ONNX-MLIR

Overview

ONNX-MLIR provides a representation and reference lowering of ONNX models within the MLIR compiler infrastructure. It enables compilation and optimization of ONNX models using MLIR's multi-level intermediate representation. The project is community-driven and open-source.

Best for

Best for
Developers who need to compile ONNX models for custom hardware or optimize inference using MLIR

Use cases

• Compiling ONNX models to efficient code for various hardware targets
• Optimizing machine learning model inference through MLIR transformation passes
• Integrating ONNX model execution into custom compiler pipelines

Notes

ONNX-MLIR provides a representation and reference lowering of ONNX models within the MLIR compiler infrastructure. It enables compilation and optimization of ONNX models using MLIR’s multi-level intermediate representation. The project is community-driven and open-source.

1,024 stars on GitHub. Last updated 2026-05-30. Licensed Apache-2.0.

Use cases

• Compiling ONNX models to efficient code for various hardware targets
• Optimizing machine learning model inference through MLIR transformation passes
• Integrating ONNX model execution into custom compiler pipelines

Pros

• Leverages MLIR’s modular design for flexible and advanced optimizations
• Supports multiple backend targets including CPU and accelerators
• Open-source project with an active community and ongoing development

Cons

• Still experimental and may not cover all ONNX operators
• Requires familiarity with MLIR infrastructure to use effectively
• Performance and stability can vary across different hardware backends

Indexed from awesome-llmops and enriched against its public facts.