Fixed
- Ensure only one
mlx_lm.serverinstance runs while switching models. - Wait for the previous server process and port to stop before launching the selected model.
- Queue a follow-up restart when the model is changed again during an active transition.