iree
A retargetable MLIR-based machine learning compiler and runtime toolkit.
What is iree?
A retargetable MLIR-based machine learning compiler and runtime toolkit.
Overview of iree
What it is
IREE (Intermediate Representation Execution Environment, pronounced as 'eerie') is an MLIR-based end-to-end compiler and runtime toolkit. It lowers machine learning models to a unified intermediate representation that can scale from datacenter deployments down to mobile and edge devices. The project is hosted under the LF AI & Data Foundation as a sandbox-stage project.
Key features
- Retargetable MLIR-based compiler and runtime toolkit - End-to-end compilation and execution of machine learning models - Lowers models to a unified intermediate representation (IR)
- Scales to meet datacenter needs and mobile/edge constraints - Supports multiple hardware targets through a hardware abstraction layer (HAL)
- Open source under Apache 2.0 License with LLVM Exceptions
Use cases
IREE is designed for deploying machine learning models across diverse hardware, from large-scale datacenter servers to resource-constrained mobile and edge devices. It has been used in benchmark submissions such as AMD's IREE-based SDXL implementation for MLPerf. The compiler and runtime are suitable for production inference workloads requiring high performance and portability.
Getting started
For installation instructions, building from source, and user guides, please refer to the repository README and the project website at https://iree.dev.
License and pricing
IREE is open source and free to use. It is licensed under the Apache 2.0 License with LLVM Exceptions.
Based on the project's public README and repository metadata.