This message was deleted.
# ask-for-helps
Slackbot
02/09/2024, 1:37 AMThis message was deleted.
m
Mariano Semelman
02/09/2024, 10:17 AMHi Mike, with latest changes to the bentoml library, I’m not sure focusing on runners still makes sense.
m
Mike Kuhlen
02/09/2024, 4:00 PMInteresting statement. I'm assuming you're referring to BentoML v1.2? Where can I read about these "latest changes"?
Mike Kuhlen
02/09/2024, 4:00 PMIf this change removed the concept of a runner, then that seems like a pretty major change to the architecture... Our production systems utilize ~20 different bentos, so moving them all over to a new runner-less version would be a pretty major undertaking for us.
Mike Kuhlen
02/09/2024, 4:01 PMFwiw, the changes I proposed in those issues are quite minor. I think they might both end up being one-liners...
c
Chaoyu
02/12/2024, 5:04 AM@Mariano Semelman - 1.2 still offers the same “Runner Architecture”, with a twist in API design to make things more flexible for many new use cases we are seeing
Chaoyu
02/12/2024, 5:06 AMthis is also how we were able to make sure everything is backwards compatible in 1.2
m
Mariano Semelman
02/12/2024, 11:38 AMOh, sorry for the confusion!
Mariano Semelman
02/12/2024, 11:39 AMIt's just that from bentoml documentation we couldn't find anything that would clarify
a
Amit Gelber
02/14/2024, 9:19 PMsame here, where all the runners code go? how can use bentoml._internal.frameworks.diffusers code without runners?
c
Chaoyu
02/15/2024, 12:16 AMYou can still use the runner code with 1.1.x compatible APIs
Chaoyu
02/15/2024, 12:20 AMThe built-in runner class has many limitations and that’s part of the reason we designed the new way of building multi-model inference graph with multiple services (same architectural concept as runners in 1.1.x), see example here: https://github.com/bentoml/BentoSD2Upscaler
We will release a blog post explaining some of the changes here in 1.2 design early next week. In the meanwhile, all existing runner/service code should be fully compatible in the new version. You don’t have to upgrade your code to get the latest bug fixes and optimizations in BentoML.
a
Amit Gelber
02/15/2024, 8:13 AMThank you for the answer, I understand. What I don't fully answer is how the integrations section will combine with the code.
for example, lora weights example with the new syntax
thanks!