Introduction
In order to run customized models on Kneron AI device, there are four stages are involved:
-
Model Development
This will not be introduced in this document.
-
PLUS Development
The software interface for sending requests to firmware and receiving results from firmware.
-
SCPU Firmware Development
The entry of the firmware. Which models should be run on NCPU Firmware and in what sequence should these models run are determined in SCPU Firmware.
-
NCPU Firmware Development
Where models actually run. Besides the model inference, the preprocess and the postprocess can be chosen to run on NCPU Firmware.
The diagram below demonstrates the inference flow for every models runnig on Kneron AI device, and how the PLUS, SCPU, and NCPU interact with each other.
Note: Multiple calls of kp_customized_inference_send()
without any call of kp_customized_inference_receive()
may cause the system stuck.