ML

The ML Interface provides methods for creating and managing machine learning contexts in web applications.

Web IDL

web-idl


enum MLDeviceType {
  "cpu",
  "gpu",
  "npu"
};

enum MLPowerPreference {
  "default",
  "high-performance",
  "low-power"
};

dictionary MLContextOptions {
  MLDeviceType deviceType = "cpu";
  MLPowerPreference powerPreference = "default";
};

[SecureContext, Exposed=(Window, DedicatedWorker)]
interface ML {
  Promise<MLContext> createContext(optional MLContextOptions options = {});
  Promise<MLContext> createContext(GPUDevice gpuDevice);
};

Methods

createContext()

Syntax

create-context.js


createContext()
createContext(options)
createContext(gpuDevice)

Parameters

options: A MLContextOptions object providing the application’s preferences
gpuDevice: A specific GPUDevice object to use with the context

Return Value

A MLContext

Description

The ML interface enables creation of machine learning contexts with specific device and power preferences. It provides two methods for creating contexts - one accepting configuration options and another accepting a specific GPU device.

Device Type Options

An enumeration defining supported device types for ML operations:

"cpu": CPU-based processing
"gpu": GPU-based processing
"npu": Neural Processing Unit-based processing

`"cpu"`

Provides maximum compatibility across devices
Variable performance based on hardware capabilities
Default option if not specified

`"gpu"`

Offers highest performance potential
May fall back to other devices for certain operations
Requires compatible graphics hardware

`"npu"`

Optimized for power efficiency
Suitable for sustained workloads
May fall back to other devices for unsupported operations

Power Preference Options

An enumeration specifying power consumption preferences:

"default": System-determined behavior
"high-performance": Prioritizes speed over power efficiency
"low-power": Prioritizes power efficiency over performance

`"default"`

Lets the user agent determine optimal behavior
Balanced approach to performance and power consumption

`"high-performance"`

Maximizes execution speed
May increase power consumption
Suitable for compute-intensive tasks

`"low-power"`

Minimizes power consumption
May reduce performance
Ideal for battery-powered devices

ML Context Options

A dictionary of configuration options for ML contexts:

web-idl


dictionary MLContextOptions {
  MLDeviceType deviceType = "cpu";
  MLPowerPreference powerPreference = "default";
}

Examples

Creating a Context with Default Options

example-1.js


try {
  const context = await navigator.ml.createContext();
  // Context ready for use
} catch (error) {
  console.error('Failed to create ML context:', error);
}

Creating a Context with Specific Options

example-2.js


const options = {
  deviceType: "gpu",
  powerPreference: "high-performance"
};
 
try {
  const context = await navigator.ml.createContext(options);
  // High-performance GPU context ready
} catch (error) {
  console.error('Failed to create GPU context:', error);
}

Using with a Specific GPU Device

example-3.js


// Assuming we have a GPUDevice instance
const gpuDevice = await navigator.gpu.requestAdapter()
  .then(adapter => adapter.requestDevice());
 
try {
  const context = await navigator.ml.createContext(gpuDevice);
  // Context created with specific GPU device
} catch (error) {
  console.error('Failed to create context with GPU device:', error);
}

Specifications

WebNN API	Status
`ML` interface	Candidate Recommendation Draft

Security Requirements

Requires a secure context
Available in Window and DedicatedWorker contexts

Browser Compatibility

The ML interface is under active development and browser support may vary.

See Browser Compatibility: WebNN API

Development Status

The ML interface and MLContextOptions are actively being developed. Current considerations include:

Implementation of fallback device mechanisms
Support for multiple devices in preferred order
Device exclusion capabilities
Enhanced error handling
Quantized operator support

ML