1 of 16

Oct 4th, 2022

Flyte Community Sync

2 of 16

  • Community Updates featuring Orchestration by Flyte
    • Hosted By Martin Stein featuring Samhita Alla [Union.ai]
  • Flyte Roadmap
    • By Eduardo Apolinario, Union.ai
  • Flyte Ecosystem Projects - UnionML / Pandera
  • Demo
    • Flyte Ecosystem: UnionML Release 0.1.4 Demo
    • By Niels Bantilan, Union.ai

​

Agenda

3 of 16

Community Updates

Hosted By Martin Stein

@MartinSteinBend

4 of 16

Community of Contributors

✨ Contributors of the Month ✨ (Month) :

  • Zev Isert @zevisert [Revela Systems]
  • Arief Rahmansyah @ariefrahmansyah [GoJek]
  • Eduardo Apolinario @eapolinario [Union.ai]
  • Ailin Yu @niliayu [Pachama]

Community Updates

Get To Know Your Contributor

5 of 16

  • @ariefrahmansyah
  • ML Engineer, Go-Jek
  • LinkedIn

​

​

​

​

  • Contributions:
    • Adding a Flytekit plugin for DBT

(available in the upcoming Flytekit release)

Arief Rahmansyah & Eduardo Apolinario

Community Updates

  • @eapolinario
  • Software Engineer, Union.ai
  • LinkedIn

​

6 of 16

  • @nilianyu
  • Senior Software Engineer, Pachama
  • Contributions:

​

https://www.linkedin.com/in/ailinyu-engineer/

Ailin Yu

Community Updates

7 of 16

Physical booth - Detroit MI (Oct 26-28)

  • Demos, Games, Giveaways
  • Office Hours
  • Booth #SU 91

Virtual booth - Online (Oct 24-28)

  • MeetingPlay portal
  • Live Office Hours - TBA

​

https://events.linuxfoundation.org/kubecon-cloudnativecon-north-america/

KubeCon / CloudNativeCon

Community Updates

8 of 16

Orchestration for Data, Machine Learning and Infrastructure

​

Community Updates

Which orchestrator should you use?

​

  • How scalable should your pipelines be? Can you tolerate sporadic scalability?
  • Do you need infrastructure automation?
  • Do you need custom environments?
  • How skilled is your team?
  • What’s your preferred learning curve?
  • How often do your pipelines change?
  • Would using an ML orchestrator be overkill?

https://www.union.ai/blog-post/orchestration-for-data-machine-learning-and-infrastructure

9 of 16

How Flyte is Simplifying the Orchestration of Machine Learning Pipelines

​

Community Updates

  • Flyte Decks
    • Visualize your data on the Flyte UI
  • Auto GPU-to-CPU conversion in PyTorch
    • No more to(“cpu”) for GPU to CPU conversion
  • ONNX Type Transformers
    • Generate deployable versions of your models
  • Spark Pipelines
    • Seamlessly pass Spark pipelines among your tasks
  • whylogs Integration
    • Log your data and ML models from within Flyte

Flyte Decks

https://blog.flyte.org/orchestration-of-machine-learning-pipelines

10 of 16

Flyte Roadmap

By Eduardo Apolinario

@curupa / eapolinario

11 of 16

  • Highlights
    • Support for Ray Tasks - see blog post
    • Longer workflow execution names (up to 63 characters now)
    • Portions of the FlyteWorkflow CRD offloaded
    • HuggingFace and dbt flytekit plugins
  • First Flyte Quarterly Planning Meeting
    • 1.3.0 slated for the end of the year
    • Human-in-the-loop tasks
    • Out-of-core backend plugins

Flyte v1.2.0

Flyte Roadmap

12 of 16

Flyte Ecosystem Projects

UnionML / Pandera

By Niels Bantilan

@cosmicBboy / cosmicBboy

13 of 16

Flyte Ecosystem Projects

Patch deployment

UnionML Release 0.1.4 Demo

Flyte Ecosystem Projects

Reactive S3 event-based Predictions

Deploy UnionML app workflows without rebuilding using fast registration

Initialize a template that’s configured to generate predictions based on S3 events

14 of 16

Links & Resources

15 of 16

Office Hours

Weekly on Wednesdays, Online

Ask Flyte maintainers anything:

Links & Resources

16 of 16

Upcoming Sync

  • Tues Oct 18 - Leveraging Flyte for transportation data science

[Wen Shu Tang- Lacuna.ai]

​

Calendar Invite - https://www.addevent.com/calendar/kE355955

YouTube channel - https://www.youtube.com/channel/UCNduEoLOToNo3nFVly-vUTQ

Newsletter Flyte Monthly - https://www.getrevue.co/profile/flyte

​

Links & Resources