RK3576 LINUX RKNN SDK 测试

Lary_Rock2024-11-08 5:04

安装Conda工具

bash 复制代码 
安装 Miniforge Conda
wget -c https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh
chmod 777 Miniforge3-Linux-x86_64.sh
bash Miniforge3-Linux-x86_64.sh

source ~/miniforge3/bin/activate # Miniforge 安装的目录


# 创建名称为 toolkit2 的 Python 3.8 环境
conda create -n toolkit2 python=3.8

# 激活toolkit2环境
conda activate toolkit2

工程代码下载

bash 复制代码 
# 新建 Projects 文件夹
mkdir Projects
# 进入该目录
cd Projects
# 下载 RKNN-Toolkit2 仓库
git clone https://github.com/airockchip/rknn-toolkit2.git --depth 1
# 下载 RKNN Model Zoo 仓库
git clone https://github.com/airockchip/rknn_model_zoo.git --depth 1
# 下载 rknn-llm 仓库
git clone https://github.com/airockchip/rknn-llm.git --depth 1

确认板端环境

安装依赖环境以及YOLOv5运行

bash 复制代码 
# 安装依赖库和 RKNN-Toolkit2
# 进入 rknn-toolkit2 目录
cd Projects/rknn-toolkit2/rknn-toolkit2
# 请根据不同的 python 版本，选择不同的 requirements 文件
# 例如 python3.8 对应 requirements_cp38.txt
pip install -r packages/requirements_cpxx.txt
# 安装 RKNN-Toolkit2
# 请根据不同的 python 版本及处理器架构，选择不同的 wheel 安装包文件：
# 其中 x.x.x 是 RKNN-Toolkit2 版本号，cpxx 是 python 版本号，请根据实际数值进行替换
pip install packages/rknn_toolkit2--x.x.x-cpxx-cpxx-manylinux_2_17_x86_64.manylinux2014_x86_64.whl

# 验证是否安装成功
# 进入 Python 交互模式
python
# 导入 RKNN 类
from rknn.api import RKNN

# 更新包列表
sudo apt update
# 安装 cmake
sudo apt install cmake


 # 进入 rknpu2 目录
cd Projects/rknn-toolkit2/rknpu2
# 推送 rknn_server 到板端
# 注：在64位Linux系统中，BOARD_ARCH对应a
# adb push runtime/Linux/rknn_server/${BOARD_ARC
adb push  runtime\Linux\rknn_server\aarch64\usr\bin\rknn_server  /usr/bin
adb push  runtime\Linux\rknn_server\aarch64\usr\bin\start_rknn.sh  /usr/bin
adb push  runtime\Linux\rknn_server\aarch64\usr\bin\restart_r knn.sh  /usr/bin
# 推送 librknnrt.so
# adb push runtime/Linux/librknn_api/${BOARD_ARC
adb push runtime\Linux\rlibrknn_api\aarch64\librknnrt.so   / usr/lib
# 进入板端
adb shell
# 赋予可执行权限
chmod +x /usr/bin/rknn_server
chmod +x /usr/bin/start_rknn.sh
chmod +x /usr/bin/restart_rknn.sh
# 重启 rknn_server 服务
restart_rknn.sh



 RKNN C Demo 使用方法
 
准备模型
# 进入 rknn_model_zoo/examples/yolov5/model 目录
cd Projects/rknn_model_zoo/examples/yolov5/model
# 运行 download_model.sh 脚本，下载 yolov5 onnx 模型
# 例如，下载好的 onnx 模型存放路径为 model/yolov5s_relu.onnx
./download_model.sh



模型转换
# 进入 rknn_model_zoo/examples/yolov5/python 目录
cd Projects/rknn_model_zoo/examples/yolov5/python
# 运行 convert.py 脚本，将原始的 ONNX 模型转成 RKNN 模型
# 用法: python convert.py model_path [rk3566|rk3588|rk3562] [i8/fp] [output_path]
python convert.py ../model/yolov5s_relu.onnx rk3576 i8 ../model/yolov5s_relu.rknn    
    
