")
+                process_btn = gr.Button("🔄 Process PDFs")
+                process_output = gr.Textbox(label="Processing Result")
+
+            with gr.Column():
+                question = gr.Textbox(label="Ask a Question")
+                ask_btn = gr.Button("🤔 Ask")
+                answer = gr.Textbox(label="Answer")
+                sources = gr.Textbox(label="Sources")
+
         process_btn.click(
-            fn=process_pdfs,
-            inputs=[pdf_files, chunk_size, chunk_overlap],
-            outputs=[process_output]
-        )
-        
-        preload_btn.click(
             fn=load_preloaded_pdfs,
             inputs=[chunk_size, chunk_overlap],
             outputs=[process_output]
         )
-        
-        extract_btn.click(
-            fn=extract_zip_to_pdfs,
-            inputs=[zip_file],
-            outputs=[extract_output]
-        )
-        
+
         ask_btn.click(
             fn=answer_question,
-            inputs=[question_input],
-            outputs=[answer_output, sources_output]
-        )
-        
-        question_input.submit(
-            fn=answer_question,
-            inputs=[question_input],
-            outputs=[answer_output, sources_output]
+            inputs=[question],
+            outputs=[answer, sources]
         )
-        
-        # Example questions - more mobile-friendly
-        with gr.Accordion("💡 Example Questions", open=False):
-            gr.Markdown("""
-            **Try asking:**
-            - What are the main topics in these documents?
-            - Can you summarize the key findings?
-            - What data is available for [specific topic]?
-            - What are the differences between X and Y?
-            """)
-        
-        # Footer with helpful info
-        gr.Markdown("""
-        ---
-        <div style="text-align: center; color: #666; font-size: 0.9em;">
-        💡 <strong>Tip:</strong> For best results, ask specific questions about your documents
-        </div>
-        """)
-    
+
     return demo
 
 if __name__ == "__main__":
-    # Check if running on HuggingFace Spaces
-    if os.getenv("SPACE_ID"):
-        demo = create_interface()
-        demo.launch(
-            server_name="0.0.0.0",
-            server_port=7860,
-            share=False
-        )
-    else:
-        # Local development
-        demo = create_interface()
-        demo.launch(share=True)
+    demo = create_interface()
+    demo.launch(share=True)