I have a model that uses ‘flatten’, and when I converted it to a Core ML model and ran it on Xcode with an iPhone XR, I noticed that ‘flatten’ was automatically converted to ‘reshape’. However, the NPU does not support ‘reshape’. howerver, I got the Resnet50 model on apple models and performance it on XCode with the same iphone XR, I can see the 'flatten' operator which run on NPU. On the other hand, when I used the following code to convert ResNet50 in PyTorch and ran it on Xcode Performance, the ‘flatten’ operation was converted to ‘reshape’, which then ran on the CPU. ? So I dont know how to keep 'flatten' operator when convert to ml model ? coreml tool 7.1 iphone XR ios 17.5.1 from torchvision import models import coremltools as ct import torch import torch.nn as nn network_name = "my_resnet50" torch_model = models.resnet50(pretrained=True) torch_model.eval() width = 224 height = 224 example_input = torch.rand(1, 3, height, width) traced_model = torch.jit.trace(torch_model, (example_input)) model = ct.convert( traced_model, convert_to = "neuralnetwork", inputs=[ ct.TensorType( name = "data", shape = example_input.shape, dtype = np.float32 ) ], outputs = [ ct.TensorType( name = "output", dtype = np.float32 ) ], compute_units = ct.ComputeUnit.CPU_AND_NE, minimum_deployment_target = ct.target.iOS14, ) model.save("my_resnet.mlmodel") ResNet50 on Resnet50.mlmodel My Convertion of ResNet50

Error when trying to generate CoreML performance report, message says The data couldn't be written because it isn't in the correct format. Here is the code to replicate the issue import numpy as np import coremltools as ct from coremltools.converters.mil import Builder as mb import coremltools.converters.mil as mil w = np.random.normal(size=(256, 128, 1)) wemb = np.random.normal(size=(1, 32000, 128)) # .astype(np.float16) rope_emb = np.random.normal(size=(1, 2048, 128)) shapes = [(1, seqlen) for seqlen in (32, 64)] enum_shape = mil.input_types.EnumeratedShapes(shapes=shapes) fixed_shape = (1, 128) max_length = 2048 dtype = np.float32 @mb.program( input_specs=[ # mb.TensorSpec(enum_shape.symbolic_shape, dtype=mil.input_types.types.int32), mb.TensorSpec(enum_shape.symbolic_shape, dtype=mil.input_types.types.int32), ], opset_version=mil.builder.AvailableTarget.iOS17, ) def flex_like(input_ids): indices = mb.fill_like(ref_tensor=input_ids, value=np.array(1, dtype=np.int32)) causal_mask = np.expand_dims( np.triu(np.full((max_length, max_length), -np.inf, dtype=dtype), 1), axis=0, ) mask = mb.gather( x=causal_mask, indices=indices, axis=2, batch_dims=1, name="mask_gather_0", ) # mask = mb.gather( # x=mask, indices=indices, axis=1, batch_dims=1, name="mask_gather_1" # ) rope = mb.gather(x=rope_emb.astype(dtype), indices=indices, axis=1, batch_dims=1, name="rope") hidden_states = mb.gather(x=wemb.astype(dtype), indices=input_ids, axis=1, batch_dims=1, name="embedding") return ( hidden_states, mask, rope, ) cml_flex_like = ct.convert( flex_like, compute_units=ct.ComputeUnit.ALL, compute_precision=ct.precision.FLOAT32, minimum_deployment_target=ct.target.iOS17, inputs=[ ct.TensorType(name="input_ids", shape=enum_shape), ], ) cml_flex_like.save("flex_like_32") If I remove the hidden states from the return it does work, and it also works if I keep the hidden states, but remove both mask, and rope, i.e, the report is generated for both programs with either these returns: return ( # hidden_states, mask, rope, ) and return ( hidden_states, # mask, # rope, ) It also works if I use a static shape instead of an EnumeratedShape I'm using macOS 15.0 and Xcode 16.0 Edit 1: Forgot to mention that although the performance report fails, the model is still able to make predictions

Hi everyone, I was wondering, on how accurate is the Hand Classification ML? For Example: Is it possible to understand the different letters of the Sign Language Alphabet or is it only capable of recognizing simple poses like a thumbs up?

