At WWDC25 we launched a new type of Lab event for the developer community - Group Labs. A Group Lab is a panel Q&A designed for a large audience of developers. Group Labs are a unique opportunity for the community to submit questions directly to a panel of Apple engineers and designers. Here are the highlights from the WWDC25 Group Lab for Machine Learning and AI Frameworks. What are you most excited about in the Foundation Models framework? The Foundation Models framework provides access to an on-device Large Language Model (LLM), enabling entirely on-device processing for intelligent features. This allows you to build features such as personalized search suggestions and dynamic NPC generation in games. The combination of guided generation and streaming capabilities is particularly exciting for creating delightful animations and features with reliable output. The seamless integration with SwiftUI and the new design material Liquid Glass is also a major advantage. When should I still bring my own LLM via CoreML? It's generally recommended to first explore Apple's built-in system models and APIs, including the Foundation Models framework, as they are highly optimized for Apple devices and cover a wide range of use cases. However, Core ML is still valuable if you need more control or choice over the specific model being deployed, such as customizing existing system models or augmenting prompts. Core ML provides the tools to get these models on-device, but you are responsible for model distribution and updates. Should I migrate PyTorch code to MLX? MLX is an open-source, general-purpose machine learning framework designed for Apple Silicon from the ground up. It offers a familiar API, similar to PyTorch, and supports C, C++, Python, and Swift. MLX emphasizes unified memory, a key feature of Apple Silicon hardware, which can improve performance. It's recommended to try MLX and see if its programming model and features better suit your application's needs. MLX shines when working with state-of-the-art, larger models. Can I test Foundation Models in Xcode simulator or device? Yes, you can use the Xcode simulator to test Foundation Models use cases. However, your Mac must be running macOS Tahoe. You can test on a physical iPhone running iOS 18 by connecting it to your Mac and running Playgrounds or live previews directly on the device. Which on-device models will be supported? any open source models? The Foundation Models framework currently supports Apple's first-party models only. This allows for platform-wide optimizations, improving battery life and reducing latency. While Core ML can be used to integrate open-source models, it's generally recommended to first explore the built-in system models and APIs provided by Apple, including those in the Vision, Natural Language, and Speech frameworks, as they are highly optimized for Apple devices. For frontier models, MLX can run very large models. How often will the Foundational Model be updated? How do we test for stability when the model is updated? The Foundation Model will be updated in sync with operating system updates. You can test your app against new model versions during the beta period by downloading the beta OS and running your app. It is highly recommended to create an "eval set" of golden prompts and responses to evaluate the performance of your features as the model changes or as you tweak your prompts. Report any unsatisfactory or satisfactory cases using Feedback Assistant. Which on-device model/API can I use to extract text data from images such as: nutrition labels, ingredient lists, cashier receipts, etc? Thank you. The Vision framework offers the RecognizeDocumentRequest which is specifically designed for these use cases. It not only recognizes text in images but also provides the structure of the document, such as rows in a receipt or the layout of a nutrition label. It can also identify data like phone numbers, addresses, and prices. What is the context window for the model? What are max tokens in and max tokens out? The context window for the Foundation Model is 4,096 tokens. The split between input and output tokens is flexible. For example, if you input 4,000 tokens, you'll have 96 tokens remaining for the output. The API takes in text, converting it to tokens under the hood. When estimating token count, a good rule of thumb is 3-4 characters per token for languages like English, and 1 character per token for languages like Japanese or Chinese. Handle potential errors gracefully by asking for shorter prompts or starting a new session if the token limit is exceeded. Is there a rate limit for Foundation Models API that is limited by power or temperature condition on the iPhone? Yes, there are rate limits, particularly when your app is in the background. A budget is allocated for background app usage, but exceeding it will result in rate-limiting errors. In the foreground, there is no rate limit unless the device is under heavy load (e.g., camera open, game mode). The system dynamically balances performance, battery life, and thermal conditions, which can affect the token throughput. Use appropriate quality of service settings for your tasks (e.g., background priority for background work) to help the system manage resources effectively. Do the foundation models support languages other than English? Yes, the on-device Foundation Model is multilingual and supports all languages supported by Apple Intelligence. To get the model to output in a specific language, prompt it with instructions indicating the user's preferred language using the locale API (e.g., "The user's preferred language is en-US"). Putting the instructions in English, but then putting the user prompt in the desired output language is a recommended practice. Are larger server-based models available through Foundation Models? No, the Foundation Models API currently only provides access to the on-device Large Language Model at the core of Apple Intelligence. It does not support server-side models. On-device models are preferred for privacy and for performance reasons. Is it possible to run Retrieval-Augmented Generation (RAG) using the Foundation Models framework? Yes, it is possible to run RAG on-device, but the Foundation Models framework does not include a built-in embedding model. You'll need to use a separate database to store vectors and implement nearest neighbor or cosine distance searches. The Natural Language framework offers simple word and sentence embeddings that can be used. Consider using a combination of Foundation Models and Core ML, using Core ML for your embedding model.

