FileGPT: Local cognitive search over a pdf

Steps to run the model locally

• clone this repository

  
    git clone https://github.com/allthatido/FileGPT.git

• Navigate to the models folder

• Download the LLama family model (GGML V3 only) you want to test. Refer to README.md in the models folder. Huggingface is a good source for these.

• install dependencies. If you get an error for the below command, install each library indivisually using pip install. Windows users may have to install latest Visual C++ libraries from here and CMake from here. If you have a Nvidia GPU use cuBLAS to speed up inference. Install llama-cpp-python as described here

  
    pip install -r requirements.txt

• Start streamlit app

  
    streamlit run app.py

• Open browser and go to http://localhost:8501/