Dive into a 48-minute podcast episode featuring Andrew Dye, a systems engineer navigating the world of machine learning. Explore the intersection of low-level engineering and MLOps as Andrew shares insights from his experience as a tech lead for ML Infrastructure at Meta and his current role as a software engineer at Union. Learn about distributed training, managing large-scale ML systems, and bridging the gap between firmware and MLOps. Discover how different engineering disciplines can work together effectively in MLOps teams, and gain valuable perspectives on the future of ML infrastructure and tooling. Hear discussions on topics such as execution patterns, rapid change adoption, consensus challenges, and the importance of abstractions in ML engineering.
Systems Engineering in Machine Learning - Navigating Low-Level Challenges
Overview
Syllabus
[] Andrew's preferred coffee
[] Introduction to Andrew Dye
[] Takeaways
[] Huge shoutout to our sponsors UnionML and UnionAI!
[] Andrew's background
[] Andrew's learning curve
[] Bridging the gap between firmware space and MLOps
[] In connection with Pytorch team
[] Things that should have learned sooner
[] Type of scale Andrew works on
[] Distributed training at Meta
[] Managing the huge search space
[] Execution patterns programs
[] Non-ML engineers dealing with ML engineers having the same skill set
[] Pace rapid change adoptation
[] Consensus challenges
[] Abstractions making sense now
[] Comparing to others
[] General principles in UnionAI tooling
[] Seeing the future
[] Inter-task checkpointing
[] Combining functionality with use cases
[] Wrap up
Taught by
MLOps.community
