Repository navigation
Conversation
KarrAcaRn
pushed a commit
to KarrAcaRn/FreeToken-ByAI
that referenced
this pull request
Oct 23, 2026
…mpling parameters when clipping next clips in _admit_user_msg (kv ladder split), so the copy-on-clip moved there. Assisted-by: Claude Opus 5.5
KarrAcaRn
pushed a commit
to KarrAcaRn/FreeToken-ByAI
that referenced
this pull request
Oct 23, 2026
FlashML-org#790 FlashML-org#788 adopted, FlashML-org#802 superseded by FlashML-org#804, FlashML-org#809 FlashML-org#792 drafts deferred; the 11 rebased drafts stay deferred at their new heads. Assisted-by: Claude Opus 5.5
This branch has not been deployed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes #787.
Offline prompts can share one
SamplingParams. With a 64-token context and prompt lengths 56/48/8, clipping the first request changes all three 32-token budgets to 8, including the caller's object.Copy the current message and its sampling parameters when clipping. Budgets remain 8/16/32; the regression covers both admission orders using the real scheduler and prefill queue.
Tested on Windows 11, Ryzen 7 9700X, RTX 5070 Ti (driver 616.56), Python 3.12.10 and Torch 2.14.1+cu130. All 16 tests in this file pass; the two new regressions fail on main
546c5fe. No checkpoint is needed for this admission-only test.AI-assisted; validation was executed through development tooling.