laya-mlx

Native MLX FP16 conversion of Laya for Apple silicon. The same ModernBERT-large encoder, decision transformer and heads, running in MLX with no PyTorch; the checkpoint Laya Studio fine-tunes from.

A 512-token context, and the choice, score and noul question types. Loads with the laya-mlx runtime on macOS 14+ with Python 3.11+: laya.load("aac6fef/laya-mlx"). Every model trained in Laya Studio on the English base names this checkpoint as its base.

What it decides

• choice — picks one option from a set
• score — places the input on an ordered scale
• noul — answers a yes/no question with one calibrated probability

At a glance

Parameters	421M
Base model	convai-innovations/laya
Maker	aac6fef
Released	2026-09-19
License	apache-2.0

Get the weights

pip install systemonemodels
systemone pull aac6fef/laya-mlx

The files are served from the maker's Hugging Face repository, aac6fef/laya-mlx, and verified against the checksums recorded here.

Read more

• Weights on Hugging Face
• laya-mlx runtime

---

This page was opened by System One for aac6fef, who can claim the organisation and take it over at any time.