[AIR] Make Checkpoint.get_preprocessor faster
#32350
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Antoni Baum <antoni.baum@protonmail.com>
krfricke
requested changes
Feb 10, 2023
python/ray/air/checkpoint.py
Outdated
| # The preprocessor will either be stored in an in-memory dict or | ||
| # written to storage. In either case, it will use the PREPROCESSOR_KEY key. | ||
|
|
||
| def _get_preprocessor(self) -> Optional["Preprocessor"]: |
There was a problem hiding this comment.
Do we need this internal function? We don't use it anywhere else?
There was a problem hiding this comment.
It's used below! I can rename it or extract it from the class.
python/ray/air/checkpoint.py
Outdated
| else: | ||
| preprocessor = load_preprocessor_from_dir(checkpoint_path) | ||
| else: | ||
| preprocessor = self._get_preprocessor() |
There was a problem hiding this comment.
Suggested change
| preprocessor = self._get_preprocessor() | |
| checkpoint_dict = self.to_dict() | |
| preprocessor = checkpoint_dict.get(PREPROCESSOR_KEY, None) |
Shouldn't this be enough?
There was a problem hiding this comment.
We may have a bytes checkpoint created from a directory, in which case this would fail.
krfricke
added
the
@author-action-required
Yard1
removed
the
@author-action-required
edoakes
pushed a commit
to edoakes/ray
that referenced
this pull request
Mar 22, 2023
Previously, `get_preprocessor` would always serialize the Checkpoint into a dictionary first. This is incredibly wasteful and causes huge memory usage and runtime with large directory-based Checkpoints. This PR changes the logic to first see if a directory Checkpoint should be loaded into a dictionary or not in order to obtain the preprocessor. Context: I had ran into it when trying to do predictions with 25 GB Hugging Face model. `HuggingFacePredictor` calls `get_preprocessor` internally, and that takes ages to complete and almost caused an OOM for me - and all of that is unnecessary as the preprocessor has to be loaded from a file anyway. Signed-off-by: Antoni Baum <antoni.baum@protonmail.com> Signed-off-by: Edward Oakes <ed.nmi.oakes@gmail.com>
elliottower
pushed a commit
to elliottower/ray
that referenced
this pull request
Apr 22, 2023
Previously, `get_preprocessor` would always serialize the Checkpoint into a dictionary first. This is incredibly wasteful and causes huge memory usage and runtime with large directory-based Checkpoints. This PR changes the logic to first see if a directory Checkpoint should be loaded into a dictionary or not in order to obtain the preprocessor. Context: I had ran into it when trying to do predictions with 25 GB Hugging Face model. `HuggingFacePredictor` calls `get_preprocessor` internally, and that takes ages to complete and almost caused an OOM for me - and all of that is unnecessary as the preprocessor has to be loaded from a file anyway. Signed-off-by: Antoni Baum <antoni.baum@protonmail.com> Signed-off-by: elliottower <elliot@elliottower.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Signed-off-by: Antoni Baum antoni.baum@protonmail.com
Why are these changes needed?
Previously,
get_preprocessorwould always serialize the Checkpoint into a dictionary first. This is incredibly wasteful and causes huge memory usage and runtime with large directory-based Checkpoints. This PR changes the logic to first see if a directory Checkpoint should be loaded into a dictionary or not in order to obtain the preprocessor.Context: I had ran into it when trying to do predictions with 25 GB Hugging Face model.
HuggingFacePredictorcallsget_preprocessorinternally, and that takes ages to complete and almost caused an OOM for me - and all of that is unnecessary as the preprocessor has to be loaded from a file anyway.Related issue number
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.