Obtain a QnA instance via one of the helper functions from get_qna and provide the instance with the user query, a system/ bot response will be returned and the instance will automatically store the chat history:
# from main.py, example usage
from document_qna.setup import get_qna
from document_qna.qna import QnA
qna_instance: QnA = get_qna() # loads documents from specified folder directly, see settings.ini
while True:
user_msg = input("Input current user message (type \"quit\" to exit): ")
if user_msg == "quit":
break
bot_msg = qna_instance(user_msg)
print(bot_msg)Check other ways of initializing a QnA object here
To restart the chat from a certain index (w.r.t. only the chat history), you can do the following:
qna_instance.restart_from_index(<index number>)Say, you have the following chat history:
[
{"role": "system", "content": "Some system prompt"}, # Is not counted
{"role": "assistant", "content": "Default message e.g. how can I help you today"}, # Is not counted
{"role": "user", "content": "blah blah blah"}, # index = 1
{"role": "assistant", "content": "blah blah blah blah"}, # index = 2
{"role": "user", "content": "yada yada yada"}, # index = 3
{"role": "assistant", "content": "yada yada yada yada"}, # index = 4
]And then you do
qna_instance.restart_from_index(2)Then the new chat history becomes:
[
{"role": "system", "content": "Some system prompt"}, # Is not counted
{"role": "assistant", "content": "Default message e.g. how can I help you today"}, # Is not counted
{"role": "user", "content": "blah blah blah"}, # index = 1
{"role": "assistant", "content": "blah blah blah blah"}, # index = 2
]A user might want to perform a chat across multiple sessions, where we won't keep the session alive.
We can store all relevant information of the instance using:
prompt, default_msg, chat_history = qna_instance.save_chat()
# Store `prompt, default_msg, chat_history` somewhere
some_storage_function(prompt, default_msg, chat_history)Then, we can retrieve the chat using:
prompt, default_msg, chat_history = some_retrieval_function()
response_func = get_inference_func() # example function, check document_qna/setup/setup.py
qna_instance = QnA(prompt=prompt, default_msg=default_message, response_func=response_func, chat_history = chat_history)
# Use this qna_instance for chatting- Place instruction documents in the documents folder
- Provide the api key of a provider of your choice in a .env folder (check (and clone) the .env.template file)
- Change nessesary settings in settings.ini (affects behavior of the
get_qna()setup function only).