mirror of
https://github.com/saymrwulf/onnxruntime.git
synced 2026-07-26 19:52:38 +00:00
### Description This PR is a preview of cherry-picks for ort-web to `rel-1.17.3` based on `rel-1.17.2`. <details> <summary>Changes of ort-web to cherry-pick</summary> The following commits are from main branch. `o` stands for pick, and `x` stands for skip. ``` o2e0a388c36[js/webgpu] Add HardSigmoid support (#19215) od226e40856[js/webgpu] set query type in onRunStart (#19202) o61610ff986[js/webgpu] Add FusedConv clip test case (#18900) oa33b5bd1fa[JS/WebGPU] Added Uniforms to SkipLayerNorm. (#18788) o591f90c0b9[js/webgpu] Fix issue of timestamp query (#19258) o7252c6e747[WebNN EP] Support WebNN async API with Asyncify (#19145) o5b06505073[js/webgpu] Fix Tanh explosion (#19201) o656ca66186[js/webgpu] Support uniforms for conv, conv transpose, conv grouped (#18753) oa3f0e2422b[js/webgpu] Support f16 uniform (#19098) o9e69606360fix f16 for attention, enable slice and flatten for more types (#19262) o624b4e2063[js/webgpu] Remove enableShapesUniforms (#19279) o90883a366a[js/webgpu] Add hardSigmoid activation for fusedConv (#19233) o85cef0af8c[js/webgpu] Support capture and replay for jsep (#18989) od73131cf0f[js/webgpu] Use DataType as uniform cpu type (#19281) odd1f6ccc45[js/webgpu] resolve codescan alert (#19343) o3a2ab1963a[js/webgpu] Refactor createTensorShapeVariables (#18883) oefc17e79de[js/webgpu] Fix the undefined push error (#19366) x50806a7dd5[js/web] support external data in npm test (#19377) occbe264a39[js/webgpu] Add LeakyRelu activation for fusedConv (#19369) o5ff27ef02a[js/webgpu] support customop FastGelu (#19392) x03be65e064[js/web] fix types exports in package.json (#19458) o06269a3952[js/webgpu] allow uint8 tensors for webgpu (#19545) odfeda9019c[JS/WebGPU] Add MatMulNBits (#19446) o1b48054e1b[js/webgpu] Create Split indices helpers by rank, not by shape (#19554) o3fe2c137ee[js] small fix to workaround formatter (#19400) x70567a4b3a[js/web] use ApiTensor insteadof onnxjs Tensor in TensorResultValidator (#19358) o6e04e36e3f[js/common] upgrade tsc in common from 4.9.5 to 5.2.2 (#19317) o58f4921686[js] changes to allow Float16Array if any polyfill is available (#19305) o57d6819212[js/web] Fix fused-conv is not included in npm test (#19581) oebd220b073Misspelling in README.md (#19433) o38c3432393Bump ip from 1.1.8 to 1.1.9 in /js/react_native (#19582) ofe82fccf1a[js/webgpu] Fix Conv2DTransposeMatMul f16 compilation failure (#19596) o76a2a487a1Bump ip from 1.1.8 to 1.1.9 in /js/react_native/e2e (#19583) o29b1106033[node] Switch to setImmediate to avoid starving the Node.js event loop (#19610) oae3d73c981[JS/WebGPU] Fix Split and Where to handle corner cases. (#19613) oaec2389ad0[js/webgpu] allows a ProgramInfo's RunData to use zero sized output (#19614) obb43a0f133[js/webgpu] minor fixes to make tinyllama work (#19564) o0edb035808[js/web] fix suite test list for zero sized tensor (#19638) o3cb81cdde2[js/common] move 'env.wasm.trace' to 'env.trace' (#19617) oe30618d055[js/webgpu] use Headless for webgpu test by default (#19702) of06164ef8b[js/web] transfer input buffer back to caller thread (#19677) xa788514027[js/web] dump debug logs for karma for diagnose purpose (#19785) o24b72d2613[JS/WebGPU] Preserve zero size input tensor dims. (#19737) o4538d31a8b[js/webgpu] expose a few properties in WebGPU API (#19857) o53de2d8cb0[js/webgpu] Enable GroupedConvVectorize path (#19791) oed250b88c3[JS/WebGPU] Optimize MatMulNBits (#19852) xe771a763c3[js/test] align web test runner flags with ort.env (#19790) o79e50aeef3[js/web] rewrite backend resolve to allow