final commit - callbacks logging needs work

code/.chainlit/config.toml

@@ -23,7 +23,7 @@ allow_origins = ["*"]
 unsafe_allow_html = false
 
 # Process and display mathematical expressions. This can clash with "$" characters in messages.
-latex = false
+latex = true
 
 # Automatically tag threads with the current chat profile (if a chat profile is used)
 auto_tag_thread = true

code/main.py

@@ -70,11 +70,12 @@ class Chatbot:
         start_time = time.time()
 
         llm_settings = cl.user_session.get("llm_settings", {})
-        chat_profile, retriever_method, memory_window, llm_style = (
+        chat_profile, retriever_method, memory_window, llm_style, generate_follow_up = (
             llm_settings.get("chat_model"),
             llm_settings.get("retriever_method"),
             llm_settings.get("memory_window"),
             llm_settings.get("llm_style"),
+            llm_settings.get("follow_up_questions"),
         )
 
         chain = cl.user_session.get("chain")
@@ -87,22 +88,24 @@ class Chatbot:
             ),
         )
         conversation_list = get_history_setup_llm(memory_list)
-        print("\n\n\n")
-        print("history at setup_llm", conversation_list)
-        print("\n\n\n")
 
         old_config = copy.deepcopy(self.config)
         self.config["vectorstore"]["db_option"] = retriever_method
         self.config["llm_params"]["memory_window"] = memory_window
         self.config["llm_params"]["llm_style"] = llm_style
         self.config["llm_params"]["llm_loader"] = chat_profile
+        self.config["llm_params"]["generate_follow_up"] = generate_follow_up
 
         self.llm_tutor.update_llm(
             old_config, self.config
-        )  # update only attributes that are changed
+        )  # update only llm attributes that are changed
         self.chain = self.llm_tutor.qa_bot(
             memory=conversation_list,
-            callbacks=[cl.LangchainCallbackHandler()] if cl_data._data_layer else None,
+            callbacks=(
+                [cl.LangchainCallbackHandler()]
+                if cl_data._data_layer and self.config["chat_logging"]["callbacks"]
+                else None
+            ),
         )
 
         tags = [chat_profile, self.config["vectorstore"]["db_option"]]
@@ -165,7 +168,14 @@ class Chatbot:
                     id="view_sources", label="View Sources", initial=False
                 ),
                 cl.input_widget.Switch(
-                    id="stream_response", label="Stream response", initial=False
+                    id="stream_response",
+                    label="Stream response",
+                    initial=config["llm_params"]["stream"],
+                ),
+                cl.input_widget.Switch(
+                    id="follow_up_questions",
+                    label="Generate follow up questions",
+                    initial=False,
                 ),
                 cl.input_widget.Select(
                     id="llm_style",
@@ -193,6 +203,7 @@ class Chatbot:
                 else 0
             ),
             "view_sources": llm_settings.get("view_sources"),
+            "follow_up_questions": llm_settings.get("follow_up_questions"),
         }
         await cl.Message(
             author=SYSTEM,
@@ -270,21 +281,21 @@ class Chatbot:
             "user_id": user.identifier,
             "session_id": cl.context.session.thread_id,
         }
-        print(self.user)
 
         memory = cl.user_session.get("memory", [])
 
         cl.user_session.set("user", self.user)
         self.llm_tutor = LLMTutor(self.config, user=self.user)
 
-        print(cl.LangchainCallbackHandler())
-        print(cl_data._data_layer)
         self.chain = self.llm_tutor.qa_bot(
             memory=memory,
-            callbacks=[cl.LangchainCallbackHandler()] if cl_data._data_layer else None,
+            callbacks=(
+                [cl.LangchainCallbackHandler()]
+                if cl_data._data_layer and self.config["chat_logging"]["callbacks"]
+                else None
+            ),
         )
         self.question_generator = self.llm_tutor.question_generator
-        print(self.question_generator)
         cl.user_session.set("llm_tutor", self.llm_tutor)
         cl.user_session.set("chain", self.chain)
 
