Faster and local Jev like model for Mac (github.com)

Apple has announced the release of a new local AI model called Laya for the Mac, optimized for Apple Silicon using Core ML and the Neural Engine. This model enhances decision-making efficiency by enabling real-time game interactions—demonstrated through the classic game Snake—while showcasing critical performance metrics like decision speed and safety interventions. The implementation achieved a game loop sustaining 49.1–50.0 decisions per second and demonstrated significant energy efficiency, achieving up to 3.19 times better energy efficiency compared to compiled MLX FP16 models. This development is significant for the AI/ML community as it illustrates the potential of localized AI models to operate efficiently on consumer hardware, reducing reliance on cloud-based solutions. The Laya model supports multilingual decision-making and works without heavy frameworks like PyTorch or Transformers, making it accessible for various applications. The model's architecture includes features such as real-time safety checks and efficient memory usage, establishing a foundation for future enhancements in AI interactivity on personal devices. Users can download and run the model offline, paving the way for broader applications in areas such as customer service and automated assistance while maintaining user privacy.