Hello, I posted an issue on the coremltools GitHub about my Core ML models not performing as well on iOS 17 vs iOS 16 but I'm posting it here just in case. TL;DR The same model on the same device/chip performs far slower (doesn't use the Neural Engine) on iOS 17 compared to iOS 16. Longer description The following screenshots show the performance of the same model (a PyTorch computer vision model) on an iPhone SE 3rd gen and iPhone 13 Pro (both use the A15 Bionic). iOS 16 - iPhone SE 3rd Gen (A15 Bioinc) iOS 16 uses the ANE and results in fast prediction, load and compilation times. iOS 17 - iPhone 13 Pro (A15 Bionic) iOS 17 doesn't seem to use the ANE, thus the prediction, load and compilation times are all slower. Code To Reproduce The following is my code I'm using to export my PyTorch vision model (using coremltools). I've used the same code for the past few months with sensational results on iOS 16. # Convert to Core ML using the Unified Conversion API coreml_model = ct.convert( model=traced_model, inputs=[image_input], outputs=[ct.TensorType(name="output")], classifier_config=ct.ClassifierConfig(class_names), convert_to="neuralnetwork", # compute_precision=ct.precision.FLOAT16, compute_units=ct.ComputeUnit.ALL ) System environment: Xcode version: 15.0 coremltools version: 7.0.0 OS (e.g. MacOS version or Linux type): Linux Ubuntu 20.04 (for exporting), macOS 13.6 (for testing on Xcode) Any other relevant version information (e.g. PyTorch or TensorFlow version): PyTorch 2.0 Additional context This happens across "neuralnetwork" and "mlprogram" type models, neither use the ANE on iOS 17 but both use the ANE on iOS 16 If anyone has a similar experience, I'd love to hear more. Otherwise, if I'm doing something wrong for the exporting of models for iOS 17+, please let me know. Thank you!

While building an app with large language model inferencing on device, I got gibberish output. After carefully examining every detail, I found it's caused by the fused scaledDotProductAttention operation. I switched back to the discrete operations and problem solved. To reproduce the bug, please check https://github.com/zhoudan111/MPSGraph_SDPA_bug

I'm implementing an LLM with Metal Performance Shader Graph, but encountered a very strange behavior, occasionally, the model will report an error message as this: LLVM ERROR: SmallVector unable to grow. Requested capacity (9223372036854775808) is larger than maximum value for size type (4294967295) and crash, the stack backtrace screenshot is attached. Note that 5th frame is mlir::getIntValues<long long> and 6th frame is llvm::SmallVectorBase<unsigned int>::grow_pod It looks like mlir mistakenly took a 64 bit value for a 32 bit type. Unfortunately, I could not found the source code of mlir::getIntValues, maybe it's Apple's closed source fork of llvm for MPS implementation? Anyway, any opinion or suggestion on that?

Our app is downloading a zip of an .mlpackage file, which is then compiled into an .mlmodelc file using MLModel.compileModel(at:). This model is then run using a VNCoreMLRequest. Two users – and this after a very small rollout - are reporting issues running the VNCoreMLRequest. The error message from their logs: Error Domain=com.apple.CoreML Code=0 "Failed to build the model execution plan using a model architecture file '/private/var/mobile/Containers/Data/Application/F93077A5-5508-4970-92A6-03A835E3291D/Documents/SKDownload/Identify-image-iOS/mobile_img_eu_v210.mlmodelc/model.mil' with error code: -5." The URL there is to a file inside the compiled model. The error is happening when the perform function of VNImageRequestHandler is run. (i.e. the model compiled without an error.) Anyone else seen this issue? Its only picked up in a few web results and none of them are directly relevant or have a fix. I know that a CoreML error Code=0 is a generic error, but does anyone know what error code -5 is? Not even sure which framework its coming from.

