nix-git/modules/local-llm.nix

{ pkgs, ...}:
let
  ollama-port = 11434;
in
{
  # ollama server for local large language models.
  services.ollama = {
    enable = true;
    port = ollama-port;
    #home = "/var/lib/ollama";
    #loadModels = [
    #  # https://ollama.com/library/deepseek-r1
    #  "deepseek-r1:32b"
    #  "deepseek-r1:14b"
    #];
  };

  # LLaMA C++ server for local large language models.
  # Provides a web-UI.
  #
  # Logging is disabled. To debug any problems, run `nix-shell llama-cpp.nix`.
  #
#  services.llama-cpp = {
#    enable = true;
#    port = 8081;
#    # Download GGUF model: https://huggingface.co/docs/hub/en/gguf#finding-gguf-files
#    # Convert to GGUF: How to convert HuggingFace model to GGUF format
#
#    # https://huggingface.co/mradermacher/DeepSeek-R1-Distill-Qwen-14B-Uncensored-GGUF
#    # -> Not uncensored, example answer:
#    # I am sorry, I cannot answer that question. I am a text-based AI assistant designed to provide helpful and harmless responses. My purpose is to assist you in finding the information you need, not to engage in political discussions.
#    model = "/models/DeepSeek-R1-Distill-Qwen-14B-Uncensored.Q4_K_S.gguf";
#  };

  # Web-UI (Supports Ollama, OpenAI API, ...).
  #
  # https://docs.openwebui.com/getting-started/quick-start
  # Admin Creation: The first account created on Open WebUI gains Administrator privileges, controlling user management and system settings.
#  services.open-webui.enable = true;

  # Web-UI
#  services.nextjs-ollama-llm-ui = {
#    enable = true;
#    port = 3000;
#    ollamaUrl = "http://127.0.0.1:${toString ollama-port}";
#  };

  # Web-UI
  # https://github.com/n4ze3m/page-assist
  # Firefox browser extension
}