Run a model without GNOME

Running without the graphical desktop leaves more GPU memory available for the model. Save your desktop work before continuing: switching targets immediately ends the GNOME session.

Switch to terminal-only mode

To switch only for the current boot:

sudo systemctl isolate multi-user.target

To make terminal-only mode the default for future boots as well:

sudo systemctl set-default multi-user.target

Log in on the text console, or connect over SSH. lllm2 detects that graphical.target is inactive and does not reserve model memory for the desktop.

Launch the model

List the available models if needed, then launch:

lllm2 models
lllm2 launch

The launch command selects a compatible installed model and engine. If the workbench database in LLLM2_STATE_DIR contains saved settings for that model and backend, it uses them; a missing or empty state directory falls back to the built-in settings. The API is served at http://127.0.0.1:1920/v1 until you press Ctrl-C.

To choose a particular installed model:

lllm2 launch --model /path/to/model.gguf

Return to GNOME

Stop the model with Ctrl-C before restoring the desktop, because GNOME needs some of the GPU memory used by a headless launch. Then run:

sudo systemctl set-default graphical.target
sudo systemctl isolate graphical.target

If you changed only the current session and left the default target graphical, the set-default command is unnecessary.