#118118th of 242 in ML frameworks and libraries

iree

MLIR-based compiler and runtime for deploying ML models from datacenter to edge

Stars
4k
License
Apache-2.0
Last commit
Nov 2026
Last release
Nov 2026
Language
C++
Stars, last 29 weeks+320
25 Apr 2026: 3,699 stars2 May 2026: 3,711 stars9 May 2026: 3,718 stars16 May 2026: 3,730 stars23 May 2026: 3,742 stars30 May 2026: 3,754 stars6 Jun 2026: 3,762 stars13 Jun 2026: 3,779 stars20 Jun 2026: 3,790 stars27 Jun 2026: 3,805 stars4 Jul 2026: 3,809 stars11 Jul 2026: 3,819 stars18 Jul 2026: 3,830 stars25 Jul 2026: 3,838 stars1 Aug 2026: 3,850 stars8 Aug 2026: 3,864 stars15 Aug 2026: 3,871 stars22 Aug 2026: 3,879 stars29 Aug 2026: 3,896 stars5 Sept 2026: 3,908 stars12 Sept 2026: 3,917 stars19 Sept 2026: 3,929 stars26 Sept 2026: 3,943 stars3 Oct 2026: 3,952 stars10 Oct 2026: 3,962 stars17 Oct 2026: 3,978 stars24 Oct 2026: 3,990 stars31 Oct 2026: 4,000 stars7 Nov 2026: 4,009 stars14 Nov 2026: 4,019 stars

3,699 on 25 Apr 20264,019 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

IREE compiles machine learning models to a unified MLIR-based IR and executes them with a runtime, targeting hardware from datacenter systems to mobile and edge devices. It is distributed as the PyPI packages iree-base-compiler and iree-base-runtime, and can be built from source. User guides and build instructions are on iree.dev.

Who it is for: Compiler and ML systems engineers deploying models to varied hardware

Strengths

  • One compiler and runtime path covering datacenter through mobile and edge targets
  • Installable from PyPI as iree-base-compiler and iree-base-runtime
  • Nightly CI on Linux arm64 and macOS arm64
  • AMD submitted an IREE-based SDXL implementation to MLPerf inference

Weaknesses

  • Compiler toolchain, not a ready-to-use model server or chat app
  • Extending or debugging it requires MLIR and compiler knowledge
  • README lists no supported model formats or hardware backends
  • Source builds need a C++ toolchain; no Docker setup provided

What it needs

  • GPU optional

What changed

  1. “ireelinalgext.sort takes its inputs as ins(). The op used to sort its outs() operands in place. Inputs now go in ins() and destination tensors in outs(), like other LinalgExt ops. Update hand-written…”

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Also in ML frameworks and libraries

See all 242
Also in ML frameworks and libraries
RankProjectScore
1transformersModel-definition library for pretrained text, vision, audio and multimodal models169.3k starsApache-2.094 out of 100
2ultralyticsPython library and CLI for training and running YOLO vision models63.5k starsAGPL-3.094 out of 100
3pytorch-image-modelsPyTorch library of image models, pretrained weights and training scripts37.3k starsApache-2.094 out of 100
#4pytorchPython tensor library with GPU support and tape-based autograd105.8k starscustom license93 out of 100
#5supervisionPython toolkit for loading, annotating and counting computer vision detections Live demo ↗ (opens in a new tab)51.7k starsMIT92 out of 100