Hi everyone, I'm working with Core ML for object tracking and have successfully trained a couple of models. However, when I try to use the reference object file in the object tracking sample code from the WWDC video, it doesn't work. I'm training the model on a single-color plastic spherical object. Could this be the reason for the issue? I also attempted to use USDZ 3D assets that resemble the real object. Do these need to be trained with the Object Capture app to work properly? Speaking of the Object Capture app, my experience hasn't been great. When I scanned my spherical object, the result was far from a sphere—it looked more like a mushy dough. Any insights or suggestions would be greatly appreciated!

Hi there, I am trying to create a Message Filter app that uses a trained Text Classification to predict scam texts (as it is common in my country and is constantly evolving). However, when I try to use the MLModel in the MessageFilterExtension class, I'm getting initialization of text classifier model with model data failed Here's how I initialize my MLModel that is created using Create ML. do { let model = try MyModel(configuration: .init()) let output = try model.prediction(text: text) guard !output.label.isEmpty else { return nil } return MessagePrediction(rawValue: output.label) } catch { return nil } Is it impossible to use CoreML in Message Filter extensions? Thank you

I have created a Hand Action Classification project following the guidelines which causes the Create ML tool to provide the very cryptic "Unexpected Error". The feature extraction phase is fine, with the error occurring during model training after the tool reports completion of the first five training iterations as it moves on to report the next ten. The project is small with 3 training classes and 346 items I have tried to vary the frame rate and action duration, with all augmentations unset, but the error still persists. Can you please confirm how I may get further error diagnostic information so that I can determine why Create ML is unable to work with my training data? Mac OS is Sonoma 14.5 on an iMac 24-inch, M1, 2021. Create ML is Version 5.0 (121.4)

I have a couple of models that I want to migrate to .mlpackage but can not find the resources of the session: https://developer.apple.com/videos/play/wwdc2024/10159/ In the video at 21:10 talk about modifications and optimizations, but in the video can not even see the dependencies of the demo. Thanks

Deploy machine learning and AI models on-device with Core ML say the performance report can see the ops run on which unit and why it cannot run on Neural Engine. I tested my model and the report shows a gray checkmark at the Neural Engine, indicating it can run on the Neural Engine. However, it's not executing on the Neural Engine but on the CPU. Why is this happening?

I'm looking for a solution to take a picture or point the camera at a piece of clothing and match that image with an image the user has stored in my app. I'm storing the data in a Core Data database as a Binary Data object. Since the user also takes the pictures they store in the database I think I cannot use pre-trained Core ML models. I would like the matching to be done on device if possible instead of going to an external service. That will probably describe the item based on what the AI sees, but then I cannot match the item with the stored images in the app. Does anyone know if this is possible with frameworks as Vision or VisionKit?

I want to try an any resolution image input Core ML model. So I wrote the model following the Core ML Tools "Set the Range for Each Dimensionas" sample code, modified as below: # Trace the model with random input. example_input = torch.rand(1, 3, 50, 50) traced_model = torch.jit.trace(model.eval(), example_input) # Set the input_shape to use RangeDim for each dimension. input_shape = ct.Shape(shape=(1, 3, ct.RangeDim(lower_bound=25, upper_bound=1920, default=45), ct.RangeDim(lower_bound=25, upper_bound=1920, default=45))) scale = 1/(0.226*255.0) bias = [- 0.485/(0.229) , - 0.456/(0.224), - 0.406/(0.225)] # Convert the model with input_shape. mlmodel = ct.convert(traced_model, inputs=[ct.ImageType(shape=input_shape, name="input", scale=scale, bias=bias)], outputs=[ct.TensorType(name="output")], convert_to="mlprogram", ) # Save the Core ML model mlmodel.save("image_resize_model.mlpackage") It converts OK but when I predict the result with an image It will get the error as below: You will not be able to run predict() on this Core ML model. Underlying exception message was: { NSLocalizedDescription = "Failed to build the model execution plan using a model architecture file '/private/var/folders/8z/vtz02xrj781dxvz1v750skz40000gp/T/model-small.mlmodelc/model.mil' with error code: -7."; } Where did I do wrong?

