Skip to main content

RKNN Toolkit Lite2

1 Toolkit Lite2 Installation​

RKNN Toolkit Lite2 is a programming interface (Python) for the Rockchip NPU platform, used for deploying RKNN models on the board.

Test Environment​

• System Version: Debian 12 • Tool Version: RKNN-Toolkit2 2.3.0 • Driver Version: NPU driver 0.8.8

Installation Steps​

Toolkit-lite2 is suitable for board-side model deployment. For more dependencies and usage information, please check Rockchip_RKNPU_User_Guide_RKNN_SDK

To get RKNN Toolkit Lite2 on the board, you can download directly from official github

  1. Get Installation Files:

    git clone https://github.com/airockchip/rknn-toolkit2.git
    cd rknn_toolkit_lite2/

  2. Install Dependencies:

    sudo apt update
    sudo apt-get install python3-dev python3-pip gcc
    sudo apt install -y python3-opencv python3-numpy python3-setuptools

  3. Install Package:

    # Debian 12 (Python 3.10)
    pip3 install packages/rknn_toolkit_lite2-2.3.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl

  4. Verify Installation:

    from rknnlite.api import RKNNLite  # No error means successful installation

2 Toolkit Lite2 Interface Usage​

Deployment Process​

  1. Create RKNNLite object
  2. Call load_rknn to import the model (must match the hardware platform)
  3. Call init_runtime to initialize the runtime environment
  4. Call inference for inference
  5. Process inference results
  6. Call release to release resources

Interface Documentation​

Refer to the user manual in the rknn_toolkit_lite2/docs directory.

3 Board-Side Inference Test​

Precautions​

• Ensure the board has the librknnrt.so runtime library installed (default path: /usr/lib) • The version must match RKNN-Toolkit2 to avoid compatibility issues (e.g., error Invalid RKNN model version)

3.1 Resnet18 Inference Test​

  1. Run Example: 
    cd examples/inference_with_lite
    python3 test.py
  2. Output Example: 
    --> Load RKNN model done
    --> Init runtime environment done
    --> Running model
    resnet18
    -----TOP 5-----
    [812]: 0.9996760487556458
    [404]: 0.00024927023332566023
    ...

4 References​

https://github.com/airockchip/rknn-toolkit2

https://github.com/rockchip-linux/rknn-toolkit2

https://github.com/rockchip-linux/rknpu2

Note
Ensure the RKNN model and runtime library versions are consistent to avoid compatibility issues.