Dev Builds » 20230209-0650

Use this dev build

NCM plays each Stockfish dev build 20,000 times against Stockfish 15. This yields an approximate Elo difference and establishes confidence in the strength of the dev builds.

Summary

Host Duration Avg Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo
ncm-dbt-01 06:59:38 583823 4016 1215 798 2003 +36.21 ± 5.05 2 255 1083 660 8 +72.13 ± 10.27
ncm-dbt-02 06:58:45 585854 3994 1197 824 1973 +32.54 ± 5.15 2 286 1050 655 4 +65.3 ± 10.47
ncm-dbt-03 06:58:21 584928 3998 1174 833 1991 +29.71 ± 5.11 6 280 1084 625 4 +60.21 ± 10.27
ncm-dbt-04 06:56:21 568975 4000 1208 834 1958 +32.58 ± 5.11 5 271 1074 645 5 +65.74 ± 10.33
ncm-dbt-05 06:59:26 577961 3992 1230 787 1975 +38.72 ± 5.1 2 252 1048 689 5 +77.87 ± 10.46
20000 6024 4076 9900 +33.95 ± 2.28 17 1344 5339 3274 26 +68.23 ± 4.63

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
430348 ncm-dbt-01 585885 16 6 5 5 +21.74 ± 73.63 0 1 5 2 0 +43.66 ± 157.0 ↓
430347 ncm-dbt-02 585548 494 142 101 251 +28.9 ± 15.48 2 39 122 84 0 +61.11 ± 30.91 ↓
430346 ncm-dbt-05 572800 492 150 97 245 +37.57 ± 13.7 0 27 139 80 0 +76.05 ± 28.45 ↓
430345 ncm-dbt-03 586266 498 146 107 245 +27.26 ± 14.23 0 37 136 76 0 +54.87 ± 29.06 ↓
430344 ncm-dbt-04 567918 500 158 114 228 +30.65 ± 14.16 0 35 136 79 0 +61.79 ± 29.05 ↓
430343 ncm-dbt-01 584285 500 161 89 250 +50.38 ± 13.29 0 20 138 92 0 +102.97 ± 28.45 ↓
430342 ncm-dbt-05 581361 500 146 99 255 +32.76 ± 14.17 0 34 135 81 0 +66.1 ± 29.17 ↓
430341 ncm-dbt-02 584579 500 139 104 257 +24.36 ± 15.13 0 45 126 78 1 +47.55 ± 30.4 ↓
430340 ncm-dbt-03 584159 500 148 122 230 +18.08 ± 14.93 2 42 134 72 0 +39.08 ± 29.38 ↓
430339 ncm-dbt-04 570348 500 156 113 231 +29.95 ± 15.02 2 36 129 83 0 +63.23 ± 29.99 ↓
430338 ncm-dbt-01 584285 500 152 102 246 +34.86 ± 14.83 0 35 133 79 3 +66.1 ± 29.44 ↓
430337 ncm-dbt-05 577889 500 153 105 242 +33.46 ± 14.74 1 35 129 85 0 +68.99 ± 29.98 ↓
430336 ncm-dbt-02 585126 500 146 114 240 +22.27 ± 14.48 0 42 134 74 0 +44.72 ± 29.37 ↓
430335 ncm-dbt-03 583950 500 148 95 257 +36.97 ± 14.17 1 28 139 81 1 +74.79 ± 28.56 ↓
430334 ncm-dbt-04 571471 500 149 106 245 +29.95 ± 14.24 0 35 138 76 1 +58.93 ± 28.79 ↓
430333 ncm-dbt-01 582402 500 146 110 244 +25.06 ± 14.16 1 34 144 70 1 +50.38 ± 27.99 ↓
430332 ncm-dbt-05 583405 500 144 115 241 +20.18 ± 14.58 1 41 136 72 0 +41.89 ± 29.11 ↓
430331 ncm-dbt-04 567086 500 164 106 230 +40.48 ± 15.04 1 33 124 91 1 +82.1 ± 30.63 ↓
430330 ncm-dbt-02 585885 500 156 108 236 +33.46 ± 14.74 0 38 126 86 0 +67.55 ± 30.39 ↓
430329 ncm-dbt-03 583363 500 159 111 230 +33.46 ± 14.61 0 37 128 85 0 +67.55 ± 30.12 ↓
430328 ncm-dbt-01 583154 500 147 99 254 +33.46 ± 15.24 1 38 124 86 1 +67.55 ± 30.65 ↓
430327 ncm-dbt-04 570108 500 154 101 245 +36.97 ± 13.49 1 22 152 73 2 +73.34 ± 26.64 ↓
430326 ncm-dbt-05 574094 500 159 101 240 +40.48 ± 14.78 0 35 122 93 0 +82.1 ± 30.9 ↓
430325 ncm-dbt-03 586774 500 148 106 246 +29.25 ± 14.46 1 35 135 79 0 +60.36 ± 29.19 ↓
430324 ncm-dbt-02 587494 500 153 99 248 +37.67 ± 13.81 0 29 138 83 0 +76.25 ± 28.69 ↓
430323 ncm-dbt-01 584453 500 160 91 249 +48.25 ± 14.57 0 28 127 93 2 +95.44 ± 30.17 ↓
430322 ncm-dbt-04 567403 500 143 93 264 +34.86 ± 14.31 0 33 135 81 1 +68.99 ± 29.16 ↓
430321 ncm-dbt-05 576946 500 154 100 246 +37.67 ± 14.87 0 35 128 85 2 +73.34 ± 30.1 ↓
430320 ncm-dbt-03 584537 500 139 95 266 +30.65 ± 14.29 0 35 137 77 1 +60.36 ± 28.92 ↓
430319 ncm-dbt-02 585843 500 158 93 249 +45.42 ± 13.87 0 25 136 88 1 +90.97 ± 28.88 ↓
430318 ncm-dbt-01 585674 500 148 93 259 +38.37 ± 14.26 0 32 131 87 0 +77.71 ± 29.68 ↓
430317 ncm-dbt-04 569350 500 142 94 264 +33.46 ± 14.61 1 34 131 84 0 +68.99 ± 29.71 ↓
430316 ncm-dbt-02 585928 500 145 109 246 +25.06 ± 14.29 0 39 136 75 0 +50.38 ± 29.09 ↓
430315 ncm-dbt-03 584201 500 129 93 278 +25.06 ± 14.29 1 35 142 71 1 +50.38 ± 28.27 ↓
430314 ncm-dbt-05 574297 500 165 87 248 +54.64 ± 14.08 0 22 130 96 2 +109.07 ± 29.63 ↓
430313 ncm-dbt-01 578342 500 152 110 238 +29.25 ± 14.06 0 35 138 77 0 +58.93 ± 28.79 ↓
430312 ncm-dbt-04 568116 500 142 107 251 +24.36 ± 14.76 0 43 129 78 0 +48.96 ± 30.01 ↓
430311 ncm-dbt-03 586181 500 157 104 239 +36.97 ± 14.57 1 31 133 84 1 +74.79 ± 29.41 ↓
430310 ncm-dbt-02 586435 500 158 96 246 +43.3 ± 14.42 0 29 132 87 2 +85.04 ± 29.5 ↓
430309 ncm-dbt-05 582903 500 159 83 258 +53.22 ± 14.01 0 23 129 97 1 +107.54 ± 29.8 ↓
430308 ncm-dbt-01 585928 500 143 99 258 +30.65 ± 13.89 0 32 143 74 1 +60.36 ± 28.08 ↓

