Quickstart

This tutorial introduces the new dict-style API through the smallest useful workflow: create an experiment, log metadata and metrics, upload a file, and finalize the run.

Create an Experiment

import litlogger

experiment = litlogger.init(name="quickstart")

litlogger.init() returns an Experiment instance. That object is the main entrypoint for the new API.

Log Metadata

Static string assignments become metadata entries:

experiment["model"] = "resnet50"
experiment["dataset"] = "cifar10"
experiment["optimizer"] = "adamw"

Log Metrics

Time-series metrics use Series through append() or extend() on experiment["key"]:

for step in range(10):
    experiment["train/loss"].append(1.0 / (step + 1), step=step)
    experiment["train/accuracy"].append(step / 10.0, step=step)

You can also batch values:

experiment["val/loss"].extend([0.5, 0.4, 0.3], start_step=100)

Log a File

Use File for static artifacts:

from litlogger import File

experiment["config"] = File("config.yaml")

Inspect Logged Data

The experiment exposes runtime views through metadata, metrics, artifacts, and url:

print(experiment.metadata)
print(experiment.metrics)
print(experiment.artifacts)
print(experiment.url)

Finalize

Use finalize() when the run is complete.

experiment.finalize()

Next Steps

• Files, Media, Models for files, media, and model logging
• Lightning Tutorial for Lightning/Fabric integration
• Complete Workflow for a log-and-fetch workflow
• Standalone Usage for the full standalone guide