MLOps Coffee Sessions #136 with Andrew Dye, Systems Engineer, Navigating the World of ML, co-hosted by David Aponte.

// Abstract
We don't hear that much about working at a very low level on this podcast, but they are still very valid. Andrew is able to give us his take on why and what you need to keep in mind when you are working at these low levels, and why it is very important when you are a Machine Learning Engineer, and how the two can play together nicely.
Most MLOps teams are formed using existing people and existing engineers. More often than not, you have to blend these various disciplines, and it works well when there's a common goal.

// Bio
Andrew is a software engineer at Union and a contributor to Flyte, a production-grade data and ML orchestration platform. Prior to that, he was a tech lead for ML Infrastructure at Meta, where he focused on ML training reliability.

// MLOps Jobs board

jobs.mlops.community  

// MLOps Swag/Merch
https://mlops-community.myshopify.com/

// Related Links

--------------- ✌️Connect With Us ✌️ -------------
Join our Slack community: https://go.mlops.community/slack
Follow us on Twitter: @mlopscommunity
Sign up for the next meetup: https://go.mlops.community/register
Catch all episodes, blogs, newsletters, and more: https://mlops.community/

Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/
Connect with David on LinkedIn: https://www.linkedin.com/in/aponteanalytics/
Connect with Andrew on LinkedIn: https://www.linkedin.com/in/andrewwdye

Timestamps:
[00:00] Andrew's preferred coffee
[03:30] Introduction to Andrew Dye
[03:33] Takeaways
[07:32] Huge shoutout to our sponsors UnionML and UnionAI!
[07:48] Andrew's background
[10:08] Andrew's learning curve
[11:10] Bridging the gap between firmware space and MLOps
[12:18] In connection with the Pytorch team
[12:54] Things that should have been learned sooner
[14:54] Type of scale Andrew works on
[17:42] Distributed training at Meta
[19:55] Managing the huge search space
[22:18] Execution patterns programs
[23:20] Non-ML engineers dealing with ML engineers having the same skill set
[26:44] Pace rapid change adoption
[29:18] Consensus challenges
[32:26] Abstractions making sense now
[34:53] Comparing to others
[39:21] General principles in UnionAI tooling
[41:54] Seeing the future
[43:54] Inter-task checkpointing
[44:52] Combining functionality with use cases
[46:17] Wrap up