NPU SPECIFICATIONS:-
Processor: Cortex [email protected] + RISC-V
NPU: 0.5TOPS, supports int4, int8 and int16
Memory: 64MB DDR2
*6TOPS@int8
*support int4/int8/FP16/BF16/TF32
*support deep learning frameworks :-TensorFlow,Caffe, Tflite,Pytorch,Onnx NN,Android NN,etc
*Memory:-4GB
sudo apt install python3-pip
harish@DESKTOP-JCC3N1V:/mnt/c/Users/haris/Downloads/p$ pip3 install -r rknn-toolkit2/rknn-toolkit2/packages/requirements_cp310-1.6.0.txt
https://www.scs.stanford.edu/~zyedidia/docs/rockchip/
Network connection test:- https://www.youtube.com/watch?v=9rojGc7OKao
https://www.youtube.com/watch?v=ILmedtqtrQ8&t=79s
https://www.youtube.com/watch?v=N_l7rZeAxKs
https://www.youtube.com/watch?v=DMWD7wfhgNY&t=102s
installation sucefull with
pip install rknn-toolkit2 -i https://mirrors.aliyun.com/pypi/simple/
(rknn) ubuntu@DESKTOP-MLTHVPE:/mnt/c/Users/haris/Downloads/p/Projects/p/rknn-toolkit2$ pip install -r rknn-toolkit2/packages/requirements_cp39-1.6.0.txt
ADB is not working
winscp is successfule with transfering data from pc to target board
wsl --shutdown Ubuntu
g++ set using windows:- setx PATH "%PATH%;D:\mingw64\bin"
tasks.json
{ "version": "2.0.0", "tasks": [ { "type": "shell", "label": "C/C++: g++.exe build active file", "command": "D:/mingw64/bin/g++.exe", "args": [ "-fdiagnostics-color=always", "-g", "${file}", "-o", "${fileDirname}\${fileBasenameNoExtension}.exe" ], "group": { "kind": "build", "isDefault": true }, "problemMatcher": ["$gcc"] } ] }
Enabling intellisense:-
"intelliSenseMode": "gcc-x64",
~/miniconda3/bin/conda init bash
source ~/.bashrc
conda --version
rkllm:-pip install --force-reinstall "setuptools<81"
A good progression is:
Symmetric INT8 quantization
Asymmetric INT8 quantization
Per-channel quantization
Per-group quantization
SmoothQuant
GPTQ
AWQ
AIMET PTQ
FP8 quantization
NF4 / QLoRA quantization
g++ path in windows terminal
set PATH=C:\msys64\ucrt64\bin;%PATH%
g++ path set in vscode terminal
$env:Path = "C:\msys64\ucrt64\bin;" + $env:Path
Resnet:-
