What it is
Runs Laya typed decision models locally on Apple Silicon with MLX, delivering 7.4-13.4 ms short decisions and no token generation. Run constrained classification, scoring, and boolean decision models fully local on Apple Silicon, preserving upstream Laya outputs while dropping PyTorch, Transformers runtime, and cloud APIs.
Key results
- 7.39 ms P50, 395.0 q/s for multilingual 322M on M3 Max
- 13.42 ms P50, 146.8 q/s for English 421M on M3 Max
- Matched upstream selected answers on 63/63 validation questions in FP16/FP32
Shows that bidirectional decision models can replace token generation for low-latency structured local inference on Apple Silicon.