MLBoy·Jun 13One Metal kernel made Apple Core AI’s MoE decode 2–3.6× faster — same qualitySparse Mixture-of-Experts models are perfect for on-device — until the framework reads every expert on every token. Here’s the kernel that…
MLBoy·Jun 127 LLMs Pre-Converted to Apple’s Core AI Format (.aimodel), Now on Hugging Facegpt-oss-20B, Qwen3, Gemma 3, Mistral — converted with Apple’s official recipes, unmodified, with hashes and measured benchmarks.
MLBoy·Jun 10I Benchmarked Apple’s New Framework Against MLX for On-Device LLMsIt really pays to actually run the comparison.A response icon2A response icon2
MLBoy·Jun 10CoreAI-Model-Zoo: Open LLMs for Apple’s New Core AI FrameworkApple has a new on-device AI framework: Core AI. It lets you convert your favorite AI and LLM models into its .aimodel format and run them…A response icon1A response icon1
MLBoy·Jun 9Read meters and nameplates with just an iPhoneAll you need is an iPhone. Let AI do the field check.
MLBoy·Jun 5Recipes for on-device VLM (image input LLM)There’s an AI you can show a photo to and ask: “what is this?” — and it will reply, “Looks like a wooden dining table. Two mugs on it; one…
MLBoy·Jun 4iPhone on-device LLM: the GPU wins the sprint, the Neural Engine wins the marathonThe follow-up: burst speed is only half the story
MLBoy·May 30Scoring Outfits with AI — Fully Offline, on an iPhoneCan a 2-billion-parameter vision model judge fashion? It can — if the domain is a “closed system.” A field report on distilling a giant…
MLBoy·May 29Let Customers Feel a Product’s Texture — Over the InternetA Gaussian Splatting × AR product viewer, relightable on iPhone
MLBoy·May 28Local LLM on iPhone: which runtime is actually fastest?Four on-device runtimes, same models, same prompt, same harness — on a real iPhone 17 Pro.