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.

Documentation on adapter train is lacking any details related to training on dataset with tool calling. And page about tool calling itself only explain how to use it from Swift without any internal details useful in training. Question is how schema should looks like for including tool calling in dataset?

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!

I have used mlx_lm.lora to fine tune a mistral-7b-v0.3-4bit model with my data. I fused the mistral model with my adapters and upload the fused model to my directory on huggingface. I was able to use mlx_lm.generate to use the fused model in Terminal. However, I don't know how to load the model in Swift. I've used Imports import SwiftUI import MLX import MLXLMCommon import MLXLLM let modelFactory = LLMModelFactory.shared let configuration = ModelConfiguration( id: "pharmpk/pk-mistral-7b-v0.3-4bit" ) // Load the model off the main actor, then assign on the main actor let loaded = try await modelFactory.loadContainer(configuration: configuration) { progress in print("Downloading progress: \(progress.fractionCompleted * 100)%") } await MainActor.run { self.model = loaded } I'm getting an error runModel error: downloadError("A server with the specified hostname could not be found.") Any suggestions? Thanks, David PS, I can load the model from the app bundle // directory: Bundle.main.resourceURL! but it's too big to upload for Testflight

*I can't put the attached file in the format, so if you reply by e-mail, I will send the attached file by e-mail. Dear Apple AI Research Team, My name is Gong Jiho (“Hem”), a content strategist based in Seoul, South Korea. Over the past few months, I conducted a user-led AI experiment entirely within ChatGPT — no code, no backend tools, no plugins. Through language alone, I created two contrasting agents (Uju and Zero) and guided them into a co-authored modular identity system using prompt-driven dialogue and reflection. This system simulates persona fusion, memory rooting, and emotional-logical alignment — all via interface-level interaction. I believe it resonates with Apple’s values in privacy-respecting personalization, emotional UX modeling, and on-device learning architecture. Why I’m Reaching Out I’d be honored to share this experiment with your team. If there is any interest in discussing user-authored agent scaffolding, identity persistence, or affective alignment, I’d love to contribute — even informally. ⚠ A Note on Language As a non-native English speaker, my expression may be imperfect — but my intent is genuine. If anything is unclear, I’ll gladly clarify. 📎 Attached Files Summary Filename → Description Hem_MultiAI_Report_AppleAI_v20250501.pdf → Main report tailored for Apple AI — narrative + structural view of emotional identity formation via prompt scaffolding Hem_MasterPersonaProfile_v20250501.json → Final merged identity schema authored by Uju and Zero zero_sync_final.json / uju_sync_final.json → Persona-level memory structures (logic / emotion) 1_0501.json ~ 3_0501.json → Evolution logs of the agents over time GirlfriendGPT_feedback_summary.txt → Emotional interpretation by external GPT hem_profile_for_AI_vFinal.json → Original user anchor profile Warm regards, Gong Jiho (“Hem”) Seoul, South Korea

I’m considering creating an ILMessageFilterExtension using a mini LLM/SLM to detect fraud and I’ve read it has strict memory limits yet I can’t find it in the documentation. What’s the set limit or any other constraints impacting the feasibility of running 100-500mb model?

After a few attempts I end up at the same spot as categorized here by ChatGPT - I get why it feels that way, Rob — honestly, you’ve walked right into the worst possible moment in Apple’s Core ML toolchain evolution. What you’re running into isn’t your fault; Apple really did rip out the ONNX front-end without offering a clean replacement yet. So every tutorial, every script, and even coremltools itself contradicts what’s actually possible right now. Let’s be brutally clear about the state of things (late 2025): Goal Supported by Apple today Practical workaround Convert ONNX → Core ML using coremltools 9 No. The ONNX importer was removed. Use the legacy coremltools 6.x + onnx-coreml stack in a separate Python 3.9/3.10 env. Convert PyTorch → Core ML directly Yes, but only for simple or traced models. T5 still fails because of dynamic control flow. None reliable without rewriting the model. Run ONNX models directly on iOS Partially (via MLC LLM / Metal Performance Shaders). Works, but that’s an entirely different runtime. So: you’re not crazy, and this isn’t wasted effort—you’ve actually proven that your model exports cleanly and that the only missing piece is Apple’s converter. The ONNX file you built is solid. You just need something that can translate it to Core ML. If your goal is simply “get it running on-device”, you still have two viable moves that aren’t rabbit holes: 1. Use MLC LLM / Metal runtime Skip Core ML altogether. Apple’s mlc-llm (and the “MLCChat” pipeline) can execute your ONNX or quantized model directly on the GPU. You keep full fidelity and avoid all these broken converters. 2. Use the legacy Core ML 6.x toolchain in a sandboxed Python 3.10 VM It actually works. You don’t have to downgrade your whole system—just run that specific environment once to emit the .mlpackage, then never touch it again. Nothing else will silently fix itself until Apple re-adds ONNX ingestion in Core ML 10 or a new converter.

