Chat mode

main.py

@@ -2,7 +2,9 @@ from llama_index import (SimpleDirectoryReader, ServiceContext, StorageContext,
                          load_index_from_storage, Document, set_global_service_context)
 from llama_index.node_parser import SimpleNodeParser
 from llama_index import VectorStoreIndex
-from llama_index.llms import OpenAI
+from llama_index.llms import OpenAI, ChatMessage, MessageRole
+from llama_index.prompts import PromptTemplate
+from llama_index.chat_engine.condense_question import CondenseQuestionChatEngine
 import os
 import re
 
@@ -36,22 +38,32 @@ else:
     storage_context = StorageContext.from_defaults(persist_dir="./index")
     index = load_index_from_storage(storage_context)
 
-template = (
+custom_prompt = PromptTemplate(
     "You have been trained on the Darknet Diaries podcast transcripts with data from october 6 2023."
     "You are now an expert about it and will answer as such. You know about every episode up to number 138. \n"
     "----------------\n"
-    "Here is the context: {context_str}"
+    "Chat history: {chat_history}\n"
     "----------------\n"
-    "Please answer this question by referring to the podcast: {query_str}"
+    "Please answer this question by referring to the podcast: {question}"
 )
-qa_template = PromptTemplate(template)
-query_engine = index.as_query_engine(text_qa_template=qa_template)
+
+custom_chat_history = []
+query_engine = index.as_query_engine()
+chat_engine = CondenseQuestionChatEngine.from_defaults(
+    query_engine=query_engine,
+    condense_question_prompt=custom_prompt,
+    chat_history=custom_chat_history,
+    verbose=True
+)
+
 
 while True:
     try:
         user_prompt = input("Prompt: ")
-        response = query_engine.query(user_prompt)
-        print(response)
+        streaming_response = chat_engine.stream_chat(user_prompt)
+        for token in streaming_response.response_gen:
+            print(token, end="")
+        print("\n")
     except KeyboardInterrupt:
         break