card: Neural Engine bundle section — 6-bit, why, speed, first-load cost, GSM8K 200

ios-static: the 6-bit k-means g8 static iOS export before AOT (replaces the 4-bit g32 IR of 2026-09-15)

ios-ane-h18p: drop the 4-bit g32 aimodelc (replaced by the 6-bit export; metadata.json points at the 6-bit one)

ios-ane-h18p: 6-bit k-means g8 static iOS export, AOT h18p neural-engine (replaces the 4-bit g32 export of 2026-09-15; device gate 3/4 exact, long answer diverges on a 0.17-margin step)

card: same-day ANE vs GPU A/B on iPhone 17 Pro

card: Neural Engine bundle section

ios-static: stock static iOS export, AOT h18p neural-engine (device gate PASS 3/3)

ios-ane-h18p: stock static iOS export, AOT h18p neural-engine (device gate PASS 3/3)

Card: macOS/iOS 27 GA wording (beta requirement dropped; beta findings dated)

MiniCPM5-2B — Core AI int8 bundle (dynamic, pipelined engine)

MiniCPM5-2B — Core AI int8 bundle (dynamic, pipelined engine)

MiniCPM5-2B — Core AI int8 bundle (dynamic, pipelined engine)

initial commit