📇 GRPO: print completions to console and update docs #2951
Merged
+100
−25
Conversation
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
added 2 commits
February 24, 2025 21:35
- Update `GRPOConfig` to replace `log_completions` with `log_completions_steps` - Add `print_prompt_completions_sample()` utility function for rich console logging - Modify `GRPOTrainer` to additionally print 5 random prompt-completion pairs every log_completions_steps steps
qgallouedec
reviewed
Feb 24, 2025
qgallouedec
reviewed
Feb 24, 2025
qgallouedec
reviewed
Feb 24, 2025
…ons_sample when rich is not available
qgallouedec
reviewed
Feb 24, 2025
qgallouedec
reviewed
Feb 24, 2025
qgallouedec
reviewed
Feb 24, 2025
qgallouedec
reviewed
Feb 24, 2025
Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com>
qgallouedec
requested changes
Feb 24, 2025
|
LGTM, once the latest nit recommendations are applied, and CI green, we're good to merge, thanks :) |
Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com>
|
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. |
|
|
qgallouedec
approved these changes
Feb 24, 2025
|
@bot /style |
|
Failing tests are because liger-kernel introduced a bug in their latest version. We can safely ignore it, I guess they'll do a patch release soon. See linkedin/Liger-Kernel#586 |
kashif
pushed a commit
that referenced
this pull request
Feb 27, 2025
* ✨ Enhance GRPO logging with configurable completions sampling - Update `GRPOConfig` to replace `log_completions` with `log_completions_steps` - Add `print_prompt_completions_sample()` utility function for rich console logging - Modify `GRPOTrainer` to additionally print 5 random prompt-completion pairs every log_completions_steps steps * GRPO trainer completions logging, move wandb checks together * Add rich availability check and use fallback in print_prompt_completions_sample when rich is not available * Update docstrings on print_prompt_completions_sample Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com> * Revert back to simple log_completions bool * GRPO log completions fully * Remove print fallback from print_prompt_completions_sample * Move accelerator main process check up for grpo log completions * Explicit variable names in print_prompt_completions_sample * Make GRPOConfig docstring match field description * Update log_completions docs again Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com> * Update GRPOConfig docs to match field * improve readibility when prompt or completions are multilines * log reward * prevent hanging, don't print without rich, print reward * style --------- Co-authored-by: Robert Veres <robert.veres@languagetool.org> Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com> Co-authored-by: Quentin Gallouédec <gallouedec.quentin@gmail.com>
jhinpan
pushed a commit
to jhinpan/trl-jin
that referenced
this pull request
Mar 12, 2025
* ✨ Enhance GRPO logging with configurable completions sampling - Update `GRPOConfig` to replace `log_completions` with `log_completions_steps` - Add `print_prompt_completions_sample()` utility function for rich console logging - Modify `GRPOTrainer` to additionally print 5 random prompt-completion pairs every log_completions_steps steps * GRPO trainer completions logging, move wandb checks together * Add rich availability check and use fallback in print_prompt_completions_sample when rich is not available * Update docstrings on print_prompt_completions_sample Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com> * Revert back to simple log_completions bool * GRPO log completions fully * Remove print fallback from print_prompt_completions_sample * Move accelerator main process check up for grpo log completions * Explicit variable names in print_prompt_completions_sample * Make GRPOConfig docstring match field description * Update log_completions docs again Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com> * Update GRPOConfig docs to match field * improve readibility when prompt or completions are multilines * log reward * prevent hanging, don't print without rich, print reward * style --------- Co-authored-by: Robert Veres <robert.veres@languagetool.org> Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com> Co-authored-by: Quentin Gallouédec <gallouedec.quentin@gmail.com>
yxliu-TAMU
pushed a commit
to mincheolseong/ECEN743-GRPO-Project-Proposal
that referenced
this pull request
Apr 20, 2025
* ✨ Enhance GRPO logging with configurable completions sampling - Update `GRPOConfig` to replace `log_completions` with `log_completions_steps` - Add `print_prompt_completions_sample()` utility function for rich console logging - Modify `GRPOTrainer` to additionally print 5 random prompt-completion pairs every log_completions_steps steps * GRPO trainer completions logging, move wandb checks together * Add rich availability check and use fallback in print_prompt_completions_sample when rich is not available * Update docstrings on print_prompt_completions_sample Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com> * Revert back to simple log_completions bool * GRPO log completions fully * Remove print fallback from print_prompt_completions_sample * Move accelerator main process check up for grpo log completions * Explicit variable names in print_prompt_completions_sample * Make GRPOConfig docstring match field description * Update log_completions docs again Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com> * Update GRPOConfig docs to match field * improve readibility when prompt or completions are multilines * log reward * prevent hanging, don't print without rich, print reward * style --------- Co-authored-by: Robert Veres <robert.veres@languagetool.org> Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com> Co-authored-by: Quentin Gallouédec <gallouedec.quentin@gmail.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.
What does this PR do?
GRPOConfigto replacelog_completionswithlog_completions_stepsprint_prompt_completions_sample()utility function for rich console loggingGRPOTrainerto additionally print 5 random prompt-completion pairs every log_completions_steps stepsFixes #2948
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.