Implementation:Microsoft Onnxruntime InferenceSession Create Nodejs

Field	Value
Implementation Name	InferenceSession_Create_Nodejs
Overview	Asynchronous creation of an ONNX Runtime inference session from a model file or buffer in Node.js.
Type	API Doc
Language	JavaScript
Domains	ML_Inference, JavaScript_Integration
Source Repository	microsoft/onnxruntime
Last Updated	2026-02-10

Overview

Asynchronous creation of an ONNX Runtime inference session from a model file or buffer in Node.js using the InferenceSession.create() factory method. This async method loads an ONNX model, parses the graph, performs graph optimizations, and returns a ready-to-use inference session.

API

ort.InferenceSession.create(arg: string | Uint8Array | ArrayBuffer, options?: { intraOpNumThreads?: number }) -> Promise<InferenceSession>

Import

const ort = require('onnxruntime-node');
// or
const { InferenceSession } = require('onnxruntime-node');

Source Code Reference

Repository: microsoft/onnxruntime
Primary Source: samples/nodejs/04_create-inference-session/index.js:L15-31

Key Parameters

Parameter	Type	Required	Description
arg	string / Uint8Array / ArrayBuffer	Yes	ONNX model source: file path string, Uint8Array buffer, or ArrayBuffer with offset/length
options.intraOpNumThreads	number	No	Number of threads for intra-operator parallelism (default: system-determined)
byteOffset	number	No	Byte offset when using ArrayBuffer (third positional argument)
byteLength	number	No	Byte length when using ArrayBuffer (fourth positional argument)

I/O Contract

Direction	Name	Type	Description
Input	arg	string / Uint8Array / ArrayBuffer	ONNX model file path or binary buffer containing the serialized model
Input	options	object (optional)	Session configuration options including intraOpNumThreads
Output	session	Promise<InferenceSession>	Promise resolving to an initialized InferenceSession ready for inference

Usage Examples

Create from File Path

const ort = require('onnxruntime-node');

async function main() {
    // Create session from a file path
    const session = await ort.InferenceSession.create('./model.onnx');
    console.log('Session created successfully');
}
main();

Create with Session Options

const ort = require('onnxruntime-node');

async function main() {
    const options = { intraOpNumThreads: 1 };
    const session = await ort.InferenceSession.create('./model.onnx', options);
    console.log('Session created with single thread');
}
main();

Create from Uint8Array Buffer

const ort = require('onnxruntime-node');
const fs = require('fs');
const util = require('util');

async function main() {
    // Read model file into a Node.js Buffer (Uint8Array)
    const buffer = await util.promisify(fs.readFile)('./model.onnx');
    const session = await ort.InferenceSession.create(buffer);
    console.log('Session created from buffer');
}
main();

Create from ArrayBuffer with Offset

const ort = require('onnxruntime-node');
const fs = require('fs');
const util = require('util');

async function main() {
    const buffer = await util.promisify(fs.readFile)('./model.onnx');
    const arrayBuffer = buffer.buffer;
    const offset = buffer.byteOffset;
    const length = buffer.byteLength;
    const session = await ort.InferenceSession.create(arrayBuffer, offset, length);
    console.log('Session created from ArrayBuffer with offset');
}
main();

Key Details

The create() method is asynchronous to avoid blocking the Node.js event loop during model loading and graph optimization.
The factory pattern is used because JavaScript constructors cannot be asynchronous.
When loading from a file path, the native layer reads the file directly without JavaScript involvement.
When loading from a buffer, the model bytes are passed to the native layer via N-API.
The session retains the optimized graph in memory for the lifetime of the session object.

Related Pages

Page Connections

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

Principle

Implementation

Heuristic

Environment