Update app.py

app.py

@@ -11,64 +11,23 @@ import json
 
 import chainlit as cl
 
+#from tiktoken import encoding_for_model
+
 from langchain import hub
 from langchain_openai import OpenAI
 from langchain.chains import LLMChain, APIChain
 from langchain_core.prompts import PromptTemplate
 from langchain.memory.buffer import ConversationBufferMemory
+from langchain.memory import ConversationTokenBufferMemory
+from langchain.memory import ConversationSummaryMemory
+
 
 from api_docs_mck import api_docs_str 
-from faq_data import ansatte_faq_data, utleiere_faq_data
-from personvernspolicy import personvernspolicy_data
+#from faq_data import ansatte_faq_data, utleiere_faq_data
+#from personvernspolicy import personvernspolicy_data
 
 OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY")
 
-FAQ_ANSATTE = [
-    "Hvordan registrerer jeg meg som bruker?",
-    "Når får jeg leieinstruks for min bestilling? Informasjon om nøkler etc.?",
-    "Det står barneseng og barnestol under fasiliteter, må dette forhåndsbestilles?",
-    "Kan jeg ta med hund eller katt?",
-    "Jeg har lagt inn en bestilling hva skjer videre?",
-    "Jeg har bestilt firmahytte, men kan ikke reise. Kan jeg endre navn på bestillingen til min kollega eller familiemedlem som vil reise i stedet for meg?",
-    "Kan jeg avbestille min reservasjon?",
-    "Jeg har bestilt utvask. Hva må jeg gjøre i tillegg til dette?",
-    "Jeg er medlem og eier en hytte! Kan jeg bli utleier i DaysOff?",
-    "Bestille opphold?"
-]   
-
-FAQ_UTLEIERE = [
-    "Hva er betingelser for utleie?",
-    "Hvor lang tid har jeg på å bekrefte en bestilling?",
-    "Hvilke kanselleringsregler gjelder?",
-    "Hvem er kundene deres?",
-    "Kan jeg legge inn rabatterte priser for å lage egne kampanjer?",
-    "Når mottar jeg betaling for leie?",
-    "Jeg fikk en e-post om ny bestilling, men jeg finner den ikke i systemet?",
-    "Hvordan registrerer jeg opptatte perioder i kalenderen?",
-    "Jeg leier ut i andre kanaler. Hvordan kan jeg synkronisere kalenderne?"
-]  
-
-PERSONVERNSPOLICY_QUESTIONS = [
-    "Hvilke personlige opplysninger samler vi inn?",
-    "Kan dere motta personlig informasjon fra tredjepart?",
-    "Hvordan bruker vi dine personlige opplysninger?",
-    "Med hvem deler vi dine personlige opplysninger?",
-    "Adferdsmessig annonsering?",
-    "Hvordan reagerer vi på « Spor ikke » forespørsler?",
-    "Hva er dine rettigheter?",
-    "Hvordan beskytter vi dataene dine?",
-    "Hvilke data brudd prosedyrer har vi på plass?",
-    "Hvem i vårt team har tilgang til dine data?",
-    "Endringer i denne policyen"
-]
-
-# If you do not know the answer, just reply truthfully that you do not have the answer rather than
-# giving inaccurate or speculative information.
-
-#(i) retrieving booking information for a given booking ID,
-#(ii) answering frequently asked questions (FAQ) about DaysOff firmahytteordning for employees and employers,
-#(iii) addressing questions related to Daysoff's personvernspolicy.
-
 daysoff_assistant_template = """
 You are a customer support assistant (’kundeservice AI assistent’) for Daysoff. 
 By default, you respond in Norwegian language, using a warm, direct, and professional tone. 
@@ -116,6 +75,19 @@ api_response_prompt = PromptTemplate(
     template=api_response_template
 )
 
+# ---------------------------------------------------------------------------------------------------------
+# 100 tokens ≃ 75 words
+# system prompt(s), total = 330 tokens
+# average api response = 250-300 tokens (current)
+# user input "reserved" = 400 tokens (300 words max. /English; Polish, Norwegian {..}?@tiktokenizer), could be reduc3d to 140 tokens ≃ 105 words
+# model output (max_tokens) = 2048
+
+# ConversationBufferMemory = maintains raw chat history; crucial for "nuanced" follow-ups (e.g. "nuanced" ~ for non-English inputs)
+# ConversationTokenBufferMemory (max_token_limit) = 1318 (gives space in chat_history for approximately 10-15 exchanges, assuming ~100 tokens/exchange) 
+# ConversationSummaryMemory = scalable approach, especially useful for extended or complex interactions, caveat: loss of granular context
+# ---------------------------------------------------------------------------------------------------------
+
+
 @cl.on_chat_start
 def setup_multiple_chains():
 
@@ -123,16 +95,29 @@ def setup_multiple_chains():
         model='gpt-3.5-turbo-instruct',
         temperature=0.7, 
         openai_api_key=OPENAI_API_KEY,
-        #max_tokens=512, 
+        max_tokens=2048, 
         top_p=0.9,  
         frequency_penalty=0.1,
         presence_penalty=0.1   
     )
+
+    # --ConversationBufferMemory
+    #conversation_memory = ConversationBufferMemory(memory_key="chat_history", 
+                                                   #max_len=30,  # --retains only the last 30 exchanges
+                                                   #return_messages=True,
+    )
     
-    conversation_memory = ConversationBufferMemory(memory_key="chat_history",
-                                                   max_len=300,
-                                                   return_messages=True,
-                                                   )
+    # --ConversationTokenBufferMemory
+    conversation_memory = ConversationTokenBufferMemory(memory_key="chat_history",
+                                                        max_token_limit=1318,
+                                                        return_messages=True,
+    )
+
+    # --ConversationSummaryMemory
+    #conversation_memory = ConversationSummaryMemory(memory_key="chat_history",
+                                                    return_messages=True,
+    )
+                                                        
     llm_chain = LLMChain(llm=llm,
                          prompt=daysoff_assistant_prompt,
                          memory=conversation_memory
@@ -160,12 +145,6 @@ async def handle_message(message: cl.Message):
     booking_pattern = r'\b[A-Z]{6}\d{6}\b' 
     endpoint_url = "https://670dccd0073307b4ee447f2f.mockapi.io/daysoff/api/V1/booking"
 
-    #faq_keywords = (  
-    #[key for key in ansatte_faq_data.keys()] + 
-    #[key for key in utleiere_faq_data.keys()] +   
-    #[key for key in personvernspolicy_data.keys()]    
-    #)
-
     if re.search(booking_pattern, user_message):  
         bestillingskode = re.search(booking_pattern, user_message).group(0)  
         question = f"Retrieve information for booking ID {endpoint_url}?search={bestillingskode}"
@@ -178,13 +157,6 @@ async def handle_message(message: cl.Message):
             },
             callbacks=[cl.AsyncLangchainCallbackHandler()])
 
-        #if any(keyword in user_message for keyword in (FAQ_ANSATTE + FAQ_UTLEIERE + PERSONVERNSPOLICY_QUESTIONS)):
-            #elif any(keyword in user_message for keyword in faq_keywords): 
-            #response = await api_chain.acall(user_message, callbacks=[cl.AsyncLangchainCallbackHandler()])
-
-        #else:    
-        #response = await llm_chain.acall(user_message, callbacks=[cl.AsyncLangchainCallbackHandler()])
-
     else:    
         response = await llm_chain.acall(user_message, callbacks=[cl.AsyncLangchainCallbackHandler()])