中文

Introduction

RKNN3 SDK provides the complete software stack for deploying AI models on RK1820/RK1828/RK3572, including:

• RKNN3-Toolkit: PC-side software development kit for model conversion, inference, performance evaluation, etc.
• RKNN3 Runtime: On-board runtime library providing C/C++ programming interfaces for deploying RKNN models and accelerating AI applications.
• RKNN3 Model Zoo: Model conversion and deployment example repository, including reference implementations for CNN / LLM / VLM and other models.

Typical Workflow: Users first convert their trained models to RKNN format using RKNN3-Toolkit on a PC, then perform inference on the development board via the RKNN3 Runtime API.

Supported Platforms

• RK1820
• RK1828
• RK3572

Note:

• For RK3588/RK3576/RK3568/RK3566/RK3562 series, RV1103/RV1106, RV1103B/RV1106B, RV1126B, RK2118, please refer to:

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

• For RK1808/RV1109/RV1126/RK3399Pro, please refer to:

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

https://github.com/airockchip/rknpu

https://github.com/airockchip/RK3399Pro_npu

• RKNN3 Model Zoo provides more conversion and deployment examples

https://github.com/airockchip/rknn3-model-zoo

Supported Models

• [x] Qwen2.5-0.5B / 1.5B / 3B / 7B
• [x] Qwen3-0.6B / 1.7B / 4B / 8B
• [x] HY-MT1.5-1.8B
• [x] Youtu-LLM-2B
• [x] GLM-Edge
• [x] Qwen2.5-VL-3B / 7B
• [x] Qwen2.5-Omni-3B (Thinker)
• [x] Qwen3-VL-2B / 4B
• [x] FastVLM-1.6B
• [x] InternVL3-2B
• [x] InternVL3.5-4B
• [x] MiMo-VL-7B-RL
• [x] Gemma4
• [x] SmolVLM
• [x] SmolVLM2
• [x] UI_TARS
• [x] PaddleOCR VL
• [x] Qwen3-Reranker-0.6B / 4B
• [x] Qwen3-Embedding-4B
• [x] gme-Qwen2-VL-2B
• [x] Qwen3-ASR
• [x] Qwen3_TTS
• [x] VITS
• [x] Whisper
• [x] SenseVoice
• [x] Zipformer
• [x] SigLIP
• [x] SigLIP2
• [x] MetaCLIP2
• [x] DINOv2
• [x] DINOv3
• [x] MobileNetV1 / V2
• [x] ResNet-50
• [x] YOLOv5 / YOLOv6 / YOLOv8
• [x] YOLO-World
• [x] Diffusion Policy
• [x] GR00T

Performance

For performance data, please refer to the Release Notes.

Notes

• RKNN3-Toolkit is not compatible with RKNN-Toolkit and RKNN-Toolkit2.

Supported Python Versions:

• Python 3.10
• Python 3.12

Latest Version: V1.0.4

Changelog

V1.0.4

• Added RK3572 platform support (Beta)
• Added Windows platform support
• Added Ethernet communication type support
• Added device suspension and wake-up support
• Added server-side auto-select communication method
• Added streaming weight loading support
• Added model encryption support
• Added LoRA support
• Added session pause/resume support
• Added session input callback support
• Added multi-session parallel running support
• Added KVCache import/export interface
• Added module version compatibility check
• Added Qwen3 TTS text-to-speech model
• Added Qwen3-ASR speech recognition model
• Added VITS audio synthesis model
• Added Zipformer model
• Added MetaCLIP2 model
• Added Gemma4 multimodal model
• Added GR00T model
• Added PaddleOCR VL model
• Added SmolVLM2 vision-language model

Feedback and Community Support

• Redmine (Recommended for reporting issues. Please contact sales or an FAE to get a Redmine account)
• QQ Group 1: 1025468710 (Full, please join group 5)
• QQ Group 2: 547021958 (Full, please join group 5)
• QQ Group 3: 469385426 (Full, please join group 5)
• QQ Group 4: 958083853 (Full, please join group 5)
• QQ Group 5: 1077888690