Lightning-AI/LitLogger v2026.03.17

Release 2026.03.17

Repository: Lightning-AI/LitLogger

Tag: v2026.03.17

Published: 2026-03-17T13:28:03Z

Prerelease: no

Release notes: This release makes the dict-style Experiment API the primary LitLogger interface for standalone usage, and begins the deprecation of older compatibility surfaces.

Highlights

• The dict-style API is now the recommended API for standalone logging.
• The older module-level helpers and method-style Experiment logging helpers are now deprecated in favor of the dict API.
• litlogger.LightningLogger is now deprecated.
• Docs and examples have been updated to recommend:
• the dict-style API for standalone and inference workloads
• upstream Lightning LitLogger for Lightning/Fabric integration

Recommended API

For standalone usage, prefer litlogger.init() plus dict-style access on the returned experiment:

import litlogger
from litlogger import File, Text

exp = litlogger.init(name="my-run")

exp["model"] = "resnet50"
exp["summary"] = Text("first run")
exp["train/loss"].append(0.42, step=0)
exp["artifacts/config"] = File("config.yaml")

exp.finalize()

Deprecations

Module-level compatibility helpers

The older standalone helpers remain available in this release, but are now deprecated in favor of the dict-style API.

This includes patterns such as:

• litlogger.log(...)
• litlogger.log_metrics(...)
• litlogger.log_file(...)
• litlogger.log_model(...)
• litlogger.log_model_artifact(...)
• litlogger.log_metadata(...)

Method-style Experiment helper API

Method-style logging on Experiment is also deprecated in favor of direct dict-style interaction.

Prefer:

• exp["metric"].append(...) instead of exp.log_metrics(...)
• exp["artifact"] = File(...) instead of exp.log_file(...)
• exp["model"] = Model(...) instead of exp.log_model(...) / exp.log_model_artifact(...)
• exp["tag"] = "value" instead of exp.log_metadata(...)

LightningLogger

litlogger.LightningLogger is now deprecated.

For Lightning/Fabric integration, use upstream Lightning loggers instead:

• lightning.pytorch.loggers.LitLogger
• pytorch_lightning.loggers.LitLogger

For standalone usage, use the dict-style LitLogger API.

Migration examples

Before

import litlogger

litlogger.init(name="my-run")
litlogger.log_metrics({"loss": 0.42}, step=0)
litlogger.log_file("config.yaml")
litlogger.finalize()

After

import litlogger
from litlogger import File

exp = litlogger.init(name="my-run")
exp["loss"].append(0.42, step=0)
exp["config"] = File("config.yaml")
exp.finalize()

Lightning/Fabric

Before

from litlogger import LightningLogger

logger = LightningLogger(name="train-run")

After

from lightning.pytorch.loggers import LitLogger

logger = LitLogger(name="train-run")

Additional improvements

• Updated examples to use the new recommended APIs.
• Updated docs to cross-reference upstream Lightning LitLogger.
• Improved resumed experiment reconstruction for artifacts and artifact series.
• Improved model-series behavior and version handling in the dict-style API.

Compatibility

This release does not remove the deprecated APIs yet. They remain available as compatibility paths, but new code should migrate to the dict-style API or upstream Lightning LitLogger now.