This message was deleted.

I think what you propose is fine. I think we should update the doc to expose that functionality.

What you proposed is OK. The example had it that way because the model instance needs to be passed as a Runner init param for the builder to recognize it. As long as you do that, how to structure the runnable doesn’t matter.

spam_detection_runner = bentoml.Runner(SpamDetectionRunnable, models=[bento_model])

Thanks @Bo and @Sean for the feedback. I am circling back to this just to close the loop. I discovered that actually passing to models, did turn out to be necessary. So:

# this did not save the model into the bento
spam_detection_runner = bentoml.Runner(SpamDetectionRunnable, 
  runnable_init_params={"model": bento_model},
)

# but this did
spam_detection_runner = bentoml.Runner(SpamDetectionRunnable, 
  runnable_init_params={"model": bento_model},
  models=[bento_model],
)

Oddly, the former doesn’t error when building the bento, but when I try running the service inside the built docker container, it complains the model wasn’t packaged in it