Fix slurm multinode example #3229
Open
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
Compute metric with evaluate from main process only, avoiding bug in multinode evaluate.
Enforce --multi_gpu on multiple nodes. Moreover, make sure that each rank gets correctly addressed based on the $SLURM_PROCID. $SLURM_NNODES has now been deprecated and replaced by $SLURM_JOB_NUM_NODES. Fixed typo in $CMD as well.
muellerzr
reviewed
Dec 2, 2024
Comment on lines
-249
to
+256
| eval_metric = metric.compute() | ||
| # Use accelerator.print to print only on the main process. | ||
| accelerator.print(f"epoch {epoch}:", eval_metric) | ||
| if accelerator.is_main_process: | ||
| # Computing metrics in a distributed manner requires calling evaluate.load() with the | ||
| # n_process and process_id arguments. However, the metric.add_batch() step will fail | ||
| # due to a bug with datasets and evaluate (see https://github.com/huggingface/evaluate/issues/542) | ||
| # and related | ||
| eval_metric = metric.compute() | ||
| # Use accelerator.print to print only on the main process. | ||
| accelerator.print(f"epoch {epoch}:", eval_metric) |
There was a problem hiding this comment.
This breaking is a first for me, let me try to repr this
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread. Please note that issues that do not follow the contributing guidelines are likely to be ignored. |
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?
The SLURM multinode submit script doesn't work for multiple reasons (see also #1239 which is still open). This PR aims at solving some of those issues, namely:
$CMDcommand$SLURM_NNODESwas recently deprecated in favour of$SLURM_JOB_NUM_NODES--multi-gpuand the rank of each process has to be set with--machine-rankcomplete_nlp_exampledoesn't handle distributed evaluation correctly. In particular, either the evaluate GLUE mrpc metric has to be loaded with the--num_processand--process_idflags (which will subsequently fail due to several breakages due to recentdatasetsevaluate#542 & related) or metric computation has to happen only on the main process. This PR goes for the second.Fixes #3206
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings. NA
Who can review?
@muellerzr or @SunMarc