Machine Learning with Wasm - WASI-NN

Explore the potential of machine learning with WebAssembly in this informative conference talk. Dive into the WASI-NN interface design and its reference implementation in Wasmtime. Learn how to leverage Web Assembly for deploying machine learning models with near-native performance across various devices, including CPU with AVX512, GPU, TPU, and FPGA. Follow a step-by-step guide to build essential components and develop an image classification application using Rust and AssemblyScript. Discover the performance advantages of WASI-NN compared to alternative approaches, and gain insights into WebAssembly environments, use cases, and API walkthroughs. Understand the concept of ML models as virtualized I/O types and the differences between Loader and Builder APIs. By the end of this talk, you'll be equipped with the knowledge to harness the power of WASI-NN for efficient machine learning deployment in WebAssembly environments.

Intro
Agenda
What is WASI?
What is the Bytecode Alliance?
Why wasi-nn?
WebAssembly Environments & Use Cases
Loader vs. Builder API
ML Model as a virtualized I/O type
API Walkthrough
Reference implementation on Wasmtime
Demo: wasi-nn
Performance Comparison (cont.)
Call for action