Calculate _num_batches_to_skip based on global_rows_processed_this_epoch#55964
Merged
matthewdeng merged 6 commits intoray-project:masterfrom Aug 28, 2025
Merged
Calculate _num_batches_to_skip based on global_rows_processed_this_epoch#55964matthewdeng merged 6 commits intoray-project:masterfrom
matthewdeng merged 6 commits intoray-project:masterfrom
Conversation
Contributor
There was a problem hiding this comment.
Code Review
This pull request refactors the logic for skipping batches during training resumption to be more robust, particularly when the number of workers changes. It achieves this by introducing _global_rows_processed_this_epoch and _num_batches_to_skip to replace the previous _restored_train_batch_idx mechanism. Additionally, it decouples checkpointing from validation by adding a new checkpoint_every_n_steps configuration. The changes are well-structured and improve the flexibility of the benchmark runner. I have identified a couple of areas for improvement in runner.py: one concerning code duplication and another related to a potential logic bug in the new checkpointing flow.
Comment on lines
+160
to
163
| global_batch_size = ( | ||
| self.benchmark_config.dataloader_config.train_batch_size | ||
| * ray.train.get_context().get_world_size() | ||
| ) |
Contributor
There was a problem hiding this comment.
This global_batch_size calculation is duplicated from the _num_batches_to_skip property (lines 128-131). To improve maintainability and avoid potential inconsistencies, consider extracting this logic into a shared helper property or method.
7121434 to
4b4155d
Compare
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Signed-off-by: Lehui Liu <lehui@anyscale.com>
Contributor
Author
|
Succeed release test run: https://buildkite.com/ray-project/release/builds/55274 |
justinvyu
approved these changes
Aug 27, 2025
tohtana
pushed a commit
to tohtana/ray
that referenced
this pull request
Aug 29, 2025
…och (ray-project#55964) 1. Previously, we use `_restored_train_batch_idx` as a run_state to determine how many batches to skip when resuming training. 2. In this PR we introduced a `_global_rows_processed_this_epoch` and `_num_batches_to_skip` instead so that it will be easier for us to calculate the num_batches to skip when resuming training with different number of workers. 3. Added one `checkpoint_every_n_steps: int = -1` config so that we can separate validation and checkpoint frequency. 4. release test run: https://buildkite.com/ray-project/release/builds/55274 Signed-off-by: Lehui Liu <lehui@anyscale.com> Signed-off-by: Masahiro Tanaka <mtanaka@anyscale.com>
tohtana
pushed a commit
to tohtana/ray
that referenced
this pull request
Aug 29, 2025
…och (ray-project#55964) 1. Previously, we use `_restored_train_batch_idx` as a run_state to determine how many batches to skip when resuming training. 2. In this PR we introduced a `_global_rows_processed_this_epoch` and `_num_batches_to_skip` instead so that it will be easier for us to calculate the num_batches to skip when resuming training with different number of workers. 3. Added one `checkpoint_every_n_steps: int = -1` config so that we can separate validation and checkpoint frequency. 4. release test run: https://buildkite.com/ray-project/release/builds/55274 Signed-off-by: Lehui Liu <lehui@anyscale.com> Signed-off-by: Masahiro Tanaka <mtanaka@anyscale.com>
gangsf
pushed a commit
to gangsf/ray
that referenced
this pull request
Sep 2, 2025
…och (ray-project#55964) 1. Previously, we use `_restored_train_batch_idx` as a run_state to determine how many batches to skip when resuming training. 2. In this PR we introduced a `_global_rows_processed_this_epoch` and `_num_batches_to_skip` instead so that it will be easier for us to calculate the num_batches to skip when resuming training with different number of workers. 3. Added one `checkpoint_every_n_steps: int = -1` config so that we can separate validation and checkpoint frequency. 4. release test run: https://buildkite.com/ray-project/release/builds/55274 Signed-off-by: Lehui Liu <lehui@anyscale.com> Signed-off-by: Gang Zhao <gang@gang-JQ62HD2C37.local>
sampan-s-nayak
pushed a commit
to sampan-s-nayak/ray
that referenced
this pull request
Sep 8, 2025
