LEIP Optimize¶
LEIP Optimize simplifies the process of quantizing and compiling or exporting machine learning models for specific hardware targets. Capable of ingesting models from a variety of sources into a unified intermediate representation, LEIP Optimize supports both TVM- and ONNX-based pipelines, making it flexible, approachable, and customizable for a range of deployment needs.
Unlike compiler tooling that requires in-depth hardware expertise and results in time-consuming optimization processes, LEIP Optimize enables rapid prototyping for a variety of model-hardware combinations.
Under the hood, LEIP Optimize uses a tool called Forge, an advanced extension of TVM and ONNX that's designed to simplify the editing, manipulation, and introspection of the Relay or ONNX Intermediate Representation (IR). You'll interact with Forge directly via the API, and you can learn about Forge, IR, and more in the Explanation section.
You should use LEIP Optimize if you already have a machine learning model and want to deploy to the edge, but lack specific hardware expertise. To begin, check out the Getting Started guide.
In this Documentation¶
-
Start here for step-by-step instructions for installing and getting started with LEIP Optimize
-
Guides that cover key operations and common tasks
-
Reference documentation for the LEIP Optimize API
-
Learn more about machine learning compilation, intermediate representation, quantization, and optimization