Notes

• Here are some links to some instructions for each step, but feel free to search for other resources on the internet if there are problems. Similarly, there are probably some missing instructions for MacOS, so try searching on the internet or find me ([email protected]) if you need further help
• The code probably runs with all Python distributions from 3.6 to 3.11, but I tested on these 2 setups and can confirm they work:
  • Python=3.11.5, pytorch=2.0.1, on CPU
  • Python=3.10.12, pytorch=2.0.1, on Colab GPUs (default versions for Google Colab)
• If you have a Linux distribution all commands given for WSL here can be used in the Linux terminal

There are multiple choices:

1. Run on Google Colab:

   1. Connect to GPUs and Mount drive

   2. Clone the repo (with GitHub access token: https://docs.github.com/en/authentication/keeping-your-account-and-data-secure/managing-your-personal-access-tokens#keeping-your-personal-access-tokens-secure)

   3. Run the train command with the “—cuda” argument. Example:

      ! python train.py \\
        --data data/en-sv/infopankki/prepared \\
        --source-lang sv \\
        --target-lang en \\
        --save-dir assignments/01/baseline/checkpoints \\
        --train-on-tiny \\
        --cuda
      

   4. Be careful about:

      1. Make sure the laptop doesn’t go into sleep mode after some inactive time (because Colab might disconnect then)

      2. There is a trick: https://stackoverflow.com/questions/57113226/how-can-i-prevent-google-colab-from-disconnecting

         1. On Windows, in Colab click on F12 or Fn + F12 to open developer tools, then open the console and enter one of the javascript functions given under the StackOverflow question. For me, this one works:

         function ConnectButton(){
           console.log("Connect pushed");
           document.querySelector("#top-toolbar > colab-connectbutton").shadowRoot.querySelector("#connect").click()
         }
         setInterval(ConnectButton,60000);
         

2. Run on local:

   1. Install VSCode and Python: https://www.datacamp.com/tutorial/setting-up-vscode-python
   2. Optional: Install WSL
      1. Install WSL on Windows (https://learn.microsoft.com/en-us/windows/wsl/install) and WSL extension on VSCode (https://marketplace.visualstudio.com/items?itemName=ms-vscode-remote.remote-wsl)
      2. ctr+shift+p to open the VScode command palette, then “WSL: Reopen folder in WSL”
   3. Python Environment:
      1. Option 1 - Installing Conda (recommended):
         1. If you have WSL (see step 2.b):
            1. Install Conda https://gist.github.com/kauffmanes/5e74916617f9993bc3479f401dfec7da
            2. Now we can run the commands on any WSL bash terminal (see the Environment Setup - conda commands)
         2. If you choose not to install WSL:
            1. Install Conda on Windows Powershell: https://docs.conda.io/projects/conda/en/latest/user-guide/install/download.html#hash-verification
            2. Then search for “Anaconda prompt” in Windows and we can run the commands under README: Environment Setup - conda on Anaconda prompt.
      2. Option 2: Using virtualenv
         1. If you have WSL (see step 2.b): Run the commands in our README given for the WSL on the wsl terminal
         2. If you choose not to install WSL: Run the commands in our README given for Powershell