# 添加到 build-linux.sh 脚本的开头位置即可
GCC_COMPILER=Projects/gcc-linaro-6.3.1-2017.05-x86_64_aarch64-linux-gnu/bin/aarch64-linux-gnu

# 用的是sdk自带的gcc-arm-10.3-2021.07-x86_64-aarch64-none-linux-gnu 
# 路径 linux/prebuilts/gcc/linux-x86/aarch64/gcc-arm-10.3-2021.07-x86_64-aarch64-none-linux-gnu
--- a/build-linux.sh
+++ b/build-linux.sh
@@ -1,5 +1,7 @@
 #!/bin/bash

+GCC_COMPILER=Projects/gcc-arm-10.3-2021.07-x86_64-aarch64-none-linux-gnu/bin/aarch64-rockchip1031-linux-gnu
+
 set -e

 echo "$0 $@"
 # 编译之前修改 库文件
   cd 3rdparty/mpp/Linux/aarch64/
   git status  ./
   rm librockchip_mpp.so
   mv librockchip_mpp.so.0  librockchip_mpp.so
   cd ..
   cd rknn_yolov5_demo/

 
 # 编译
 ./build-linux.sh -t rk3576 -a aarch64 -b Release
 (toolkit2) ip3@ip3-MOMENTPLUS-M2:~/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo$ ./build-linux.sh -t rk3576 -a aarch64 -b Release
./build-linux.sh -t rk3576 -a aarch64 -b Release
/home/ip3/work/linux/Projects/gcc-arm-10.3-2021.07-x86_64-aarch64-none-linux-gnu/bin/aarch64-rockchip1031-linux-gnu
===================================
TARGET_SOC=RK3576
TARGET_ARCH=aarch64
BUILD_TYPE=Release
BUILD_DIR=/home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/build/build_RK3576_linux_aarch64_Release
CC=/home/ip3/work/linux/Projects/gcc-arm-10.3-2021.07-x86_64-aarch64-none-linux-gnu/bin/aarch64-rockchip1031-linux-gnu-gcc
CXX=/home/ip3/work/linux/Projects/gcc-arm-10.3-2021.07-x86_64-aarch64-none-linux-gnu/bin/aarch64-rockchip1031-linux-gnu-g++
===================================
-- Configuring done
-- Generating done
-- Build files have been written to: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/build/build_RK3576_linux_aarch64_Release
Consolidate compiler generated dependencies of target rknn_yolov5_video_demo
Consolidate compiler generated dependencies of target rknn_yolov5_demo
[ 60%] Built target rknn_yolov5_video_demo
[100%] Built target rknn_yolov5_demo
[ 40%] Built target rknn_yolov5_demo
[100%] Built target rknn_yolov5_video_demo
Install the project...
-- Install configuration: "Release"
-- Up-to-date: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/./rknn_yolov5_demo
-- Up-to-date: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/lib/librknnrt.so
-- Up-to-date: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/lib/librga.so
-- Up-to-date: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/./model/RK3576
-- Up-to-date: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/./model/RK3576/yolov5s-640-640.rknn
-- Up-to-date: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/./model/bus.jpg
-- Up-to-date: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/./model/coco_80_labels_list.txt
-- Up-to-date: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/./rknn_yolov5_video_demo
-- Up-to-date: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/lib/librockchip_mpp.so
-- Up-to-date: /home/ip3/work/linux/Projects/rknn-toolkit2/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/lib/libmk_api.so


  
 
