How to build and consume WebGPU ( 2023-08-19 )

how-to-build-onnxruntime-web-webgpu.md

How to build

Prerequisites

• Checkout branch main from https://github.com/microsoft/onnxruntime to C:\code\onnxruntime
• Install Ninja: pip install ninja or download from github (make sure it's in PATH)

Build using the build_jsep.bat

Use the file js\build_jsep.bat. see its content for usage instruction

C:\code\onnxruntime>js\build_jsep d st

If it's the first time to build, perform a clean build

C:\code\onnxruntime>js\build_jsep d st clean

Build manually

1. Use the following command line to build wasm:

   C:\code\onnxruntime>.\build.bat --config Release --build_wasm --enable_wasm_simd --use_jsep --target onnxruntime_webassembly --skip_tests

2. Prepare js/web:

   C:\code\onnxruntime>cd js
   C:\code\onnxruntime\js>npm ci
   C:\code\onnxruntime\js>cd common
   C:\code\onnxruntime\js\common>npm ci
   C:\code\onnxruntime\js\common>cd ..\web
   C:\code\onnxruntime\js\web>npm ci
   C:\code\onnxruntime\js\web>npm run pull:wasm
   C:\code\onnxruntime\js\web>copy /Y ..\..\build\Windows\Release\ort-wasm-simd.js .\lib\wasm\binding\ort-wasm-simd.jsep.js
   C:\code\onnxruntime\js\web>copy /Y ..\..\build\Windows\Release\ort-wasm-simd.wasm .\dist\ort-wasm-simd.jsep.wasm
   

To run npm tests, do:

C:\code\onnxruntime\js\web>npm test -- model <path-to-ort-model-test-folder> -b=webgpu --wasm-number-threads=1 --debug

To build web artifacts:

C:\code\onnxruntime\js\web>npm run build

Following files are built and needed for distribution:

• ort.webgpu.min.js
• ort.webgpu.min.js.map ( optional, for debug purpose )
• ort-wasm-simd.jsep.wasm
• ort-wasm-simd-threaded.jsep.wasm

To run E2E, build a NPM package and consume:

C:\code\onnxruntime\js\common>npm pack
C:\code\onnxruntime\js\web>npm pack

2 Package files are needed for npm install. Please install them together:

• js/common/onnxruntime-common-*.tgz
• js/web/onnxruntime-web-*.tgz

How to consume

Make sure the files mentioned above are put in dist folder.

• Consume from <script> tag:

  <script src="./dist/ort.webgpu.min.js"></script>

• Consume from package:

  import {Tensor, InferenceSession} from "onnxruntime-web/webgpu";