NOTICE

laya-apple
Copyright 2026 laya-apple contributors

Independent project. Not an official Convai Innovations, Apple, or MLX release.

This product includes software adapted from the following Apache-2.0 projects:

Laya — https://github.com/NandhaKishorM/laya
  Copyright Convai Innovations and Laya contributors.
  Revision: 573e5b62696ba441230cd6be71d593331b5d23af, updated to v0.3.20
  (23a17522aa4942da6cce53a995a275760320b691)
  Adapted: prompt construction, option rendering, question validation, calibration
  temperatures and clamping, result schema (laya_apple/prompt.py, laya_apple/result.py).
  Vendored from v0.3.20: language/script detection (laya_apple/lang.py, from
  laya/lang.py). Adapted from v0.3.20: the /v1/systemone wire format, limits and
  language routing of laya/serve.py and laya/router.py (laya_apple/serve.py,
  laya_apple/router.py).
  Adapted from v0.3.20: the opt-in embedding shortlist of laya/shortlist.py
  (laya_apple/shortlist.py).

laya-mlx — https://github.com/mizorewww/laya-mlx
  Copyright 2026 laya-mlx contributors.
  Revision: 0a859518634112655cb97c745dbf04f5191aaf13
  Adapted: native MLX ModernBERT and Laya decision heads, weight-name mapping
  (laya_apple/models/modernbert_mlx.py); Rust-tokenizer wrapper; MLX encoder mean-pooling
  for the shortlist embedding helper (laya_apple/shortlist.py, from laya_mlx/shortlist.py).
  Modifications: packaging only; module docstring and imports.

laya-coreml — https://github.com/mizorewww/laya-coreml
  Copyright 2026 laya-coreml contributors.
  Revision: 4619e0483f07adf39068532e85b42ec2347edb83
  Adapted: export-only PyTorch DecisionModel (laya_apple/conversion/torch_reference.py),
  channel-first BC1S Neural Engine graph (laya_apple/conversion/bc1s.py, from
  experiments/ane_engineering/model.py), host embedding / action-head runtime
  (laya_apple/backends/coreml_ane.py, from laya_coreml/ane.py), Core ML compute-plan
  inspection (laya_apple/artifacts.py).
  Modifications: generalised to ModernBERT-large checkpoints, strict artifact
  verification, fixed-shape bucket management.

Model weights are not included. They are downloaded from the upstream Hugging Face
repositories at pinned revisions and remain under their respective licenses.