Web Neural Network API

1. Introduction

The Web Neural Network API defines a web-friendly hardware-agnostic abstraction layer that makes use of Machine Learning capabilities of operating systems and underlying hardware platforms without being tied to platform-specific capabilities. The abstraction layer addresses the requirements of key Machine Learning JavaScript frameworks and also allows web developers familiar with the ML domain to write custom code without the help of libraries.

For an illustrated introduction, please see the explainer.

2. Use cases

2.1. Application Use Cases

This section illustrates application-level use cases for neural network inference hardware acceleration. All applications in those use cases can be built on top of pre-trained deep neural network (DNN) [models].

Note: Please be aware that some of the use cases described here, are by their very nature, privacy-invasive. Developers who are planning to use the API for such use cases should ensure that the API is being used to benefit users, for purposes that users understand, and approve. They should apply the Ethical Principles for Web Machine Learning [webmachinelearning-ethics] and implement appropriate privacy risk mitigations such as transparency, data minimisation, and users controls.

2.1.1. Person Detection

A user opens a web-based video conferencing application, but she temporarily leaves from her room. The application is watching whether she is in front of her PC by using object detection (for example, using object detection approaches such as [SSD] or [YOLO] that use a single DNN) to detect regions in a camera input frame that include persons.

When she comes back, the application automatically detects her and notifies other online users that she is active now.

2.1.2. Semantic Segmentation

A user joins a teleconference via a web-based video conferencing application at her desk since no meeting room in her office is available. During the teleconference, she does not wish that her room and people in the background are visible. To protect the privacy of the other people and the surroundings, the application runs a machine learning model such as [DeepLabv3+], [MaskR-CNN] or [SegAny] to semantically split an image into segments and replaces segments that represent other people and background with another picture.

2.1.3. Skeleton Detection

A web-based video conferencing application tracks a pose of user’s skeleton by running a machine learning model, which allows for real-time human pose estimation, such as [PoseNet] to recognize her gesture and body language. When she raises her hand, her microphone is automatically unmuted and she can start speaking on the teleconference.

2.1.4. Face Recognition

There are multiple people in the conference room and they join an online meeting using a web-based video conferencing application. The application detects faces of participants by using object detection (for example, using object detection approaches such as [SSD]) and checks whether each face was present at the previous meeting or not by running a machine learning model such as [FaceNet], which verifies whether two faces would be identical or not.

2.1.5. Facial Landmark Detection

A user wants to find new glasses that beautifully fits her on an online glasses store. The online store offers web-based try-on simulator that runs a machine learning model such as Face Alignment Network [FAN] to detect facial landmarks like eyes, nose, mouth, etc. When she chooses a pair of glasses, the simulator properly renders the selected glasses on the detected position of eyes on her facial image.

2.1.6. Style Transfer

A user is looking for cosmetics on an online store and wondering which color may fit her face. The online store shows sample facial makeup images of cosmetics, and offers makeup simulator that runs a machine learning model like [ContextualLoss] or [PairedCycleGAN] to transfer the makeup style of the sample makeup image to her facial image. She can check how the selected makeup looks like on her face by the simulator.

2.1.7. Super Resolution

A web-based video conferencing is receiving a video stream from its peer, but the resolution of the video becomes lower due to network congestion. To prevent degradation of the perceived video quality, the application runs a machine learning model for super-resolution such as [SRGAN] to generate higher-resolution video frames.

2.1.8. Image Captioning

For better accessibility, a web-based presentation application provides automatic image captioning by running a machine learning model such as [im2txt] which predicts explanatory words of the presentation slides.

2.1.9. Text-to-image

Images are a core part of modern web experiences. An ability to generate images based on text input in a privacy-preserving manner enables visual personalization and adaptation of web applications and content. For example, a web application can use as an input a natural language description on the web page or a description provided by the user within a text prompt to produce an image matching the text description. This text-to-image use case enabled by latent diffusion model architecture [LDM] forms the basis for additional text-to-image use cases. For example, inpainting where a portion of an existing image on the web page is selectively modified using the newly generated content, or the converse, outpainting, where an original image is extended beyond its original dimensions filling the empty space with generated content.

2.1.10. Machine Translation

Multiple people from various countries are talking via a web-based real-time text chat application. The application translates their conversation by using a machine learning model such as [GNMT] or [OpenNMT], which translates every text into different language.

2.1.11. Emotion Analysis

A user is talking to her friend via a web-based real-time text chat application, and she is wondering how the friend feels because she cannot see the friend’s face. The application analyses the friend’s emotion by using a machine learning model such as [DeepMoji], which infers emotion from input texts, and displays an emoji that represents the estimated emotion.

2.1.12. Video Summarization

A web-based video conferencing application records received video streams, and it needs to reduce recorded video data to be stored. The application generates the short version of the recorded video by using a machine learning model for video summarization such as [Video-Summarization-with-LSTM].

2.1.13. Noise Suppression

A web-based video conferencing application records received audio streams, but usually the background noise is everywhere. The application leverages real-time noise suppression using Recurrent Neural Network such as [RNNoise] for suppressing background dynamic noise like baby cry or dog barking to improve audio experiences in video conferences.

2.1.14. Speech Recognition

Speech recognition, also known as speech to text, enables recognition and translation of spoken language into text. Example applications of speech recognition include transcription, automatic translation, multimodal interaction, real-time captioning and virtual assistants. Speech recognition improves accessibility of auditory content and makes it possible to interact with such content in a privacy-preserving manner in a textual form. Examples of common use cases include watching videos or participating in online meetings using real-time captioning. Models such as [Whisper] approach humans in their accuracy and robustness and are well positioned to improve accessibility of such use cases.

2.1.15. Text Generation

Various text generation use cases are enabled by large language models (LLM) that are able to perform tasks where a general ability to predict the next item in a text sequence is required. This class of models can translate texts, answer questions based on a text input, summarize a larger body of text, or generate text output based on a textual input. LLMs enable better performance compared to older models based on RNN, CNN, or LSTM architectures and further improve the performance of many other use cases discussed in this section. Examples of LLMs include [t5-small], [m2m100_418M], [gpt2], and [llama-2-7b].

