Hi I'm trying to pack model(pytorch/torchscript) weights with my bento using custom wrapper around mlflow In previous version of bento i was able to do this like that

model = CustomMlflowClient.download_model(...)
BentoService.pack('model', model)

And the model will be saved with bento How can i do the same in bentoml 1.0 with custom lib? I tried to define custom runnable and load the model in `___init_`__ but it does not work as expected because model is not packed with bento So it means that the weights will be downloaded each time when the new pod rises instead of build time