I was trying the latest coremltools-8.0b1 beta on macOS 15 Beta with the intent to try using the new stateful models api in CoreML. But the conversion would always fail with the error: /AppleInternal/Library/BuildRoots/<snip>/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShadersGraph/mpsgraph/MetalPerformanceShadersGraph/Core/Files/MPSGraphExecutable.mm:162: failed assertion `Error: the minimum deployment target for macOS is 14.0.0' Here's a minimal repro, which works fine with both the stable version of coremltools (7.2) and the beta version (8.0b1) on macOS Sonoma 14.5, but fails with both versions of coremltools on macOS 15.0 Beta and Xcode 16.0 Beta. Which means that this most likely isn't an issue with coremltools, but with the native compilation toolchain. from collections import OrderedDict import coremltools as ct import numpy as np import torch import torch.nn as nn class ResidualAttentionBlock(nn.Module): def __init__(self, d_model: int, n_head: int, attn_mask: torch.Tensor = None): super().__init__() self.attn = nn.MultiheadAttention(d_model, n_head) self.ln_1 = nn.LayerNorm(d_model) self.mlp = nn.Sequential( OrderedDict( [ ("c_fc", nn.Linear(d_model, d_model * 4)), ("gelu", nn.GELU()), ("c_proj", nn.Linear(d_model * 4, d_model)), ] ) ) self.ln_2 = nn.LayerNorm(d_model) self.attn_mask = attn_mask def attention(self, x: torch.Tensor): self.attn_mask = ( self.attn_mask.to(dtype=x.dtype, device=x.device) if self.attn_mask is not None else None ) return self.attn(x, x, x, need_weights=False, attn_mask=self.attn_mask)[0] def forward(self, x: torch.Tensor): x = x + self.attention(self.ln_1(x)) x = x + self.mlp(self.ln_2(x)) return x class Transformer(nn.Module): def __init__( self, width: int, layers: int, heads: int, attn_mask: torch.Tensor = None ): super().__init__() self.width = width self.layers = layers self.resblocks = nn.Sequential( *[ResidualAttentionBlock(width, heads, attn_mask) for _ in range(layers)] ) def forward(self, x: torch.Tensor): return self.resblocks(x) transformer = Transformer(width=512, layers=12, heads=8) emb_tokens = torch.rand((1, 512)) ct_model = ct.convert( torch.jit.trace(transformer.eval(), emb_tokens), convert_to="mlprogram", minimum_deployment_target=ct.target.macOS14, inputs=[ct.TensorType(name="embIn", shape=[1, 512])], outputs=[ct.TensorType(name="embOutput", dtype=np.float32)], )

I have several CoreML models that I've set up to run in sequence where one of the outputs from each model is passed as one of the inputs to the next. For the most part, there is very little overhead in between each sub-model "chunk": However a couple of the models (eg the first two above) spend a noticeable amount of time in "Prepare Neural Engine Request". From Instruments, it seems like this is spent doing some sort of model loading. Given that I'm calling these models in sequence and in a fixed order, is there some way to reduce or amortize this cost? Thanks!

I have made a text classifier model but I want to train it on device too. When text is classified wrong, user can make update the model on device. Code : // // SpamClassifierHelper.swift // LearningML // // Created by Himan Dhawan on 7/1/24. // import Foundation import CreateMLComponents import CoreML import NaturalLanguage enum TextClassifier : String { case spam = "spam" case notASpam = "ham" } class SpamClassifierModel { // MARK: - Private Type Properties /// The updated Spam Classifier model. private static var updatedSpamClassifier: SpamClassifier? /// The default Spam Classifier model. private static var defaultSpamClassifier: SpamClassifier { do { return try SpamClassifier(configuration: .init()) } catch { fatalError("Couldn't load SpamClassifier due to: \(error.localizedDescription)") } } // The Spam Classifier model currently in use. static var liveModel: SpamClassifier { updatedSpamClassifier ?? defaultSpamClassifier } /// The location of the app's Application Support directory for the user. private static let appDirectory = FileManager.default.urls(for: .applicationSupportDirectory, in: .userDomainMask).first! class var urlOfModelInThisBundle : URL { let bundle = Bundle(for: self) return bundle.url(forResource: "SpamClassifier", withExtension:"mlmodelc")! } /// The default Spam Classifier model's file URL. private static let defaultModelURL = urlOfModelInThisBundle /// The permanent location of the updated Spam Classifier model. private static var updatedModelURL = appDirectory.appendingPathComponent("personalized.mlmodelc") /// The temporary location of the updated Spam Classifier model. private static var tempUpdatedModelURL = appDirectory.appendingPathComponent("personalized_tmp.mlmodelc") // MARK: - Public Type Methods static func predictLabelFor(_ value: String) throws -> (predication :String?, confidence : String) { let spam = try NLModel(mlModel: liveModel.model) let result = spam.predictedLabel(for: value) let confidence = spam.predictedLabelHypotheses(for: value, maximumCount: 1).first?.value ?? 0 return (result,String(format: "%.2f", confidence * 100)) } static func updateModel(newEntryText : String, spam : TextClassifier) throws { guard let modelURL = Bundle.main.url(forResource: "SpamClassifier", withExtension: "mlmodelc") else { fatalError("Could not find model in bundle") } // Create feature provider for the new image let featureProvider = try MLDictionaryFeatureProvider(dictionary: ["label": MLFeatureValue(string: newEntryText), "text": MLFeatureValue(string: spam.rawValue)]) let batchProvider = MLArrayBatchProvider(array: [featureProvider]) let updateTask = try MLUpdateTask(forModelAt: modelURL, trainingData: batchProvider, configuration: nil, completionHandler: { context in let updatedModel = context.model let fileManager = FileManager.default do { // Create a directory for the updated model. try fileManager.createDirectory(at: tempUpdatedModelURL, withIntermediateDirectories: true, attributes: nil) // Save the updated model to temporary filename. try updatedModel.write(to: tempUpdatedModelURL) // Replace any previously updated model with this one. _ = try fileManager.replaceItemAt(updatedModelURL, withItemAt: tempUpdatedModelURL) loadUpdatedModel() print("Updated model saved to:\n\t\(updatedModelURL)") } catch let error { print("Could not save updated model to the file system: \(error)") return } }) updateTask.resume() } /// Loads the updated Spam Classifier, if available. /// - Tag: LoadUpdatedModel private static func loadUpdatedModel() { guard FileManager.default.fileExists(atPath: updatedModelURL.path) else { // The updated model is not present at its designated path. return } // Create an instance of the updated model. guard let model = try? SpamClassifier(contentsOf: updatedModelURL) else { return } // Use this updated model to make predictions in the future. updatedSpamClassifier = model } }

