Installing Miniconda

TensorFlow and other third-party packages on which DLRM data preprocessing and code training depend can be managed in a unified manner using Miniconda, avoiding incompatibility issues.

1. Use PuTTY to log in to the server as the root user.
2. Create a Miniconda installation directory.

   1	mkdir -p /path/to/miniconda3

3. Download miniconda.sh to /path/to/miniconda3.

   1	wget https://repo.anaconda.com/miniconda/Miniconda3-py39_24.3.0-0-Linux-aarch64.sh -O /path/to/miniconda3/miniconda.sh

4. Install Miniconda.

   bash /path/to/miniconda3/miniconda.sh -b -u -p /path/to/miniconda3

   Delete miniconda.sh, which you will no longer use.

   rm -rf  /path/to/miniconda3/miniconda.sh

5. (Optional) Run the following commands to initialize Miniconda:

   1 2	/path/to/miniconda3/bin/conda init bash /path/to/miniconda3/bin/conda init zsh

   After the installation is complete, the conda directory (/path/to/miniconda3/condabin) is added to the PATH environment variable.

6. Close the window of connecting to the server and reconnect to the server. The temporary proxy needs to be reconfigured. See Method 1: Configuring a Temporary Network Proxy.
7. Check whether Miniconda is installed.

   1

   conda -V

   

   If the command output shows that the version is 24.3.0, the installation is successful.

8. Configure the conda proxy environment.
   1. Create a .condarc file.

      1	vi /root/.condarc

   2. Press i to enter the insert mode and add the following content to the .condarc file.

      Set the IP address and port as required.

      channels:
        - defaults
      
      proxy_servers:
      
        http: http://Proxy_IP_address:Proxy_port
      
        https: http://Proxy_IP_address:Proxy_port
      
      
      ssl_verify: false
      
      show_channel_urls: true
      
      allow_other_channels: true
      

   3. Press Esc, type :wq!, and press Enter to save the settings and exit.