backend: more post-ldm-removal cleanup

[keturn]

No description provided.

[lstein]

Nice to see so much code disappearing.

[lstein]

In model_manager.py dangling call to _model_to_cpu here:

    def offload_model(self, model_name: str) -> None:
        """
        Offload the indicated model to CPU. Will call
        _make_cache_room() to free space if needed.
        """
        if model_name not in self.models:
            return

        print(f">> Offloading {model_name} to CPU")
        model = self.models[model_name]["model"]
        model.offload_all()
        self.models[model_name]["model"] = self._model_to_cpu(model)

        gc.collect()
        if self._has_cuda():
            torch.cuda.empty_cache()

Also, pyflakes is complaining about undefined name 'model_description' I'll commit fixes.

[lstein]

Seems to be working now.

[keturn]

Thanks for catching that!