rknn-toolkit2 2.3.2

Software development kit (SDK) developed by Rockchip for their Neural Processing Unit (NPU) platforms, such as the RK3566, RK3588, and RV1103/RV1106 series

Sorting out versions

airockchip/rknn-toolkit2 → Current, official upstream.

This is where RKNN-Toolkit2 lives now. It carries active releases (e.g., v2.3.2) with broader platform coverage (RK3588, RK3576, RK356x, RV110x, RK2118, etc.), lists Python 3.6–3.12 support, and links out to RKNN-LLM if you’re deploying LLMs. If you’re building/compiling or following docs, start here.

PyPI’s rknn-toolkit2 page now points to the airockchip repo, reinforcing that it’s the canonical source,

Pelochus/ezrknn-toolkit2 → Community convenience fork.

A fork aimed at making installs on SBCs (e.g., Orange Pi 5 / Radxa Rock 5) easier via an install.sh. It focuses on ARM64, Debian/Ubuntu and intentionally skips extras like the simulator/server; it’s not the canonical repo, just an easier on-device setup path.

rockchip-linux/rknn-toolkit2 → Old home, deprecated.

That repo now says it’s no longer maintained and points you to the new airockchip/rknn-toolkit2

Install

# Git Repo
git clone https://github.com/airockchip/rknn-toolkit2.git

# Python install
pip install rknn-toolkit2

Examples :

rknn-llm/examples/DeepSeek-R1-Distill-Qwen-1.5B_Demo

Dependencies

sudo apt install cmake python3-scipy python3-h5py libprotobuf-dev protobuf-compiler zlib1g-dev libjpeg-dev python3-dev python3-opencv