2.1.16. Detecting fake video

A user is exposed to realistic fake videos generated by ‘deepfake’ on the web. The fake video can swap the speaker’s face into the president’s face to incite a user politically or to manipulate user’s opinion. The deepfake detection applications such as [FaceForensics++] analyze the videos and protect a user against the fake videos or images. When she watches a fake video on the web, the detection application alerts her of the fraud video in real-time.

2.2. Framework Use Cases

This section collects framework-level use cases for a dedicated low-level API for neural network inference hardware acceleration. It is expected that Machine Learning frameworks will be key consumers of the Web Neural Network API (WebNN API) and the low-level details exposed through the WebNN API are abstracted out from typical web developers. However, it is also expected that web developers with specific interest and competence in Machine Learning will want to interface with the WebNN API directly instead of a higher-level ML framework.

2.2.1. Custom Layer

A web application developer wants to run a DNN model on the WebNN API. However, she has found that some of activation functions like [LeakyReLU], [ELU], etc. are not included in the WebNN API. To address this issue, she constructs custom layers of the additional activation functions on top of the WebNN API. Note that the scope of custom layers may include convolution, normalization, etc. as well as activation.

2.2.2. Network Concatenation

A web application uses a DNN model, and its model data of upper convolutional layers and lower fully-connected layers are stored in separate files, since model data of the fully-connected layers are periodically updated due to fine tuning at the server side.

Therefore, the application downloads both partial model files at first and concatenates them into a single model. When the model is updated, the application downloads fine-tuned part of the model and replace only the fully-connected layers with it.

2.2.3. Performance Adaptation

A web application developer has a concern about performance of her DNN model on mobile devices. She has confirmed that it may run too slow on mobile devices which do not have GPU acceleration. To address this issue, her web application refers to the WebNN API to confirm whether acceleration is available or not, so that the application can display the warning for devices without acceleration.

After several weeks, she has developed a tiny DNN model that can even run on CPU. In order to accommodate CPU execution, she modifies the application so that the application loads the tiny model in the case of CPU-only devices.

2.2.4. Operation Level Execution

A JavaScript ML framework is responsible for loading, interpreting and executing a ML model. During the model execution phase, the framework iterates through the operations of the model and executes each operation on the hardware device, like CPU, GPU or ML accelerator. To avoid the unnecessary data copying across devices, the framework selects the same device to execute the operations. For a compute intensive operation, such as convolution 2D or matrix multiplication, the framework uses WebNN API to execute it with the ML-specific acceleration available on that selected device.

2.2.5. Integration with real-time video processing

The user experience of WebRTC-based video conferencing is enhanced using real-time video processing. For example, background blur implemented using a § 2.1.2 Semantic Segmentation model blurs the background in the user’s live camera feed. To satisfy the performance requirements of this use case, the WebNN API integrates with primitives from other Web APIs that make up the media pipeline to allow WebNN API-based transformation of real-time video streams.

3. Security Considerations

This API is disabled by default in all cross-origin frames using the § 6.4 Permissions Policy Integration. This prevents third-party content from using this API unless the embedding page explicitly sets a policy that grants permission.

This API allows creation of an MLContext from a GPUDevice defined by WebGPU specification. See WebGPU Security Considerations for more information regarding security characteristics of this context.

This API provides an abstraction across GPU, CPU, and dedicated ML accelerator hardware. When using a GPU, denial of service considerations similar to WebGPU apply. When using a CPU or a dedicated ML accelerator, the types of potential resource contention are different and mitigations will be implementation and configuration dependent. Implementations should use whatever mechanisms are available from the platform to prevent sites from using an unfair amount of system resources. These compute units are shared resources, and the use of any compute API will affect overall performance on a fully-loaded system.

Once the graph is fully constructed and compiled, the input shapes into each of the operations in the graph are inferred and finalized. The bounds checking occurs when the compute method is invoked that executes the graph against the actual data. No actual data is bound to the compiled graph before this stage. It is the implementation’s responsibility to make sure proper bounds checking occurs against the shapes of the data already inferred by that time.

Document operations susceptible to out-of-bounds access as a guidance to implementers.

Implementations must defend against control-flow attacks based on changes to data considered to be constant. For example, optimizations in the underlying platform may assume that a weight remains unchanged throughout a computation. If the API allowed the contents of buffers holding weights to change during a computation then those optimization assumptions would be invalidated, causing undefined behavior in the underlying platform. The API mitigates this category of attacks from script by always copying or transferring buffers, but implementations should consider additional defenses such as process isolation of data assumed to be constant.

As a future-proofing measure, the API design allows certain operations that can be generically emulated to be deprecated for security, performance, or other reasons without breaking compatibility. This is made possible by high-level functions that are defined in terms of smaller primitive operations defined in this specifications. This enables a native implementation of a high-level function to be replaced with a polyfill implementation.

Investigate side channel attack feasibility considering the current state where CPU is shared between processes running renderers.

In order to not allow an attacker to target a specific implementation that may contain a flaw, the § 6.2 Device Selection mechanism is a hint only, and the concrete device selection is left to the implementation - a user agent could for instance choose never to run a model on a device with known vulnerabilities. As a further mitigation, no device enumeration mechanism is defined.

Hinting partially mitigates the concern. Investigate additional mitigations.

The API design minimizes the attack surface for the compiled computational graph. The MLGraphBuilder interface that hosts the various operations is a data definition API and as such doesn’t execute anything, only constructs data. What follows, is that the potential for an attack is limited to when binding the data to the graph before executing it by invoking the MLContext.dispatch() method. This enables implementers to focus on hardening the MLContext.dispatch() method. For example, by making sure it honors the boundary of data and fails appropriately when the bounds are not respected.

Purpose-built Web APIs for measuring high-resolution time mitigate against timing attacks using techniques such as resolution reduction, adding jitter, detection of abuse and API call throttling [hr-time-3]. The practical deployment of WebNN implementations are likely to bring enough jitter to make timing attacks impractical (e.g. because they would use IPC) but implementers are advised to consider and test their implementations against timing attacks.

3.1. Guidelines for new operations

This section is non-normative.

4. Privacy Considerations

This API enhances privacy compared to cloud-based inference, since input data such as locally sourced images or video streams stay within the browser’s sandbox.

This API exposes the minimum amount of information necessary to address the identified § 2 Use cases for the best performance and reliability of results.

No information from the underlying platform is exposed directly. An execution time analysis may reveal indirectly the performance of the underlying platform’s neural network hardware acceleration capabilities relative to another underlying platform.

Note: The group is soliciting further input on the proposed execution time analysis fingerprinting vector and will augment this section with more information and mitigations to inform the implementers of this API.

Unlike WebGPU, this API does not intrinsically support custom shader authoring; and as a result is not prone to timing attacks that rely on shader caches, or other persistent data. The API builds upon pre-existing shaders and lower level primitives of the browser or the underlying OS. Web developers who interface with GPUDevice are expected to be aware of WebGPU compilation cache considerations.

The WebGPU API identifies machine-specific artifacts as a privacy consideration. Similarly, the WebNN API’s compute unit scheduling may under certain circumstances introduce a fingerprint. However, similarly to WebGPU, such fingerprints are identical across most or all of the devices of each vendor, mitigating the concern. Furthermore, software implementations can be used to further eliminate such artifacts.

The WebNN API defines two developer-settable preferences to help inform § 6.2 Device Selection and allow the implementation to better select the most appropriate underlying execution device for the workload. An MLDeviceType normatively indicates the kind of device and is one of: "cpu", "gpu", "npu". If this type cannot be satisfied, an "OperationError" DOMException is thrown, thus this type can in some cases add two bits of entropy to the fingerprint. An MLPowerPreference indicates preference as related to the power consumption and is considered a hint only and as such does not increase entropy of the fingerprint.

MLContextOptions is under active development, and the design is expected to change, informed by further implementation experience and new use cases from the wider web community. [Issue #623]

If a future version of this specification introduces support for a new MLDeviceType that can only support a subset of MLOperandDataTypes, that could introduce a new fingerprint.

In general, implementers of this API are expected to apply WebGPU Privacy Considerations to their implementations where applicable.

5. Ethical Considerations

The Working Group has started documenting ethical issues associated with using Machine Learning on the Web, to help identify what mitigations its normative specifications should take into account. The Working Group publishes and maintains an Ethical Principles for Web Machine Learning document [webmachinelearning-ethics] open to contributions from the wider community via a dedicated GitHub repository.

6. Programming Model

6.1. Overview

At the heart of neural networks is a computational graph of mathematical operations. These operations are the building blocks of modern machine learning technologies in computer vision, natural language processing, and robotics. The WebNN API is a specification for constructing, compiling, and executing computational graphs of neural networks.

The MLGraph interface represents a compiled computational graph that is immutable (that is, a model).

The MLGraphBuilder interface serves as a builder (factory) to construct a computational graph (its graph) that is then compiled to create an MLGraph.

In WebNN, a computational graph is composed of operators which act on data, and are the nodes of the graph. MLOperands are a representation of data that flows within the computational graph, and are the edges of the graph. MLOperands include a computational graph's input values for inference, constants (including trained weights) used for inference, intermediate values (often referred to as activations) computed during inference, as well as the output values of inference. An operator's input is one or more MLOperands. An operator's output is one or more MLOperands. Operators have operator-specific parameters that control their behavior, which can include zero or more activation functions.

A key part of the MLGraphBuilder interface are methods such as gemm() and relu() which create an operator which represents the actual operation to perform on the input data when the computation is run, and return a new MLOperand holding the operator. Methods that create an MLOperand connect any inputs and activations to the operator. Each method invocation returns a distinct new value, without changing the value of any other MLOperand.

An operator has a label, a string which may be included in diagnostics such as exception messages. When an operator is created its label is initialized in an implementation-defined manner and may include the passed label.

Note: Implementations are encouraged to use the label provided by developers to enhance error messages and improve debuggability, including both synchronous errors during graph construction and for errors that occur during the asynchronous build() method.

Consider adding a mechanism for reporting errors during dispatch(). [Issue #778]

At inference time, every MLOperand will be bound to a tensor (the actual data), which are essentially multidimensional arrays. The representation of the tensors is implementation dependent, but it typically includes the array data stored in some buffer (memory) and some metadata describing the array data (such as its shape).

Operations within the computational graph have functional semantics. This allows the implementation to potentially share the array data between multiple tensors. For example, the implementation of operations such as reshape, or slice may return a view of its input tensor that shares the same buffer as the input tensor. (In the case of reshape, the entire data is shared, while in the case of slice, a part of the input data is shared.) The implementation may use views, as above, for intermediate values.

Before the execution, the computation graph that is used to compute one or more specified outputs needs to be converted, compiled, and optimized. The key purpose of the compilation step is to enable optimizations that span two or more operations, such as operation or loop fusion. The user agent may also perform these optimizations during graph conversion.

The MLGraphBuilder.build() method compiles the graph in the background without blocking the calling thread, and returns a Promise that resolves to an MLGraph. Each MLGraphBuilder can build at most one MLGraph.

The MLGraph underlying implementation will be composed of platform-specific representations of operators and operands which correspond to the MLGraphBuilder's operators and MLOperands, but which are not script-visible and may be compositions or decompositions of the graph as constructed by script.

Once the MLGraph is constructed, the MLContext.dispatch() method performs the execution of the graph asynchronously either on a parallel timeline in a separate worker thread for the CPU execution or on a GPU timeline in a GPU command queue. This method returns immediately without blocking the calling thread while the actual execution is offloaded to a different timeline. The caller supplies the input values using MLNamedTensors, binding the input MLOperands to their values. The caller also supplies MLNamedTensors for output MLOperands which will contain the result of graph execution, if successful, which may be read back to script using the MLContext.readTensor(tensor) method. This type of execution supports CPU, GPU, and NPU devices.

6.2. Device Selection

An MLContext interface represents a global state of neural network execution. One of the important context states is the underlying execution device that manages the resources and facilitates the compilation and the eventual execution of the neural network graph. In addition to the default method of creation with MLContextOptions, an MLContext could also be created from a specific GPUDevice that is already in use by the application.

In a situation when a GPU context executes a graph with a constant or an input in the system memory as an ArrayBufferView, the input content is automatically uploaded from the system memory to the GPU memory, and downloaded back to the system memory of an ArrayBufferView output buffer at the end of the graph execution. This data upload and download cycles will only occur whenever the execution device requires the data to be copied out of and back into the system memory, such as in the case of the GPU. It doesn’t occur when the device is a CPU device. Additionally, the result of the graph execution is in a known layout format. While the execution may be optimized for a native memory access pattern in an intermediate result within the graph, the output of the last operation of the graph must convert the content back to a known layout format at the end of the graph in order to maintain the expected behavior from the caller’s perspective.

When an MLContext is created with MLContextOptions, the user agent selects and creates the underlying execution device by taking into account the application’s MLPowerPreference and MLDeviceType options.

6.3. Task Source

The ML task source is a task source to be used for all tasks related to asynchronous compilation and execution of MLGraphs and creation of MLContexts.

6.4. Permissions Policy Integration

This specification defines a policy-controlled feature identified by the string "webnn". Its default allowlist is 'self'.

7. API

7.1. The navigator.ml interface

An ML object is available in the Window and DedicatedWorkerGlobalScope contexts through the Navigator and WorkerNavigator interfaces respectively and is exposed via navigator.ml.

interface mixin NavigatorML {
  [SecureContext, SameObject] readonly attribute ML ml;
};
Navigator includes NavigatorML;
WorkerNavigator includes NavigatorML;

7.2. ML interface

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);
};

7.2.1. MLContextOptions

MLContextOptions is under active development, and the design is expected to change, informed by further implementation experience and new use cases from the wider web community. The Working Group is considering additional API controls to allow the definition of a fallback device, multiple devices in a preferred order, or an exclusion of a specific device. Other considerations under discussion include error handling, ultimate fallback, and quantized operators. Feedback is welcome on any of these design considerations from web developers, library authors, OS and hardware vendors, and other stakeholders via GitHub: [Issue #623]

The deviceType option is an MLDeviceType and indicates the application’s preference for the kind of device used for the context. It is one of the following:

"cpu"

Provides the broadest compatibility and usability across all client devices with varying degrees of performance.

"gpu"

Provides the broadest range of achievable performance across graphics hardware platforms from consumer devices to professional workstations. The underlying platform implementation may fall back to other devices for certain operators and parts of the graph.

"npu"

Provides power efficiency for sustained workloads across hardware platforms with purpose-built accelerators. The underlying platform implementation may fall back to other devices for certain operators and parts of the graph.

The powerPreference option is an MLPowerPreference and indicates the application’s preference as related to power consumption. It is one of the following:

"default"

Let the user agent select the most suitable behavior.

"high-performance"

Prioritizes execution speed over power consumption.

"low-power"

Prioritizes power consumption over other considerations such as execution speed.

7.2.2. createContext()

7.3. MLContext interface

typedef record<USVString, MLTensor> MLNamedTensors;

[SecureContext, Exposed=(Window, DedicatedWorker)]
interface MLContext {
  undefined dispatch(MLGraph graph, MLNamedTensors inputs, MLNamedTensors outputs);  

  Promise<MLTensor> createTensor(MLTensorDescriptor descriptor);

  Promise<ArrayBuffer> readTensor(MLTensor tensor);
  Promise<undefined> readTensor(MLTensor tensor, AllowSharedBufferSource outputData);

  undefined writeTensor(MLTensor tensor, AllowSharedBufferSource inputData);

  MLOpSupportLimits opSupportLimits();
};

The context type is the type of the execution context that manages the resources and facilitates the compilation and execution of the neural network graph:

"default"

Context created per user preference options.

"webgpu"

Context created from WebGPU device.

7.3.1. dispatch()

Schedules the computational workload of a compiled MLGraph on the MLContext's [[timeline]].

Note: dispatch() itself provides no signal that graph execution has completed. Rather, callers can await the results of reading back the output tensors. See § 7.3.1.1 Examples below.

7.3.1.1. Examples

const descriptor = {dataType: 'float32', shape: [2, 2]};
const context = await navigator.ml.createContext();
const builder = new MLGraphBuilder(context);

// 1. Create a computational graph 'C = 0.2 * A + B'.
const constant = builder.constant(descriptor, new Float32Array(4).fill(0.2));
const A = builder.input('A', descriptor);
const B = builder.input('B', descriptor);
const C = builder.add(builder.mul(A, constant), B);

// 2. Compile the graph.
const graph = await builder.build({'C': C});

// 3. Create reusable input and output tensors.
const [inputTensorA, inputTensorB, outputTensorC] =
    await Promise.all([
      context.createTensor({
        dataType: A.dataType, shape: A.shape, writable: true
      }),
      context.createTensor({
        dataType: B.dataType, shape: B.shape, writable: true
      }),
      context.createTensor({
        dataType: C.dataType, shape: C.shape, readable: true
      })
    ]);

// 4. Initialize the inputs.
context.writeTensor(inputTensorA, new Float32Array(4).fill(1.0));
context.writeTensor(inputTensorB, new Float32Array(4).fill(0.8));

// 5. Execute the graph.
const inputs = {
  'A': inputTensorA,
  'B': inputTensorB
};
const outputs = {
  'C': outputTensorC
};
context.dispatch(graph, inputs, outputs);
    
// 6. Read back the computed result.
const result = await context.readTensor(outputTensorC);
console.log('Output value:', new Float32Array(result));  // [1, 1, 1, 1]

7.3.2. createTensor()

Creates an MLTensor associated with this MLContext.

7.3.3. readTensor(tensor)

Reads back the [[data]] of an MLTensor from the MLContext.[[timeline]] to script.

7.3.4. readTensor(tensor, outputData)

Bring-your-own-buffer variant of readTensor(tensor). Reads back the [[data]] of an MLTensor into the provided buffer.

7.3.5. writeTensor()

Writes data to the [[data]] of an MLTensor on the MLContext's [[timeline]].

Note: Similar to dispatch(), writeTensor() itself provides no signal that the write has completed. To inspect the contents of a tensor, callers can await the results of reading back the tensor.

7.3.6. opSupportLimits()

7.3.6.1. MLOpSupportLimits dictionary

dictionary MLOpSupportLimits {
  MLInputOperandLayout preferredInputLayout;
  MLSupportLimits input;
  MLSupportLimits constant;
  MLSupportLimits output;
};

preferredInputLayout, of type MLInputOperandLayout

Preferred input layout for layout dependent operators like conv2d().

input, of type MLSupportLimits

Support limits for input MLOperands for an MLGraph.

constant, of type MLSupportLimits

Support limits for constant MLOperands for an MLGraph.

output, of type MLSupportLimits

Support limits for output MLOperands for an MLGraph.

7.3.6.2. MLSupportLimits dictionary

dictionary MLSupportLimits {
  sequence<MLOperandDataType> dataTypes;
};

dataTypes, of type sequence<MLOperandDataType>

Supported data types.

7.3.6.3. MLBinarySupportLimits dictionary

dictionary MLBinarySupportLimits {
  MLSupportLimits a;
  MLSupportLimits b;
  MLSupportLimits output;
};

a, of type MLSupportLimits

MLSupportLimits for a operand.

b, of type MLSupportLimits

MLSupportLimits for b operand.

output, of type MLSupportLimits

MLSupportLimits for output operand.

7.3.6.4. MLSingleInputSupportLimits dictionary

dictionary MLSingleInputSupportLimits {
  MLSupportLimits input;
  MLSupportLimits output;
};

input, of type MLSupportLimits

MLSupportLimits for input operand.

output, of type MLSupportLimits

MLSupportLimits for output operand.

7.4. MLGraph interface

[SecureContext, Exposed=(Window, DedicatedWorker)]
interface MLGraph {};

7.5. MLOperandDescriptor dictionary

An MLOperandDescriptor describes the shape (dimensions) and data type of an operand. They are used to describe the inputs and constants for an MLGraph, and every MLOperand has an internal MLOperandDescriptor.

enum MLInputOperandLayout {
  "nchw",
  "nhwc"
};

enum MLOperandDataType {
  "float32",
  "float16",
  "int32",
  "uint32",
  "int64",
  "uint64",
  "int8",
  "uint8"
};

dictionary MLOperandDescriptor {
  required MLOperandDataType dataType;
  required sequence<[EnforceRange] unsigned long> shape;
};

dataType, of type MLOperandDataType

The operand data type.

shape, of type sequence<[EnforceRange] unsigned long>

The list of dimensions of the operand. It is empty for scalar operands.

A valid dimension is an integer greater than zero and in the range of long. Implementations may impose a smaller upper bound.

Should 0-size dimensions be supported? [Issue #391]

7.6. MLOperand interface

An MLOperand represents an intermediary graph being constructed as a result of compositing parts of an operation into a fully composed operation.

For instance, an MLOperand can represent a constant feeding to an operation or the result from combining multiple constants together into an operation. See also § 6 Programming Model.

[SecureContext, Exposed=(Window, DedicatedWorker)]
interface MLOperand {
  readonly attribute MLOperandDataType dataType;
  readonly attribute FrozenArray<unsigned long> shape;
};

dictionary MLOperatorOptions {
  USVString label = "";
};

typedef (bigint or unrestricted double) MLNumber;

An MLOperand's dataType is its [[descriptor]].dataType.

An MLOperand's shape is its [[descriptor]].shape.

An MLOperand's rank is its shape's size.

The dataType getter steps are to return this's dataType.

The shape getter steps are to return this's shape.

Since the [[builder]] object is bound by the MLGraphBuilder() constructor to an MLContext object, an MLOperand is also always bound to the same MLContext object.

If an operation supports only a subset of MLOperandDataTypes, the allowed data types for each of the operation’s input operands, including both positional arguments and options, are given as either an explicit list of MLOperandDataTypes, or a constraint that the operand’s dataType must be the same as the dataType of another input operand, or any to allow any MLOperandDataType.

If an operation requires input operands with a particular rank, the allowed ranks for each of the operation’s input operands, including both positional arguments and options, are given as an explicit rank (e.g. 1), or N to allow any dimensionality, or the same as another operand. More specific constraints are common, such as when an input operand’s shape must be unidirectionally broadcastable to or bidirectionally broadcastable with another input operand; in these cases, the allowed ranks are listed as a range, with specific validation given as steps in the operation.

MLOperatorOptions has the following members:

label, of type USVString, defaulting to ""

Optionally provided when an operator is created using MLGraphBuilder methods that create MLOperands. The implementation may use this value to initialize the operator's label.

7.6.1. Creating an MLOperand

To validate operand given MLGraphBuilder builder and MLOperand operand, return true if operand.[[builder]] is builder, and false otherwise.

7.6.1.1. MLNumber

MLNumber is used when specifying the type of a numeric option for an MLOperand which can be of any MLOperandDataType, including both 64-bit integer types ("uint64" and "int64") and 32-bit floating point ("float32"). Implementations process the value according to the corresponding MLOperandDataType. For example, if clamp(input, options) is called with an MLOperand with dataType "uint32", the MLNumber parameters are explicitly cast to unsigned long.

Support for unions of bigint and numeric types is new in [WEBIDL], and implementation support is also limited. Prototype implementations are encouraged to provide feedback for this approach. [Issue #whatwg/webidl#1388]

7.7. MLTensorDescriptor dictionary

An MLTensorDescriptor describes the characteristics and capabilities of an MLTensor.

dictionary MLTensorDescriptor : MLOperandDescriptor {
  boolean readable = false;
  boolean writable = false;
};

readable, of type boolean, defaulting to false

Whether the tensor’s contents can be read via readTensor(tensor) or readTensor(tensor, outputData).

writable, of type boolean, defaulting to false

Whether the tensor’s contents can be written to via writeTensor().

7.8. MLTensor interface

The MLTensor interface represents a tensor which may be used as an input or output to an MLGraph. The memory backing an MLTensor should be allocated in an implementation-defined fashion according to the requirements of the MLContext and the MLTensorDescriptor used to create it. Operations involving the [[data]] of an MLTensor occur on the [[timeline]] of its associated MLContext.

The implementation-defined requirements of how an MLTensor is allocated may include constraints such as that the memory is allocated with a particular byte alignment or in a particular memory pool.

[SecureContext, Exposed=(Window, DedicatedWorker)]
interface MLTensor {
  readonly attribute MLOperandDataType dataType;
  readonly attribute FrozenArray<unsigned long> shape;
  readonly attribute boolean readable;
  readonly attribute boolean writable;

  undefined destroy();
};

An MLTensor's dataType is its [[descriptor]]'s dataType.

