Repolode
Submit
airockchip

airockchip/rknn-toolkit2

SDK for converting trained AI models to RKNN format and running inference on Rockchip NPU boards via Python and C APIs.

View on GitHub
Stars
3,463
Forks
409
Open issues
473
Last commit
10 hours ago

What is it?

RKNN-Toolkit2 is Rockchip's software development kit for deploying AI models to its NPU-equipped chips. A developer runs the toolkit on a PC to convert a trained model into the RKNN format, then runs inference on the target board using the Python API (RKNN-Toolkit-Lite2) or the C/C++ RKNN Runtime API.

The PC toolkit also covers model conversion, inference simulation and performance evaluation. The project lists support for RK3588, RK3576, RK3566/RK3568, RK3562, several RV11xx chips and RK2118. The RKNPU kernel driver that talks to the hardware is described as open source and is part of Rockchip's kernel code.

The repository notes that RKNN-Toolkit2 is not compatible with the older RKNN-Toolkit used for RK1808, RV1109, RV1126 and RK3399Pro. For LLM deployment, Rockchip points to a separate RKNN-LLM SDK.

Key features

  • 01Converts trained models into the RKNN format for Rockchip NPUs.
  • 02Provides a Python inference API (RKNN-Toolkit-Lite2) for boards.
  • 03Offers a C/C++ runtime API for deploying RKNN models.
  • 04Includes model conversion and performance evaluation on PC.
  • 05Supports multiple Rockchip chip families, from RK3588 to RV1103/RV1106.

Use cases

  • Deploying computer vision models on Rockchip-based edge devices.
  • Benchmarking a converted model's performance before shipping it on an RK3588 board.
  • Running RKNN inference from a Python or C++ application on an embedded Linux board.

Quick start

  1. 1

    Download the packages, docker image and examples from the RKNPU2_SDK link in the README

  2. 2

    Get more example projects from the rknn_model_zoo repository

Things to know

  • Targets Rockchip NPU platforms only, so inference requires supported Rockchip boards.
  • The license is not clearly specified (NOASSERTION), so check the terms before commercial use.
  • RKNN-Toolkit2 is not compatible with the older RKNN-Toolkit, and older chips such as RK1808 and RV1126 use separate repositories.

Topics

#npu#model-conversion#edge-ai#embedded#inference#rockchip

FAQ

What is rknn-toolkit2?
SDK for converting trained AI models to RKNN format and running inference on Rockchip NPU boards via Python and C APIs. RKNN-Toolkit2 is Rockchip's software development kit for deploying AI models to its NPU-equipped chips. A developer runs the toolkit on a PC to convert a trained model into the RKNN format, then runs inference on the target board using the Python API (RKNN-Toolkit-Lite2) or the C/C++ RKNN Runtime API.
Is rknn-toolkit2 free to use? What license does it have?
rknn-toolkit2 is open source under the Other license. Check the license terms before using it in commercial products.
Is rknn-toolkit2 still maintained?
Actively maintained. The last commit to rknn-toolkit2 was on Oct 20, 2026.
What can I use rknn-toolkit2 for?
Deploying computer vision models on Rockchip-based edge devices. Benchmarking a converted model's performance before shipping it on an RK3588 board. Running RKNN inference from a Python or C++ application on an embedded Linux board.
How do I install rknn-toolkit2?
Download the packages, docker image and examples from the RKNPU2_SDK link in the README Get more example projects from the rknn_model_zoo repository
Where can I get rknn-toolkit2?
The source code and installation instructions are on GitHub: https://github.com/airockchip/rknn-toolkit2

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects