Lightning-AI/pytorch-lightning

Make logging_mode (`on_step` or `on_epoch`) available to loggers

Open

#20,502 opened on Dec 17, 2024

View on GitHub
 (5 comments) (0 reactions) (0 assignees)Python (26,687 stars) (3,233 forks)batch import
featurehelp wanted

Description

Description & Motivation

Lightning lets you log on_step, or on_epoch. However, even when using on_epoch, the global step is logged on the x-axis, not the epoch number.

Loggers can manually extract epoch from the logged metrics and pass that as the step. However, this would break on_step logging.

Making the logging mode for a logging call available would let loggers conditionally use either the global_step or current_epoch as the step for that particular metric.

Pitch

No response

Alternatives

No response

Additional context

No response

cc @lantiga @borda

Contributor guide

Tech stack
pythonpytorch
Domain
machine learningdeveloper experience
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
half day
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
fresh
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
PythonPyTorchLightning logging API
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
60
Research direction
Investigate the Logger base class in the Lightning codebase, specifically how metrics are passed to loggers (e.g., the `log` method). Look at the `LightningLoggerBase` class and how `global step` and `current epoch` are currently used. Determine where the logging mode is available and how it can be exposed to loggers. Review any related issues or comments in this thread for additional context.