add documentation for updating winmlrunner (#371)
This commit is contained in:
Родитель
7727c51fe9
Коммит
7b8d947c2e
|
@ -1,6 +1,6 @@
|
|||
# Updating WinMLDashboard
|
||||
|
||||
WinMLDashboard should be updated in cadence with updates to WinML and OnnxRuntime.
|
||||
WinMLDashboard should be updated in cadence with updates to Windows Machine Learning and ONNX Runtime.
|
||||
|
||||
## Steps to update WinMLDashboard:
|
||||
1) a. Update the following dependencies in requirements.txt to their latest stable versions:
|
||||
|
@ -23,7 +23,7 @@ WinMLDashboard should be updated in cadence with updates to WinML and OnnxRuntim
|
|||
|
||||
2) Update MicrosoftMLRunner binary to the latest release. Using MicrosoftMLRunner rather than WinMLRunner allows us to utilize opset support from the latest WinML/Onnxruntime nuget release. The latest MicrosoftMLRunner release is available [here](https://github.com/microsoft/Windows-Machine-Learning/releases).
|
||||
3) Update that the latest WinML, DirectML, and Onnxruntime binaries to their latest versions. Microsoft.AI.MachineLearning.dll and onnxruntime.dll are available in this [package](https://www.nuget.org/packages/Microsoft.AI.MachineLearning/), and DirectML.dll is available in this [package](https://www.nuget.org/packages/Microsoft.AI.DirectML/). These binaries are also present in the MicrosoftMLRunner package and can be used from there.
|
||||
4) Update the opset options in the UI to the latest supported by WinML. This is done by adding to the ONNXVersionOptions object in [View.tsx](src/view/convert/View.tsx). WinML's latest supported opset information is available [here](https://github.com/microsoft/onnxruntime/blob/master/docs/Versioning.md).
|
||||
4) Update the opset options in the UI to the latest supported by Windows Machine Learning. This is done by adding to the ONNXVersionOptions object in [View.tsx](src/view/convert/View.tsx). The latest supported opset information for Windows Machine Learning is available [here](https://github.com/microsoft/onnxruntime/blob/master/docs/Versioning.md).
|
||||
5) Verify that the CI automated pipeline has succeeded when opening a pull request. This pipeline builds WinMLDashboard and runs basic UI tests.
|
||||
5) Manually ensure that conversion from each framework succeeds.Test collateral for each framework can be found in test/convert_collateral. Make sure to recreate the local python environment after updating dependency versions. For the tensorflow output names specify: MobilenetV1/Predictions/Reshape_1:0
|
||||
6) Issue a new official release of WinMLDashboard on the release page of this repository.
|
||||
6) Issue a new official release of WinMLDashboard on the repository release page [here](https://github.com/microsoft/Windows-Machine-Learning/releases).
|
|
@ -0,0 +1,12 @@
|
|||
# Updating WinMLRunner
|
||||
|
||||
WinMLRunner should be updated in cadence with updates to Windows Machine Learning and ONNX Runtime.
|
||||
|
||||
## Steps to update WinMLRunner:
|
||||
1) Update the Microsoft.AI.MachineLearning package in packages.config to its latest release.
|
||||
2) Ensure all relevant vcxproj files reference the updated package path. These files are:
|
||||
a) WinMLRunner.vcxproj
|
||||
b) WinMLRunnerScenarios.vcxproj
|
||||
c) WinMLRunnerStaticLib.vcxproj
|
||||
3) Ensure WinMLRunner builds and perform some basic selfhosting, like running any of the models in the SharedContent folder.
|
||||
4) Issue a new official release of WinMLRunner on the repository release page [here](https://github.com/microsoft/Windows-Machine-Learning/releases).
|
Загрузка…
Ссылка в новой задаче