Repository navigation
New Text Classification Model on ARM Fails to Train #7488
Description
System Information (please complete the following information):
- OS & Version: Windows 11 for ARM on a Surface Laptop 7
- ML.NET Version: ML.NET Model Builder 2022 v17.19.2.2511501
- .NET Version: v9
Describe the bug
When creating a new ML.NET Text Classification Model in Visual Studio. The wizard errors when trying to train the model.
To Reproduce
Steps to reproduce the behavior:
- File -> New Model on a blank .NET 9 project.
- Select Text Classification
- Enter a CSV file.
- Click on train
Expected behavior
I should be able to train the model. ARM is supposed to work.
Screenshots, Code, Sample Projects
Additional context
Full logs below:
MLModel1-67B6PD.txt
Activity
RehanSaeed commentedon Aug 26, 2025
Just tested on latest VS updates and still seems totally broken. I've seen a blog post about this working, is this supported?
ericstj commentedon Oct 21, 2025
@JakeRadMSFT @LittleLittleCloud -- is it possible that modelbuilder is deploying the wrong binaries in this case?
I do think a standalone app should be able to work here.
LittleLittleCloud commentedon Oct 22, 2025
Unfortunately text classification uses torchsharp trainer behind and Torchsharp doesn't support arm device yet. So it's currently expected that the text classification training doesn't work on arm device.
Looks like there's ongoing PR to support arm device, we can probably support arm64 device after that PR get merged
dotnet/TorchSharp#1493
matouskozak commentedon Sep 22, 2026
Triage: Windows arm64 support was added to torchsharp in dotnet/TorchSharp#1549, we need to investigate if it fixed the reported issue.
rosebyte commentedon Nov 4, 2026
Investigation showed the TorchSharp part is fine. Native loading, gradients, 200 CPU training steps, and model save/load passed. However, we need to fix the ML.NET part which failed while loading pretrained NAS-BERT weights, before training started.