AIM326-R – Implement ML workflows with Kubernetes and Amazon SageMaker – re:Invent 2019 – Key Takeaways
The Key
How K8s and SageMaker work together, with SageMaker Operator for K8s
The Takeaways
Lots of things to worry when running ML workloads on K8s, security, scalability, multiple libraries, orchestration, etc.
The K8s operator maps SageMaker API as K8s resource
Lyft case study
Data and ML becoming one, so need a single tool for both
Flyte is such an open source tool created by Lyft
Flyte supports SageMaker (though it’s confusing that I saw this recently)
Demo
NOTE: I found this guy elaborate code details fairly well, you should watch it if you want some deep dive
It’s cool to see user submits a task in Flyte, and drill down on each steps, and finally can out-link to SageMaker jobs and view all the way to the deepest detail