Add new args to BatchSizeFinder
#21163
Draft
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Fixes #21154
Fixes #20447
Fixes #20231
Fixes #20614
Adds two new arguments to batch size finder:
margin
which formode="binsearch"
will reduce the found batch size toint(found_batch_size * (1-margin))
to make sure there is some room for GPU allocation afterwards. The first 3 issues linked are different scenarios related to the fact that binsearch sometimes can find a batch size that just barely fits and then when training starts it crashes immediately because the found batch size does not fit in memory anymore.The second argument added
max_val
is a simple way for users to specify a maximum batch size that can be found. One case for this arg is to make sure the algorithm does not search for a batch size larger than the dataset.Before submitting
PR review
Anyone in the community is welcome to review the PR.
Before you start reviewing, make sure you have read the review guidelines. In short, see the following bullet-list:
Reviewer checklist
📚 Documentation preview 📚: https://pytorch-lightning--21163.org.readthedocs.build/en/21163/