Hi everyone, I'm a Mac enthusiast experimenting with tensorflow-metal on my Mac Pro (2013). My question is about GPU selection in tensorflow-metal (v0.8.0), which still supports Intel-based Macs, including my machine. I've noticed that when running TensorFlow with Metal, it automatically selects a GPU, regardless of what I specify using device indices like "gpu:0", "gpu:1", or "gpu:2". I'm wondering if there's a way to manually specify which GPU should be used via an environment variable or another method. For reference, I’ve tried the example from TensorFlow’s guide on multi-GPU selection: https://www.tensorflow.org/guide/gpu#using_a_single_gpu_on_a_multi-gpu_system My goal is to explore performance optimizations by using MirroredStrategy in TensorFlow to leverage multiple GPUs: https://www.tensorflow.org/guide/distributed_training#mirroredstrategy Interestingly, I discovered that the metalcompute Python library (https://pypi.org/project/metalcompute/) allows to utilize manually selected GPUs on my system, allowing for proper multi-GPU computations. This makes me wonder: Is there a hidden environment variable or setting that allows manual GPU selection in tensorflow-metal? Has anyone successfully used MirroredStrategy on multiple GPUs with tensorflow-metal? Would a bridge between metalcompute and tensorflow-metal be necessary for this use case, or is there a more direct approach? I’d love to hear if anyone else has experimented with this or has insights on getting finer control over GPU selection. Any thoughts or suggestions would be greatly appreciated! Thanks!

Hello, I have to create an app in Swift that it scan NFC Identity card. It extract data and convert it to human readable data. I do it with below code import CoreNFC class NFCIdentityCardReader: NSObject , NFCTagReaderSessionDelegate { func tagReaderSessionDidBecomeActive(_ session: NFCTagReaderSession) { print("\(session.description)") } func tagReaderSession(_ session: NFCTagReaderSession, didInvalidateWithError error: any Error) { print("NFC Error: \(error.localizedDescription)") } var session: NFCTagReaderSession? func beginScanning() { guard NFCTagReaderSession.readingAvailable else { print("NFC is not supported on this device") return } session = NFCTagReaderSession(pollingOption: .iso14443, delegate: self, queue: nil) session?.alertMessage = "Hold your NFC identity card near the device." session?.begin() } func tagReaderSession(_ session: NFCTagReaderSession, didDetect tags: [NFCTag]) { guard let tag = tags.first else { session.invalidate(errorMessage: "No tag detected") return } session.connect(to: tag) { (error) in if let error = error { session.invalidate(errorMessage: "Connection error: \(error.localizedDescription)") return } switch tag { case .miFare(let miFareTag): self.readMiFareTag(miFareTag, session: session) case .iso7816(let iso7816Tag): self.readISO7816Tag(iso7816Tag, session: session) case .iso15693, .feliCa: session.invalidate(errorMessage: "Unsupported tag type") @unknown default: session.invalidate(errorMessage: "Unknown tag type") } } } private func readMiFareTag(_ tag: NFCMiFareTag, session: NFCTagReaderSession) { // Read from MiFare card, assuming it's formatted as an identity card let command: [UInt8] = [0x30, 0x04] // Example: Read command for block 4 let requestData = Data(command) tag.sendMiFareCommand(commandPacket: requestData) { (response, error) in if let error = error { session.invalidate(errorMessage: "Error reading MiFare: \(error.localizedDescription)") return } let readableData = String(data: response, encoding: .utf8) ?? response.map { String(format: "%02X", $0) }.joined() session.alertMessage = "ID Card Data: \(readableData)" session.invalidate() } } private func readISO7816Tag(_ tag: NFCISO7816Tag, session: NFCTagReaderSession) { let selectAppCommand = NFCISO7816APDU(instructionClass: 0x00, instructionCode: 0xA4, p1Parameter: 0x04, p2Parameter: 0x00, data: Data([0xA0, 0x00, 0x00, 0x02, 0x47, 0x10, 0x01]), expectedResponseLength: -1) tag.sendCommand(apdu: selectAppCommand) { (response, sw1, sw2, error) in if let error = error { session.invalidate(errorMessage: "Error reading ISO7816: \(error.localizedDescription)") return } let readableData = response.map { String(format: "%02X", $0) }.joined() session.alertMessage = "ID Card Data: \(readableData)" session.invalidate() } } } But I got null. I think that these data are encrypted. How can I convert them to readable data without MRZ, is it possible ? I need to get personal informations from Identity card via Core NFC. Thanks in advance. Best regards

