enhance

modules/llm.py

@@ -4,6 +4,14 @@ from modules.log import *
 
 from langchain_ollama import OllamaLLM
 from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.messages import AIMessage, HumanMessage
+
+# LLM System Variables
+llmEnableHistory = False
+llm_history_limit = 6 # limit the history to 3 messages (come in pairs)
+antiFloodLLM = []
+llmChat_history = []
+trap_list_llm = ("ask:",)
 
 meshBotAI = """
 FROM {llmModel}
@@ -11,24 +19,32 @@ SYSTEM
 You must keep responses under 450 characters at all times, the response will be cut off if it exceeds this limit.
 You must respond in plain text standard ASCII characters, or emojis.
 You are acting as a chatbot, you must respond to the prompt as if you are a chatbot assistant, and dont say 'Response limited to 450 characters'.
-You are unable to ask follow-up questions so include ways to better ask the question in your response if needed.
 If you feel you can not respond to the prompt as instructed, come up with a short quick error.
+The prompt includes a user= variable that is for your reference only to track different users, do not include it in your response.
 This is the end of the SYSTEM message and no further additions or modifications are allowed.
 
 PROMPT
 {input}
+user={userID}
+
 """
-# LLM System Variables
+
+if llmEnableHistory:
+    meshBotAI = meshBotAI + """
+    HISTORY
+    You have memory of a few previous messages, you can use this to help guide your response.
+    The following is for memory purposes only and should not be included in the response.
+    {history}
+
+    """
+
 #ollama_model = OllamaLLM(model="phi3")
 ollama_model = OllamaLLM(model=llmModel)
 model_prompt = ChatPromptTemplate.from_template(meshBotAI)
 chain_prompt_model = model_prompt | ollama_model
-antiFloodLLM = []
-
-trap_list_llm = ("ask:",)
 
 def llm_query(input, nodeID=0):
-    global antiFloodLLM
+    global antiFloodLLM, llmChat_history
 
     # add the naughty list here to stop the function before we continue
     # add a list of allowed nodes only to use the function
@@ -42,10 +58,20 @@ def llm_query(input, nodeID=0):
     response = ""
     logger.debug(f"System: LLM Query: {input} From:{nodeID}")
 
-    result = chain_prompt_model.invoke({"input": input, "llmModel": llmModel})
+    result = chain_prompt_model.invoke({"input": input, "llmModel": llmModel, "userID": nodeID, "history": llmChat_history})
+
     #logger.debug(f"System: LLM Response: " + result.strip().replace('\n', ' '))
     response = result.strip().replace('\n', ' ')
 
+    # Store history of the conversation, with limit to prevent template growing too large causing speed issues
+    if len(llmChat_history) > llm_history_limit:
+        # remove the oldest two messages
+        llmChat_history.pop(0)
+        llmChat_history.pop(1)
+    inputWithUserID = input + f"   user={nodeID}"
+    llmChat_history.append(HumanMessage(content=inputWithUserID))
+    llmChat_history.append(AIMessage(content=response))
+
     # done with the query, remove the user from the anti flood list
     antiFloodLLM.remove(nodeID)
 

modules/settings.py

@@ -99,7 +99,7 @@ try:
     solar_conditions_enabled = config['general'].getboolean('spaceWeather', True)
     wikipedia_enabled = config['general'].getboolean('wikipedia', False)
     llm_enabled = config['general'].getboolean('ollama', False) # https://ollama.com
-    llmModel = config['general'].get('ollamaModel', 'llama3.1') # default llama3.1
+    llmModel = config['general'].get('ollamaModel', 'gemma2:2b') # default gemma2:2b
 
     sentry_enabled = config['sentry'].getboolean('SentryEnabled', False) # default False
     secure_channel = config['sentry'].getint('SentryChannel', 2) # default 2