Repository navigation
Preserve CMLFQ learning across rebinds and correct TP cost units - #15
Merged
Merged
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Rebinding a rollout engine currently drops CMLFQ's live prefix evidence, recent-trajectory window, and rebuild counters. Preserve independent copies of that learned state across both engine reconstruction and reconfiguration, so periodic rebuilds keep using the accumulated online window.
Also correct the analytic TP estimates used for resource and routing decisions: divide aggregate KV capacity by aggregate token storage, include replicated or uneven KV heads, and use per-worker bandwidth for per-worker decode reads.
Validation (re-run on 2026-10-20): local CPU CI passed all 101 tests, including 21 cases covering rebind/rebuild behavior, state isolation, and TP capacity/bandwidth units. No model or GPU runtime is required for these regressions.