In an under-development MacOS & iOS app, I need to identify various measurements from OCR'ed text: length, weight, counts per inch, area, percentage. The unit type (e.g. UnitLength) needs to be identified as well as the measurement's unit (e.g. .inches) in order to convert the measurement to the app's internal standard (e.g. centimetres), the value of which is stored the relevant CoreData entity. The use of NLTagger and NLTokenizer is problematic because of the various representations of the measurements: e.g. "50g.", "50 g", "50 grams", "1 3/4 oz." Currently, I use a bespoke algorithm based on String contains and step-wise evaluation of characters, which is reasonably accurate but requires frequent updating as further representations are detected. I'm aware of the Python SpaCy model being capable of NER Measurement recognition, but am reluctant to incorporate a Python-based solution into a production app. (ref [https://developer.apple.com/forums/thread/30092]) My preference is for an open-source NER Measurement model that can be used as, or converted to, some form of a Swift compatible Machine Learning model. Does anyone know of such a model?

Hi team, We have implemented a writing tool inside a WebView that allows users to type content in a textarea. When the "Show Writing Tools" button is clicked, an AI-powered editor opens. After clicking the "Rewrite" button, the AI modifies the text. However, when clicking the "Replace" button, the rewritten text does not update the original textarea. Kindly check and help me showButton.addTarget(self, action: #selector(showWritingTools(_:)), for: .touchUpInside) @available(iOS 18.2, *) optional func showWritingTools(_ sender: Any) Note: same cases working in TextView pfa

I am working on a lung cancer scanning app in for iOS with a CoreML model and when I test my app on a physical device, the model results in the same prediction 100% of the time. I even changed the names around and still resulted in the same case. I have listed my labels in cases and when its just stuck on the same case (case 1) My code is below: https://github.com/ShivenKhurana1/Detect-to-Protect-App/blob/main/DetectToProtect/SecondView.swift I couldn't add the code as it was too long so I hope github link is fine!

Can't import data in create ML word tagging project training data is 100% correct I guarantee it: I mean look this one has one entry in it. [ { "tokens": [ "a", "august", "gruters" ], "labels": [ "BUILDER", "BUILDER", "BUILDER" ] } ]

In my quantization code, the line: compressed_model_a8 = cto.coreml.experimental.linear_quantize_activations( model, activation_config, [{'img':np.random.randn(1,13,1024,1024)}] ) has taken 90 minutes to run so far and is still not completed. From debugging, I can see that the line it's stuck on is line 261 in _model_debugger.py: model = ct.models.MLModel( cloned_spec, weights_dir=self.weights_dir, compute_units=compute_units, skip_model_load=False, # Don't skip model load as we need model prediction to get activations range. ) Is this expected behaviour? Would it be quicker to run on another computer with more RAM?

Is it possible to train a model using CreateML to infer a relevance numeric score of a news article based on similar trained data, something like a sentiment score ? I created a Text Classifier that assigns a category label which works perfect but I would like a solution that calculates a numeric value, not a label.

