integrated langfuse

requirements.txt

@@ -2,6 +2,7 @@ docarray==0.39.1
 faiss-cpu==1.7.4
 gradio==4.8.0
 langchain==0.0.348
+langfuse==2.5.0
 openai==1.3.8
 pypdf==3.17.2
 tiktoken==0.5.2

src/conversation.py

@@ -3,9 +3,12 @@ from langchain.chains import ConversationalRetrievalChain
 from langchain.chat_models import ChatOpenAI
 from langchain.embeddings import OpenAIEmbeddings
 from langchain.prompts import PromptTemplate
+from langfuse.callback import CallbackHandler
 import os
 
 openai_api_key = os.environ.get("OPENAI_API_KEY")
+langfuse_public_key = os.environ.get("LANGFUSE_PUBLIC_KEY")
+langfuse_secret_key = os.environ.get("LANGFUSE_SECRET_KEY")
 
 class Conversation_RAG:
     def __init__(self, model_name="gpt-3.5-turbo"):
@@ -17,7 +20,6 @@ class Conversation_RAG:
         return vectordb
     
     def create_model(self, max_new_tokens=512, temperature=0.1):
-
         llm = ChatOpenAI(
             openai_api_key=openai_api_key,
             model_name=self.model_name,
@@ -35,6 +37,8 @@ class Conversation_RAG:
         data: {question}\n
         """
 
+        handler = CallbackHandler(langfuse_public_key, langfuse_secret_key)
+
         QCA_PROMPT = PromptTemplate(input_variables=["instruction", "context", "question"], template=template)
 
         qa = ConversationalRetrievalChain.from_llm(
@@ -43,6 +47,7 @@ class Conversation_RAG:
             retriever=vectordb.as_retriever(search_kwargs={"k": k_context}),
             combine_docs_chain_kwargs={"prompt": QCA_PROMPT},
             get_chat_history=lambda h: h,
-            verbose=True
+            verbose=True,
+            callbacks=[handler]
         )
         return qa