This is a Nix flake easing the use of NVIDIA's NeMo framework on NixOS.
Run a custom Python script as a background service, using any NeMo model, on any NeMo version (in theory), with the optional container runtime of your choice (Podman or systemd-nspawn).
It also exposes all official NeMo images and models currently available, whose hashes are updated every day by a GitHub action.
Import the provided module in your NixOS configuration:
# flake.nix
{
inputs = {
nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";
nemo = {
url = "github:m-bdf/NeMo.nix";
inputs.nixpkgs.follows = "nixpkgs";
};
};
outputs = { self, nixpkgs, nemo }: {
nixosConfigurations.default =
nixpkgs.lib.nixosSystem {
modules = [
nemo.nixosModules.default
./configuration.nix
# other modules
];
};
};
}Then set the services.nemo.enable option to true,
and specify a NeMo image, an optional NeMo model,
and a Python script to run as a background service.
Here is an example for automatic speech recognition:
# configuration.nix
{ pkgs, ... }:
let
script = ''
from watchfiles import watch, Change
print("Watching for changes in /usr/share/asr...")
for changes in watch("asr", recursive=False):
for (change, path) in changes:
if change is Change.added and path.endswith(".wav"):
transcript = model.transcribe(path)[0]
print(transcript.text)
'';
in
{
services.nemo = {
enable = true;
image = pkgs.nemo.images.dev;
model = pkgs.nemo.models.canary-1b-flash;
mounts.asr = "/usr/share/asr";
libs = ps: with ps; [ watchfiles ];
inherit script;
};
}Activate this configuration (nixos-rebuild switch),
add a .wav file to the /usr/share/asr directory,
and you should see its transcription appear automagically
in the output of the nemo systemd service (journalctl -fu nemo.service).