Laya has been adapted to work on Apple M4 chips via CoreML

Developers have presented a way to run the multilingual model Laya on Mac computers with M4 processors using the CoreML framework. The optimization allows calculations to be performed completely locally without accessing cloud servers. In tests, the system demonstrated performance at the level of 45 solutions per second. To achieve such performance, a specialized shell script was prepared, which converts the model weights and configures the execution environment for efficient use of the Apple Neural Engine. The integration process involves preparing the model to work in CoreML format, which minimizes delays in processing requests. The use of local resources allows data confidentiality to be maintained, as all decision-making operations take place directly on the user's device. This solution is aimed at developers who require high response speed when working with intelligent agents in offline mode. Detailed instructions for setting up the environment and running the script are available in the public repository, where the technical nuances of optimization for the Apple Silicon architecture are also discussed. Transferring models to local hardware with hardware acceleration is becoming a standard for tasks that require minimal latency. The use of CoreML allows the capabilities of the M4 chips to be maximally utilized, providing high throughput at low power consumption compared to running on CPU or discrete GPU.