Luis Ceze, CEO of OctoML, speaks about the creation of the Apache TVM at the University of Washington, how they’ve built a vibrant community around the open source project, and how bringing a commercial product to market is key to a sustainable open source ecosystem. Also catch Luis sharing a sneak preview of the Octomizer, a fully managed software-as-a-service that you can use to optimize, benchmark, and package your machine learning models for a number of different hardware platforms and language runtimes.
Podcast: Play in new window | Download
Episode Notes
(00:19) Intro
(00:51) Creation of Apache TVM at University of Washington
(2:41) Early members of the community
(3:08) The customer pain
(3:47) The community today
(4:16) Two groups of users: hardware vendors and machine learning data scientists and engineers
(5:28) Awareness among machine learning data scientists (integration with popular frameworks, effectiveness with models that customers care about, connect with adjacent communities)
(6:46) Common use cases and customer needs (latency, throughput per dollar, hardware coverage, operator coverage)
(9:59) Building a commercial product with a committed open source project
(12:05) Using the Octomizer
(13:22) Evaluating building vs. buying
(14:26) Ideas for business models around open source software
(16:04) Network effects
(17:34) Connecting different pursuits
(18:55) Challenges of building a business around an open source project
(20:01) What not to do!
(21:05) Community really likes the commercialization (‘this is serious!’)