multiple EPs (#19735) oacb0df2280Fix #19931 broken Get Started link of "ONNX Runtime JavaScript API" page (#19932) ob29849a287[js/common] fix typedoc warnings (#19933) oafdab62f53Bump follow-redirects from 1.15.4 to 1.15.6 in /js/web (#19949) o28ad6c3955Bump follow-redirects from 1.15.4 to 1.15.6 in /js/node (#19951) o7e0d424934accumulate in fp32 for Reduce* (#19868) o4c6a6a37f7[js/webgpu] Fix NAN caused by un-initialized buffer in instance-norm (#19387) o01c7aaf6aa[js/webgpu] allow setting env.webgpu.adapter (#19940) oc45cff60cf[js/webgpu] fix maxpool / fp16 (#19981) ``` </details> <details> <summary>Cherry-pick commandlines</summary> ```sh git cherry-pick2e0a388c36git cherry-pickd226e40856git cherry-pick61610ff986git cherry-picka33b5bd1fagit cherry-pick591f90c0b9git cherry-pick7252c6e747git cherry-pick5b06505073git cherry-pick656ca66186git cherry-picka3f0e2422bgit cherry-pick9e69606360git cherry-pick624b4e2063git cherry-pick90883a366agit cherry-pick85cef0af8c#<<<<< Note: conflicts git cherry-pickd73131cf0fgit cherry-pickdd1f6ccc45git cherry-pick3a2ab1963agit cherry-pickefc17e79degit cherry-pickccbe264a39git cherry-pick5ff27ef02agit cherry-pick06269a3952git cherry-pickdfeda9019cgit cherry-pick1b48054e1bgit cherry-pick3fe2c137eegit cherry-pick6e04e36e3fgit cherry-pick58f4921686git cherry-pick57d6819212git cherry-pickebd220b073git cherry-pick38c3432393git cherry-pickfe82fccf1agit cherry-pick76a2a487a1git cherry-pick29b1106033git cherry-pickae3d73c981git cherry-pickaec2389ad0git cherry-pickbb43a0f133git cherry-pick0edb035808git cherry-pick3cb81cdde2git cherry-picke30618d055git cherry-pickf06164ef8bgit cherry-pick24b72d2613git cherry-pick4538d31a8bgit cherry-pick53de2d8cb0git cherry-picked250b88c3git cherry-pick79e50aeef3git cherry-pickacb0df2280git cherry-pickb29849a287git cherry-pickafdab62f53git cherry-pick28ad6c3955git cherry-pick7e0d424934git cherry-pick4c6a6a37f7git cherry-pick01c7aaf6aagit cherry-pickc45cff60cf``` </details> <details> <summary>Cherry-pick conflicts</summary> -85cef0af8c#18989 this change is for enabling graph capture feature for JSEP, and it is done after ROCM EP enabled graph capture feature. However, the ROCM EP graph capture feature is not cherry-picked in rel-1.17.2. </details> --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: Jiajia Qin <jiajia.qin@intel.com> Co-authored-by: Xu Xing <xing.xu@intel.com> Co-authored-by: satyajandhyala <satya.k.jandhyala@gmail.com> Co-authored-by: Yang Gu <yang.gu@intel.com> Co-authored-by: Wanming Lin <wanming.lin@intel.com> Co-authored-by: Jiajie Hu <jiajie.hu@intel.com> Co-authored-by: Guenther Schmuelling <guschmue@microsoft.com> Co-authored-by: Matttttt <18152455+martholomew@users.noreply.github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Segev Finer <segev208@gmail.com> Co-authored-by: Belem Zhang <belem.zhang@intel.com>
320 lines
12 KiB
TypeScript
320 lines
12 KiB
TypeScript
// Copyright (c) Microsoft Corporation. All rights reserved.
|
|
// Licensed under the MIT License.
|
|
|
|
import {Tensor, TypedTensor} from './tensor.js';
|
|
|
|
export type ImageFormat = 'RGB'|'RGBA'|'BGR'|'RBG';
|
|
export type ImageTensorLayout = 'NHWC'|'NCHW';
|
|
|
|
// the following region contains type definitions for constructing tensor from a specific location.
|
|
|
|
// #region types for constructing a tensor from a specific location
|
|
|
|
/**
|
|
* represent common properties of the parameter for constructing a tensor from a specific location.
|
|
*/
|
|
interface CommonConstructorParameters<T> extends Pick<Tensor, 'dims'> {
|
|
/**
|
|
* Specify the data type of the tensor.