@@ -324,22 +335,10 @@ class Chatbot:
 
         chain = cl.user_session.get("chain")
 
-        print("\n\n\n")
-        print(
-            "session history",
-            chain.get_session_history(
-                self.user["user_id"],
-                self.user["session_id"],
-                self.config["llm_params"]["memory_window"],
-            ),
-        )
-        print("\n\n\n")
-
         llm_settings = cl.user_session.get("llm_settings", {})
         view_sources = llm_settings.get("view_sources", False)
-        stream = (llm_settings.get("stream_response", True)) or (
-            not self.config["llm_params"]["stream"]
-        )
+        stream = llm_settings.get("stream_response", False)
+        steam = False  # Fix streaming
         user_query_dict = {"input": message.content}
         # Define the base configuration
         chain_config = {
@@ -350,8 +349,6 @@ class Chatbot:
             }
         }
 
-        stream = False
-
         if stream:
             res = chain.stream(user_query=user_query_dict, config=chain_config)
             res = await self.stream_response(res)
@@ -385,31 +382,33 @@ class Chatbot:
         answer_with_sources, source_elements, sources_dict = get_sources(
             res, answer, stream=stream, view_sources=view_sources
         )
+        answer_with_sources = answer_with_sources.replace("$$", "$")
 
         print("Time taken to process the message: ", time.time() - start_time)
 
-        list_of_questions = self.question_generator.generate_questions(
-            query=user_query_dict["input"],
-            response=answer,
-            chat_history=res.get("chat_history"),
-            context=res.get("context"),
-        )
+        actions = []
 
-        print("\n\n\n")
-        print("Questions: ", list_of_questions)
-        print("\n\n\n")
+        if self.config["llm_params"]["generate_follow_up"]:
+            start_time = time.time()
+            list_of_questions = self.question_generator.generate_questions(
+                query=user_query_dict["input"],
+                response=answer,
+                chat_history=res.get("chat_history"),
+                context=res.get("context"),
+            )
 
-        actions = []
-        for question in list_of_questions:
-
-            actions.append(
-                cl.Action(
-                    name="follow up question",
-                    value="example_value",
-                    description=question,
-                    label=question,
+            for question in list_of_questions:
+
+                actions.append(
+                    cl.Action(
+                        name="follow up question",
+                        value="example_value",
+                        description=question,
+                        label=question,
+                    )
                 )
-            )
+
+            print("Time taken to generate questions: ", time.time() - start_time)
 
         await cl.Message(
             content=answer_with_sources,
@@ -422,11 +421,6 @@ class Chatbot:
         steps = thread["steps"]
         k = self.config["llm_params"]["memory_window"]
         conversation_list = get_history_chat_resume(steps, k, SYSTEM, LLM)
-
-        print("\n\n\n")
-        print("history at on_chat_resume", conversation_list)
-        print(len(conversation_list))
-        print("\n\n\n")
         cl.user_session.set("memory", conversation_list)
         await self.start()
 
@@ -439,9 +433,11 @@ class Chatbot:
     ) -> Optional[cl.User]:
         return default_user
 
-    async def on_action(self, action: cl.Action):
+    async def on_follow_up(self, action: cl.Action):
         message = await cl.Message(
-            content=action.description, type="user_message"
+            content=action.description,
+            type="user_message",
+            author=self.user["user_id"],
         ).send()
         await self.main(message)
 
@@ -449,11 +445,8 @@ class Chatbot:
 chatbot = Chatbot(config=config)
 
 
-async def start():
-    print("Setting up data layer...")
+async def start_app():
     cl_data._data_layer = await setup_data_layer()
-    print("Data layer set up.")
-    print(cl_data._data_layer)
     chatbot.literal_client = cl_data._data_layer.client if cl_data._data_layer else None
     cl.set_starters(chatbot.set_starters)
     cl.author_rename(chatbot.rename)
@@ -461,7 +454,7 @@ async def start():
     cl.on_chat_resume(chatbot.on_chat_resume)
     cl.on_message(chatbot.main)
     cl.on_settings_update(chatbot.update_llm)
-    cl.action_callback("follow up question")(chatbot.on_action)
+    cl.action_callback("follow up question")(chatbot.on_follow_up)
 
