Implementation:Onnx Onnx Save Model

Knowledge Sources	ONNX ONNX API Reference
Domains	Serialization, Model_Persistence
Last Updated	2026-02-10 00:00 GMT

Overview

Concrete tool for serializing ONNX models to disk provided by the ONNX top-level package.

Description

The save_model function serializes a ModelProto to a file, supporting multiple formats and optional external data handling. The serialization format is auto-detected from the file extension or can be specified explicitly. When save_as_external_data is True, the function first converts large tensors to external data references (via convert_model_to_external_data) and then writes the tensor data to separate files (via write_external_data_tensors) before saving the model protobuf.

Usage

Import this function whenever you need to persist an ONNX model to disk. It is the standard way to save models after creation, validation, conversion, or composition. Use the external data parameters when working with models whose tensor data exceeds 2GB.

Code Reference

Source Location

Repository: onnx
File: onnx/__init__.py
Lines: 295-345

Signature

def save_model(
    proto: ModelProto | bytes,
    f: IO[bytes] | str | os.PathLike,
    format: str | None = None,
    *,
    save_as_external_data: bool = False,
    all_tensors_to_one_file: bool = True,
    location: str | None = None,
    size_threshold: int = 1024,
    convert_attribute: bool = False,
) -> None:
    """Saves the ModelProto to the specified path.

    Args:
        proto: In-memory ModelProto or serialized bytes.
        f: File-like object or file path string.
        format: Serialization format (auto-detected from extension if None).
        save_as_external_data: If True, save large tensors to external files.
        all_tensors_to_one_file: Save all tensors to one external file.
        location: External file name (relative to model path).
        size_threshold: Minimum tensor size in bytes to externalize (default: 1024).
        convert_attribute: Also convert attribute tensors to external data.
    """

Import

import onnx

I/O Contract

Inputs

Name	Type	Required	Description
proto	ModelProto or bytes	Yes	Model to save
f	IO[bytes] or str or PathLike	Yes	Output file path or file-like object
format	str or None	No	Serialization format (auto-detected from extension)
save_as_external_data	bool	No	Save large tensors externally (default: False)
all_tensors_to_one_file	bool	No	Single external file for all tensors (default: True)
location	str or None	No	External data file name (relative path)
size_threshold	int	No	Minimum tensor size to externalize in bytes (default: 1024)
convert_attribute	bool	No	Also externalize attribute tensors (default: False)

Outputs

Name	Type	Description
return	None	Model file written to disk (and optional external data files)

Usage Examples

Basic Save

import onnx

# Save as binary protobuf (default)
onnx.save_model(model, "model.onnx")

# Save as JSON
onnx.save_model(model, "model.json", format="json")

Save with External Data

import onnx

# Save large model with external data
onnx.save_model(
    model,
    "large_model.onnx",
    save_as_external_data=True,
    all_tensors_to_one_file=True,
    location="weights.bin",
    size_threshold=1024,
)

Related Pages

Page Connections

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

Principle

Implementation

Heuristic

Environment