-
Notifications
You must be signed in to change notification settings - Fork 3.6k
Closed
Labels
designIncludes a design discussionIncludes a design discussiondiscussionIn a discussion stageIn a discussion stagetunerwon't fixThis will not be worked onThis will not be worked on
Description
Description & Motivation
The PR #16462 introduced a breaking change with Lightning 2.0 that moves the auto_lr_find
and auto_scale_batch_size
features out of the Trainer:
trainer = Trainer()
tuner = Tuner(trainer)
tuner.lr_find(...)
tuner.scale_batch_size(...)
There is now an opportunity to also rename the methods, if we feel like there are better, more intuitive terms out there.
Pitch
Learning Rate tuner:
# current
tuner.lr_find()
# A
tuner.find_lr()
# B
tuner.find_learning_rate()
# C
(your suggestion in the comment)
Batch Size tuner:
# current
tuner.scale_batch_size()
# A (@Borda)
tuner.maximize_batch_size()
# B
tuner.find_batch_size()
# C
(your suggestion in the comment)
Alternatives
Keep as is :)
Additional context
No response
Metadata
Metadata
Assignees
Labels
designIncludes a design discussionIncludes a design discussiondiscussionIn a discussion stageIn a discussion stagetunerwon't fixThis will not be worked onThis will not be worked on