Unlike Jetson Nano, Xavier NX can install SSD. This allows for much larger storage than the Jetson Nano.
When using the Jetson Nano, I made and used various SD cards depending on the purpose. However, in Xavier NX, you can create a development environment using ample SSD storage without replacing the SD card.
The development language will probably use Python and C/C++ heavily. The most difficult part of building a development environment in Xavier NX is the Python development environment. Many machine learning software use the Python language. And it requires a specific Python version, Numpy version. Therefore, if you install two or more machine learning software on one system, the version conflict problem of Python and Python modules occurs. The following figure summarizes the TensorFlow version and required Python, compiler, and CUDA versions. You may need more than one version of Python if you want to use more than one TensorFlow.
To solve this problem, we use Python virtualization. There are two main ways to use Python virtualization. First of all, you can create a virtual environment using venv, a built-in module of Python. And the other way is to install and use Anaconda. However, using venv has one drawback. venv can manage the versions of modules that are installed by creating multiple virtual environments, but cannot change the Python version.
However, Anaconda has the advantage of specifying the desired Python version while creating a virtual environment. Therefore, you can create a virtual environment much more flexibly than venv. Therefore, many Python developers prefer Anaconda over Python's built-in venv.
Therefore, I will create a virtual environment using Anaconda. And we will install software such as PyTorch, Tensorflow, and YOLO in this virtual environment.
Anaconda
Anaconda is a conditional free and open source software name that facilitates package/dependency management and distribution of Python and R languages, suitable for scientific research and machine learning fields.
It was originally free, but starting in 2020, it is free only for individual users, universities, non-profit organizations, and small and medium-sized enterprises (SMEs) with less than 200 employees, and has been changed to a fee for the government and companies with 200 or more employees.
Anaconda includes most of the packages suitable for scientific research and machine learning applications. Therefore, there is a drawback that the size is quite large. Miniconda is a solution to this problem. Unlike anaconda, only the minimum required for operation is provided, and the user has to find and install the necessary packages himself. However, for this reason, the file is smaller and lighter than Anaconda.
Miniconda is a free minimal installer for conda. It is a small, bootstrap version of Anaconda that includes only conda, Python, the packages they depend on, and a small number of other useful packages, including pip, zlib and a few others. Use the conda install command to install 720+ additional conda packages from the Anaconda repository.
I'm going to install Anaconda on the Xavier NX. Because I am using an SSD with sufficient capacity.
Installing Anaconda on Xavier NX
There are instructions for installing Anaconda at https://www.anaconda.com/products/individual#Downloads.
Install Anaconda
After downloading the installation script, give permission for execution and install it.
wget https://repo.anaconda.com/archive/Anaconda3-2022.10-Linux-aarch64.sh
chmod 755 Anaconda3-2022.10-Linux-aarch64.sh
./Anaconda3-2022.10-Linux-aarch64.sh
In the previous article, I installed the SSD while installing JetPack 5.1. And I mounted /home/user (spypiggy) on the SSD. Therefore, installing all possible software into /home/spypiggy is a good way to utilize SSD storage.
When you run the Anaconda3-2022.10-Linux-aarch64.sh script, it asks for the installation path as follows. You must select the mounted path to obtain the effect of installing the SSD.
Do you accept the license terms? [yes|no] [no] >>> yes Anaconda3 will now be installed into this location: /home/spypiggy/anaconda3 - Press ENTER to confirm the location - Press CTRL-C to abort the installation - Or specify a different location below [/home/spypiggy/anaconda3] >>> PREFIX=/home/spypiggy/anaconda3 Unpacking payload ...
When Anaconda is installed, the following content is added to the ~/.bashrc file. So, whenever you open a new shell in the future, these lines will be executed.
# >>> conda initialize >>> # !! Contents within this block are managed by 'conda init' !! __conda_setup="$('/home/spypiggy/anaconda3/bin/conda' 'shell.bash' 'hook' 2> /dev/null)" if [ $? -eq 0 ]; then eval "$__conda_setup" else if [ -f "/home/spypiggy/anaconda3/etc/profile.d/conda.sh" ]; then . "/home/spypiggy/anaconda3/etc/profile.d/conda.sh" else export PATH="/home/spypiggy/anaconda3/bin:$PATH" fi fi unset __conda_setup # <<< conda initialize <<<
And in the new shell prompt, (base) is added as follows:
(base) spypiggy@spypiggy-desktop:~$
If you do not want to initialize anaconda automatically, you can delete the conda initialization part from your ~/.bashrc file. And only one line related to the following path is left in the ~/.bashrc file.
