Repolode
Submit
iree-org

iree-org/iree

An MLIR-based compiler and runtime that lowers ML models to a unified IR for deployment from datacenters to mobile and edge devices.

View on GitHubWebsite
Stars
3,986
Forks
1,038
Open issues
1,721
Last commit
6 hours ago

What is it?

IREE is an end-to-end compiler and runtime for machine learning models, built on MLIR. It lowers models into a unified intermediate representation that can scale up to datacenter workloads and down to the constraints of mobile and edge deployments.

The project's topics point to supported inputs and targets, including models from PyTorch, TensorFlow, JAX and ONNX, and backends such as CUDA, ROCm, Vulkan and SPIR-V. The name stands for Intermediate Representation Execution Environment.

The project is licensed under Apache 2.0 with LLVM exceptions and includes Python packages (iree-base-compiler and iree-base-runtime) on PyPI, plus release notes on GitHub.

Key features

  • 01Lowers ML models from frameworks such as PyTorch, TensorFlow, JAX and ONNX into a unified MLIR-based IR.
  • 02Targets GPU and accelerator backends including CUDA, ROCm and Vulkan.
  • 03Generates SPIR-V code for portable GPU execution.
  • 04Provides a runtime for executing compiled models.
  • 05Ships Python packages for the compiler and runtime.

Use cases

  • Deploying a trained model to datacenter GPUs and to mobile or edge devices from one toolchain.
  • Targeting Vulkan or ROCm hardware for ML inference.
  • Researching compiler optimizations for ML workloads on MLIR.

Quick start

  1. 1

    Open the project website for user guides and build-from-source instructions

Things to know

  • The README gives no direct install command, so you must build from source by following the website guides.
  • It is a compiler and runtime stack, so getting started takes more effort than a typical library.

Topics

#compiler#mlir#machine-learning#runtime#gpu#inference#edge-ai#cuda#jax#onnx

FAQ

What is iree?
An MLIR-based compiler and runtime that lowers ML models to a unified IR for deployment from datacenters to mobile and edge devices. IREE is an end-to-end compiler and runtime for machine learning models, built on MLIR. It lowers models into a unified intermediate representation that can scale up to datacenter workloads and down to the constraints of mobile and edge deployments.
Is iree free to use? What license does it have?
iree is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is iree still maintained?
Actively maintained. The last commit to iree was on Oct 21, 2026.
What can I use iree for?
Deploying a trained model to datacenter GPUs and to mobile or edge devices from one toolchain. Targeting Vulkan or ROCm hardware for ML inference. Researching compiler optimizations for ML workloads on MLIR.
How do I install iree?
Open the project website for user guides and build-from-source instructions
Where can I get iree?
The source code and installation instructions are on GitHub: https://github.com/iree-org/iree

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects