Workflows

This guide maps the LitLogger feature set to concrete workflows so users can quickly choose the right API surface.

Standalone New API

Recommended for new code.

• initialize with litlogger.init.init()
• work with the returned Experiment
• log metadata with experiment["key"] = "value"
• log metrics with append() and extend() on experiment["key"]
• log files, media, and models with the file-like wrappers

Legacy Module-Level API

Supported for existing standalone scripts, but the dict-style API is the main user-facing API going forward.

• log_metrics()
• log_file()
• log_model()
• log_model_artifact()
• log_metadata()

Lightning and Fabric

Use lightning.pytorch.loggers.LitLogger with a Trainer or Fabric loop. The local LightningLogger alias is deprecated.

Files, Media, and Models

Use the dedicated wrappers when you want data attached to a run:

• File
• Image
• Text
• Model

Resume and Retrieval

Re-initialize with the same experiment name to continue logging or inspect existing metadata, metrics, artifacts, and media.

End-to-End Log and Fetch

See Complete Workflow for the full train-and-fetch flow.

Inference Logging

Use LitLogger inside a LitServe endpoint to track inference metrics and request-level telemetry.