This message was deleted.
# ask-for-help
s
This message was deleted.
👀 1
🍱 1
b
You can have the training and serving project separately
The bentofile is mainly describing what the serving will need to do and what behaviors it will have
❤️ 1
e
Awesome! So as long as the python requirements in
bentofile.yaml
match the python requirements in teh `setup.cfg`/
requirements.txt
of the training project, it should work?
c
Yes, you can in fact point to your existing requirements.txt file in your bentofile as well
😮 1
BentoML does a bit extra here - it will lock the entire dependency trees’ package versions and make sure to reproduce identical environments between training and serving
e
That is fantastic 😄, no need for a Pipfile.lock
Ah, but it sounds from this like bentoml really encourages you to keep your model and service in the same python project--that way the
bentofile.yaml
can just point to the
requirements.txt
which results in a single place where you can define requirements for both training and serving
It sounds like it wouldn't be DRY to split up training and serving because you'd have to keep the requirements for both projects in sync.
b
I think in some way, we feel the person who builds the model, also owns the service. We see that sometimes nuances are lost when transiting from one to another.