Implementation:Microsoft Onnxruntime PlotMetadata

Knowledge Sources	Microsoft_Onnxruntime
Domains	Documentation, Python, Examples
Last Updated	2026-02-10 04:00 GMT

Overview

A Python example script demonstrating how to inspect ONNX model metadata using both the onnx library and ONNX Runtime.

Description

The `plot_metadata.py` script is a Sphinx gallery documentation example showing two methods for accessing ONNX model metadata:

1. Using the onnx library: Loads the model via `onnx.load()` and prints `ModelProto` fields:

  - `doc_string`: Model documentation string.
  - `domain`: Model domain.
  - `ir_version`: ONNX IR version.
  - `metadata_props`: Custom metadata key-value pairs.
  - `model_version`: Model version number.
  - `producer_name`: Name of the tool that produced the model.
  - `producer_version`: Version of the producer tool.

2. Using ONNX Runtime: Creates an `InferenceSession` and calls `get_modelmeta()` to access:

  - `custom_metadata_map`: Custom metadata dictionary.
  - `description`: Model description.
  - `domain`: Model domain.
  - `graph_name`: Name of the computation graph.
  - `producer_name`: Producer tool name.
  - `version`: Model version.

The example uses the bundled `logreg_iris.onnx` dataset model, a logistic regression trained on Iris data and converted with sklearn-onnx.

Usage

Use this script as a reference for extracting model provenance and metadata from ONNX models, which is important for model governance and deployment tracking.

Code Reference

Source Location

• Repository: Microsoft_Onnxruntime
• File: docs/python/examples/plot_metadata.py
• Lines: 1-46

Signature

from onnxruntime.datasets import get_example
import onnx
import onnxruntime as rt

example = get_example("logreg_iris.onnx")
model = onnx.load(example)
# Access: model.doc_string, model.domain, model.ir_version, etc.

sess = rt.InferenceSession(example, providers=rt.get_available_providers())
meta = sess.get_modelmeta()
# Access: meta.custom_metadata_map, meta.description, meta.domain, etc.

Import

from onnxruntime.datasets import get_example
import onnx
import onnxruntime as rt

I/O Contract

Method	Input	Output	Description
onnx.load	file path	ModelProto	Loads ONNX model with all metadata fields
sess.get_modelmeta	(none)	ModelMetadata	Returns ORT-accessible model metadata

Usage Examples

from onnxruntime.datasets import get_example
import onnxruntime as rt

example = get_example("logreg_iris.onnx")
sess = rt.InferenceSession(example, providers=rt.get_available_providers())
meta = sess.get_modelmeta()

print(f"Producer: {meta.producer_name}")
print(f"Domain: {meta.domain}")
print(f"Custom metadata: {meta.custom_metadata_map}")

Related Pages

• Implementation:Microsoft_Onnxruntime_PlotCommonErrors
• Implementation:Microsoft_Onnxruntime_PlotPipeline