RAG(File search) used too much token

[glossy tiger]

Anyone using vector database? I used RAG and it took me 5K tokens (I only asked like 30 token, response like 100 tokens) and my database is only 2.5K tokens. My config is chunk size: 200, chunk overlap: 0

RAG(File search) used too much token

[ocean oxide]

inspect what its doing https://platform.openai.com/docs/assistants/tools/file-search#improve-file-search-result-relevance-with-chunk-ranking

[glossy tiger]

I did check, but I need less token getting used, I cannot afford if it use that many token(rn I upload 2 files out of 25 files we have)
If I upload 25 files it going to take something like 20K token per message

If I upload the file data in the system part it only have 1108 token
if use vector storage it will be 3.5K token

I want it decrease to 300-500 token

[ocean oxide]

set max_num_results

[glossy tiger]

I can't I am not using assistant, I am using resp api

[ocean oxide]

responses api has the same property

[glossy tiger]

playground don't have

[ocean oxide]

then you need to wait for them to add it or use the actual api

[glossy tiger]

I find out vector_store.search_results.page return whole txt file(which is not a part of txt) even I limit the result to 1 it still use whole txt

so do I have to break the content in to mutiple txt not like how does DB like