An MLTensor's shape is its [[descriptor]]'s shape.

The dataType getter steps are to return this's dataType.

The shape getter steps are to return this's shape.

The readable getter steps are to return this.[[descriptor]].readable.

The writable getter steps are to return this.[[descriptor]].writable.

7.8.1. Creating an MLTensor

An MLTensor is created by its associated MLContext.

7.8.2. destroy()

Releases the resources associated with the MLTensor. This method is idempotent.

Note: Since no further operations can be enqueued using this tensor, implementations can free any additional resource allocations associated with this tensor once all previously submitted operations using it are complete.

7.9. MLGraphBuilder interface

The MLGraphBuilder interface defines a set of operations as identified by the § 2 Use cases that can be composed into a computational graph. It also represents the intermediate state of a graph building session.

typedef record<USVString, MLOperand> MLNamedOperands;

[SecureContext, Exposed=(Window, DedicatedWorker)]
interface MLGraphBuilder {
  // Construct the graph builder from the context.
  constructor(MLContext context);

  // Create an operand for a graph input.
  MLOperand input(USVString name, MLOperandDescriptor descriptor);

  // Create an operand for a graph constant.
  MLOperand constant(MLOperandDescriptor descriptor,
                     AllowSharedBufferSource buffer);

  // Create a scalar operand from the specified number of the specified type.
  MLOperand constant(MLOperandDataType type, MLNumber value);

  // Compile the graph up to the specified output operands asynchronously.
  Promise<MLGraph> build(MLNamedOperands outputs);
};

7.9.1. MLGraphBuilder constructor

7.9.2. input operands

Create a named MLOperand based on a descriptor, that can be used as an input.

7.9.3. constant operands

7.9.3.1. constant(descriptor, buffer)

7.9.3.2. constant(type, value)

7.9.4. build method

NOTE: Specifying an input operand or constant operand as a graph output results in an error, as this is usually an incorrect usage of the API. Callers can work around this by introducing an identity() operator.

7.9.5. argMin/argMax operations

dictionary MLArgMinMaxOptions : MLOperatorOptions {
  boolean keepDimensions = false;
  MLOperandDataType outputDataType = "int32";
};

partial interface MLGraphBuilder {
  MLOperand argMin(MLOperand input, [EnforceRange] unsigned long axis,
                   optional MLArgMinMaxOptions options = {});
  MLOperand argMax(MLOperand input, [EnforceRange] unsigned long axis,
                   optional MLArgMinMaxOptions options = {});
};

partial dictionary MLOpSupportLimits {
  MLSingleInputSupportLimits argMin;
  MLSingleInputSupportLimits argMax;
};

MLArgMinMaxOptions has the following members:

keepDimensions, of type boolean, defaulting to false

If true, retains reduced dimensions with size 1.

outputDataType, of type MLOperandDataType, defaulting to "int32"

An MLOperandDataType. The output data type.

MLOpSupportLimits has the following members for argMin() and argMax():

argMin, of type MLSingleInputSupportLimits

Support limits for operator argMin().

argMax, of type MLSingleInputSupportLimits

Support limits for operator argMax().

7.9.6. batchNormalization

dictionary MLBatchNormalizationOptions : MLOperatorOptions {
  MLOperand scale;
  MLOperand bias;
  [EnforceRange] unsigned long axis = 1;
  double epsilon = 1e-5;
};

partial interface MLGraphBuilder {
  MLOperand batchNormalization(MLOperand input, MLOperand mean, MLOperand variance,
                               optional MLBatchNormalizationOptions options = {});
};

dictionary MLBatchNormalizationSupportLimits {
  MLSupportLimits input;
  MLSupportLimits mean;
  MLSupportLimits variance;
  MLSupportLimits scale;
  MLSupportLimits bias;
  MLSupportLimits output;
};

partial dictionary MLOpSupportLimits {
  MLBatchNormalizationSupportLimits batchNormalization;
};

MLBatchNormalizationOptions has the following members:

scale, of type MLOperand

The 1-D tensor of the scaling values whose size is equal to the size of the input dimension denoted by axis.

bias, of type MLOperand

The 1-D tensor of the bias values whose size is equal to the size of the input dimension denoted by axis.

axis, of type unsigned long, defaulting to 1

The index to the feature count dimension of the input shape for which the mean and variance values are. Its value must be in the range [0, N-1] where N is the rank of the input tensor. The default value is 1, corresponding to the channel ("c") dimension in the "nchw" data layout.

epsilon, of type double, defaulting to 1e-5

A small value to prevent computational error due to divide-by-zero.

MLBatchNormalizationSupportLimits has the following members:

input, of type MLSupportLimits

MLSupportLimits for input operand.

mean, of type MLSupportLimits

MLSupportLimits for mean operand.

variance, of type MLSupportLimits

MLSupportLimits for variance operand.

scale, of type MLSupportLimits

MLSupportLimits for scale operand.

bias, of type MLSupportLimits

MLSupportLimits for bias operand.

output, of type MLSupportLimits

MLSupportLimits for output operand.

MLOpSupportLimits has the following members for batchNormalization():

batchNormalization, of type MLBatchNormalizationSupportLimits

Support limits for operator batchNormalization().

function batchNormalization(builder, input, mean, variance, options) {
  const shape = [1, input.shape[options.axis], 1, 1];
  return builder.add(
    builder.mul(
      builder.reshape(options.scale, shape),
      builder.div(
        builder.sub(input, builder.reshape(mean, shape)),
        builder.sqrt(builder.add(
          builder.reshape(variance, shape),
          builder.constant(input.dataType, options.epsilon))))),
    builder.reshape(options.bias, shape));
}

7.9.7. cast

partial interface MLGraphBuilder {
  MLOperand cast(MLOperand input,
                 MLOperandDataType type,
                 optional MLOperatorOptions options = {});
};

partial dictionary MLOpSupportLimits {
  MLSingleInputSupportLimits cast;
};

MLOpSupportLimits has the following members for cast():

cast, of type MLSingleInputSupportLimits

Support limits for operator cast().

Casting between MLOperandDataTypes is specified for some cases and implementation-defined in other cases, according to the following table:

NOTE: For example, casting -1 from "int8" to "uint8" is specified to yield 255. But casting -1 from "float32" to "uint8" is implementation-defined.