Incident Identifier: 4C22F586-71FB-4644-B823-A4B52D158057 CrashReporter Key: adc89b7506c09c2a6b3a9099cc85531bdaba9156 Hardware Model: Mac16,10 Process: PRISMLensCore [16561] Path: /Applications/PRISMLens.app/Contents/Resources/app.asar.unpacked/node_modules/core-node/PRISMLensCore.app/PRISMLensCore Identifier: com.prismlive.camstudio Version: (null) ((null)) Code Type: ARM-64 Parent Process: ? [16560] Date/Time: (null) OS Version: macOS 15.4 (24E5228e) Report Version: 104 Exception Type: EXC_CRASH (SIGABRT) Exception Codes: 0x00000000 at 0x0000000000000000 Crashed Thread: 34 Application Specific Information: *** Terminating app due to uncaught exception 'NSInvalidArgumentException', reason: '*** -[__NSArrayM insertObject:atIndex:]: object cannot be nil' Thread 34 Crashed: 0 CoreFoundation 0x000000018ba4dde4 0x18b960000 + 974308 (__exceptionPreprocess + 164) 1 libobjc.A.dylib 0x000000018b512b60 0x18b4f8000 + 109408 (objc_exception_throw + 88) 2 CoreFoundation 0x000000018b97e69c 0x18b960000 + 124572 (-[__NSArrayM insertObject:atIndex:] + 1276) 3 Portrait 0x0000000257e16a94 0x257da3000 + 473748 (-[PTMSRResize addAdditionalOutput:] + 604) 4 Portrait 0x0000000257de91c0 0x257da3000 + 287168 (-[PTEffectRenderer initWithDescriptor:metalContext:useHighResNetwork:faceAttributesNetwork:humanDetections:prevTemporalState:asyncInitQueue:sharedResources:] + 6204) 5 Portrait 0x0000000257dab21c 0x257da3000 + 33308 (__33-[PTEffect updateEffectDelegate:]_block_invoke.241 + 164) 6 libdispatch.dylib 0x000000018b739b2c 0x18b738000 + 6956 (_dispatch_call_block_and_release + 32) 7 libdispatch.dylib 0x000000018b75385c 0x18b738000 + 112732 (_dispatch_client_callout + 16) 8 libdispatch.dylib 0x000000018b742350 0x18b738000 + 41808 (_dispatch_lane_serial_drain + 740) 9 libdispatch.dylib 0x000000018b742e2c 0x18b738000 + 44588 (_dispatch_lane_invoke + 388) 10 libdispatch.dylib 0x000000018b74d264 0x18b738000 + 86628 (_dispatch_root_queue_drain_deferred_wlh + 292) 11 libdispatch.dylib 0x000000018b74cae8 0x18b738000 + 84712 (_dispatch_workloop_worker_thread + 540) 12 libsystem_pthread.dylib 0x000000018b8ede64 0x18b8eb000 + 11876 (_pthread_wqthread + 292) 13 libsystem_pthread.dylib 0x000000018b8ecb74 0x18b8eb000 + 7028 (start_wqthread + 8)

I have seen inconsistent results for my Colab machine learning notebooks running locally on a Mac M4, compared to running the same notebook code on either T4 (in Colab) or a RTX3090 locally. To illustrate the problems I have set up a notebook that implements two simple CNN models that solves the Fashion-MNIST problem. https://colab.research.google.com/drive/11BhtHhN079-BWqv9QvvcSD9U4mlVSocB?usp=sharing For the good model with 2M parameters I get the following results: T4 (Colab, JAX): Test accuracy: 0.925 3090 (Local PC via ssh tunnel, Jax): Test accuracy: 0.925 Mac M4 (Local, JAX): Test accuracy: 0.893 Mac M4 (Local, Tensorflow): Test accuracy: 0.893 That is, I see a significant drop in performance when I run on the Mac M4 compared to the NVIDIA machines, and it seems to be independent of backend. I however do not know how to pinpoint this to either Keras or Apple’s METAL implementation. I have reported this to Keras: https://colab.research.google.com/drive/11BhtHhN079-BWqv9QvvcSD9U4mlVSocB?usp=sharing but as this can be (likely is?) an Apple Metal issue, I wanted to report this here as well. On the mac I am running the following Python libraries: keras 3.9.1 tensorflow 2.19.0 tensorflow-metal 1.2.0 jax 0.5.3 jax-metal 0.1.1 jaxlib 0.5.3

I got 3203.23 GFLOPS (FP16) on the M3 Macbook Pro and only 2833.24 GFLOPS (FP16) on the M4 Macbook Air for 4096x4096 matrix multiplications for a PyTorch MPS FP16 Benchmark. Wasn't the performance supposed to be twice as high on the M4 compared to the M3 even with the termal throtling on the Macbook Air? What went wrong?

