Update app.py

app.py

@@ -177,45 +177,23 @@ def create_deepseek_interface():
                    f"- **Search Term**: {search_query}\n" + \
                    f"- **Parameters**: {params}\n"
     
-    # Test function without streaming
-    def test_simple_response(message, history):
-        """Simple test function to verify UI is working"""
-        print(f"Test function called with message: {message}")
-        # Convert to messages format
-        new_message = {"role": "user", "content": message}
-        response = {"role": "assistant", "content": f"Echo: {message}"}
-        return history + [new_message, response], ""
-    
-    # Function to call DeepSeek API with streaming
-    def query_deepseek_streaming(message, history, use_deep_research):
-        print(f"\n=== Starting query_deepseek_streaming ===")
+    # Simple non-streaming version
+    def query_deepseek_no_stream(message, history, use_deep_research):
+        """Non-streaming version that returns complete response"""
+        print(f"\n=== Starting query_deepseek_no_stream ===")
         print(f"Message: {message}")
         print(f"History length: {len(history)}")
         print(f"Deep Research: {use_deep_research}")
         
         if not api_key:
             error_msg = "Environment variable FW_API_KEY is not set. Please check the environment variables on the server."
-            # Convert to messages format
-            new_history = history + [
-                {"role": "user", "content": message},
-                {"role": "assistant", "content": error_msg}
-            ]
-            yield new_history, error_msg
-            return
+            return history + [(message, error_msg)], ""
         
         search_context = ""
         search_info = ""
         
-        # Create new history with the user message in messages format
-        new_history = history + [{"role": "user", "content": message}]
-        
         if use_deep_research:
             try:
-                # Add assistant message with search status
-                new_history.append({"role": "assistant", "content": "🔍 Extracting optimal keywords and searching the web..."})
-                yield new_history, ""
-                
-                # Execute search
                 print(f"Deep Research activated: Starting search for '{message}'")
                 search_results = search_with_serphouse(message)
                 print(f"Search results received: {search_results[:100]}...")
@@ -234,25 +212,13 @@ When citing search results, mention the source, and ensure your answer reflects
             except Exception as e:
                 print(f"Exception occurred during Deep Research: {str(e)}")
                 search_info = f"🔍 Deep Research feature error: {str(e)}\n\n"
-        else:
-            # Add empty assistant message to start
-            new_history.append({"role": "assistant", "content": ""})
-            yield new_history, ""
-        
-        # Update last assistant message with search info
-        if search_info:
-            new_history[-1]["content"] = search_info
-            yield new_history, ""
         
         # Prepare conversation history for API request
         messages = []
-        
-        # Convert history to API format
-        for msg in history:
-            if msg["role"] == "user":
-                messages.append({"role": "user", "content": msg["content"]})
-            elif msg["role"] == "assistant":
-                messages.append({"role": "assistant", "content": msg["content"]})
+        for user_msg, assistant_msg in history:
+            if user_msg and assistant_msg:  # Only add non-empty messages
+                messages.append({"role": "user", "content": user_msg})
+                messages.append({"role": "assistant", "content": assistant_msg})
         
         # Add system message with search context if available
         if search_context:
@@ -272,7 +238,7 @@ When citing search results, mention the source, and ensure your answer reflects
             "frequency_penalty": 0,
             "temperature": 0.6,
             "messages": messages,
-            "stream": True
+            "stream": False  # No streaming
         }
         headers = {
             "Accept": "application/json",
@@ -280,73 +246,26 @@ When citing search results, mention the source, and ensure your answer reflects
             "Authorization": f"Bearer {api_key}"
         }
         
-        print(f"Calling DeepSeek API...")
+        print(f"Calling DeepSeek API (no streaming)...")
         print(f"Number of messages in conversation: {len(messages)}")
         
         try:
-            response = requests.post(url, headers=headers, json=payload, stream=True)
+            # Add temporary message while waiting
+            temp_history = history + [(message, search_info + "⏳ Generating response...")]
+            
+            response = requests.post(url, headers=headers, json=payload)
             response.raise_for_status()
             
             print(f"API Response Status: {response.status_code}")
             
-            # Full response text
-            full_response = search_info
-            chunk_count = 0
-            
-            # Process streaming response
-            for line in response.iter_lines():
-                if line:
-                    line_text = line.decode('utf-8')
-                    
-                    # Remove 'data: ' prefix
-                    if line_text.startswith("data: "):
-                        line_text = line_text[6:]
-                    
-                    # Check for stream end message
-                    if line_text == "[DONE]":
-                        print(f"Stream completed. Total chunks: {chunk_count}")
-                        break
-                    
-                    try:
-                        # Parse JSON
-                        chunk = json.loads(line_text)
-                        
-                        # Check if there's content in the chunk
-                        if "choices" in chunk and len(chunk["choices"]) > 0:
-                            delta = chunk["choices"][0].get("delta", {})
-                            chunk_content = delta.get("content", "")
-                            
-                            if chunk_content:
-                                chunk_count += 1
-                                full_response += chunk_content
-                                
-                                # Update last assistant message
-                                new_history[-1]["content"] = full_response
-                                
-                                # Debug: Print first few chunks
-                                if chunk_count <= 5:
-                                    print(f"Chunk {chunk_count}: {repr(chunk_content[:50])}")
-                                
-                                # Yield less frequently to avoid overwhelming Gradio
-                                # Only yield every 10 chunks or if it's a significant chunk
-                                if chunk_count % 10 == 0 or len(chunk_content) > 50:
-                                    yield new_history, ""
-                                    time.sleep(0.01)  # Small delay to prevent connection errors
-                        
-                    except json.JSONDecodeError as e:
-                        print(f"JSON decode error: {e}")
-                        print(f"Line content: {repr(line_text[:100])}")
-                        continue
-            
-            # Final update
-            if not full_response or full_response == search_info:
-                # If no response was generated, add an error message
-                full_response = search_info + "Error: No response generated from the model. Please try again."
-                new_history[-1]["content"] = full_response
+            result = response.json()
+            full_response = search_info + result["choices"][0]["message"]["content"]
             
