Find answers from the community

Updated 12 months ago

Hello,

Hello,

do you have to use the chatgpt api just once to create a vector index file ?

Or has it to be done every time? because that would be expensive 😄

i am building chatbot with langchain and need some knoledge based on local documents instead of using the cloud

edit: i think yes you just generate it one time.
BUT do you upload/pass it every time (at every request/input of a user) to the ChatGPT API then?

10 comments

WWhiteFang_Jr

Yea you can You can persist the data in LlamaIndex and in next iteration check if the data is present and if so then load that data to avoid indexing again

https://docs.llamaindex.ai/en/stable/getting_started/starter_example.html#storing-your-index

LLaurenzFeistc53

thank you @WhiteFang_Jr . i just edited my text:

BUT do you upload/pass the data/json every time (at every request/input of a user) to the ChatGPT API after that? i think yes right? or maybe just at beginning of the conversation into the prompt

LLaurenzFeistc53

@WhiteFang_Jr ?

WWhiteFang_Jr

No, you don't have to pass it to chatgpt from your side. You can use chat engines from llamaindex and it will take care of maintaining the context of the conversation.

https://docs.llamaindex.ai/en/stable/module_guides/deploying/chat_engines/root.html

LLaurenzFeistc53

Thank you. So llama index has a NLP component to match the right questions to the users input?

LLaurenzFeistc53

in a q&a json for example

LLaurenzFeistc53

@WhiteFang_Jr ?

WWhiteFang_Jr

Yes. Basically llamaindex uses similarity algo

Flow in simple terms is like that

you provide the docs
docs are chunked into parts
vectors are created for these chunks
when you ask a query , most relatable chunk gets picked up
chunk + your query is passed to LLM to generate answrr

WWhiteFang_Jr

All of this will be handled by llamaindex

LLaurenzFeistc53

Thank you very much! 😉

Add a reply