Laya on Core ML with 99.5% of operations on the Neural Engine
FluidInference's port, benchmarked at 3.7 ms per decision on an M5 Pro.
BUILDS
6 builds made with System One models.
Agents, routers, scorers and games — built on models that decide. Every entry links to the models it uses, so you can pull the same one and start from there.
6 builds
Laya on Axera edge NPUs in under 70 ms
AXERA-TECH converted all three checkpoints for its AX650 and AX8850 chips.
EdgeJev: offline Laya on a 4-core CPU in 15.6 ms
A convert, quantize and deploy toolchain that turns Laya and related models into INT8 ONNX for offline CPUs, measured at 15.6 ms per question.
Laya Warehouse Safety: a robot that knows when to wait
In a replayable grid with moving workers and forklifts, Laya chooses advance, shift or wait and flags collision risk, under a deterministic safety shield.
LayaKit: one Core ML decision per call from Swift
A Swift package derived from laya-coreml that loads the general or Neural Engine bundles and predicts in a single call.
Laya-CoreML: ~5 ms decisions on the Neural Engine
Validated Core ML ports of every Laya checkpoint: 4.98 ms median on an M3 Max and 2.78× better energy use than MLX.
Many of the first builds here were collected by madewithlaya.com. Every entry links to its creator’s own post, repository or site.