 # 进入 rknn_model_zoo 目录
cd Projects/rknn_model_zoo
# 运行 build-linux.sh 脚本
# 用法:./build-linux.sh -t <target> -a <arch> -d <build_demo_name> [-b <build_type>] [-m]
# -t : target (rk356x/rk3588) # 平台类型，rk3568/rk3566 都统一为rk356x
# -a : arch (aarch64/armhf) # 板端系统架构
# -d : demo name # 对应 examples 目录下子文件夹的名称，如yolov5、mobilenet
# -b : build_type(Debug/Release)
# -m : enable address sanitizer, build_type need set to Debug
chmod +x build-linux.sh
./build-linux.sh -t rk3576 -a aarch64 -d yolov5

(toolkit2) ip3@ip3-MOMENTPLUS-M2:~/work/linux/Projects/rknn_model_zoo$ ./build-linux.sh -t rk3576 -a aarch64 -d yolov5
./build-linux.sh -t rk3576 -a aarch64 -d yolov5
/home/ip3/work/linux/Projects/gcc-arm-10.3-2021.07-x86_64-aarch64-none-linux-gnu/bin/aarch64-rockchip1031-linux-gnu
===================================
BUILD_DEMO_NAME=yolov5
BUILD_DEMO_PATH=examples/yolov5/cpp
TARGET_SOC=rk3576
TARGET_ARCH=aarch64
BUILD_TYPE=Release
ENABLE_ASAN=OFF
DISABLE_RGA=OFF
INSTALL_DIR=/home/ip3/work/linux/Projects/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov5_demo
BUILD_DIR=/home/ip3/work/linux/Projects/rknn_model_zoo/build/build_rknn_yolov5_demo_rk3576_linux_aarch64_Release
CC=/home/ip3/work/linux/Projects/gcc-arm-10.3-2021.07-x86_64-aarch64-none-linux-gnu/bin/aarch64-rockchip1031-linux-gnu-gcc
CXX=/home/ip3/work/linux/Projects/gcc-arm-10.3-2021.07-x86_64-aarch64-none-linux-gnu/bin/aarch64-rockchip1031-linux-gnu-g++
===================================
-- Configuring done
-- Generating done
-- Build files have been written to: /home/ip3/work/linux/Projects/rknn_model_zoo/build/build_rknn_yolov5_demo_rk3576_linux_aarch64_Release
Consolidate compiler generated dependencies of target imageutils
Consolidate compiler generated dependencies of target fileutils
Consolidate compiler generated dependencies of target imagedrawing
Consolidate compiler generated dependencies of target audioutils
[ 25%] Built target imagedrawing
[ 41%] Built target fileutils
[ 50%] Built target audioutils
[ 66%] Built target imageutils
Consolidate compiler generated dependencies of target rknn_yolov5_demo
[100%] Built target rknn_yolov5_demo
[ 16%] Built target imageutils
[ 33%] Built target fileutils
[ 50%] Built target imagedrawing
[ 83%] Built target rknn_yolov5_demo
[100%] Built target audioutils
Install the project...
-- Install configuration: "Release"
-- Installing: /home/ip3/work/linux/Projects/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov5_demo/./rknn_yolov5_demo
-- Set runtime path of "/home/ip3/work/linux/Projects/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov5_demo/./rknn_yolov5_demo" to "$ORIGIN/lib"
-- Installing: /home/ip3/work/linux/Projects/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov5_demo/./model/bus.jpg
-- Installing: /home/ip3/work/linux/Projects/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov5_demo/./model/coco_80_labels_list.txt
-- Installing: /home/ip3/work/linux/Projects/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov5_demo/model/yolov5s_relu.rknn
-- Installing: /home/ip3/work/linux/Projects/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov5_demo/lib/librknnrt.so
-- Installing: /home/ip3/work/linux/Projects/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov5_demo/lib/librga.so

#编译之后的文件
(toolkit2) ip3@ip3-MOMENTPLUS-M2:~/work/linux/Projects/rknn_model_zoo/install$ tree
.
└── rk3576_linux_aarch64
    └── rknn_yolov5_demo
        ├── lib
        │   ├── librga.so
        │   └── librknnrt.so
        ├── model
        │   ├── bus.jpg
        │   ├── coco_80_labels_list.txt
        │   └── yolov5s_relu.rknn
        └── rknn_yolov5_demo



