Using a 5800h and rtx3060 laptop i constructed a rag pipline to do basically pdf Chat qith a local llama 7b 4bit quantized Modell in llama_index using llama.cpp as backend. I use an emmbeding and a vector store through postgresql. Under wsl.

With a context of 4k and 256 token output length generating an answer takes about 2-6min which seems relatively long. I wanted to know if that is expected or if i need to go on the hunt for what makes my code inefficient.

Also what kinds of speed ups would other gpus bring ?

Would be very happy to get some thoughts on the matter :)

  • Noxusequal
    cake
    OPB
    link
    fedilink
    English
    arrow-up
    1
    ·
    1 year ago

    I know that cuda is used vram is full and i get the message in the beginning. What is your hardware setup ?

    Do you also use llama_index and then langchain or did you build it more or less from llama_cpp and langchain without llama_index ?

    • harrroB
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 year ago

      I’m using langchain with qdrant as the vector store.

      VRAM is full

      How is a 7B model maxing out your VRAM? A 7B model at 4bit and 4k context should not use the 12GB VRAM on a 3060.

      • Noxusequal
        cake
        OPB
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 year ago

        Its a 3060 laptop so only 6gb and model plus embedding etc. Is at like 5.8gb