Kaggle GPUs from Local VS Code

Feb 10, 2024

Kaggle gives you free GPUs. With a VS Code tunnel, you can use them from your own editor.

1. Run this in a Kaggle notebook cell

In the notebook settings, turn on Internet and Persistence.

This uses the VS Code CLI’s code tunnel, which connects through Microsoft’s relay, so no SSH or open ports are needed. The cell downloads the CLI, signs in with GitHub, and starts a tunnel named kaggle-tunnel.

!mkdir -p /opt/vscode-cli
!curl -sSL "https://code.visualstudio.com/sha/download?build=stable&os=cli-alpine-x64" -o /opt/vscode-cli/vscode_cli.tar.gz
!tar -xzf /opt/vscode-cli/vscode_cli.tar.gz -C /opt/vscode-cli && rm /opt/vscode-cli/vscode_cli.tar.gz
!/opt/vscode-cli/code tunnel user login --provider github
!/opt/vscode-cli/code tunnel --accept-server-license-terms --name kaggle-tunnel

It prints a GitHub device code. Open the link and enter the code.

2. Connect

• Local VS Code: install the Remote - Tunnels extension, choose “Connect to Tunnel”, sign in with GitHub, and pick kaggle-tunnel.
• Browser: the cell prints a link like this, just open it: https://vscode.dev/tunnel/kaggle-tunnel/kaggle/working

Once connected, open the /kaggle folder and work there. With Persistence on, everything in it stays even after the instance shuts down.