weights-and-biases

What it does

This skill enables AI agents to interact with Weights & Biases (W&B) for experiment tracking, model management, and visualization. It allows agents to log metrics, store artifacts, and analyze training runs directly within the workflow.

When to use it

• Tracking hyperparameters and loss curves during iterative model development.
• Storing and versioning large model checkpoints or generated datasets as artifacts.
• Comparing performance across multiple training runs to identify optimal configurations.
• Generating interactive dashboards for real-time monitoring of experiment health.

Key capabilities

• Logging custom metrics and scalars to W&B experiments.
• Uploading files, models, and data as tracked artifacts.
• Creating and managing W&B projects programmatically.
• Visualizing training history and comparing run distributions.

Example prompts

• "Log the current accuracy and loss values for this training run in a new W&B experiment."
• "Upload the latest model checkpoint to the 'production' project and tag it with version 2.1."
• "Compare the training curves of the last three runs and highlight the one with the lowest final loss."

Tips & gotchas

Ensure your API key is securely configured in the agent's environment before attempting to log data or create projects. Large artifact uploads may require additional authentication scopes or bandwidth considerations depending on your W&B plan.