Laya-MLX: open typed decision model runs locally at 7–14 ms per decision, zero tokens
JiliJeanlouis · x · 2026-09-20
- Developer mizorewww ported the open-source typed-decision system Laya to Apple MLX as laya-mlx (576 GitHub stars), with no PyTorch, Transformers runtime, or cloud API.
- Benchmarks: 13.4 ms median end-to-end for a short English typed decision on M3 Max, 7.4 ms with the multilingual checkpoint, 0 output tokens, 1 GB max memory.
- A demo shows the model playing Snake locally at roughly 60 decisions per second, with a cycle safety layer to correct unsafe proposals.
- Weights are on Hugging Face (aac6fef/laya-mlx); quick start via pip install laya-mlx.
Related event: Open-Source Laya Ported to Apple MLX for 7ms On-Device Decisions(3 posts)→
More from coding & agent
- Devin claims "Jason approved my PR," roasted for being stuck in training data — tekbog · 2026-11-06
- Veteran dev: AI-built TraceDraw beats CorelDRAW 2026, predicts software industry upheaval by 2036 — Maxiol777 · 2026-11-06
- A practical rule for personal AI agents: automate the boring, gate the consequential — No_Appeal_5223 · 2026-11-06
- 3D artist runs Blender rigging via MCP on 8GB VRAM with a 35B local model — NicholasCureton · 2026-11-06
- CodeChart: open-source skill makes any coding agent start summaries with a change map — alexmazawd · 2026-11-06
- RAX lets coding agents unscrew a real jar with dexterous hands: 6 chained skills, 140° rotation — du_yilun · 2026-11-06