refactor CheckpointConnector.restore_weights #7862
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
Codecov Report
@@ Coverage Diff @@
## master #7862 +/- ##
======================================
- Coverage 91% 91% -0%
======================================
Files 204 204
Lines 13630 13643 +13
======================================
- Hits 12414 12389 -25
- Misses 1216 1254 +38 |
awaelchli
requested review from
Borda,
carmocca,
SeanNaren and
tchaton
as code owners
awaelchli
commented
Jun 7, 2021
Comment on lines
55
to
60
| dir_path_hpc = str(self.trainer.weights_save_path) | ||
| max_suffix = self.max_ckpt_in_folder(dir_path_hpc, "hpc_ckpt_") | ||
| if max_suffix is not None: | ||
| checkpoint_path = f'{dir_path_hpc}/hpc_ckpt_{max_suffix}.ckpt' | ||
| self.hpc_load(checkpoint_path, self.trainer._device_type == DeviceType.GPU) | ||
| rank_zero_info(f'restored hpc model from: {checkpoint_path}') |
There was a problem hiding this comment.
reviewers: this part is now in the property hpc_resume_path
Comment on lines
78
to
85
| # Try to read the checkpoint file at `checkpoint_path`. If not exist, do not restore checkpoint. | ||
| fs = get_filesystem(checkpoint_path) | ||
| if not fs.exists(checkpoint_path): | ||
| raise FileNotFoundError(f"Checkpoint at {checkpoint_path} not found. Aborting training.") | ||
|
|
||
| checkpoint, load_optimizer_states = self.trainer.training_type_plugin.restore_model_state_from_ckpt_path( | ||
| checkpoint_path, map_location=lambda storage, loc: storage | ||
| ) |
There was a problem hiding this comment.
reviewers: this part has moved to resume_start()
awaelchli
commented
Jun 7, 2021
Borda
reviewed
Jun 7, 2021
carmocca
approved these changes
Jun 9, 2021
ananthsub
approved these changes
Jun 9, 2021
11 tasks
carmocca
reviewed
Jun 16, 2021
|
|
||
| def resume_end(self) -> None: | ||
| """ Signal the connector that all states have resumed and memory for the checkpoint object can be released. """ | ||
| rank_zero_info(f"Restored all states from the checkpoint file at {self.resume_checkpoint_path}") |
There was a problem hiding this comment.
@awaelchli we shouldn't print this if not self.resume_checkpoint_path
There was a problem hiding this comment.
I know and it's addressed in #7652
sorry for the inconvenience
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.
What does this PR do?
Part of #7652
Splits the function restore_weights in to two:
resume_start()resume_end()The checkpoint loaded in
resume_start()will be temporarily cached and then destroyed inresume_end().This is in preparation of #7652 where we will call
resume_start()andresume_end()individually to split the restoring of trainer state to multiple stages.Discussion regarding partial restoring of trainer state in stages can be found in #7535
Before submitting
PR review
Anyone in the community is free to review the PR once the tests have passed.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:
Did you have fun?
Make sure you had fun coding 🙃