7.9.8. clamp

dictionary MLClampOptions : MLOperatorOptions {
  MLNumber minValue;
  MLNumber maxValue;
};

partial interface MLGraphBuilder {
  MLOperand clamp(MLOperand input, optional MLClampOptions options = {});
};

partial dictionary MLOpSupportLimits {
  MLSingleInputSupportLimits clamp;
};

MLClampOptions has the following members:

minValue, of type MLNumber

The minimum value of the range. When it is not specified, the clamping is not performed on the lower limit of the range.

maxValue, of type MLNumber

The maximum value of the range. When it is not specified, the clamping is not performed on the upper limit of the range.

MLOpSupportLimits has the following member for clamp():

clamp, of type MLSingleInputSupportLimits

Support limits for operator clamp().

function clamp(builder, input, options) {
  if (options.minValue === undefined) {
    if (options.maxValue === undefined) {
      return input;
    } else {
      return builder.min(
        input, builder.constant(input.dataType, options.maxValue));
    }
  } else {
    if (options.maxValue === undefined) {
      return builder.max(
        input, builder.constant(input.dataType, options.minValue));
    } else {
      return builder.min(
        builder.max(input, builder.constant(input.dataType, options.minValue)),
        builder.constant(input.dataType, options.maxValue));
    }
  }
}

7.9.9. concat

partial interface MLGraphBuilder {
  MLOperand concat(sequence<MLOperand> inputs,
                   [EnforceRange] unsigned long axis,
                   optional MLOperatorOptions options = {});
};

dictionary MLConcatSupportLimits {
  MLSupportLimits inputs;
  MLSupportLimits output;
};

partial dictionary MLOpSupportLimits {
  MLConcatSupportLimits concat;
};

MLConcatSupportLimits has the following members:

inputs, of type MLSupportLimits

MLSupportLimits for all input operands.

output, of type MLSupportLimits

MLSupportLimits for output operand.

MLOpSupportLimits has the following member for concat():

concat, of type MLConcatSupportLimits

Support limits for operator concat().

7.9.10. conv2d

enum MLConv2dFilterOperandLayout {
  "oihw",
  "hwio",
  "ohwi",
  "ihwo"
};

dictionary MLConv2dOptions : MLOperatorOptions {
  sequence<[EnforceRange] unsigned long> padding;
  sequence<[EnforceRange] unsigned long> strides;
  sequence<[EnforceRange] unsigned long> dilations;
  [EnforceRange] unsigned long groups = 1;
  MLInputOperandLayout inputLayout = "nchw";
  MLConv2dFilterOperandLayout filterLayout = "oihw";
  MLOperand bias;
};

partial interface MLGraphBuilder {
  MLOperand conv2d(MLOperand input,
                   MLOperand filter,
                   optional MLConv2dOptions options = {});
};

dictionary MLConv2dSupportLimits {
  MLSupportLimits input;
  MLSupportLimits filter;
  MLSupportLimits bias;
  MLSupportLimits output;
};

partial dictionary MLOpSupportLimits {
  MLConv2dSupportLimits conv2d;
};

MLConv2dOptions has the following members:

padding, of type sequence<[EnforceRange] unsigned long>

A list of length 4: [beginningHeight, endingHeight, beginningWidth, endingWidth]. Specifies the additional rows and columns added to the beginning and ending of each spatial dimension of the convolution input. The default value is [0, 0, 0, 0].

strides, of type sequence<[EnforceRange] unsigned long>

A list of length 2: [strideHeight, strideWidth]. Specifies the stride of the sliding window for each spatial dimension of the convolution input. The default value is [1, 1].

dilations, of type sequence<[EnforceRange] unsigned long>

A list of length 2: [dilationHeight, dilationWidth]. Specifies the dilation factor for each spatial dimension applied on the convolution filter (kernel). The default value is [1, 1].

groups, of type unsigned long, defaulting to 1

The number of groups that input channels and output channels are divided into.

inputLayout, of type MLInputOperandLayout, defaulting to "nchw"

Specifies the layout format of the input and output tensor as follows:

• "nchw"

  • input tensor: [batches, inputChannels, height, width]

  • output tensor: [batches, outputChannels, height, width]

• "nhwc":

  • input tensor: [batches, height, width, inputChannels]

  • output tensor: [batches, height, width, outputChannels]

filterLayout, of type MLConv2dFilterOperandLayout, defaulting to "oihw"

Specifies the layout format of the filter tensor as follows:

• "oihw": [outputChannels, inputChannels/groups, height, width]

• "hwio": [height, width, inputChannels/groups, outputChannels]

• "ohwi": [outputChannels, height, width, inputChannels/groups]

• "ihwo": [inputChannels/groups, height, width, outputChannels]

bias, of type MLOperand

An additional 1-D tensor with the shape of [outputChannels] whose values are to be added to the convolution result.

MLConv2dSupportLimits has the following members:

input, of type MLSupportLimits

MLSupportLimits for input operand.

filter, of type MLSupportLimits

MLSupportLimits for filter operand.

bias, of type MLSupportLimits

MLSupportLimits for bias operand.

output, of type MLSupportLimits

MLSupportLimits for output operand.

MLOpSupportLimits has the following member for conv2d():

conv2d, of type MLConv2dSupportLimits

Support limits for operator conv2d().