"On the latest iOS 18 beta 2, the OCR API，the Translate App and Live Text performs very poorly in recognizing Japanese."

Hi everyone, I attempted to use the MultivariateLinearRegressor from the Create ML Components framework to fit some multi-dimensional data linearly (4 dimensions in my example). I aim to obtain multi-dimensional output points (2 points in my example). However, when I fit the model with my training data and test it, it appears that only the first element of my training data is used for training, regardless of whether I use CreateMLComponents.AnnotatedBatch or [CreateMLComponents.AnnotatedFeature, CoreML.MLShapedArray>] as input. let sourceMatrix: [[Double]] = [ [0,0.1,0.2,0.3], [0.5,0.2,0.6,0.2] ] let referenceMatrix: [[Double]] = [ [0.2,0.7], [0.9,0.1] ] Here is a test code to test the function (ios 18.0 beta, Xcode 16.0 beta) In this example I train the model to learn 2 multidimensional points (4 dimensions) and here are the results of the predictions: ▿ 2 elements ▿ 0 : AnnotatedPrediction<MLShapedArray<Double>, MLShapedArray<Double>> ▿ prediction : 0.20000000298023224 0.699999988079071 ▿ _storage : <StandardStorage<Double>: 0x600002ad8270> ▿ annotation : 0.2 0.7 ▿ _storage : <StandardStorage<Double>: 0x600002b30600> ▿ 1 : AnnotatedPrediction<MLShapedArray<Double>, MLShapedArray<Double>> ▿ prediction : 0.23158159852027893 0.9509953260421753 ▿ _storage : <StandardStorage<Double>: 0x600002ad8c90> ▿ annotation : 0.9 0.1 ▿ _storage : <StandardStorage<Double>: 0x600002b55f20> 0.23158159852027893 0.9509953260421753 is totally random and should be far more closer to [0.9,0.1]. Here is the test code : ( i run it on "My mac, Designed for Ipad") ContentView.swift import CoreImage import CoreImage.CIFilterBuiltins import UIKit import CoreGraphics import Accelerate import Foundation import CoreML import CreateML import CreateMLComponents func createMLShapedArray(from array: [Double], shape: [Int]) -> MLShapedArray<Double> { return MLShapedArray<Double>(scalars: array, shape: shape) } func calculateTransformationMatrixWithNonlinearity(sourceRGB: [[Double]], referenceRGB: [[Double]], degree: Int = 3) async throws -> MultivariateLinearRegressor<Double>.Model { let annotatedFeatures2 = zip(sourceRGB, referenceRGB).map { (featureArray, targetArray) -> AnnotatedFeature<MLShapedArray<Double>, MLShapedArray<Double>> in let featureMLShapedArray = createMLShapedArray(from: featureArray, shape: [featureArray.count]) let targetMLShapedArray = createMLShapedArray(from: targetArray, shape: [targetArray.count]) return AnnotatedFeature(feature: featureMLShapedArray, annotation: targetMLShapedArray) } // Flatten the sourceRGBPoly into a single-dimensional array var flattenedArray = sourceRGB.flatMap { $0 } let featuresMLShapedArray = createMLShapedArray(from: flattenedArray, shape: [2, 4]) flattenedArray = referenceRGB.flatMap { $0 } let targetMLShapedArray = createMLShapedArray(from: flattenedArray, shape: [2, 2]) // Create AnnotatedFeature instances /* let annotatedFeatures2: [AnnotatedFeature<MLShapedArray<Double>, MLShapedArray<Double>>] = [ AnnotatedFeature(feature: featuresMLShapedArray, annotation: targetMLShapedArray) ]*/ let annotatedBatch = AnnotatedBatch(features: featuresMLShapedArray, annotations: targetMLShapedArray) var regressor = MultivariateLinearRegressor<Double>() regressor.configuration.learningRate = 0.1 regressor.configuration.maximumIterationCount=5000 regressor.configuration.batchSize=2 let model = try await regressor.fitted(to: annotatedBatch,validateOn: nil) //var model = try await regressor.fitted(to: annotatedFeatures2) // Proceed to prediction once the model is fitted let predictions = try await model.prediction(from: annotatedFeatures2) // Process or use the predictions print(predictions) print("Predictions:", predictions) return model } struct ContentView: View { var body: some View { VStack {} .onAppear { Task { do { let sourceMatrix: [[Double]] = [ [0,0.1,0.2,0.3], [0.5,0.2,0.6,0.2] ] let referenceMatrix: [[Double]] = [ [0.2,0.7], [0.9,0.1] ] let model = try await calculateTransformationMatrixWithNonlinearity(sourceRGB: sourceMatrix, referenceRGB: referenceMatrix, degree: 2 ) print("Model fitted successfully:", model) } catch { print("Error:", error) } } } } }

