Explore an introduction to optimizing machine learning models using TVMC in this 58-minute tinyML Talks local webcast. Dive into the Apache TVM deep learning compiler stack and learn how to build optimized models for various platforms. Discover the high-level workings of TVM, including its runtime, intermediate representation, and tensor expressions. Gain practical knowledge on using TVMC, understanding concepts like Auto TVM, Relay, and supported platforms. Follow along with a hands-on demonstration using a resnet50 model and learn about remote procedure calls for Raspberry Pi deployment. Acquire valuable insights into improving deep learning system performance through this comprehensive overview of TVMC and its applications.
Overview
Syllabus
Intro
Welcome
Agenda
Why should you use TVMC
How does TVMC work
Architecture diagram
Runtime
Intermediate Representation
TIR
Target
Tensor Expression
Matrix Multiplication
Tensor Expressions
Tile Schedule
Vector Operations
Scheduling
Performance
Topi
What is Auto TVMC
Feedback loop
Relay
What is Relay
Fused Operators
Supported Platforms
Supported Front Ends
Results
Installing TVMC
How to use TVMC
What is resnet50
Roadmap
Downloading the model
Processing the model
Raspberry Pi
Remote Procedure Calls
TVMC
Wrapup
Sponsors
Taught by
tinyML
