Daisuke Majima (MLBoy)·Sep 24Jev-style text classification (System One), run locally. How accurate can it get?Jev is a text classification model that TypeSafe AI offers in the cloud. You send a text and a question with a set of options, and it…
Daisuke Majima (MLBoy)·Sep 23System One inference with no API fees (on the iPhone and the Mac)I think System One inference, the kind Jev made popular, suits on-device use. Telling which utterance needs a reply, filling a form from…
Daisuke Majima (MLBoy)·Sep 15Run Gemma or Qwen behind FoundationModelsSame LanguageModelSession, your chosen model behind it.A response icon1A response icon1
Daisuke Majima (MLBoy)·Sep 10Adding AI doesn’t have to mean charging a subscriptionDaisuke Majima (MLBoy)
Daisuke Majima (MLBoy)·Sep 8Apple Core AI vs MLX: which is faster on iPhone and Mac for the same model?Last verified: 2026–09–05. Every number below comes from one open harness, apple-silicon-llm-bench, on one iPhone 17 Pro (iOS 27 beta) and…A response icon1A response icon1
Daisuke Majima (MLBoy)·Sep 8LiteRT vs TensorFlow Lite: what actually changed?Last verified: 2026–09–05 — LiteRT 2.2.0, LiteRT-LM 0.16.1, litert-torch 0.9.4, ai-edge-litert 2.2.0. Every version and link below was read…
Daisuke Majima (MLBoy)·Aug 3Calling your own AI from iPhone’s Visual IntelligenceCalling your own AI from iPhone’s Visual Intelligence
Daisuke Majima (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…
Daisuke Majima (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.