Repository navigation
Avoid duplicate network gates and model promotions - #13
Merged
arhancanli merged 1 commit intoOct 5, 2026
Merged
arhancanli merged 1 commit into
arhancanli merged 1 commit into
Conversation
arhancanli
marked this pull request as ready for review
October 5, 2026 19:19
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fine-tuning can select its verified initial checkpoint when no epoch improves held-out loss. If that export equals the explicit champion NNUE, the controller now saves the completed training snapshot and starts no duplicate network gate. It also refuses a new model version when a separately requested H1 result has identical network bytes. Candidate/checkpoint artifacts remain retained. Distinct NNUE candidates and search/executable comparisons keep their existing strength gates.
Two controller regressions cover complete trained-data accounting and queue preservation without a gate, and rejection of an identical-NNUE H1 without publication or a version increment. All 64 tests passed locally with ResourceWarnings treated as errors, preserving the active 1,600-game checkpoint, queue and settings. The previously validated real Metal zero-LR control exports exactly the current champion bytes. Exact-source CI passed release workspace tests, full perft/SPRT calibration, bench and WASM at d29d211; 64 Python cases ran, with nine optional Torch cases explicitly skipped in CI. No playing-engine strength gain or promotion is claimed.