Implementation:Onnx Onnx Helper Make Node

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

Overview

Concrete tool for constructing computation operator nodes provided by the ONNX helper module.

Description

The make_node function constructs a NodeProto protobuf message representing a single computation operation in the ONNX graph. It accepts the operator type, input/output edge names, and optional attributes as keyword arguments. Attributes are converted to AttributeProto messages via the internal make_attribute function, supporting integers, floats, strings, tensors, graphs, and their repeated variants.

Usage

Import this function when you need to define computation steps in an ONNX model. Call it once per operation, passing the operator name from the ONNX spec (e.g., "Relu", "Conv", "Add"), the input tensor names, and the output tensor names. Operator-specific parameters (like kernel_shape for Conv) are passed as keyword arguments.

Code Reference

Source Location

• Repository: onnx
• File: onnx/helper.py
• Lines: 133-179

Signature

def make_node(
    op_type: str,
    inputs: Sequence[str],
    outputs: Sequence[str],
    name: str | None = None,
    doc_string: str | None = None,
    domain: str | None = None,
    overload: str | None = None,
    **kwargs: Any,
) -> NodeProto:
    """Construct a NodeProto.

    Args:
        op_type: The name of the operator to construct (e.g., "Relu", "Conv").
        inputs: List of input edge names.
        outputs: List of output edge names.
        name: Optional unique identifier for the node.
        doc_string: Optional documentation string.
        domain: Optional operator domain (default is "" for standard ONNX ops).
        overload: Optional field for resolving model-local function calls.
        **kwargs: Operator attributes passed to make_attribute.
    """

Import

from onnx import helper

I/O Contract

Inputs

Name	Type	Required	Description
op_type	str	Yes	ONNX operator name (e.g., "Relu", "Conv", "MatMul")
inputs	Sequence[str]	Yes	List of input tensor edge names
outputs	Sequence[str]	Yes	List of output tensor edge names
name	str or None	No	Optional unique node identifier
domain	str or None	No	Operator domain (default: "" for ai.onnx)
**kwargs	Any	No	Operator-specific attributes (e.g., kernel_shape, pads)

Outputs

Name	Type	Description
return	NodeProto	Protobuf message representing a single computation node

Usage Examples

Simple Activation Node

from onnx import helper

# Create a Relu activation node
relu_node = helper.make_node(
    "Relu",           # op_type
    ["X"],            # inputs
    ["Y"],            # outputs
    name="relu0"      # optional name
)

Node with Attributes

from onnx import helper

# Create a Conv node with kernel shape and padding attributes
conv_node = helper.make_node(
    "Conv",
    inputs=["X", "W", "B"],
    outputs=["Y"],
    kernel_shape=[3, 3],
    pads=[1, 1, 1, 1],
    strides=[1, 1],
)

# Create an Add node (element-wise addition)
add_node = helper.make_node(
    "Add",
    inputs=["A", "B"],
    outputs=["C"],
)

Related Pages

Implements Principle

• Principle:Onnx_Onnx_Operator_Node_Construction

Requires Environment

• Environment:Onnx_Onnx_Python_Runtime_Environment