* add device map override functionality
* fix syntax
* update readme
* Edit inference file so it keeps memory reserved
* prepare release branch
* make it so that the model is optionally unloaded between generations
* remove newline
* update assets and example workflow
* update pyproject
* reload the model if settings have changed; use model if it's already … (#16)
* reload the model if settings have changed; use model if it's already loaded even if keep_model_in_memory is false
* update readme
* prepare relase 1.2.0
* Add moe drop tokens (#18)
* moe_drop_tokens option
* log kwargs, change readme
* make moe_drop_tokens the default
* edit readme
---------
Co-authored-by: timkhronos <timkhronos@gmail.com>