7.9.11. convTranspose2d

enum MLConvTranspose2dFilterOperandLayout {
  "iohw",
  "hwoi",
  "ohwi"
};

dictionary MLConvTranspose2dOptions : MLOperatorOptions {
  sequence<[EnforceRange] unsigned long> padding;
  sequence<[EnforceRange] unsigned long> strides;
  sequence<[EnforceRange] unsigned long> dilations;
  sequence<[EnforceRange] unsigned long> outputPadding;
  sequence<[EnforceRange] unsigned long> outputSizes;
  [EnforceRange] unsigned long groups = 1;
  MLInputOperandLayout inputLayout = "nchw";
  MLConvTranspose2dFilterOperandLayout filterLayout = "iohw";
  MLOperand bias;
};

partial interface MLGraphBuilder {
  MLOperand convTranspose2d(MLOperand input, MLOperand filter,
                            optional MLConvTranspose2dOptions options = {});
};

partial dictionary MLOpSupportLimits {
  MLConv2dSupportLimits convTranspose2d;
};

MLConvTranspose2dOptions has the following members:

padding, of type sequence<[EnforceRange] unsigned long>

A list of length 4: [beginningHeight, endingHeight, beginningWidth, endingWidth]. Specifies the additional rows and columns added to the beginning and ending of each spatial dimension of the convolution input. The default value is [0, 0, 0, 0].

strides, of type sequence<[EnforceRange] unsigned long>

A list of length 2: [strideHeight, strideWidth]. Specifies the stride of the sliding window for each spatial dimension of the convolution input. The default value is [1, 1].

dilations, of type sequence<[EnforceRange] unsigned long>

A list of length 2: [dilationHeight, dilationWidth]. Specifies the dilation factor for each spatial dimension applied on the convolution filter (kernel). The default value is [1, 1].

outputPadding, of type sequence<[EnforceRange] unsigned long>

A list of length 2. Specifies the padding values applied to each spatial dimension of the output tensor. The explicit padding values are needed to disambiguate the output tensor shape for transposed convolution when the value of the options.strides is greater than 1.

Note that these values are only used to disambiguate output shape when needed; it does not necessarily cause any padding value to be written to the output tensor.

The default value is [0, 0].

outputSizes, of type sequence<[EnforceRange] unsigned long>

A list of length 2. Specifies the sizes of the last two dimensions of the output tensor. When the output sizes are explicitly specified, the output padding values in outputPadding are ignored.

If not specified, the output sizes are automatically computed.

groups, of type unsigned long, defaulting to 1

The number of groups that input channels and output channels are divided into.

inputLayout, of type MLInputOperandLayout, defaulting to "nchw"

Specifies the layout format of the input and output tensor as follows:

• "nchw"

  • input tensor: [batches, inputChannels, height, width]

  • output tensor: [batches, outputChannels, height, width]

• "nhwc":

  • input tensor: [batches, height, width, inputChannels]

  • output tensor: [batches, height, width, outputChannels]

filterLayout, of type MLConvTranspose2dFilterOperandLayout, defaulting to "iohw"

Specifies the layout format of the filter tensor as follows:

• "iohw": [inputChannels, outputChannels/groups, height, width]

• "hwoi": [height, width, outputChannels/groups, inputChannels]

• "ohwi": [outputChannels/groups, height, width, inputChannels]

bias, of type MLOperand

An additional 1-D tensor with the shape of [outputChannels] whose values are to be added to the convolution result.

MLOpSupportLimits has the following member for convTranspose2d():

convTranspose2d, of type MLConv2dSupportLimits

Support limits for operator convTranspose2d().

7.9.12. Element-wise binary operations

The operation will be broadcast according to [numpy-broadcasting-rule]. The input tensors must be bidirectionally broadcastable. The rank of the output tensor is the maximum rank of the input tensors. For each dimension of the output tensor, its size is the maximum size along that dimension of the input tensors.

partial interface MLGraphBuilder {
  MLOperand add(MLOperand a, MLOperand b, optional MLOperatorOptions options = {});
  MLOperand sub(MLOperand a, MLOperand b, optional MLOperatorOptions options = {});
  MLOperand mul(MLOperand a, MLOperand b, optional MLOperatorOptions options = {});
  MLOperand div(MLOperand a, MLOperand b, optional MLOperatorOptions options = {});
  MLOperand max(MLOperand a, MLOperand b, optional MLOperatorOptions options = {});
  MLOperand min(MLOperand a, MLOperand b, optional MLOperatorOptions options = {});
  MLOperand pow(MLOperand a, MLOperand b, optional MLOperatorOptions options = {});
};

partial dictionary MLOpSupportLimits {
  MLBinarySupportLimits add;
  MLBinarySupportLimits sub;
  MLBinarySupportLimits mul;
  MLBinarySupportLimits div;
  MLBinarySupportLimits max;
  MLBinarySupportLimits min;
  MLBinarySupportLimits pow;
};

MLOpSupportLimits has the following members for element-wise binary operations:

add, of type MLBinarySupportLimits

Support limits for operator add().

sub, of type MLBinarySupportLimits

Support limits for operator sub().

mul, of type MLBinarySupportLimits

Support limits for operator mul().

div, of type MLBinarySupportLimits

Support limits for operator div().

max, of type MLBinarySupportLimits

Support limits for operator max().

min, of type MLBinarySupportLimits

Support limits for operator min().

pow, of type MLBinarySupportLimits

Support limits for operator pow().

7.9.13. Element-wise logical operations

For multiple-operand operations, the operation will be broadcast according to [numpy-broadcasting-rule]. The input tensors must be bidirectionally broadcastable. The rank of the output tensor is the maximum rank of the input tensors. For each dimension of the output tensor, its size is the maximum size along that dimension of the input tensors.

partial interface MLGraphBuilder {
  MLOperand equal(MLOperand a,
                  MLOperand b,
                  optional MLOperatorOptions options = {});
  MLOperand greater(MLOperand a,
                    MLOperand b,
                    optional MLOperatorOptions options = {});
  MLOperand greaterOrEqual(MLOperand a,
                           MLOperand b,
                           optional MLOperatorOptions options = {});
  MLOperand lesser(MLOperand a,
                   MLOperand b,
                   optional MLOperatorOptions options = {});
  MLOperand lesserOrEqual(MLOperand a,
                          MLOperand b,
                          optional MLOperatorOptions options = {});
  MLOperand logicalNot(MLOperand a, optional MLOperatorOptions options = {});
};

dictionary MLLogicalNotSupportLimits {
  MLSupportLimits a;
  MLSupportLimits output;
};

partial dictionary MLOpSupportLimits {
  MLBinarySupportLimits equal;
  MLBinarySupportLimits greater;
  MLBinarySupportLimits greaterOrEqual;
  MLBinarySupportLimits lesser;
  MLBinarySupportLimits lesserOrEqual;
  MLLogicalNotSupportLimits logicalNot;
};

MLLogicalNotSupportLimits has the following members:

a, of type MLSupportLimits

MLSupportLimits for a operand.

output, of type MLSupportLimits

MLSupportLimits for output operand.

MLOpSupportLimits has the following members for element-wise logical operations: