MicroTVM: A Tensor Compiler for Bare Metal

Explore the world of microTVM, a tensor compiler for bare metal devices, in this 58-minute tinyML Talks Tokyo meetup webcast. Dive into the challenges of deploying machine learning models on resource-constrained devices and learn how microTVM addresses these issues. Discover the inner workings of microTVM, including model compilation, autotvm, and quantization techniques. Follow along with a demonstration of the compiler in action, covering host-driven and standalone approaches. Gain insights into future directions for microTVM and opportunities for involvement in this open-source project. Whether you're a developer, researcher, or enthusiast in the field of tiny machine learning, this talk provides valuable knowledge on optimizing tensor-oriented models for bare metal environments.

Introduction
Sponsors
Next tinyML Talks
Local Committee
Speaker Introduction
Deployment Challenge
Why is Bare Meta so hard
What is TBM
TBM traction
MicroTVM approach
What is microTVM
Model compilation
Autotbm
Parameters
Running a model
Host Driven
Standalone
Quantization
Explaining the target
Running the compiler
Building a hostdriven binary
Future directions
pending questions
closing