| name | trackio |
| description | Integration with Trackio for ML experiment tracking, dashboard deployments, and Hugging Face Spaces sync. |
Trackio Skill
Use this skill to integrate trackio into machine learning workflows for experiment tracking, model evaluation, and Hugging Face Spaces dashboard deployment.
Key Directives
- Drop-in replacement for wandb: Trackio's API mirrors Weights & Biases in most cases (
import trackio as wandb).
- Local by default:
trackio.init() stores data locally unless a space_id is provided.
- Dashboard Access: Use
trackio show or trackio.show() to launch the Gradio dashboard locally.
- Hugging Face Spaces: Sync local runs to HF Spaces using
trackio sync --project "my-project" --space-id "org/space_id". This uploads all runs and data to the Space.
Available References
Detailed documentation is available in the references/ directory. When working on specific tasks, use view_file to read the relevant reference:
quickstart.md: Basic logging and dashboard launching.track.md: Detailed logging of metrics, artifacts, and media.launch.md: Automated job launching.self_hosted_server.md: Setting up a self-hosted server.deploy_embed.md: Deploying dashboards and freezing snapshots.manage.md: Project and workspace management.python_api.md: Comprehensive Python SDK reference.cli_commands.md: CLI tools for syncing and managing.storage_schema.md: Underlying storage format for logged data.api_mcp_server.md: MCP server integration.alerts.md: Configuring Slack/email alerts.ml_agents.md: Tracking LLM agents and interactions.environment_variables.md: Env var configuration.transformers_integration.md: Hugging Face Transformers Trainer integration.trl_integration.md: Transformer Reinforcement Learning integration.rapidfireai_integration.md: RapidFireAI integration.