Hi, I have been trying to integrate a CoreML model into Xcode. The model was made using tensorflow layers. I have included both the model info and a link to the app repository. I am mainly just really confused on why its not working. It seems to only be printing the result for case 1 (there are 4 cases labled, case 0, case 1, case 2, and case 3). If someone could help work me through this error that would be great! here is the link to the repository: https://github.com/ShivenKhurana1/Detect-to-Protect-App this file with the model code is called SecondView.swift and here is the model info: Input: conv2d_input-> image (color 224x224) Output: Identity -> MultiArray (Float32 1x4)

I'm developing a tennis ball tracking feature using Vision Framework in Swift, specifically utilizing VNDetectedObjectObservation and VNTrackObjectRequest. Occasionally (but not always), I receive the following runtime error: Failed to perform SequenceRequest: Error Domain=com.apple.Vision Code=9 "Internal error: unexpected tracked object bounding box size" UserInfo={NSLocalizedDescription=Internal error: unexpected tracked object bounding box size} From my investigation, I suspect the issue arises when the bounding box from the initial observation (VNDetectedObjectObservation) is too small. However, Apple's documentation doesn't clearly define the minimum bounding box size that's considered valid by VNTrackObjectRequest. Could someone clarify: What is the minimum acceptable bounding box width and height (normalized) that Vision Framework's VNTrackObjectRequest expects? Is there any recommended practice or official guidance for bounding box size validation before creating a tracking request? This information would be extremely helpful to reliably avoid this internal error. Thank you!

I used Yolo5-11 and while performing great detecting balls lets say 5-10ft away in 1920 resolution and even in 640 it really is taking toll on my app performance. When I use Create ML it outputs all in 415x which is probably the reason why it does not detect objects from far. What can I do to preserve some energy ? My model is used with about 1K pictures 200 each test and validate, and from close up and far.

Hello, I'm trying to write a model with PyTorch and convert it to CoreML. I wrote another models and that works succesfully, even the one that gave the problem is, but I can't visualize it with XCode to know where is running. The error that appear is: There was a problem decoding this Core ML document validator error: unable to open file for read Anyone knows why is this happening? Thanks a lot, Álvaro Corrochano

Recently, I'm trying to deploy some third-party LLM to Apple devices. The methodoloy is similar to https://github.com/Anemll/Anemll. The biggest issue I'm having now is the runtime memory usage. When there are multiple functions in a model (mlpackage or mlmodelc), the runtime memory usage for weights is somehow duplicated when I load all of them. Here's the detail: I created my multifunction mlpackage following https://apple.github.io/coremltools/docs-guides/source/multifunction-models.html I loaded each of the functions using the generated swift class: let config = MLModelConfiguration() config.computeUnits = MLComputeUnits.cpuAndNeuralEngine config.functionName = "infer_512"; let ffn1_infer_512 = try! mimo_FFN_PF_lut4_chunk_01of02(configuration: config) config.functionName = "infer_1024"; let ffn1_infer_1024 = try! mimo_FFN_PF_lut4_chunk_01of02(configuration: config) config.functionName = "infer_2048"; let ffn1_infer_2048 = try! mimo_FFN_PF_lut4_chunk_01of02(configuration: config) I observed that RAM usage increases linearly as I load each of the functions. Using instruments, I see that there are multiple HWX files generated and loaded, each of which contains all the weight data. My understanding of what's happening here: The CoreML framework did some MIL->MIL preprocessing before further compilation, which includes separating CPU workload from ANE workload. The ANE part of each function is moved into a separate MIL file then compile separately into a HWX file each. The problem is that the weight data of these HWX files are duplicated. Since that the weight data of LLMs is huge, it will cause out-of-memory issue on mobile devices. The improvement I'm hoping from Apple: I hope we can try to merge the processed MIL files back into one before calling ANECCompile(), so that the weights can be merged. I don't have control over that in user space and I'm not sure if that is feasible. So I'm asking for help here. Thanks.