# >>> conda initialize >>>export PATH="/home/spypiggy/anaconda3/bin:$PATH"# <<< conda initialize <<<
Creating a new conda virtual environment
JetPack 5.1 comes with Python 3.8 installed . However, in Anaconda, you can choose from a variety of Python versions, such as:
(base) spypiggy@spypiggy-NX:/usr/local$ conda search python Loading channels: done # Name Version Build Channel python 3.7.10 he65049a_0 pkgs/main python 3.7.10 he65049a_1 pkgs/main python 3.7.10 he65049a_3 pkgs/main python 3.7.11 hc137634_0 pkgs/main python 3.7.13 h6c906d0_1 pkgs/main python 3.7.13 hc137634_0 pkgs/main python 3.7.15 h6c906d0_0 pkgs/main python 3.7.15 h89984f6_1 pkgs/main python 3.7.16 h89984f6_0 pkgs/main python 3.8.8 he65049a_4 pkgs/main python 3.8.8 he65049a_5 pkgs/main python 3.8.8 he65049a_7 pkgs/main python 3.8.11 hc137634_0 pkgs/main python 3.8.11 hc137634_1 pkgs/main python 3.8.12 hc137634_0 pkgs/main python 3.8.13 h6c906d0_1 pkgs/main python 3.8.13 hc137634_0 pkgs/main python 3.8.15 h6c906d0_0 pkgs/main python 3.8.15 h89984f6_2 pkgs/main python 3.8.16 h89984f6_2 pkgs/main python 3.8.16 h89984f6_3 pkgs/main python 3.9.1 he65049a_6 pkgs/main python 3.9.1 he65049a_7 pkgs/main python 3.9.4 he65049a_0 pkgs/main python 3.9.4 he65049a_2 pkgs/main python 3.9.5 hc137634_3 pkgs/main python 3.9.6 hc137634_0 pkgs/main python 3.9.6 hc137634_1 pkgs/main python 3.9.7 hc137634_1 pkgs/main python 3.9.11 hc137634_1 pkgs/main python 3.9.11 hc137634_2 pkgs/main python 3.9.12 hc137634_0 pkgs/main python 3.9.12 hc137634_1 pkgs/main python 3.9.13 h6c906d0_1 pkgs/main python 3.9.13 h6c906d0_2 pkgs/main python 3.9.15 h6c906d0_0 pkgs/main python 3.9.15 h89984f6_2 pkgs/main python 3.9.16 h89984f6_0 pkgs/main python 3.9.16 h89984f6_1 pkgs/main python 3.9.16 h89984f6_2 pkgs/main python 3.10.0 hc137634_0 pkgs/main python 3.10.0 hc137634_2 pkgs/main python 3.10.0 hc137634_3 pkgs/main python 3.10.0 hc137634_4 pkgs/main python 3.10.0 hc137634_5 pkgs/main python 3.10.3 hc137634_5 pkgs/main python 3.10.4 hc137634_0 pkgs/main python 3.10.6 h6c906d0_0 pkgs/main python 3.10.6 h6c906d0_1 pkgs/main python 3.10.8 h6c906d0_0 pkgs/main python 3.10.8 h89984f6_1 pkgs/main python 3.10.9 h89984f6_0 pkgs/main python 3.10.9 h89984f6_1 pkgs/main python 3.10.9 h89984f6_2 pkgs/main python 3.11.0 h89984f6_2 pkgs/main python 3.11.0 h89984f6_3 pkgs/main
I will install Python 3.8 which is the same as Python installed in JetPack 5.1. Only minor versions will be upgraded from 3.8.10 to 3.8.16.