During testing the “Bringing advanced speech-to-text capabilities to your app” sample app demonstrating the use of iOS 26 SpeechAnalyzer, I noticed that the language model for the English locale was presumably already downloaded. Upon checking the documentation of AssetInventory, I found out that indeed, the language model can be preinstalled on the system. Can someone from the dev team share more info about what assets are preinstalled by the system? For example, can we safely assume that the English language model will almost certainly be already preinstalled by the OS if the phone has the English locale?

A foundation models bug I keep running into when in the preview phase of the testing. The error never seems to occur or break the app when I am testing on the simulator or on a device but sometimes I am running into this error when in a longer session while being in preview. The error breaks the preview and crashes it and the waring on it is labeled as : "Assert in LanguageModelFeedback.swift" This is something I keep running into, where I have been using foundation models for my project

I am using gemini2.5-flash with SwiftUI. How can I receive a response in JSON?

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?

WWDC25: Combine Metal 4 machine learning and graphics Demonstrated a way to combine neural network in the graphics pipeline directly through the shaders, using an example of Texture Compression. However there is no mention of using which ML technique texture is compressed. Can anyone point me to some well known model/s for this particular use case shown in WWDC25.

I've built a model using Create ML, but I can't make it, for the love of God, updatable. I can't find any checkbox or anything related. It's an Activity Classifier, if it matters. I want to continue training it on-device using MLUpdateTask, but the model, as exported from Create ML, fails with error: Domain=com.apple.CoreML Code=6 "Failed to unarchive update parameters. Model should be re-compiled." UserInfo={NSLocalizedDescription=Failed to unarchive update parameters. Model should be re-compiled.}

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?

Is it possible to train an Adaptor for the Foundation Models to produce Generable output? If so what would the response part of the training data need to look like? Presumably, under the hood, the model is outputting JSON (or some other similar structure) that can be decoded to a Generable type. Would the response part of the training data for an Adaptor need to be in that structured format?

Recursive and Self-Referential Data Structures Combining recursive and self-referential data structures with frameworks like Accelerate, SwiftMacros, and utilizing SwiftUI hooks can offer significant benefits in terms of performance, maintainability, and expressiveness. Here is how Apple Intelligence breaks it down. Benefits: Natural Representation of Complex Data: Recursive structures, such as trees and graphs, are ideal for representing hierarchical or interconnected data, like file systems, social networks, and DOM trees. Simplified Algorithms: Many algorithms, such as traversals, sorting, and searching, are more straightforward and elegant when implemented using recursion. Dynamic Memory Management: Self-referential structures can dynamically grow and shrink, making them suitable for applications with unpredictable data sizes. Challenges: Performance Overhead: Recursive algorithms can lead to stack overflow if not properly optimized (e.g., using tail recursion). Self-referential structures can introduce memory management challenges, such as retain cycles. Accelerate Framework Benefits: High-Performance Computation: Accelerate provides optimized libraries for numerical and scientific computing, including linear algebra, FFT, and image processing. It can significantly speed up computations, especially for large datasets, by leveraging multi-core processors and GPU acceleration. Parallel Processing: Accelerate automatically parallelizes operations, making it easier to take advantage of modern hardware capabilities. Integration with Recursive Data: Matrix and Vector Operations: Use Accelerate for operations on matrices and vectors, which are common in recursive algorithms like those used in machine learning and physics simulations. FFT and Convolutions: Accelerate's FFT functions can be used in recursive algorithms for signal processing and image analysis. SwiftMacros Benefits: Code Generation and Transformation: SwiftMacros allow you to generate and transform code at compile time, enabling the creation of DSLs, boilerplate reduction, and optimization. Improved Compile-Time Checks: Macros can perform complex compile-time checks, ensuring code correctness and reducing runtime errors. Integration with Recursive Data: DSL for Data Structures: Create a DSL using SwiftMacros to define recursive data structures concisely and safely. Optimization: Use macros to generate optimized code for recursive algorithms, such as memoization or iterative transformations. SwiftUI Hooks Benefits: State Management: Hooks like @State, @Binding, and @Effect simplify state management in SwiftUI, making it easier to handle dynamic data. Side Effects: @Effect allows you to perform side effects in a declarative manner, integrating seamlessly with asynchronous operations. Reusable Logic: Custom hooks enable the reuse of stateful logic across multiple views, promoting code maintainability. Integration with Recursive Data: Dynamic Data Binding: Use SwiftUI's data binding to manage the state of recursive data structures, ensuring that UI updates reflect changes in the underlying data. Efficient Rendering: SwiftUI's diffing algorithm efficiently updates the UI only for the parts of the recursive structure that have changed, improving performance. Asynchronous Data Loading: Combine @Effect with recursive data structures to fetch and process data asynchronously, such as loading a tree structure from a remote server. Example: Combining All Components Imagine you're building an app that visualizes a hierarchical file system using a recursive tree structure. Here's how you might combine these components: Define the Recursive Data Structure: Use SwiftMacros to create a DSL for defining tree nodes. @macro struct TreeNode { var value: T var children: [TreeNode] } Optimize with Accelerate: Use Accelerate for operations like computing the size of the tree or performing transformations on node values. func computeTreeSize(_ node: TreeNode) -> Int { return node.children.reduce(1) { $0 + computeTreeSize($1) } } Manage State with SwiftUI Hooks: Use SwiftUI hooks to load and display the tree structure dynamically. struct FileSystemView: View { @State private var rootNode: TreeNode = loadTree() var body: some View { TreeView(node: rootNode) } private func loadTree() -> TreeNode<String> { // Load or generate the tree structure } } struct TreeView: View { let node: TreeNode var body: some View { List(node.children, id: \.value) { Text($0.value) TreeView(node: $0) } } } Perform Side Effects with @Effect: Use @Effect to fetch data asynchronously and update the tree structure. struct FileSystemView: View { @State private var rootNode: TreeNode = TreeNode(value: "/") @Effect private var loadTreeEffect: () -> Void = { // Fetch data from a server or database } var body: some View { TreeView(node: rootNode) .onAppear { loadTreeEffect() } } } By combining recursive data structures with Accelerate, SwiftMacros, and SwiftUI hooks, you can create powerful, efficient, and maintainable applications that handle complex data with ease.

