Machine Learning

HAL (Hardware Abstraction Layer) modules for the Machine Learning framework provide an interface for hardware-accelerated neural network inference.

ML Hardware Acceleration

At the Hardware Abstraction Layer, the ML acceleration feature handles inference requests from Tizen’s Machine Learning API and delegates them to various hardware acceleration devices. This HAL is not used directly by application developers; instead, it serves as a backend for NNStreamer.

The process is as follows:

  • An application uses the Tizen ML API.
  • The underlying framework (NNStreamer’s tizen-hal tensor_filter subplugin) calls the ML HAL API.
  • The ML HAL API routes the request to a specific HAL ML Accelerator Backend.
  • The backend executes the task on the target device and returns the result.

You may check the HAL usage in the tensor_filter implementation here.

Edit this page