reasoning-budget: do not re-fire the intro message in once mode - #8
Merged
Merged
Conversation
The sampler re-entered INTRO_FORCING for every new reasoning block because the intro token list was never consumed. Add an intro_once flag (set by the server when --reasoning-budget-intro-mode is once) that clears the list after the intro completes, so the message fires at most once per request. Mode every keeps the previous per-block behavior and its test; add a regression test for once mode. Assisted-by: zai/glm-5.3-flash
dzannotti
approved these changes
Aug 29, 2026
dzannotti
approved these changes
Aug 29, 2026
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The intro message forced at reasoning-block start re-fires for every new block in the same request because the sampler never consumes the intro token list. With
--reasoning-budget-intro-mode oncethis pollutes the chain of thought on multi-block agent turns. The mode previously only affected the server-side prompt scan.Adds an
intro_onceflag to the sampler (set when the mode isonce); the intro list is cleared once it has fired, so the message appears at most once per request. Modeeverykeeps the per-block behavior (existing test unchanged) and a regression test covers the once-mode case.Verified: test-reasoning-budget (24 cases incl. new one) and test-chat pass on a Vulkan Release build.
AI usage disclosure: implemented with AI assistance (GLM, Assisted-by trailer on the commit); reviewed and tested by the submitter.