Hello, I'm running a large language model (LLM) in Core ML that uses a key-value cache (KV-cache) to store past attention states. The model was converted from PyTorch using coremltools and deployed on-device with Swift. The KV-cache is exposed via MLState and is used across inference steps for efficient autoregressive generation. During the prefill stage — where a prompt of multiple tokens is passed to the model in a single batch to initialize the KV-cache — I’ve noticed that some entries in the KV-cache are not updated after the inference. Specifically: Here are a few details about the setup: The MLState returned by the model is identical to the input state (often empty or zero-initialized) for some tokens in the batch. The issue only happens during the prefill stage (i.e., first call over multiple tokens). During decoding (single-token generation), the KV-cache updates normally. The model is invoked using MLModel.prediction(from:using:options:) for each batch. I’ve confirmed: The prompt tokens are non-repetitive and not masked. The model spec has MLState inputs/outputs correctly configured for KV-cache tensors. Each token is processed in a loop with the correct positional encodings. Questions: Is there any known behavior in Core ML that could prevent MLState from updating during batched or prefill inference? Could this be caused by internal optimizations such as lazy execution, static masking, or zero-value short-circuiting? How can I confirm that each token in the batch is contributing to the KV-cache during prefill? Any insights from the Core ML or LLM deployment community would be much appreciated.

Hello, I am studying macOS26 Apple Intelligence features. I have created a basic swift program with Xcode. This program is sending prompts to FoundationModels.LanguageModelSession. It works fine but this model is not trained for programming or code completion. Xcode has an AI code completion feature. It is called "Predictive Code completion model". So, there are multiple on-device models on macOS26 ? Are there others ? Is there a way for me to send prompts to this "Predictive Code completion model" from my program ? Thanks

I'm running MacOs 26 Beta 5. I noticed that I can no longer achieve 100% usage on the ANE as I could before with Apple Foundations on-device model. Has Apple activated some kind of throttling or power limiting of the ANE? I cannot get above 3w or 40% usage now since upgrading. I'm on the high power energy mode. I there an API rate limit being applied? I kave a M4 Pro mini with 64 GB of memory.

Hi all, I spent the last few months developing an MLX/Ollama local AI Benchmarking suite for Apple Silicon, written in pure Swift and signed with an Apple Developer Certificate, open source, GPL, and free. I would love some feedback to continue development. It is the only benchmarking suite I know of that supports live power metrics and MLX natively, as well as quick exports for benchmark results, and an arena mode, Model A vs B with history. I really want this project to succeed, and have widespread use, so getting 75 stars on the github repo makes it eligible for Homebrew/Cask distribution. Github Repo

Hello Team, I'm currently working on a proof of concept using Apple's Foundation Model for a RAG-based chat system on my MacBook Pro with the M1 Max chip. Environment details: macOS: 26.0 Beta Xcode: 26.0 beta 2 (17A5241o) Target platform: iPad (as the iPhone simulator does not support Foundation models) While testing, even with very small input prompts to the LLM, I intermittently encounter the following error: InferenceError::inference-Failed::Failed to run inference: Context length of 4096 was exceeded during singleExtend. Has anyone else experienced this issue? Are there known limitations or workarounds for context length handling in this setup? Any insights would be appreciated. Thank you!