Commit

Commit ID 05dea2ca4657dec10637bb53c4ad583f680e0677
Author Linmiao Xu
Date 2023-02-09 06:50:27 UTC
Update default net to nn-1337b1adec5b.nnue Created by retraining the master net on a dataset composed of: * Most of the previous best dataset filtered to remove positions likely having only one good move * Adding training data from Leela T77 dec2021 rescored with 16tb of 7-piece tablebases Trained with end lambda 0.7 and max epoch 900. Positions with ply <= 28 were removed from most of the previous best dataset before training began. A new nnue-pytorch trainer param for skipping early plies was used to skip plies <= 24 in the unfiltered and additional Leela T77 parts of the dataset. ``` python easy_train.py \ --experiment-name leela96-dfrc99-T80octnovT79aprmayT60novdec-eval-filt-v2-T78augsep-12tb-T77dec-16tb-lambda7-sk24 \ --training-dataset /data/leela96-dfrc99-T80octnovT79aprmayT60novdec-eval-filt-v2-T78augsep-12tb-T77dec-16tb.binpack \ --nnue-pytorch-branch linrock/nnue-pytorch/easy-train-early-fen-skipping \ --early-fen-skipping 24 \ --gpus "0," \ --start-from-engine-test-net True \ --start-lambda 1.0 \ --end-lambda 0.7 \ --gamma 0.995 \ --lr 4.375e-4 \ --tui False \ --seed $RANDOM \ --max_epoch 900 ``` The depth6 multipv2 search filtering method is the same as the one used for filtering recent best datasets, with a lower eval difference threshold to remove slightly more positions than before. These parts of the dataset were filtered: * 96% of T60T70wIsRightFarseerT60T74T75T76.binpack * 99% of dfrc_n5000.binpack * T80 oct + nov 2022 data, no positions with castling flags, rescored with ~600gb 7p tablebases * T79 apr + may 2022 data, rescored with 12tb 7p tablebases * T60 nov + dec 2021 data, rescored with 12tb 7p tablebases These parts of the dataset were not filtered. Positions with ply <= 24 were skipped during training: * T78 aug + sep 2022 data, rescored with 12tb 7p tablebases * 84% of T77 dec 2021 data, rescored with 16tb 7p tablebases The code and exact evaluation thresholds used for data filtering can be found at: https://github.com/linrock/Stockfish/tree/tools-filter-multipv2-eval-diff-t2/src/filter The exact training data used can be found at: https://robotmoon.com/nnue-training-data/ Local elo at 25k nodes per move: nn-epoch859.nnue : 3.5 +/ 1.2 Passed STC: LLR: 2.95 (-2.94,2.94) <0.00,2.00> https://tests.stockfishchess.org/tests/view/63dfeefc73223e7f52ad769f Total: 219744 W: 58572 L: 58002 D: 103170 Ptnml(0-2): 609, 24446, 59284, 24832, 701 Passed LTC: https://tests.stockfishchess.org/tests/view/63e268fc73223e7f52ade7b6 LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 91256 W: 24528 L: 24121 D: 42607 Ptnml(0-2): 48, 8863, 27390, 9288, 39 closes https://github.com/official-stockfish/Stockfish/pull/4387 bench 3841998
Copyright 2011–2025 Next Chess Move LLC
Feedback & Support Terms of Service Privacy Policy Desktop Version