"I am facing an issue with RAG while using Qdrant vector database and ChatGPT limits."

“I created a project using RAG, and I used Qdrant as a vector database. However, it did not work efficiently when I tried to insert a large point. How can I solve this problem? Also, when I retrieved my points from Qdrant, it was expensive to send all the data to ChatGPT, and I struggled to deal with ChatGPT’s limits. I tried to summarize the data, but it was not effective. What should I do to send my data with the same meaning to LLM?”

What’s your “limit”/“top” parameter when you search?

1 Like

I apologize I can’t understand your question. but, In regards to the search, I needed all the retrieved data and I wanted to make it as concise as possible without losing its meaning. This was important so that LLM would understand what I was looking for. At the same time, I wanted to avoid exceeding the limits of ChatGPT and incurring additional costs.

you can use tiktoken to calculate the max number of documents you can add

well that, that is an engineeing problem.

you can:

  • improve the quality of your corpus
  • improve the quality of your query
  • improve the quality of your prompt
  • use a cheaper model

but without further information, it’s hard to say :thinking:

1 Like

Thank you for your concern. Your information has been very helpful. Can you recommend the best vector database to work with?

np :slight_smile:

Really depends on what you want to do. FAISS is an option, Milvus is often a good choice, Qdrant probably isn’t too bad either. Sometimes it’s enough to just use a for loop if you have < 1000 docs.

I think I can achieve good results with faiss if I try again. Thank you once again.

1 Like