# 推送文件到板端
# 进入 rknn_model_zoo 目录
cd Projects/rknn_model_zoo
# 推送整个 rknn_yolov5_demo 文件夹到板端
# 注：rknn_yolov5_demo 文件夹下有一个同名的可执行文件 rknn_yolov5_demo
# 注：使用不同的模型和平台时，建议直接在 install 下找对应的路径
adb push install/rk3576_linux_aarch64/rknn_yolov5_demo /data/



# 板端运行 Demo
# 进入板端
adb shell
# 进入 rknn_yolov5_demo 目录
cd /data/rknn_yolov5_demo/
# 设置依赖库环境
export LD_LIBRARY_PATH=./lib
# 运行可执行文件
# 用法: ./rknn_yolov5_demo <model_path> <input_path>
./rknn_yolov5_demo model/yolov5s_relu.rknn  model/bus.jpg


# 查看结果
# 拉取到本地当前目录
adb pull /data/rknn_yolov5_demo/out.png .

RKLLM部署

bash 复制代码 
# 代码 rknn-llm
# 环境 激活toolkit2环境
conda activate toolkit2

# 切换到rkllm-toolkit目录
cd rknn-llm/rkllm-toolkit/

# 安装rkllm_toolkit(文件请根据具体版本修改)，会自动下载RKLLM-Toolkit工具所需要的相关依赖包。
pip install packages/rkllm_toolkit-1.0.1-cp38-cp38-linux_x86_64.whl

#模型下载
cd Projects 
mkdir modem

Qwen-1_8B-Chat_downloader.py
from modelscope import snapshot_download
#model_dir = snapshot_download('qwen/Qwen-7B-Chat',cache_dir='./modem')
model_dir = snapshot_download('qwen/Qwen-1_8B-Chat',cache_dir='./modem')

运行Qwen-1_8B-Chat_downloader.py
python Qwen-1_8B-Chat_downloader.py



# 模型转换
# 在rknn-llm/rkllm-toolkit/examples/下新建 huggingface目录
# 新建download.py 代码如下:
from rkllm.api import RKLLM
# 模型路径
modelpath = '/home/ip3/work/linux/Projects/modem/qwen/Qwen-1_8B-Chat'
# 初始化RKLLM对象
llm = RKLLM()
# 模型加载
ret = llm.load_huggingface(model = modelpath)
if ret != 0:
    print('Load model failed!')
    exit(ret)
# 模型的量化构建
ret = llm.build(do_quantization=True, optimization_level=1, quantized_dtype='w8a8', target_platform='rk3576',num_npu_core=2)
if ret != 0:
    print('Build model failed!')
    exit(ret)
# 导出rkllm模型
ret = llm.export_rkllm("./qwen.rkllm")
if ret != 0:
    print('Export model failed!')
    exit(ret)



# RKLLM Runtime 编译
cd  Projects/rknn-llm/rkllm-runtime/examples/rkllm_api_demo

#设置编译工具连接

+++ b/rkllm-runtime/examples/rkllm_api_demo/build-linux.sh
@@ -1,10 +1,12 @@
 #!/bin/bash
+
 # Debug / Release / RelWithDebInfo
 if [[ -z ${BUILD_TYPE} ]];then
     BUILD_TYPE=Release
 fi

-GCC_COMPILER_PATH=~/gcc-arm-10.2-2020.11-x86_64-aarch64-none-linux-gnu/bin/aarch64-none-linux-gnu
+GCC_COMPILER_PATH=~/work/linux/Projects/gcc-arm-10.3-2021.07-x86_64-aarch64-none-linux-gnu/bin/aarch64-none-linux-gnu

