BentoML
BentoML copied to clipboard
bentoml
SnapML support
Is your feature request related to a problem? Please describe. Add support for SnapML scoring service.
Describe the solution you'd like Create a model artifact type that can exploit SnapML runtime. SnapML is a no-cost available ML runtime (available via conda and pip install) that supports and optimizes traditional ML models, including those trained in sklearn, xgboost, and other frameworks.
Details on SnapML can be found here: https://www.zurich.ibm.com/snapml/ It is available via pip and conda for x86, Power and s390x environments.
Describe alternatives you've considered N/A
Additional context SnapML will support additional optimization targets on s390x hardware, and adding BentoML will enable customers on s390x hardware (linux environments) to utilize BentoML and get the benefit of these optimizations coupled with the BentoML capabilities. I believe there is potential to drive more BentoML adoption for on-premise use cases with s390x enterprise clients and this capability will help facilitate that.
If the community is willing to accept this capability, we will contribute the support and tests.
Hi @andrewsi-z, I'm not familiar with the runtime. Since we are in the process of releasing 1.0, all ML related supports should be on-hold, but I would love to hear more about the usage of SnapML. We will definitely bring this up in our discussion.
@aarnphm great - excited for v1!
Is there anything in particular we can provide to help an evaluation?
Hi @andrewsi-z yes definitely, would love to get you on our beta test list and hear your feedback. I will ping you in slack.
@andrewsi-z we have released a stable BentoML framework API in 1.0. You are welcome to contribute SnapML as one of the natively supported frameworks.
Hey @parano @ssheng
I am looking to contribute. Is this something you guys are interested in as a feature and can I take a look at it? Or is it not prioritized yet and not that important?
@harshitsinghai77 apologize for the delay, I believe @andrewsi-z is not actively working on this, feel free to take it and contribute!