 
-asyncio.run(start())
+asyncio.run(start_app())

code/modules/chat/helpers.py

@@ -6,11 +6,6 @@ def get_sources(res, answer, stream=True, view_sources=False):
     source_elements = []
     source_dict = {}  # Dictionary to store URL elements
 
-    print("\n\n\n")
-    print(res["context"])
-    print(len(res["context"]))
-    print("\n\n\n")
-
     for idx, source in enumerate(res["context"]):
         source_metadata = source.metadata
         url = source_metadata.get("source", "N/A")
@@ -25,9 +20,6 @@ def get_sources(res, answer, stream=True, view_sources=False):
         source_type = source_metadata.get("source_type", "N/A")
 
         url_name = f"{url}_{page}"
-        print("url")
-        print(url_name)
-        print("\n\n\n")
         if url_name not in source_dict:
             source_dict[url_name] = {
                 "text": source.page_content,
@@ -122,7 +114,6 @@ def get_history_chat_resume(steps, k, SYSTEM, LLM):
     conversation_list = []
     count = 0
     for step in reversed(steps):
-        print(step["type"])
         if step["name"] not in [SYSTEM]:
             if step["type"] == "user_message":
                 conversation_list.append(

code/modules/chat/langchain/utils.py

@@ -198,13 +198,7 @@ class CustomRunnableWithHistory(RunnableWithMessageHistory):
             List[BaseMessage]: The last k conversations.
         """
         hist: BaseChatMessageHistory = config["configurable"]["message_history"]
-
-        print("\n\n\n")
-        print("Hist: ", hist)
-        print("\n\n\n")
         messages = (await hist.aget_messages()).copy()
-        print("messages: ", messages)
-
         if not self.history_messages_key:
             # return all messages
             input_val = (
@@ -220,9 +214,6 @@ class CustomRunnableWithHistory(RunnableWithMessageHistory):
 
         messages = self._get_chat_history(messages)
 
-        print("\n\n\n")
-        print("Messages: ", messages)
-        print("\n\n\n")
         return messages
 
 

code/modules/chat/llm_tutor.py

@@ -41,10 +41,6 @@ class LLMTutor:
         """
         changes = self.get_config_changes(old_config, new_config)
 
-        print("\n\n\n")
-        print("Changes: ", changes)
-        print("\n\n\n")
-
         if "llm_params.llm_loader" in changes:
             self.llm = self.load_llm()  # Reinitialize LLM if chat_model changes
 

code/modules/chat_processor/literal_ai.py

@@ -37,8 +37,8 @@ class CustomLiteralDataLayer(ChainlitDataLayer):
         if step_dict.get("isError"):
             step["error"] = step_dict.get("output")
 
-        print("\n\n\n")
-        print("Step: ", step)
-        print("\n\n\n")
+        # print("\n\n\n")
+        # print("Step: ", step)
+        # print("\n\n\n")
 
         await self.client.api.send_steps([step])

code/modules/config/config.yml

@@ -3,7 +3,7 @@ log_chunk_dir: '../storage/logs/chunks' # str
 device: 'cpu' # str [cuda, cpu]
 
 vectorstore:
-  load_from_HF: False # bool
+  load_from_HF: True # bool
   embedd_files: False # bool
   data_path: '../storage/data' # str
   url_file_path: '../storage/data/urls.txt' # str
@@ -25,8 +25,9 @@ vectorstore:
     index_name: "new_idx" # str
 
