Watch recordings featuring Merantix and Woven Planet
Sandra Youssef
📣 Community Update
Autonomous Vehicle - Data Processing & MLOps at Scale With Flyte [Miguel Toledo & Varsha Parthasarathy - Woven Planet]
Parameterizing and Executing Flyte Workflows With Hydra-Core [Fabio Grätz - Merantix]
Notes (Sandra):
Community Update: Slides / Recording
Union.ai's Ketan Umare welcomes everyone and starts off with Flyte’s first hackathon with the MLOps Community - Engineering Labs. Registration ends tomorrow Wed 23rd, kicking off 4 weeks of building MLOps and data products.
Several noteworthy features of Flyte that have been highlighted by members of the open source community on social media include:
The model of workflow execution
Kubernetes pod tasks
Highly intuitive user-friendly SDKs
Automated caching
Ketan then announces topics and speakers of the upcoming community sync, as well as weekly office hours, before diving into items on our roadmap for plugins, Flytekit, and FlyteConsole.
Woven Planet: Slides / Recording
Varsha Parthasarathy begins by introducing Woven Planet, a software-first subsidiary of Toyota first conceived at Lyft, focused on building a self-driving system. Both Varsha and Miguel are part of the ML infrastructure team. Varsha next presents an overview of Level 5, with a high-level map of where Flyte fits in the data management process, some insights into Level 5 projects, and use cases. She then shows an examples of their dataset curation and data annotation pipelines written on Flyte.
Miguel Toledo, Level 5’s in-house Flyte expert, shares some MLOps use cases, and how the entire data-to-car deployment pipeline was built using tools built on top of Flyte. He then explains their ML planner, followed by some of the add-ons built on Flyte with Woven Planet being largely a monorepo integration, such as Bazel and Wicker.
Merantix Hydra-Core: Slides / Recording
Dr Fabio Grätz introduces Merantix Labs, an ML solutions provider based in Germany, and gives an overview of their requirements to be able to work with Hydra, followed by their Continuous Delivery for ML (CD4ML) infrastructure setup.
He then explains Hydra-Core, a framework developed by Meta AI, which configures yaml files in hierarchical configurations, used by Merantix for in-house frameworks to parameterize model training.
Dr Fabio next runs an extensive demo of executing Flyte workflows with Hydra, and wraps up with future plans for hyperparameter optimization and the outlook for Flyte. A discussion follows.
Our next OS Community Sync will feature LatchBio's Kenny Workman on Tuesday March 8th, so mark your calendars!
See you then!