…och (ray-project#55964) 1. Previously, we use `_restored_train_batch_idx` as a run_state to determine how many batches to skip when resuming training. 2. In this PR we introduced a `_global_rows_processed_this_epoch` and `_num_batches_to_skip` instead so that it will be easier for us to calculate the num_batches to skip when resuming training with different number of workers. 3. Added one `checkpoint_every_n_steps: int = -1` config so that we can separate validation and checkpoint frequency. 4. release test run: https://buildkite.com/ray-project/release/builds/55274 Signed-off-by: Lehui Liu <lehui@anyscale.com> Signed-off-by: sampan <sampan@anyscale.com>
jugalshah291
pushed a commit
to jugalshah291/ray_fork
that referenced
this pull request
Sep 11, 2025
…och (ray-project#55964) 1. Previously, we use `_restored_train_batch_idx` as a run_state to determine how many batches to skip when resuming training. 2. In this PR we introduced a `_global_rows_processed_this_epoch` and `_num_batches_to_skip` instead so that it will be easier for us to calculate the num_batches to skip when resuming training with different number of workers. 3. Added one `checkpoint_every_n_steps: int = -1` config so that we can separate validation and checkpoint frequency. 4. release test run: https://buildkite.com/ray-project/release/builds/55274 Signed-off-by: Lehui Liu <lehui@anyscale.com> Signed-off-by: jugalshah291 <shah.jugal291@gmail.com>
wyhong3103
pushed a commit
to wyhong3103/ray
that referenced
this pull request
Sep 12, 2025
…och (ray-project#55964) 1. Previously, we use `_restored_train_batch_idx` as a run_state to determine how many batches to skip when resuming training. 2. In this PR we introduced a `_global_rows_processed_this_epoch` and `_num_batches_to_skip` instead so that it will be easier for us to calculate the num_batches to skip when resuming training with different number of workers. 3. Added one `checkpoint_every_n_steps: int = -1` config so that we can separate validation and checkpoint frequency. 4. release test run: https://buildkite.com/ray-project/release/builds/55274 Signed-off-by: Lehui Liu <lehui@anyscale.com> Signed-off-by: yenhong.wong <yenhong.wong@grabtaxi.com>
dstrodtman
pushed a commit
that referenced
this pull request
Oct 6, 2025
…och (#55964) 1. Previously, we use `_restored_train_batch_idx` as a run_state to determine how many batches to skip when resuming training. 2. In this PR we introduced a `_global_rows_processed_this_epoch` and `_num_batches_to_skip` instead so that it will be easier for us to calculate the num_batches to skip when resuming training with different number of workers. 3. Added one `checkpoint_every_n_steps: int = -1` config so that we can separate validation and checkpoint frequency. 4. release test run: https://buildkite.com/ray-project/release/builds/55274 Signed-off-by: Lehui Liu <lehui@anyscale.com> Signed-off-by: Douglas Strodtman <douglas@anyscale.com>
landscapepainter
pushed a commit
to landscapepainter/ray
that referenced
this pull request
Nov 17, 2025
…och (ray-project#55964) 1. Previously, we use `_restored_train_batch_idx` as a run_state to determine how many batches to skip when resuming training. 2. In this PR we introduced a `_global_rows_processed_this_epoch` and `_num_batches_to_skip` instead so that it will be easier for us to calculate the num_batches to skip when resuming training with different number of workers. 3. Added one `checkpoint_every_n_steps: int = -1` config so that we can separate validation and checkpoint frequency. 4. release test run: https://buildkite.com/ray-project/release/builds/55274 Signed-off-by: Lehui Liu <lehui@anyscale.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Why are these changes needed?
_restored_train_batch_idxas a run_state to determine how many batches to skip when resuming training._global_rows_processed_this_epochand_num_batches_to_skipinstead so that it will be easier for us to calculate the num_batches to skip when resuming training with different number of workers.checkpoint_every_n_steps: int = -1config so that we can separate validation and checkpoint frequency.Related issue number
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.method in Tune, I've added it in
doc/source/tune/api/under thecorresponding
.rstfile.