Update app.py

app.py

@@ -247,23 +247,42 @@ def respond(message, history, model, temperature, num_calls, use_web_search, sel
     logging.info(f"User Query: {message}")
     logging.info(f"Model Used: {model}")
     logging.info(f"Search Type: {'Web Search' if use_web_search else 'PDF Search'}")
-
     logging.info(f"Selected Documents: {selected_docs}")
 
     try:
         if use_web_search:
-            for main_content, sources in get_response_with_search(message, model, num_calls=num_calls, temperature=temperature):
-                response = f"{main_content}\n\n{sources}"
-                first_line = response.split('\n')[0] if response else ''
-#                logging.info(f"Generated Response (first line): {first_line}")
-                yield response
+            search_results = duckduckgo_search(message)
+            context = "\n".join(f"{result['title']}\n{result['body']}\nSource: {result['href']}\n" 
+                                for result in search_results if 'body' in result)
+            
+            prompt = f"""Using the following context:
+            {context}
+            Write a detailed and complete research document that fulfills the following user request: '{message}'
+            After writing the document, please provide a list of sources used in your response."""
+
+            if model == "@cf/meta/llama-3.1-8b-instruct":
+                for response in get_response_from_cloudflare(prompt="", context=context, query=message, num_calls=num_calls, temperature=temperature, search_type="web"):
+                    yield response
+            else:
+                client = InferenceClient(model, token=huggingface_token)
+                response = ""
+                for i in range(num_calls):
+                    for msg in client.chat_completion(
+                        messages=[{"role": "user", "content": prompt}],
+                        max_tokens=10000,
+                        temperature=temperature,
+                        stream=True,
+                    ):
+                        if msg.choices and msg.choices[0].delta and msg.choices[0].delta.content:
+                            chunk = msg.choices[0].delta.content
+                            response += chunk
+                            yield response
         else:
             embed = get_embeddings()
             if os.path.exists("faiss_database"):
                 database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
                 retriever = database.as_retriever()
                 
-                # Filter relevant documents based on user selection
                 all_relevant_docs = retriever.get_relevant_documents(message)
                 relevant_docs = [doc for doc in all_relevant_docs if doc.metadata["source"] in selected_docs]
                 
@@ -273,30 +292,34 @@ def respond(message, history, model, temperature, num_calls, use_web_search, sel
 
                 context_str = "\n".join([doc.page_content for doc in relevant_docs])
             else:
-                context_str = "No documents available."
                 yield "No documents available. Please upload PDF documents to answer questions."
                 return
             
             if model == "@cf/meta/llama-3.1-8b-instruct":
-                # Use Cloudflare API
-                for partial_response in get_response_from_cloudflare(prompt="", context=context_str, query=message, num_calls=num_calls, temperature=temperature, search_type="pdf"):
-                    first_line = partial_response.split('\n')[0] if partial_response else ''
-                    logging.info(f"Generated Response (first line): {first_line}")
-                    yield partial_response
+                for response in get_response_from_cloudflare(prompt="", context=context_str, query=message, num_calls=num_calls, temperature=temperature, search_type="pdf"):
+                    yield response
             else:
-                # Use Hugging Face API
-                for partial_response in get_response_from_pdf(message, model, selected_docs, num_calls=num_calls, temperature=temperature):
-                    first_line = partial_response.split('\n')[0] if partial_response else ''
-                    logging.info(f"Generated Response (first line): {first_line}")
-                    yield partial_response
+                prompt = f"""Using the following context from the PDF documents:
+                {context_str}
+                Write a detailed and complete response that answers the following user question: '{message}'"""
+                
+                client = InferenceClient(model, token=huggingface_token)
+                response = ""
+                for i in range(num_calls):
+                    for msg in client.chat_completion(
+                        messages=[{"role": "user", "content": prompt}],
+                        max_tokens=10000,
+                        temperature=temperature,
+                        stream=True,
+                    ):
+                        if msg.choices and msg.choices[0].delta and msg.choices[0].delta.content:
+                            chunk = msg.choices[0].delta.content
+                            response += chunk
+                            yield response
+
     except Exception as e:
         logging.error(f"Error with {model}: {str(e)}")
-        if "microsoft/Phi-3-mini-4k-instruct" in model:
-            logging.info("Falling back to Mistral model due to Phi-3 error")
-            fallback_model = "mistralai/Mistral-7B-Instruct-v0.3"
-            yield from respond(message, history, fallback_model, temperature, num_calls, use_web_search, selected_docs)
-        else:
-            yield f"An error occurred with the {model} model: {str(e)}. Please try again or select a different model."
+        yield f"An error occurred with the {model} model: {str(e)}. Please try again or select a different model."
 
 logging.basicConfig(level=logging.DEBUG)
 
