To work with RKNN through python you will need to use their packages, provided in official repository.

To simplify use I forked repository and only left packages files as well as default license. No need to copy whole repository and easier to find those files.

Straightforward folder structure:

• lite - rknn-toolkit-lite2
• arm64 - rknn-toolkit2 arm64
• x86_64 - rknn-toolkit2 x86_64

Go to file you want to use based on python version and architecture. Copy raw link to the filed and then install it with python.

I use UV package manger and would recommend to everybody else. Here is an example of command to install RKNN toolkit 2 on arm64:

uv add https://github.com/roman-koshchei/rknn-packages/raw/40a3df681ee7ea1d3540569264b53b447ac2d2f2/arm64/rknn_toolkit2-2.3.2-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl

You can also notice I use commit id instead of master branch, because it helps reproducibility.

If you want, you can same way install it from official repository, no need to trust me!

You can also just download the wheel file to your local machine and even put it inside of the repository. That's also a good way.