TECHNOLOGY PROFILE · TECH-0179
ONNX
Open representation for machine-learning computation graphs.
Reference link only — not reviewed this edition. This profile preserves the prepared catalog’s original descriptions. Advantages depend on the workload; comparisons are editorial, not benchmark results.
Where it has an edge
A common graph format separates supported model deployment from the original training framework.
Best-fit problems
Exchanging compatible models across runtimes and hardware backends.
Tradeoffs & cautions
Conversion and operator-version compatibility need numerical correctness tests.
What to test before choosing
Round-trip representative artifacts/messages; check compatibility, security, precision and version evolution.
Runtime & hardware
Across applications and runtimes
Access & licensing
Open specifications/implementations; usage terms vary
Execution
Format / protocol / component
Lifecycle
Not independently audited
Comparison basis
Editorial problem-fit judgement; no head-to-head benchmark performed
Connected concepts
Alternatives to investigate
TorchScript · StableHLO · GGUF