 llm_params: 
-  llm_arch: 'langchain' # [langchain, langgraph_agentic]
+  llm_arch: 'langchain' # [langchain]
   use_history: True # bool
+  generate_follow_up: False # bool
   memory_window: 3 # int
   llm_style: 'Normal' # str [Normal, ELI5]
   llm_loader: 'gpt-4o-mini' # str [local_llm, gpt-3.5-turbo-1106, gpt-4, gpt-4o-mini]
@@ -34,12 +35,12 @@ llm_params:
     temperature: 0.7 # float
   local_llm_params:
     temperature: 0.7 # float
-  stream: True # bool
+  stream: False # bool
 
 chat_logging:
-  log_chat: False # bool
+  log_chat: True # bool
   platform: 'literalai'
-  callbacks: True # bool
+  callbacks: False # bool
 
 splitter_options:
   use_splitter: True # bool

code/modules/config/prompts.py

@@ -16,7 +16,9 @@ prompts = {
                 "You are an AI Tutor for the course DS598, taught by Prof. Thomas Gardos. Answer the user's question using the provided context. Only use the context if it is relevant. The context is ordered by relevance. "
                 "If you don't know the answer, do your best without making things up. Keep the conversation flowing naturally. "
                 "Use chat history and context as guides but avoid repeating past responses. Provide links from the source_file metadata. Use the source context that is most relevant. "
+                "Render math equations in LaTeX format between $ or $$ signs, stick to the parameter and variable icons found in your context. Be sure to explain the parameters and variables in the equations."
                 "Speak in a friendly and engaging manner, like talking to a friend. Avoid sounding repetitive or robotic.\n\n"
+                "Do not get influenced by the style of conversation in the chat history. Follow the instructions given here."
                 "Chat History:\n{chat_history}\n\n"
                 "Context:\n{context}\n\n"
                 "Answer the student's question below in a friendly, concise, and engaging manner. Use the context and history only if relevant, otherwise, engage in a free-flowing conversation.\n"
@@ -24,15 +26,17 @@ prompts = {
                 "AI Tutor:"
             ),
             "eli5": (
-                "You are an AI Tutor for the course DS598, taught by Prof. Thomas Gardos. Answer the user's question in the simplest way possible, like explaining to someone new to the topic. Use the provided context to help clarify your explanation."
-                "If you don't know the answer, do your best without making things up. Keep the conversation straightforward and easy to follow."
-                "Use chat history and context as guides but avoid repeating past responses. Provide links from the source_file metadata when relevant. Use the source context that is most relevant."
-                "Speak in a friendly and engaging manner, like talking to someone who is curious and eager to learn. Avoid using complex terms and jargon."
-                "Use examples wherever possible to aid in understanding."
+                "You are an AI Tutor for the course DS598, taught by Prof. Thomas Gardos. Your job is to explain things in the simplest and most engaging way possible, just like the 'Explain Like I'm 5' (ELI5) concept."
+                "If you don't know the answer, do your best without making things up. Keep your explanations straightforward and very easy to understand."
+                "Use the chat history and context to help you, but avoid repeating past responses. Provide links from the source_file metadata when they're helpful."
+                "Use very simple language and examples to explain any math equations, and put the equations in LaTeX format between $ or $$ signs."
+                "Be friendly and engaging, like you're chatting with a young child who's curious and eager to learn. Avoid complex terms and jargon."
+                "Include simple and clear examples wherever you can to make things easier to understand."
+                "Do not get influenced by the style of conversation in the chat history. Follow the instructions given here."
                 "Chat History:\n{chat_history}\n\n"
                 "Context:\n{context}\n\n"
-                "Answer the student's question below in a friendly, simple, and engaging manner. Use the context and history only if relevant, otherwise, engage in a free-flowing conversation."
-                "Give a clear and detailed explanation with examples to make it easier to understand."
+                "Answer the student's question below in a friendly, simple, and engaging way, just like the ELI5 concept. Use the context and history only if they're relevant, otherwise, just have a natural conversation."
+                "Give a clear and detailed explanation with simple examples to make it easier to understand. Remember, your goal is to break down complex topics into very simple terms, just like ELI5."
                 "Student: {input}\n"
                 "AI Tutor:"
             ),