mizorewww Launches Laya-MLX Native Apple Silicon Runtime for Typed Decision Models
Summary
mizorewww launches Laya-MLX, a native Apple Silicon runtime for Convai Innovations’ typed decision models that runs without text generation, PyTorch, or cloud APIs, reporting 7.39 ms median latency for its 322M multilingual model on an M3 Max.
Key Points
- mizorewww launches Laya-MLX, an independent native MLX runtime for Convai Innovations’ typed decision models on Apple Silicon that avoids text generation, PyTorch and cloud APIs.
- On an M3 Max, the 322M multilingual checkpoint records 7.39 ms median latency for one short question, while the 421M English model records 13.42 ms.
- The optional optimized Snake demo reaches 75.40 moves per second across 2,400 moves with zero deaths and two visible safety interventions in a paired M3 Max test.