nixos/hosts/architect/llm.nix
2024-05-23 23:45:48 +01:00

101 lines
2.1 KiB
Nix

{ pkgs, ... }:
let
frontendDomain = "pino.giugl.io";
backendDomain = "ollama.giugl.io";
frontendPort = 3030;
ollamaPort = 11434;
listenAddress = "127.0.0.1:${toString ollamaPort}";
ollamaPkg = pkgs.unstablePkgs.ollama;
in
{
environment = {
systemPackages = [ ollamaPkg pkgs.aichat ];
variables = {
OLLAMA_ORIGINS = "*";
OLLAMA_FLASH_ATTENTION="1";
};
};
services.ollama = {
inherit listenAddress;
enable = true;
acceleration = "cuda";
package = ollamaPkg;
};
architect.vhost.${frontendDomain} = {
dnsInterfaces = [ "tailscale" ];
locations."/" = {
host = "127.0.0.1";
port = frontendPort;
allowLan = true;
allowWAN = true;
extraConfig = ''
proxy_read_timeout 600s;
'';
};
};
architect.vhost.${backendDomain} = {
dnsInterfaces = [ "tailscale" ];
locations."/" = {
host = "127.0.0.1";
port = 11434;
allowLan = true;
allowWAN = true;
extraConfig = ''
proxy_buffering off;
proxy_read_timeout 600s;
proxy_set_header Host localhost:${toString ollamaPort};
'';
};
};
virtualisation.oci-containers = {
containers = {
# ollama = {
# image = "ollama/ollama:latest";
# autoStart = true;
# extraOptions = [
# "--pull=always"
# "--gpus=all"
# ];
# environment = {
# OLLAMA_ORIGINS = "*";
# };
# volumes = [
# "/ollama:/root/.ollama"
# ];
# ports = [
# "${listenAddress}:${toString ollamaPort}"
# "172.17.0.1:${toString ollamaPort}:${toString ollamaPort}"
# ];
# };
ollama-webui = {
image = "ghcr.io/open-webui/open-webui:main";
autoStart = true;
ports = [
"127.0.0.1:${toString frontendPort}:8080"
];
environment = {
OLLAMA_BASE_URL = "https://${backendDomain}";
};
extraOptions = [
"--pull=always"
];
volumes = [
"/var/lib/ollama-webui:/app/backend/data"
];
};
};
};
}