01 · The idea
What stayed with us
CoreML-LLM is a community example of taking owned weights toward Apple's Neural Engine. It informs our research boundary; it is not a shipped PostTrainLLM engine.
02 · The local translation
What we did with it
Our ANE dossier uses community implementations to understand conversion, layout, and execution constraints. CoreML-LLM offers one concrete way to study an ANE-oriented stack while our native work keeps model ownership and exact evaluation separate from the serving layer.
03 · The boundary
Where the comparison stops
We have not adopted CoreML-LLM as the PostTrainLLM app engine or established a same-model, same-Mac comparison. Upstream speed claims require their exact hardware, model, and settings before they can inform a local choice.