dkdc-lm 0.2.1

Local LLM inference management via llama-server

dkdc-lm

Language model service.

Install

Recommended:

curl -LsSf https://dkdc.sh/lm/install.sh | sh

Pre-built binaries are available for Linux and macOS via Python (uv). Windows users should install via cargo or use macOS/Linux.

uv:

uv tool install dkdc-lm

cargo:

cargo install dkdc-lm-cli

Verify installation:

lm --version

You can use uvx to run it without installing:

uvx --from dkdc-lm lm

Usage

CLI

lm start                              # Start with default model (Gemma 4 26B-A4B)
lm start --builtin gemma-4-e4b-it     # Start with smaller model
lm start --hf ggml-org/some-model-GGUF  # Start with HuggingFace model
lm start --model /path/to/model.gguf  # Start with local model
lm status                             # Check server status
lm logs                               # View recent logs
lm attach                             # Attach to tmux session
lm stop                               # Stop server

Rust

use dkdc_lm::{start, stop, status, resolve_builtin, DEFAULT_PORT};

let args = resolve_builtin("gemma-4-26b-a4b-it")?;
start(&args, DEFAULT_PORT, -1, 4096)?;

let (tmux, http) = status(DEFAULT_PORT);
println!("tmux: {}, http: {}", tmux, http);

stop()?;

Python

import dkdc_lm

args = dkdc_lm.resolve_builtin(dkdc_lm.default_builtin())
dkdc_lm.start(args)

tmux_running, http_responding = dkdc_lm.status()
print(f"tmux: {tmux_running}, http: {http_responding}")

dkdc_lm.stop()