WebNN Flags in Chromium based Browsers
Flags
Visit via about://flags/
#web-machine-learning-neural-network
| Name | Enables WebNN API |
|---|---|
| Description | Enables the Web Machine Learning Neural Network (WebNN) API. Spec at https://www.w3.org/TR/webnn/  – Mac, Windows, Linux, ChromeOS, Android. |
| Expiry Milestone | 150 |
| Command Line Usage | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --enable-features=WebMachineLearningNeuralNetwork |
%LOCALAPPDATA%meansC:\Users\<username>\AppData\Local\
#experimental-web-machine-learning-neural-network
| Name | Enables experimental WebNN API features |
|---|---|
| Description | Enables additional, experimental features in Web Machine Learning Neural Network (WebNN) API. Requires the “WebNN API” flag to be enabled. – Mac, Windows, Linux, ChromeOS, Android. |
| Expiry Milestone | 150 |
| Command Line Usage | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --enable-features=EnablesExperimentalWebNNAPIFeatures |
#webnn-coreml
| Name | Core ML backend for WebNN |
|---|---|
| Description | Enables using Core ML for GPU and NPU inference with the WebNN API. Disabling this flag enables a fallback to TFLite. |
| Expiry Milestone | 146 |
#webnn-directml
| Name | DirectML backend for WebNN |
|---|---|
| Description | Enables using DirectML for GPU and NPU inference with the WebNN API. Disabling this flag enables a fallback to TFLite. – Windows. |
| Expiry Milestone | 146 |
Switches (Command Line Flags)
--use-redist-ort
| Name | Use redistributable ONNX Runtime |
|---|---|
| Description | Try to use a redistributable onnxruntime.dll. |
--webnn-use-ort
| Name | WebNN Use ONNX Runtime |
|---|---|
| Description | Introduces webnn_use_ort build flag and enable it for Windows. Replaces the DirectML backend with the ONNX Runtime backend. |
--webnn-ort-use-openvino
| Name | WebNN ONNX Runtime Use OpenVINO |
|---|---|
| Description | Enables the OpenVINO Execution Provider, OpenVINO EP will be |
| used for GPU and NPU devices. ONNX Runtime will run with the CPU EP by default. | |
| Usage (No Sandbox) | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --no-sandbox --webnn-use-ort --webnn-ort-use-openvino --use-redist-ort |
| Usage (Within Sandbox) | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --allow-third-party-modules --webnn-use-ort --webnn-ort-use-openvino |
--use-redist-dml
| Name | Use redistributable DirectML |
|---|---|
| Description | Try to use a redistributable DirectML.dll. Used for testing WebNN against newer DirectML release before it is integrated into Windows OS. Please see more info about DirectML releases at: https://learn.microsoft.com/en-us/windows/ai/directml/dml-version-history . |
| Usage | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --use-redist-dml |
--disable_dml_meta_commands_for_gpu
| Name | Disable DirectML Metacommands for GPU |
|---|---|
| Description | Run with DirectML fallback shaders if disabling Metacommands |
| Usage (true) | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --disable_dml_meta_commands_for_gpu=1 |
| Usage (false) | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --disable_dml_meta_commands_for_gpu=0 |
--disable_webnn_for_gpu
| Name | Disable WebNN for GPU |
|---|---|
| Description | If --disable_webnn_for_gpu workaround is enabled for the GPU device, we need to check to see if there is a NPU device available before setting the WebNN gpu feature status. If there is a NPU device, check the --disable_webnn_for_npu workaround. |
| Usage (true) | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --disable_webnn_for_gpu=1 |
--disable_webnn_for_npu
| Name | Disable WebNN for NPU |
|---|---|
| Description | |
| Usage (true) | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --disable_webnn_for_npu=1 |
| Usage (false) | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --disable_webnn_for_npu=0 |
--webnn-coreml-dump-model
| Name | WebNN Core ML Dump Model |
|---|---|
| Description | Copy the generated Core ML model to the folder specified by --webnn-coreml-dump-model. |
| Note | The folder needs to be accessible from the GPU sandbox or use --no-sandbox. |
| Usage | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --no-sandbox --webnn-coreml-dump-model=/tmp/CoreMLModels |
--webnn-tflite-dump-model
| Name | WebNN TFLite Dump Model |
|---|---|
| Description | Save the generated TFLite model file to the folder specified by --webnn-tflite-dump-model. |
| Note | The folder needs to be accessible from the GPU process sandbox or --no-sandbox must be used. |
| Usage | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --no-sandbox --webnn-tflite-dump-model=/tmp/tflite_models |
Other Switches
--no-sandbox
| Name | No Sandbox |
|---|---|
| Description | The --no-sandbox flag disables Chromium’s security sandbox, which isolates browser processes to prevent malicious code execution. While sometimes used in development/testing environments, using this flag in regular browsing is dangerous as it removes essential security protections. |
| Note | Running with --no-sandbox is an insecure configuration. After performing the test, you should quit Chrome and re-launch normally. |
| Usage | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --no-sandbox |
--allow-third-party-modules
| Name | Allow Third-party Modules |
|---|---|
| Description | The --allow-third-party-modules flag permits third-party software to inject DLLs or modules into Chrome’s processes. |
| Note | While useful for legitimate software like antivirus or accessibility tools, it can pose security risks by allowing potentially malicious code injection. |
| Usage | "%LOCALAPPDATA%\Google\Chrome SxS\Application\chrome.exe" --allow-third-party-modules |
Last updated on