|
|
*/
|
|
readonly type: T;
|
|
}
|
|
|
|
/**
|
|
* represent the parameter for constructing a tensor from a GPU resource.
|
|
*/
|
|
interface GpuResourceConstructorParameters<T extends Tensor.Type> {
|
|
/**
|
|
* an optional callback function to download data from GPU to CPU.
|
|
*
|
|
* If not provided, the tensor treat the GPU data as external resource.
|
|
*/
|
|
download?(): Promise<Tensor.DataTypeMap[T]>;
|
|
|
|
/**
|
|
* an optional callback function that will be called when the tensor is disposed.
|
|
*
|
|
* If not provided, the tensor treat the GPU data as external resource.
|
|
*/
|
|
dispose?(): void;
|
|
}
|
|
|
|
/**
|
|
* represent the parameter for constructing a tensor from a pinned CPU buffer
|
|
*/
|
|
export interface CpuPinnedConstructorParameters<T extends Tensor.CpuPinnedDataTypes = Tensor.CpuPinnedDataTypes> extends
|
|
CommonConstructorParameters<T> {
|
|
/**
|
|
* Specify the location of the data to be 'cpu-pinned'.
|
|
*/
|
|
readonly location: 'cpu-pinned';
|
|
/**
|
|
* Specify the CPU pinned buffer that holds the tensor data.
|
|
*/
|
|
readonly data: Tensor.DataTypeMap[T];
|
|
}
|
|
|
|
/**
|
|
* represent the parameter for constructing a tensor from a WebGL texture
|
|
*/
|
|
export interface TextureConstructorParameters<T extends Tensor.TextureDataTypes = Tensor.TextureDataTypes> extends
|
|
CommonConstructorParameters<T>, GpuResourceConstructorParameters<T> {
|
|
/**
|
|
* Specify the location of the data to be 'texture'.
|
|
*/
|
|
readonly location: 'texture';
|
|
/**
|
|
* Specify the WebGL texture that holds the tensor data.
|
|
*/
|
|
readonly texture: Tensor.TextureType;
|
|
}
|
|
|
|
/**
|
|
* represent the parameter for constructing a tensor from a WebGPU buffer
|
|
*/
|
|
export interface GpuBufferConstructorParameters<T extends Tensor.GpuBufferDataTypes = Tensor.GpuBufferDataTypes> extends
|
|
CommonConstructorParameters<T>, GpuResourceConstructorParameters<T> {
|
|
/**
|
|
* Specify the location of the data to be 'gpu-buffer'.
|
|
*/
|
|
readonly location: 'gpu-buffer';
|
|
/**
|
|
* Specify the WebGPU buffer that holds the tensor data.
|
|
*/
|
|
readonly gpuBuffer: Tensor.GpuBufferType;
|
|
}
|
|
|
|
// #endregion
|
|
|
|
// the following region contains type definitions of each individual options.
|
|
// the tensor factory functions use a composition of those options as the parameter type.
|
|
|
|
// #region Options fields
|
|
|
|
export interface OptionsFormat {
|
|
/**
|
|
* Describes the image format represented in RGBA color space.
|
|
*/
|
|
format?: ImageFormat;
|
|
}
|
|
|
|
export interface OptionsTensorFormat {
|
|
/**
|
|
* Describes the image format of the tensor.
|
|
*
|
|
* NOTE: this is different from option 'format'. While option 'format' represents the original image, 'tensorFormat'
|
|
* represents the target format of the tensor. A transpose will be performed if they are different.
|
|
*/
|
|
tensorFormat?: ImageFormat;
|
|
}
|
|
|
|
export interface OptionsTensorDataType {
|
|
/**
|
|
* Describes the data type of the tensor.
|
|
*/
|
|
dataType?: 'float32'|'uint8';
|
|
}
|
|
|
|
export interface OptionsTensorLayout {
|
|
/**
|
|
* Describes the tensor layout when representing data of one or more image(s).
|
|
*/
|
|
tensorLayout?: ImageTensorLayout;
|
|
}
|
|
|
|
export interface OptionsDimensions {
|
|
/**
|
|
* Describes the image height in pixel
|
|
*/
|
|
height?: number;
|
|
/**
|
|
* Describes the image width in pixel
|
|
*/
|
|
width?: number;
|
|
}
|
|
|
|
export interface OptionResizedDimensions {
|
|
/**
|
|
* Describes the resized height. If omitted, original height will be used.
|
|
*/
|
|
resizedHeight?: number;
|
|
/**
|
|
* Describes resized width - can be accessed via tensor dimensions as well
|
|
*/
|
|
resizedWidth?: number;
|
|
}
|
|
|
|
export interface OptionsNormalizationParameters {
|
|
/**
|
|
* Describes normalization parameters when preprocessing the image as model input.
|
|
*
|
|
* Data element are ranged from 0 to 255.
|
|
*/
|
|
norm?: {
|
|
/**
|
|
* The 'bias' value for image normalization.
|
|
* - If omitted, use default value 0.
|
|
* - If it's a single number, apply to each channel
|
|
* - If it's an array of 3 or 4 numbers, apply element-wise. Number of elements need to match the number of channels
|
|
* for the corresponding image format
|
|
*/
|
|
bias?: number|[number, number, number]|[number, number, number, number];
|
|
/**
|
|
* The 'mean' value for image normalization.
|
|
* - If omitted, use default value 255.
|
|
* - If it's a single number, apply to each channel
|
|
* - If it's an array of 3 or 4 numbers, apply element-wise. Number of elements need to match the number of channels
|
|
* for the corresponding image format
|
|
*/
|
|
mean?: number | [number, number, number] | [number, number, number, number];
|
|
};
|
|
}
|
|
|
|
// #endregion
|
|
|
|
// #region Options composition
|
|
|
|
export interface TensorFromImageDataOptions extends OptionResizedDimensions, OptionsTensorFormat, OptionsTensorLayout,
|
|
OptionsTensorDataType, OptionsNormalizationParameters {}
|
|
|
|
export interface TensorFromImageElementOptions extends OptionResizedDimensions, OptionsTensorFormat,
|
|
OptionsTensorLayout, OptionsTensorDataType,
|
|
OptionsNormalizationParameters {}
|
|
|
|
export interface TensorFromUrlOptions extends OptionsDimensions, OptionResizedDimensions, OptionsTensorFormat,
|
|
OptionsTensorLayout, OptionsTensorDataType,
|
|
OptionsNormalizationParameters {}
|
|
|
|
export interface TensorFromImageBitmapOptions extends OptionResizedDimensions, OptionsTensorFormat, OptionsTensorLayout,
|
|
OptionsTensorDataType, OptionsNormalizationParameters {}
|
|
|
|
export interface TensorFromTextureOptions<T extends Tensor.TextureDataTypes> extends
|
|
Required<OptionsDimensions>, OptionsFormat, GpuResourceConstructorParameters<T>/* TODO: add more */ {}
|
|
|
|
export interface TensorFromGpuBufferOptions<T extends Tensor.GpuBufferDataTypes> extends
|
|
Pick<Tensor, 'dims'>, GpuResourceConstructorParameters<T> {
|
|
/**
|
|
* Describes the data type of the tensor.
|
|
*/
|
|
dataType?: T;
|
|
}
|
|
|
|
// #endregion
|
|
|
|
/**
|
|
* type TensorFactory defines the factory functions of 'Tensor' to create tensor instances from existing data or
|
|
* resources.
|
|
*/
|
|
export interface TensorFactory {
|
|
/**
|
|
* create a tensor from an ImageData object
|
|
*
|
|
* @param imageData - the ImageData object to create tensor from
|
|
* @param options - An optional object representing options for creating tensor from ImageData.
|
|
*
|
|
* The following default settings will be applied:
|
|
* - `tensorFormat`: `'RGB'`
|
|
* - `tensorLayout`: `'NCHW'`
|
|
* - `dataType`: `'float32'`
|
|
* @returns A promise that resolves to a tensor object
|
|
*/
|
|
fromImage(imageData: ImageData, options?: TensorFromImageDataOptions):
|
|
Promise<TypedTensor<'float32'>|TypedTensor<'uint8'>>;
|
|
|
|
/**
|
|
* create a tensor from a HTMLImageElement object
|
|
*
|
|
* @param imageElement - the HTMLImageElement object to create tensor from
|
|
* @param options - An optional object representing options for creating tensor from HTMLImageElement.
|
|
*
|
|
* The following default settings will be applied:
|
|
* - `tensorFormat`: `'RGB'`
|
|
* - `tensorLayout`: `'NCHW'`
|
|
* - `dataType`: `'float32'`
|
|
* @returns A promise that resolves to a tensor object
|
|
*/
|
|
fromImage(imageElement: HTMLImageElement, options?: TensorFromImageElementOptions):
|
|
Promise<TypedTensor<'float32'>|TypedTensor<'uint8'>>;
|
|
|
|
/**
|
|
* create a tensor from URL
|
|
*
|
|
* @param urlSource - a string as a URL to the image or a data URL containing the image data.
|
|
* @param options - An optional object representing options for creating tensor from URL.
|
|
*
|
|
* The following default settings will be applied:
|
|
* - `tensorFormat`: `'RGB'`
|
|
* - `tensorLayout`: `'NCHW'`
|
|
* - `dataType`: `'float32'`
|
|
* @returns A promise that resolves to a tensor object
|
|
*/
|
|
fromImage(urlSource: string, options?: TensorFromUrlOptions): Promise<TypedTensor<'float32'>|TypedTensor<'uint8'>>;
|
|
|
|
/**
|
|
* create a tensor from an ImageBitmap object
|
|
*
|
|
* @param bitmap - the ImageBitmap object to create tensor from
|
|
* @param options - An optional object representing options for creating tensor from URL.
|
|
*
|
|
* The following default settings will be applied:
|
|
* - `tensorFormat`: `'RGB'`
|
|
* - `tensorLayout`: `'NCHW'`
|
|
* - `dataType`: `'float32'`
|
|
* @returns A promise that resolves to a tensor object
|
|
*/
|
|
fromImage(bitmap: ImageBitmap, options: TensorFromImageBitmapOptions):
|
|
Promise<TypedTensor<'float32'>|TypedTensor<'uint8'>>;
|
|
|
|
/**
|
|
* create a tensor from a WebGL texture
|
|
*
|
|
* @param texture - the WebGLTexture object to create tensor from
|
|
* @param options - An optional object representing options for creating tensor from WebGL texture.
|
|
*
|
|
* The options include following properties:
|
|
* - `width`: the width of the texture. Required.
|
|
* - `height`: the height of the texture. Required.
|
|
* - `format`: the format of the texture. If omitted, assume 'RGBA'.
|
|
* - `download`: an optional function to download the tensor data from GPU to CPU. If omitted, the GPU data
|
|
* will not be able to download. Usually, this is provided by a GPU backend for the inference outputs. Users don't
|
|
* need to provide this function.
|
|
* - `dispose`: an optional function to dispose the tensor data on GPU. If omitted, the GPU data will not be disposed.
|
|
* Usually, this is provided by a GPU backend for the inference outputs. Users don't need to provide this function.
|
|
*
|
|
* @returns a tensor object
|
|
*/
|
|
fromTexture<T extends Tensor.TextureDataTypes = 'float32'>(
|
|
texture: Tensor.TextureType, options: TensorFromTextureOptions<T>): TypedTensor<'float32'>;
|
|
|
|
/**
|
|
* create a tensor from a WebGPU buffer
|
|
*
|
|
* @param buffer - the GPUBuffer object to create tensor from
|
|
* @param options - An optional object representing options for creating tensor from WebGPU buffer.
|
|
*
|
|
* The options include following properties:
|
|
* - `dataType`: the data type of the tensor. If omitted, assume 'float32'.
|
|
* - `dims`: the dimension of the tensor. Required.
|
|
* - `download`: an optional function to download the tensor data from GPU to CPU. If omitted, the GPU data
|
|
* will not be able to download. Usually, this is provided by a GPU backend for the inference outputs. Users don't
|
|
* need to provide this function.
|
|
* - `dispose`: an optional function to dispose the tensor data on GPU. If omitted, the GPU data will not be disposed.
|
|
* Usually, this is provided by a GPU backend for the inference outputs. Users don't need to provide this function.
|
|
*
|
|
* @returns a tensor object
|
|
*/
|
|
fromGpuBuffer<T extends Tensor.GpuBufferDataTypes>(
|
|
buffer: Tensor.GpuBufferType, options: TensorFromGpuBufferOptions<T>): TypedTensor<T>;
|
|
|
|
/**
|
|
* create a tensor from a pre-allocated buffer. The buffer will be used as a pinned buffer.
|
|
*
|
|
* @param type - the tensor element type.
|
|
* @param buffer - a TypedArray corresponding to the type.
|
|
* @param dims - specify the dimension of the tensor. If omitted, a 1-D tensor is assumed.
|
|
*
|
|
* @returns a tensor object
|
|
*/
|
|
fromPinnedBuffer<T extends Exclude<Tensor.Type, 'string'>>(
|
|
type: T, buffer: Tensor.DataTypeMap[T], dims?: readonly number[]): TypedTensor<T>;
|
|
}
|