Weights & Biases ML Cache & Run Artifacts (wandb)

Bottom Line: It is 100% safe to delete local wandb/ run folders and caches. Wiping local run artifacts will NOT delete experiment tracking data synced to your online Weights & Biases cloud dashboard.

What is the wandb Folder?

Located inside machine learning project directories at ./wandb (and globally at ~/.config/wandb), these folders are created by Weights & Biases (wandb).

When training machine learning models with PyTorch or TensorFlow, wandb logs training loss metrics, hyperparameter configs, generated images, and intermediate model checkpoint artifacts locally before syncing them to the cloud.

If training multi-epoch deep learning models, local wandb/ run folders take up 5 GB to 25 GB+.

Can You Delete wandb Files?

✅ Yes, it is completely safe to delete.

• Cloud-synced safety: Experiments, loss curves, and logged charts live safely in your online Weights & Biases dashboard.
• No code risk: Your training code, dataset loaders, and model definitions remain 100% safe.

How to Safely Clean Up wandb Cache

Method 1: Using wandb CLI Cache Cleanup (Recommended)

1. Open your terminal.
2. Run wandb CLI to clean local artifact cache:

   wandb artifact cache cleanup 1GB

Method 2: Manual Project Run Deletion

# macOS / Linux
find . -type d -name "wandb" -exec rm -rf {} +

# Windows (PowerShell)
Get-ChildItem -Include wandb -Recurse -Force | Remove-Item -Recurse -Force

Related Guides

• Learn about deleting PyTorch model cache.
• How to clean up Hugging Face cache.
• Check if you can clean TensorFlow / PyTorch venvs.