RK3582 NPU Platform Specific Instructions
The RK3582 NPU supports mixed operations of INT4/INT8/INT16/FP16 with computing power up to 5TOPs. Users can easily convert network models based on frameworks such as TensorFlow/MXNet/PyTorch/Caffe, and can use RKNN and RKLLM frameworks for accelerated model inference.
To prevent users from misunderstanding when consulting documentation about the RK3582 NPU series, here is a unified explanation regarding specifying NPU platform models for RK3582
Currently, the RKNN Official Manual states:
Supported chips: RK3566/RK3568, RK3588, RK3562, RV1103/RV1106 series
Currently, the RKLLM Official Manual states:
Supported chips: RK3588/RK3576
Explanation
For products equipped with the RK3582 SoC, such as ROCK5C Lite and E54C, when using RKNN and RKLLM for model conversion or inference, the platform to specify for models is rk3588
, meaning that models for RK3588 can be used on the RK3582 platform.
Documentation regarding RK3588 NPU usage is universally applicable to RK3582.