onnx/onnx-mlir

Health v2 · maintenance only

Flags: prerelease_only

Adoption not part of the score

1046 stars · 449 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

ONNX-MLIR is a compiler infrastructure built on LLVM/MLIR that lowers ONNX neural network graphs into executable code, MLIR files, LLVM bytecode, or C/Java libraries. It includes an ONNX dialect, an onnx-mlir driver, and Python/C/C++/Java runtime support for executing compiled models.

Use cases

• compile onnx models to native code
• run onnx models without python runtime
• lower neural network graphs to llvm bytecode
• generate c or java libraries from onnx models
• optimize deep learning model inference with mlir
• deploy onnx models on ibm telum accelerator
• integrate onnx dialect into custom compiler pipelines

When to choose

• you need ahead-of-time compilation of ONNX models to native code
• you want minimal runtime dependencies for model inference
• you target CPUs or IBM Telum accelerators
• you are building compiler tooling around MLIR for ML workloads

When to avoid

• you just need to run ONNX models quickly with existing runtimes like ONNX Runtime
• you need GPU training or broad GPU inference support
• you want a high-level training framework rather than a compiler

Facets

library · maturity active

compiler machine-learning llm-inference machine-learning compilers developer-tools windows cpp python jvm onnx mlir llvm model-compilation code-generation inference-runtime linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="onnx/onnx-mlir")