TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Show HN: MLflow to KFServing image builds with Kaniko on K8s

3 pointsby lewqalmost 4 years ago

1 comment

lewqalmost 4 years ago
Are you on an internal ML platform team? We&#x27;d love to know if self-service (for data scientists), multi-tenant model -&gt; container image builds are a pain point for you.<p>I was lucky enough to be involved in the creation of <a href="https:&#x2F;&#x2F;chassis.ml&#x2F;" rel="nofollow">https:&#x2F;&#x2F;chassis.ml&#x2F;</a> - which we&#x27;re open sourcing today! Basically, it solves the problem of &quot;how do you get your MLflow models into KFServing&quot; - and in the process solved some interesting problems around doing rootless image builds in a multi-tenant K8s cluster.<p>In other words Chassis is an API server which muxes between MLflow models, k8s jobs + kaniko, KFServing + Modzy APIs, with a Python SDK on top for data scientists to drive it.<p>Demo: <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=d_8OIfQOa3I" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=d_8OIfQOa3I</a><p>Test drive! <a href="https:&#x2F;&#x2F;chassis.ml&#x2F;#test-drive" rel="nofollow">https:&#x2F;&#x2F;chassis.ml&#x2F;#test-drive</a><p>More broadly, we&#x27;re working on a standard for making model serving more portable, so you can &quot;build once, run many&quot; i.e. build container images that run in a variety of different platforms, to avoid lock-in: <a href="https:&#x2F;&#x2F;openmodel.ml&#x2F;" rel="nofollow">https:&#x2F;&#x2F;openmodel.ml&#x2F;</a>