nnue_train: WDL lambda の cosine-cycle スケジュールと result 欠損行の除外 - #406
Merged
Merged
Conversation
- `CycleWDL`: `--wdl` を base に `--wdl-cycle-delta` を half-cos で warmup→cooldown (`--wdl-cycle-warmup-pct`、`--wdl-schedule-superbatch`)。既存 Constant/Linear は不変 - `--wdl-ignore-draws`: WDL ラベル 0.5 の行 (引き分け / 対局結果なし) を lambda 0 で学習。 cuda-oxide kernel・CPU reference・native CUDA・launch 4 箇所に `ignore_draws` を追加 (未指定時は全経路 bit-identical) - `--validation-wdl`: held-out validation / eval-only の lambda を固定 - experiment.json params と docs (training-schedule / held-out-validation / wrm-loss-tuning) を更新 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
WHY
学習ターゲット
target = lambda * result + (1 - lambda) * WRM(score)の lambda は定数 (--wdl) か線形 taper でしか動かせなかった。「学習の中盤だけ勝敗の重みを上げ、終盤は純スコアに戻す」cos 形のスケジュール (nnue-pytorchlambda_utils.pyの cycle と同形) を superbatch 単位で使えるようにする。付随して: (1) 教師 PSV では対局に属さない局面 (policy 展開行) の result が 0 = 引き分けとして入るため、
wdl == 0.5の行を lambda 0 として扱う opt-in (--wdl-ignore-draws)。(2) lambda が動く run で検証側の lambda を固定する opt-in (--validation-wdl)。変更
CycleWDLスケジューラ (half-cos warmup → cooldown → base) +--wdl-cycle-delta/--wdl-cycle-warmup-pct/--wdl-schedule-superbatch--wdl-ignore-draws: oxide kernel / CPU reference (loss_wdl, loss_wrm) / native .cu 3 kernel + launch 4 箇所。未指定時は bit-identical--validation-wdl: superbatch 末 validation と eval-only の lambda 固定検証
bash scripts/local-ci.shPASS ×2 (sh11235、GPU 同等性テスト・native parity 込み)🤖 Generated with Claude Code