Implementation Status of WebNN Operations
|
WebNN Spec |
Web Platform Tests |
Chromium Implementation | |||||
|---|---|---|---|---|---|---|---|
| LiteRT 1 | DirectML 2 | Core ML 3 | |||||
Windows
ChromeOS
Android
Linux
|
Windows
|
macOS
|
|||||
| Operations |
 4
|
Operations |
5
|
Operations |
6
|
||
LiteRT
DirectML
Core ML
✅ Supported ()
⏳ Partly Implemented ()
🚀 Work in Progress ()
❌ Not Supported
✅ Supported ()
⏳ Partly Implemented ()
🚀 Work in Progress ()
❌ Not Supported
✅ Supported ()
⏳ Partly Implemented ()
🚀 Work in Progress ()
❌ Not Supported
JavaScript ML Frameworks Integration Status
|
WebNN Spec |
JavaScript ML Frameworks Integration | |||
|---|---|---|---|---|
 Execution Provider 7 |
External Delegate 8 |
|||
| Operations | EP Version | Operations | Delegate Version | |
ONNX Runtime Web
Execution Provider
Execution Provider
LiteRT
External Delegate
External Delegate
✅ Supported ()
⏳ Partly Implemented ()
🚀 Work in Progress ()
❌ Not Supported
✅ Supported ()
⏳ Partly Implemented ()
🚀 Work in Progress ()
❌ Not Supported
The total number of WebNN ops is 78. These tables currently lists ops that are implemented or work in progress by multiple backends and JavaScript machine learning frameworks.
[1] LiteRT Builtin Options
[2] DirectML API
[3] Core ML operators
[4] This feature is experimental. Can be enabled by setting #web-machine-learning-neural-network flag to Enabled.
[5] This feature is experimental. Can be enabled by setting #web-machine-learning-neural-network flag to Enabled. Supported on GPUs on Windows 11 21H2 or higher.
[6] This feature is experimental. Can be enabled by setting #web-machine-learning-neural-network flag to Enabled.
[7] ONNX Operator Schemas and WebNN EP Helper
[8] LiteRT built-in operators kTfLiteBuiltin*
🏅🏅🏅 Interested in contributing to this implementation status page? See contributing guidelines.