spypiggy@spypiggy-NX:~/Downloads$ conda create -n yolov8 python=3.8.16 Collecting package metadata (current_repodata.json): done Solving environment: done ## Package Plan ## environment location: /home/spypiggy/anaconda3/envs/yolov8 added / updated specs: - python=3.8.16 The following packages will be downloaded: package | build ---------------------------|----------------- certifi-2022.12.7 | py38hd43f75c_0 149 KB libffi-3.4.2 | h419075a_6 135 KB pip-23.0.1 | py38hd43f75c_0 2.5 MB python-3.8.16 | h89984f6_3 12.5 MB setuptools-65.6.3 | py38hd43f75c_0 1.1 MB wheel-0.38.4 | py38hd43f75c_0 63 KB ------------------------------------------------------------ Total: 16.5 MB The following NEW packages will be INSTALLED: _libgcc_mutex pkgs/main/linux-aarch64::_libgcc_mutex-0.1-main None _openmp_mutex pkgs/main/linux-aarch64::_openmp_mutex-5.1-51_gnu None ca-certificates pkgs/main/linux-aarch64::ca-certificates-2023.01.10-hd43f75c_0 None certifi pkgs/main/linux-aarch64::certifi-2022.12.7-py38hd43f75c_0 None ld_impl_linux-aar~ pkgs/main/linux-aarch64::ld_impl_linux-aarch64-2.38-h8131f2d_1 None libffi pkgs/main/linux-aarch64::libffi-3.4.2-h419075a_6 None libgcc-ng pkgs/main/linux-aarch64::libgcc-ng-11.2.0-h1234567_1 None libgomp pkgs/main/linux-aarch64::libgomp-11.2.0-h1234567_1 None libstdcxx-ng pkgs/main/linux-aarch64::libstdcxx-ng-11.2.0-h1234567_1 None ncurses pkgs/main/linux-aarch64::ncurses-6.4-h419075a_0 None openssl pkgs/main/linux-aarch64::openssl-1.1.1t-h2f4d8fa_0 None pip pkgs/main/linux-aarch64::pip-23.0.1-py38hd43f75c_0 None python pkgs/main/linux-aarch64::python-3.8.16-h89984f6_3 None readline pkgs/main/linux-aarch64::readline-8.2-h998d150_0 None setuptools pkgs/main/linux-aarch64::setuptools-65.6.3-py38hd43f75c_0 None sqlite pkgs/main/linux-aarch64::sqlite-3.40.1-hf533066_0 None tk pkgs/main/linux-aarch64::tk-8.6.12-h241ca14_0 None wheel pkgs/main/linux-aarch64::wheel-0.38.4-py38hd43f75c_0 None xz pkgs/main/linux-aarch64::xz-5.2.10-h998d150_1 None zlib pkgs/main/linux-aarch64::zlib-1.2.13-h998d150_0 None Proceed ([y]/n)? y Downloading and Extracting Packages wheel-0.38.4 | 63 KB | ############################################################################## | 100% libffi-3.4.2 | 135 KB | ############################################################################## | 100% certifi-2022.12.7 | 149 KB | ############################################################################## | 100% pip-23.0.1 | 2.5 MB | ############################################################################## | 100% setuptools-65.6.3 | 1.1 MB | ############################################################################## | 100% python-3.8.16 | 12.5 MB | ############################################################################## | 100% Preparing transaction: done Verifying transaction: done Executing transaction: done # # To activate this environment, use # # $ conda activate yolov8 # # To deactivate an active environment, use # # $ conda deactivate Retrieving notices: ...working... done
You can see that packages such as openssl, ncurses, pip and sqlite are installed together with Python 3.9.7. Packages included in Anaconda can be added with the conda install command, and packages not included can be added with the pip install command.
You can check the newly created anaconda virtual environment with the conda env list command.
spypiggy@spypiggy-desktop:~$ conda env list # conda environments: # base * /home/spypiggy/anaconda3 yolov8 /home/spypiggy/anaconda3/envs/yolov8
Wrapping Up
Machine learning packages each require their own version of Python and packages. When multiple packages are run on one Xavier NX, package version problems inevitably occur. The best way to solve this problem is to use Anaconda . Anaconda has the advantage of specifying the Python version at the time of creating a virtual environment, so various Python versions can be run on a single host. In the next article, I will installPyTorch and yolov8 in the Anaconda environment.
댓글 없음:
댓글 쓰기