wandb-util

A utility tool providing extended functionality for W&B (Weights & Biases).

Installation

pip install wandb-util

Features

run - Manage W&B runs

# List runs in a project
wandb-util -e <entity> -p <project> run list [--limit 10]

log - Manage W&B logs

# Display run summary metrics (filtered by prefix)
wandb-util -e <entity> -p <project> log metrics <run_id> [--prefix eval/]

# Show training history
wandb-util -e <entity> -p <project> log metrics <run_id> --history [--last 10]

# Filter by specific keys
wandb-util -e <entity> -p <project> log metrics <run_id> --keys loss,accuracy

# Display run output (stdout/stderr)
wandb-util -e <entity> -p <project> log output <run_id>

# Show only last N lines of output
wandb-util -e <entity> -p <project> log output <run_id> --tail 50

# Save output to file
wandb-util -e <entity> -p <project> log output <run_id> --output run.log

artifact - Manage W&B artifacts

# List artifacts for a run
wandb-util -e <entity> -p <project> artifact list <run_id>

# Download an artifact
wandb-util -e <entity> -p <project> artifact download <artifact_path>

Global Options

• -e, --entity TEXT - W&B entity (username or team). Defaults to the entity inferred from ~/.netrc credentials
• -p, --project TEXT - W&B project name [required]

Authentication

wandb-util automatically reads API credentials from ~/.netrc:

machine api.wandb.ai
  login user
  password <your_api_token>

License

MIT