I‘ve created text classification project and selected BERT algorithm With 100 iterations for json file. Json file is valid but training always cancels on 37 iteration… Because tool does not provide any cancellation reasons I have no clue why it happens. Can I check reasons somehow? Or do anyone knows possible reasons or solutions for this?

Hi, the following model does not run on ANE. Inspecting with deCoreML I see the error ane: Failed to retrieved zero_point. import numpy as np import coremltools as ct from coremltools.converters.mil import Builder as mb import coremltools.converters.mil as mil B, CIN, COUT = 512, 1024, 1024 * 4 @mb.program( input_specs=[ mb.TensorSpec((B, CIN), mil.input_types.types.fp16), ], opset_version=mil.builder.AvailableTarget.iOS18 ) def prog_manual_dequant( x, ): qw = np.random.randint(0, 2 ** 4, size=(COUT, CIN), dtype=np.int8).astype(mil.mil.types.np_uint4_dtype) scale = np.random.randn(COUT, 1).astype(np.float16) offset = np.random.randn(COUT, 1).astype(np.float16) # offset = np.random.randint(0, 2 ** 4, size=(COUT, 1), dtype=np.uint8).astype(mil.mil.types.np_uint4_dtype) dqw = mb.constexpr_blockwise_shift_scale(data=qw, scale=scale, offset=offset) return mb.linear(x=x, weight=dqw) cml_qmodel = ct.convert( prog_manual_dequant, compute_units=ct.ComputeUnit.CPU_AND_NE, compute_precision=ct.precision.FLOAT16, minimum_deployment_target=ct.target.iOS18, ) Whereas if I use an offset with the same dtype as the weights (uint4 in this case), it does run on ANE Tested on coremltools 8.0b1, on macOS 15.0 beta 2/Xcode 15 beta 2, and macOS 15.0 beta 3/Xcode 15 beta 3.

I have created and trained a Hand Pose classifier model and am trying to test it. I have noticed in the WWDC2021 "Classify hand poses and actions with Create ML" the preview windows has a prediction result that gives you the prediction based on the live preview or the images. Mine does not have that. When i try to import pictures or do the live test there is no result. Its just the wireframe view and under it there is nothing. How do I fix this please? Thanks.