1. Did you update? ` --- cloned and reinstalled
2. Local / cloud --- local
3. GPUs used? --- 1 gpu
4. Notebook? --- n/a
5. Unsloth version? --- fresh install
6. Trainer? --- n/a

Used very large models in Studio's Tauri app (and also web-ui directly) and changing model settings which causes any and all models to unload correctly before reloading the same or different model.

But when switching to unsloth run to customize llama-server flags then try changing a model's settings, the model gets loaded a second time without unloading the first copy. The memory spike causes system halt and or crash. (smaller model example to demonstrate)