Post not yet marked as solved
Hello, could one of you please file a bug report on http://feedbackassistant.apple.com/ along with a sysdiagnoses from the device?
Post not yet marked as solved
Hello, could you please submit a bug report on http://feedbackassistant.apple.com/ along with the version of the OS you are observing this on?
Post not yet marked as solved
This was an issue on the server side for key generation logic. It should be resolved now. It should work with any Xcode 13.0 on any macOS Monterey. Please file a bug report on https://feedbackassistant.apple.com if you are still facing the issue. Sorry about the inconvenience.
Post not yet marked as solved
However, I was surprized that also just one model (left or right) detected both type of feet: left and right.
I am not sure I completely understand this. But, I would expect a trained image classifier CreateML model that is trained with left and right foot images tagged as "left" and "right" to predict left and right foots correctly. The model should be robust enough to do that.
Post not yet marked as solved
I would like to understand what you are trying to accomplish when you say "Save a trained MetalPerformanceShadersGraph model in any other way for deployment;" so I can assist you better. Are you looking to decouple code from model weights? And, deploy weights out-of-band?
Post not yet marked as solved
Hello! Default model configuration (MLModelConfiguration()) tries to run the model on CPU, GPU and NeuralEngine. NeuralEngine is not supported in the Simulator, which could explain the issue. Could you please file a bug report with the model and some sample code on http://feedbackassistant.apple.com/
Post not yet marked as solved
Hi, unfortunately, this use case is not supported. Model has to be encrypted with a key that was generated with the same development team as the one that signed the app that is running on the device. Please file a feature request on http://feedbackassistant.apple.com/ for this and we can look into this.
Post not yet marked as solved
Hi! CoreML does not support inference with MTLTexture natively. You could convert it to a CVPixelBuffer and then feed that into CoreML for inference.
https://developer.apple.com/documentation/coreml/mlfeaturevalue/2879346-init
I'd also encourage you to file a feature request on http://feedbackassistant.apple.com/ so, we could prioritize supporting MTLTexture natively.
Post not yet marked as solved
Hi @dshakbhazyan, this is not expected. Do you have any additional logs? Could you please capture a sysdiagnose and attach it to an issue on http://feedbackassistant.apple.com/? We will take a closer look and get back to you.
https://download.developer.apple.com/iOS/iOS_Logs/sysdiagnose_Logging_Instructions.pdf
Post not yet marked as solved
We root-caused this to a mismatch in SDKs in Xcode 12.0 GM. The official note on https://developer.apple.com/download/release/:
"Xcode 12 includes everything you need to create amazing apps for all Apple platforms. Includes the latest SDKs for iOS, watchOS, and tvOS, and macOS Catalina."
CoreML compiler in Xcode 12.0 GM is generating code that has symbols available only on macOS BigSur causing the compilation issue. This was fixed in Xcode 12.2 and above.
Hi @mylesDoyle, CoreML does allocate memory within predictionFromFeatures but, they are released when the autorelease pool pops. To explicitly trigger the release, you could wrap the prediction call inside @autoreleasepool { }. Let us know if you run into issues after that.
Post not yet marked as solved
Hi, what is the OS version on your iPhone 12? Do you mind sharing a simple repro case along with the model?
Thank you for the report. We confirmed it was a problem in the model access code generator. Until we address the issue in an upcoming Xcode, a workaround is:
Open the current model access code by selecting the .mlmodel file and clicking “Model Class” and exporting it to keep.
In “Target Membership”, select “No Generated Classes”.
Add the saved code-gen Swift file in step 1 back to the project.
Remove lazy properties with MLShapedArray . If you need these properties, you can change to a computed property.
Post not yet marked as solved
Hi, what is the difference between the demo app and the main app in this case? Could you please check if the main app is code signed?
Post not yet marked as solved
Big Sur version 11.3.1 with Xcode 12.5 (12E262), no success.
Could you please once more try after activating iCloud in System Preferences Apple ID? There are no known issues on Version 12.5 (12E262) and I confirmed today that I can generate encryption keys and use that to encrypt my CoreML models.
