GRPOTrainer/async: fix prefix EOS slicing for tool suffix (with Qwen3/3.5 type of chat templates)#5330
Merged
Merged
Conversation
…(mirror _get_tool_suffix_ids)
qgallouedec
reviewed
Mar 20, 2026
qgallouedec
reviewed
Mar 20, 2026
Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com>
Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com>
qgallouedec
reviewed
Mar 20, 2026
qgallouedec
approved these changes
Mar 20, 2026
qgallouedec
left a comment
Member
There was a problem hiding this comment.
LGTM! thanks a lot for the fast fix!
|
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 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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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?
Hi,
this should be a Fix #5317
I went with option 1 (model agnostic imho) mentioned by @mattbui in the issue, and mirrored it to the new exp async GRPO.
Generated a test case, in case.
Before submitting
Pull Request section?
to it if that's the case.
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.
Note
Medium Risk
Changes how tool-result token suffixes are computed in both
GRPOTrainerand experimental async GRPO; mistakes here could break tool-calling flows or cause subtle tokenization mismatches for certain chat templates.Overview
Fixes tool-call prompt reconstruction for chat templates (notably Qwen3/Qwen3.5) that emit an EOS token followed by extra trailing tokens (e.g. newline). When deriving tool-result
suffix_ids, the code now trims the prefix to the last EOS before slicing and validates the EOS-trimmed prefix matches the full tokenization, preventing incorrect suffix extraction.Adds a regression test (
test_get_tool_suffix_ids_eos_newline_boundary) covering Qwen3 and Qwen3.5 tokenizers (with version guard) to ensure_get_tool_suffix_idsreturns the expected suffix under this EOS+newline boundary condition.Written by Cursor Bugbot for commit a8e0f3c. This will update automatically on new commits. Configure here.