Implementation:Onnx Onnx Checker Check Model

Knowledge Sources	ONNX ONNX Checker API
Domains	Validation, Model_Integrity
Last Updated	2026-02-10 00:00 GMT

Overview

Concrete tool for validating ONNX model correctness provided by the ONNX checker module, wrapping a C++ validation engine.

Description

The check_model function validates an ONNX model against the specification. It is a Python wrapper around the C++ checker implementation (onnx_cpp2py_export.checker) exposed via nanobind. The function accepts either a ModelProto object (for models under 2GB) or a file path string (required for models over 2GB). When a ModelProto is passed, it is serialized to bytes and forwarded to the C++ checker. When a file path is passed, the C++ engine reads the file directly, avoiding the 2GB protobuf size limit.

The function raises checker.ValidationError on any validation failure, making it suitable for use in assertion-style checks.

Usage

Import this function whenever you need to verify model correctness. Call it after model construction, after version conversion, after model composition, or before deployment. Use full_check=True when you also want shape inference validation. For models with external data exceeding 2GB, pass the file path instead of the ModelProto.

Code Reference

Source Location

Repository: onnx
File: onnx/checker.py
Lines: 121-169

Signature

def check_model(
    model: onnx.ModelProto | str | bytes | os.PathLike,
    full_check: bool = False,
    skip_opset_compatibility_check: bool = False,
    check_custom_domain: bool = False,
) -> None:
    """Check the consistency of a model.

    Args:
        model: Model to check. If model is a path, the function checks
            via the path-based C++ API. For models >2GB, path is required.
        full_check: If True, also runs shape inference check.
        skip_opset_compatibility_check: If True, skip opset validation.
        check_custom_domain: If True, validate all domains (not just built-in).

    Raises:
        ValidationError: If the model fails validation.
        ValueError: If the serialized model exceeds 2GB without using a path.
    """

Import

from onnx import checker

I/O Contract

Inputs

Name	Type	Required	Description
model	ModelProto or str or bytes or PathLike	Yes	Model object or file path (path required for >2GB)
full_check	bool	No	Also run shape inference check (default: False)
skip_opset_compatibility_check	bool	No	Skip opset validation (default: False)
check_custom_domain	bool	No	Validate all domains (default: False)

Outputs

Name	Type	Description
return	None	Returns nothing on success; raises ValidationError on failure

Usage Examples

Basic Validation

from onnx import checker

# Validate an in-memory model
checker.check_model(model)

# Validate with shape inference
checker.check_model(model, full_check=True)

Path-based Validation for Large Models

from onnx import checker

# Required for models >2GB with external data
checker.check_model("large_model.onnx")

# With full shape inference check
checker.check_model("large_model.onnx", full_check=True)

Error Handling

from onnx import checker

try:
    checker.check_model(model)
    print("Model is valid")
except checker.ValidationError as e:
    print(f"Validation failed: {e}")

Related Pages

Implements Principle

Principle:Onnx_Onnx_Model_Validation

Requires Environment

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment