This message was deleted.

You can configure the BentoML home directory via

BENTOML_HOME

environment. Not sure what you mean by train and save Bento models within a docker container, but in terms of training models, its probably best to just train locally and then follow the workflow. See https://docs.bentoml.org/en/latest/tutorial.html for quick overview of the workflow. Sorry if I wasn’t able to answer all of your concern.

After saving the model to bentoml you can always export it to a file using

bentoml.models.export_model()

You could write to a mounted volume in your container and then use the exported model. It can be loaded into bentoml again. Alternatively, you can also build and export the bento inside the container. Like shown here. Again to a shared mount and then later containerize it in a build container.