Development environment preparation

3. Development environment preparation#

This section describes the development environment preparations before using the Pulsar2 toolchain.

Pulsar2 uses Docker container for toolchain integration. Users can load Pulsar2 image files through Docker, and then perform model conversion, compilation, simulation, etc. Therefore, in the development environment preparation stage, you only need to correctly install the Docker environment. Supported systems are MacOS, Linux, Windows.

3.1. Install the Docker development environment#

The Docker development environment can be installed on MacOS, Linux, and Windows operating systems. For the minimum configuration requirements and specific installation procedures for the installation environment under different operating systems, please refer to the following links:

After Docker is successfully installed, enter sudo docker -v

$ sudo docker -v
Docker version 20.10.7, build f0df350

The above content is displayed, indicating that Docker has been installed successfully. The following will introduce the installation and startup of the Pulsar2 toolchain Image.

3.2. Install Pulsar2 toolchain#

Take the system version of Ubuntu 20.04 and the tool chain ax_pulsar2_${version}.tar.gz as an example to explain how to install the Pulsar2 tool chain.

Hint

In actual operation, be sure to replace ${version} with the corresponding toolchain version number.

How to get the toolchain: https://huggingface.co/AXERA-TECH/Pulsar2/tree/main/4.2

3.2.1. Load Docker Image#

Execute sudo docker load -i ax_pulsar2_${version}.tar.gz to import the docker image file. The following log will be printed if the image file is imported correctly:

$ sudo docker load -i ax_pulsar2_${version}.tar.gz
Loaded image: pulsar2:${version}

Once completed, execute sudo docker image ls

$ sudo docker image ls
REPOSITORY   TAG          IMAGE ID       CREATED         SIZE
pulsar2      ${version}   xxxxxxxxxxxx   9 seconds ago   3.27GB

You can see that the toolchain image has been successfully loaded, and then you can start the container based on this image.

3.2.2. Start the toolchain image#

Execute the following command to start the Docker container. After successful operation, enter the bash environment

$ sudo docker run -it --net host --rm -v $PWD:/data pulsar2:${version}

3.2.2.1. Version Query#

pulsar2 version is used to get the version information of the tool.

Example Results

root@xxx:/data# pulsar2 version
version: ${version}
commit: xxxxxxxx

3.2.2.2. Data preparation#

Hint

The original model , data , image , simulation tool required for model compilation and simulation running are already provided in the quick_start_example folder Click to download the example file Then unzip the downloaded file and copy it to the /data path of docker.

root@xxx:~/data# ls
config  dataset  model  output  pulsar2-run-helper

  • model: stores the original ONNX model mobilenetv2-sim.onnx (the calculation graph of onnxsim has been optimized in advance)

  • dataset: stores the compressed package of the dataset required for offline quantitative calibration (PTQ Calibration) (supports common compression formats such as tar, tar.gz, gz, etc.)

  • config: stores the configuration file config.json that depends on the operation

  • output: stores the result output

  • pulsar2-run-helper: a tool that supports axmodel to perform simulation in the X86 environment

After data preparation is completed, the directory tree structure is as follows:

root@xxx:/data# tree -L 2
.
├── config
│   ├── mobilenet_v2_build_config.json
│   └── yolov5s_config.json
├── dataset
│   ├── coco_4.tar
│   └── imagenet-32-images.tar
├── model
│   ├── mobilenetv2-sim.onnx
│   └── yolov5s.onnx
├── output
└── pulsar2-run-helper
    ├── cli_classification.py
    ├── cli_detection.py
    ├── models
    ├── pulsar2_run_helper
    ├── requirements.txt
    ├── setup.cfg
    ├── sim_images
    ├── sim_inputs
    └── sim_outputs