@@ -481,57 +504,40 @@ document_selector = gr.CheckboxGroup(label="Select documents to query")
 
 use_web_search = gr.Checkbox(label="Use Web Search", value=True)
 
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Dropdown(choices=MODELS, label="Select Model", value=MODELS[3]),
-        gr.Slider(minimum=0.1, maximum=1.0, value=0.2, step=0.1, label="Temperature"),
-        gr.Slider(minimum=1, maximum=5, value=1, step=1, label="Number of API Calls"),
-        use_web_search,
-        document_selector  # Add the document selector to the chat interface
-    ],
-    title="AI-powered Web Search and PDF Chat Assistant",
-    description="Chat with your PDFs or use web search to answer questions (Please use toggle under Additional Inputs to swithc between PDF and Web Search, Default Value Web Search)",
-    theme=gr.themes.Soft(
-        primary_hue="orange",
-        secondary_hue="amber",
-        neutral_hue="gray",
-        font=[gr.themes.GoogleFont("Exo"), "ui-sans-serif", "system-ui", "sans-serif"]
-    ).set(
-        body_background_fill_dark="#0c0505",
-        block_background_fill_dark="#0c0505",
-        block_border_width="1px",
-        block_title_background_fill_dark="#1b0f0f",
-        input_background_fill_dark="#140b0b",
-        button_secondary_background_fill_dark="#140b0b",
-        border_color_accent_dark="#1b0f0f",
-        border_color_primary_dark="#1b0f0f",
-        background_fill_secondary_dark="#0c0505",
-        color_accent_soft_dark="transparent",
-        code_background_fill_dark="#140b0b"
-    ),
-    css=css,
-    examples=[
-        ["Tell me about the contents of the uploaded PDFs."],
-        ["What are the main topics discussed in the documents?"],
-        ["Can you summarize the key points from the PDFs?"]
-    ],
-    cache_examples=False,
-    analytics_enabled=False,
-)
-
-# Add file upload functionality
-with demo:
-    gr.Markdown("## Upload PDF Documents")
+with gr.Blocks(css=css, theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# AI-powered Web Search and PDF Chat Assistant")
+    gr.Markdown("Chat with your PDFs or use web search to answer questions")
+
+    use_web_search = gr.Checkbox(label="Use Web Search", value=True)
+    document_selector = gr.CheckboxGroup(label="Select documents to query")
+
+    chat_interface = gr.ChatInterface(
+        respond,
+        additional_inputs=[
+            gr.Dropdown(choices=MODELS, label="Select Model", value=MODELS[3]),
+            gr.Slider(minimum=0.1, maximum=1.0, value=0.2, step=0.1, label="Temperature"),
+            gr.Slider(minimum=1, maximum=5, value=1, step=1, label="Number of API Calls"),
+            use_web_search,
+            document_selector
+        ],
+        title="Chat Interface",
+        description="Ask questions about your PDFs or use web search",
+        examples=[
+            ["Tell me about the contents of the uploaded PDFs."],
+            ["What are the main topics discussed in the documents?"],
+            ["Can you summarize the key points from the PDFs?"]
+        ],
+        cache_examples=False,
+        analytics_enabled=False,
+    )
 
     with gr.Row():
         file_input = gr.Files(label="Upload your PDF documents", file_types=[".pdf"])
         parser_dropdown = gr.Dropdown(choices=["pypdf", "llamaparse"], label="Select PDF Parser", value="llamaparse")
         update_button = gr.Button("Upload Document")
-    
+
     update_output = gr.Textbox(label="Update Status")
-    
-    # Update both the output text and the document selector
+
     update_button.click(update_vectors, 
                         inputs=[file_input, parser_dropdown], 
                         outputs=[update_output, document_selector])
@@ -539,12 +545,12 @@ with demo:
     gr.Markdown(
     """
     ## How to use
-    1. Upload PDF documents using the file input at the top.
-    2. Select the PDF parser (pypdf or llamaparse) and click "Upload Document" to update the vector store.
-    3. Select the documents you want to query using the checkboxes.
-    4. Ask questions in the chat interface. 
-    5. Toggle "Use Web Search" to switch between PDF chat and web search.
-    6. Adjust Temperature and Number of API Calls to fine-tune the response generation.
+    1. Use the "Use Web Search" checkbox to switch between PDF chat and web search.
+    2. Upload PDF documents using the file input at the bottom.
+    3. Select the PDF parser (pypdf or llamaparse) and click "Upload Document" to update the vector store.
+    4. Select the documents you want to query using the checkboxes.
+    5. Adjust Temperature and Number of API Calls to fine-tune the response generation.
+    6. Type your questions in the chat interface and press Enter to get responses.
     7. Use the provided examples or ask your own questions.
     """
     )