Sign in to confirm you’re not a bot
This helps protect our community. Learn more

Avoid these 3 mistakes to ensure your model reaches production | ODBRK52

Microsoft Developer
592K subscribers
<__slot-el>
<__slot-el>
243 views 2 years ago
The vast majority of fully trained models never make it to production, and those that do take an average of 12 weeks per development cycle. Whether due to performance issues, high costs to run, or pipeline dependencies that create roadblocks at every phase of deployment, nobody wants to endure delays or watch models die on the vine. In this session, you’ll learn how to spot (and avoid) the three most common pitfalls when it comes to getting ML models over the hurdle and into production. Jump to: 00:00 Introduction 00:42 Tensor from Hell 02:20 Dead and Delayed Models 03:15 Mistake 1: Not Fully Optimizing Your Model for Use Case or Hardware Target 04:36 The Fix 07:03 Mistake 2: Inadvertently Selecting the Wrong Combination or Framework, Environment, and Hardware Target 08:57 The Fix 10:48 Mistake 3: Handing off the Model with No Transition Plan 13:12 The Fix: Treat Models Like the Software They Are 13:47 Treat Models as Software Sooner, Build Smarter Apps Faster 15:08 Closing Notes Microsoft Build 2022 ...more
...more
Avoid these 3 mistakes to ensure your model reaches production | ODBRK52
2Likes
243Views
2022May 27
The vast majority of fully trained models never make it to production, and those that do take an average of 12 weeks per development cycle. Whether due to performance issues, high costs to run, or pipeline dependencies that create roadblocks at every phase of deployment, nobody wants to endure delays or watch models die on the vine. In this session, you’ll learn how to spot (and avoid) the three most common pitfalls when it comes to getting ML models over the hurdle and into production. Jump to: 00:00 Introduction 00:42 Tensor from Hell 02:20 Dead and Delayed Models 03:15 Mistake 1: Not Fully Optimizing Your Model for Use Case or Hardware Target 04:36 The Fix 07:03 Mistake 2: Inadvertently Selecting the Wrong Combination or Framework, Environment, and Hardware Target 08:57 The Fix 10:48 Mistake 3: Handing off the Model with No Transition Plan 13:12 The Fix: Treat Models Like the Software They Are 13:47 Treat Models as Software Sooner, Build Smarter Apps Faster 15:08 Closing Notes Microsoft Build 2022

Follow along using the transcript.

Microsoft Developer

592K subscribers