             print(f"Final response length: {len(full_response)}")
-            # Always yield the final state
-            yield new_history, ""
+            print(f"First 100 chars of response: {full_response[:100]}...")
+            
+            # Return the complete response
+            return history + [(message, full_response)], ""
             
         except requests.exceptions.RequestException as e:
             error_msg = f"API error: {str(e)}"
@@ -354,15 +273,23 @@ When citing search results, mention the source, and ensure your answer reflects
                 error_msg = "Authentication failed. Please check your FW_API_KEY environment variable."
             
             print(f"Request error: {error_msg}")
-            new_history[-1]["content"] = search_info + error_msg
-            yield new_history, error_msg
+            return history + [(message, search_info + error_msg)], ""
         except Exception as e:
             error_msg = f"Unexpected error: {str(e)}"
             print(f"Unexpected error: {error_msg}")
             import traceback
             traceback.print_exc()
-            new_history[-1]["content"] = search_info + error_msg
-            yield new_history, error_msg
+            return history + [(message, search_info + error_msg)], ""
+    
+    # Function to call DeepSeek API with streaming
+    def query_deepseek_streaming(message, history, use_deep_research):
+        print(f"\n=== Starting query_deepseek_streaming ===")
+        print(f"Message: {message}")
+        print(f"History length: {len(history)}")
+        print(f"Deep Research: {use_deep_research}")
+        
+        # For now, use non-streaming version
+        return query_deepseek_no_stream(message, history, use_deep_research)
     
     # Create Gradio interface
     with gr.Blocks(theme="soft", fill_height=True) as demo:
@@ -382,8 +309,8 @@ When citing search results, mention the source, and ensure your answer reflects
                 chatbot = gr.Chatbot(
                     height=500,
                     show_label=False,
-                    container=True,
-                    type="messages"  # Explicitly set to messages format
+                    container=True
+                    # Use default format
                 )
                 
                 # Add Deep Research toggle and status display
@@ -421,7 +348,6 @@ When citing search results, mention the source, and ensure your answer reflects
                 # Clear conversation button
                 with gr.Row():
                     clear = gr.ClearButton([msg, chatbot], value="🧹 Clear Conversation")
-                    test_btn = gr.Button("🧪 Test Echo", variant="secondary")
                     
                 # Example queries
                 gr.Examples(
@@ -436,19 +362,17 @@ When citing search results, mention the source, and ensure your answer reflects
                 # Error message display
                 error_box = gr.Markdown("")
         
-        # Connect buttons to functions
-        # Test button for debugging
-        test_btn.click(
-            test_simple_response,
-            inputs=[msg, chatbot],
-            outputs=[chatbot, error_box]
-        )
+        # Function to handle submission
+        def handle_submit(message, history, use_deep_research):
+            if not message.strip():
+                return history, ""
+            return query_deepseek_no_stream(message, history, use_deep_research)
         
+        # Connect buttons to functions
         submit.click(
-            query_deepseek_streaming,
+            handle_submit,
             inputs=[msg, chatbot, use_deep_research],
-            outputs=[chatbot, error_box],
-            queue=True  # Enable queue for streaming
+            outputs=[chatbot, error_box]
         ).then(
             lambda: "",
             None,
@@ -457,10 +381,9 @@ When citing search results, mention the source, and ensure your answer reflects
         
         # Allow Enter key submission
         msg.submit(
-            query_deepseek_streaming,
+            handle_submit,
             inputs=[msg, chatbot, use_deep_research],
-            outputs=[chatbot, error_box],
-            queue=True  # Enable queue for streaming
+            outputs=[chatbot, error_box]
         ).then(
             lambda: "",
             None,
@@ -471,12 +394,13 @@ When citing search results, mention the source, and ensure your answer reflects
 
 # Run interface
 if __name__ == "__main__":
+    print("Starting Deepseek Interface...")
     demo = create_deepseek_interface()
-    demo.queue()  # Enable queue with default settings
+    print("Launching Gradio interface...")
     demo.launch(
         debug=True, 
         share=False, 
         server_name="0.0.0.0", 
         server_port=7860, 
-        show_error=True  # Show detailed errors
+        show_error=True
     )