Implementation:Onnx Onnx Helper Make Model

Knowledge Sources	ONNX ONNX Helper API
Domains	Model_Construction, Intermediate_Representation
Last Updated	2026-02-10 00:00 GMT

Overview

Concrete tool for constructing complete ONNX model protobuf messages provided by the ONNX helper module.

Description

The make_model function constructs a ModelProto protobuf message from a GraphProto and optional keyword arguments. It automatically sets the IR version to the current onnx.IR_VERSION and, if no opset imports are specified, adds a default import for the latest ai.onnx opset version. The companion make_opsetid function creates OperatorSetIdProto messages for custom opset imports.

Usage

Import this function after assembling a GraphProto via make_graph. The resulting ModelProto is ready for validation with checker.check_model and serialization with save_model. Use opset_imports to pin a specific opset version rather than relying on the default.

Code Reference

Source Location

• Repository: onnx
• File: onnx/helper.py
• Lines: 294-326 (make_model), 243-255 (make_opsetid)

Signature

def make_model(graph: GraphProto, **kwargs: Any) -> ModelProto:
    """Construct a ModelProto.

    Args:
        graph: GraphProto from make_graph.
        **kwargs: Model-level attributes including:
            opset_imports (Sequence[OperatorSetIdProto]): Opset version declarations.
            functions (Sequence[FunctionProto]): Model-local function definitions.
            ir_version (int): Override IR version (default: onnx.IR_VERSION).
            producer_name (str): Producer name metadata.
            model_version (int): Model version number.
            doc_string (str): Model documentation.
    """

def make_opsetid(domain: str, version: int) -> OperatorSetIdProto:
    """Construct an OperatorSetIdProto.

    Args:
        domain: The domain of the operator set (e.g., "" for ai.onnx).
        version: Version of the operator set.
    """

Import

from onnx import helper

I/O Contract

Inputs

Name	Type	Required	Description
graph	GraphProto	Yes	The computation graph from make_graph
opset_imports	Sequence[OperatorSetIdProto]	No	Opset version declarations (default: latest ai.onnx)
functions	Sequence[FunctionProto]	No	Model-local function definitions
ir_version	int	No	ONNX IR version (default: onnx.IR_VERSION)
producer_name	str	No	Name of the producing tool
model_version	int	No	Model version number

Outputs

Name	Type	Description
return	ModelProto	Complete ONNX model protobuf ready for validation and serialization

Usage Examples

Default Opset

from onnx import helper

# Build model with default opset (latest)
model = helper.make_model(graph)
print(f"IR version: {model.ir_version}")
print(f"Opset: {model.opset_import[0].version}")

Explicit Opset Version

from onnx import helper

# Pin to opset 17 for compatibility
model = helper.make_model(
    graph,
    opset_imports=[helper.make_opsetid("", 17)],
    producer_name="my_tool",
    model_version=1,
)

Multiple Domains

from onnx import helper

# Model using both standard and ML domain operators
model = helper.make_model(
    graph,
    opset_imports=[
        helper.make_opsetid("", 21),           # ai.onnx domain
        helper.make_opsetid("ai.onnx.ml", 5),  # ML domain
    ],
)

Related Pages

Implements Principle

• Principle:Onnx_Onnx_Model_Building

Requires Environment

• Environment:Onnx_Onnx_Python_Runtime_Environment