(toolkit2) ip3@ip3-MOMENTPLUS-M2:~/work/linux/Projects/rknn-llm/rkllm-runtime/examples/rkllm_api_demo$ ./build-linux.sh
-- Configuring done
-- Generating done
-- Build files have been written to: /home/ip3/work/linux/Projects/rknn-llm/rkllm-runtime/examples/rkllm_api_demo/build/build_linux_aarch64_Release
Consolidate compiler generated dependencies of target llm_demo
Consolidate compiler generated dependencies of target multimodel_demo
[ 50%] Built target llm_demo
[100%] Built target multimodel_demo


(toolkit2) ip3@ip3-MOMENTPLUS-M2:~/work/linux/Projects/rknn-llm/rkllm-runtime/examples/rkllm_api_demo/build$ tree  -L 2
.
└── build_linux_aarch64_Release
    ├── CMakeCache.txt
    ├── CMakeFiles
    ├── cmake_install.cmake
    ├── llm_demo
    ├── Makefile
    └── multimodel_demo

2 directories, 5 files


# 推送板端

# 注意：要去到llm_demo 和 librkllmrt.so所在的目录下去执行
adb push llm_demo /data  # 将llm_demo使用ssh传输到开发板的家目录
adb push ./rkllm-runtime/runtime/Linux/librkllm_api/aarch64/librkllmrt.so /data/lib
adb push  Projects/rknn-llm/rkllm-toolkit/examples/huggingface/qwen.rkllm  /data

运行
adb shell  && cd data
cd librkllmrt.so所在的目录
export LD_LIBRARY_PATH=./lib

# 修改最大可打开文件的数量
ulimit -HSn 10240

root@linaro-alip:/data# ./llm_demo qwen.rkllm  128 4080
rkllm init start
I rkllm: rkllm-runtime version: 1.1.2, rknpu driver version: 0.9.8, platform: RK3576

# 或者运行绑定cpu运行测试
# taskset f0 ./llm_demo /path/qwen.rkllm
相关推荐
HIT_Weston12 分钟前
18、【Ubuntu】【远程开发】技术方案分析：私网ip掩码
linux·tcp/ip·ubuntu
cccccc语言我来了16 分钟前
(Linux （6）：从包管理到工具探索，构建系统操作基础认知)
linux·运维·服务器
8K超高清39 分钟前
高校巡展：中国传媒大学+河北传媒学院
大数据·运维·网络·人工智能·传媒
ben9518chen1 小时前
嵌入式Linux C语言程序设计九
linux·c语言
wuk9981 小时前
CentOS7环境搭建L2TP服务器
运维·服务器
恒创科技HK1 小时前
香港1核2G云服务器当网站服务器够用不？
运维·服务器
IT 小阿姨（数据库）2 小时前
PostgreSQL 之上的开源时序数据库 TimescaleDB 详解
运维·数据库·sql·postgresql·开源·centos·时序数据库
颜大哦2 小时前
linux安装mysql
linux·运维·mysql·adb
学习3人组2 小时前
Node.js 网站服务器开发
运维·服务器·node.js
来知晓2 小时前
Linux：WSL内存空间管理之清完内存C盘可用空间不增问题解决
linux·运维·服务器
热门推荐
01GitHub 镜像站点02UV安装并设置国内源03综合整理：pdf预览显示：你尝试预览的文件可能对你的计算机有害。如果你信任此文件以及其来源，请打开此文件以看其内容，如何解决以正常预览文件04BongoCat - 跨平台键盘猫动画工具05Linux下V2Ray安装配置指南06安娜的档案(Anna’s Archive) 镜像网站/国内最新可访问入口（持续更新）07jdk21下载、安装（Windows、Linux、macOS）08npm使用国内淘宝镜像的方法09PyCharm 社区版全平台安装指南10NVIDIA显卡驱动、CUDA、cuDNN 和 TensorRT 版本匹配指南