Dev Builds » 20230329-1937

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 06:59:37 583364 4002 1240 793 1969 +38.97 ± 5.05 2 242 1072 677 8 +77.85 ± 10.32
ncm-dbt-02 06:58:25 584493 4000 1195 804 2001 +34.07 ± 5.14 0 283 1049 662 6 +67.73 ± 10.47
ncm-dbt-03 06:59:13 585708 4008 1271 829 1908 +38.47 ± 5.17 7 254 1038 700 5 +78.27 ± 10.53
ncm-dbt-04 06:57:50 570334 4000 1267 804 1929 +40.4 ± 5.1 1 246 1052 691 10 +80.27 ± 10.44
ncm-dbt-05 06:58:26 582893 3990 1244 777 1969 +40.85 ± 5.13 3 242 1044 697 9 +81.76 ± 10.48
20000 6217 4007 9776 +38.55 ± 2.29 13 1267 5255 3427 38 +77.16 ± 4.67

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
410280 ncm-dbt-01 584117 2 1 1 0 0.0 ± 34.02 0 0 1 0 0 0.0 ± 30.47
410279 ncm-dbt-03 584790 8 3 1 4 +88.62 ± 93.84 0 0 2 2 0 +190.67 ± 458.56
410278 ncm-dbt-05 584076 490 147 92 251 +39.16 ± 14.27 0 30 130 85 0 +79.35 ± 29.74
410277 ncm-dbt-04 570068 500 169 113 218 +39.08 ± 14.04 0 28 140 80 2 +76.25 ± 28.4
410276 ncm-dbt-02 585674 500 162 107 231 +38.37 ± 14.66 0 34 128 87 1 +76.25 ± 30.1
410275 ncm-dbt-01 583405 500 151 88 261 +44.01 ± 14.99 0 32 126 89 3 +85.04 ± 30.35
410274 ncm-dbt-03 580987 500 159 93 248 +46.13 ± 13.63 1 21 139 89 0 +95.44 ± 28.38
410273 ncm-dbt-05 581402 500 161 96 243 +45.42 ± 13.3 0 21 144 84 1 +90.97 ± 27.66
410272 ncm-dbt-01 580862 500 157 106 237 +35.56 ± 14.35 0 33 134 82 1 +70.44 ± 29.29
410271 ncm-dbt-02 586308 500 145 105 250 +27.85 ± 14.49 0 39 132 79 0 +56.07 ± 29.61
410270 ncm-dbt-04 569949 500 167 100 233 +46.84 ± 13.95 0 26 131 93 0 +95.44 ± 29.59
410269 ncm-dbt-03 586435 500 157 116 227 +28.55 ± 14.54 0 39 131 80 0 +57.5 ± 29.74
410268 ncm-dbt-05 580696 500 153 107 240 +32.05 ± 15.15 1 38 126 84 1 +64.66 ± 30.39
410267 ncm-dbt-02 584537 500 143 111 246 +22.27 ± 14.48 0 42 134 74 0 +44.72 ± 29.37
410266 ncm-dbt-01 583992 500 159 98 243 +42.6 ± 14.12 0 28 134 87 1 +85.04 ± 29.21
410265 ncm-dbt-03 587113 500 160 100 240 +41.89 ± 15.74 2 35 115 97 1 +86.52 ± 31.85
410264 ncm-dbt-04 569829 500 136 96 268 +27.85 ± 14.36 0 35 143 69 3 +51.8 ± 28.13
410263 ncm-dbt-05 582235 500 150 98 252 +36.26 ± 14.26 0 32 135 82 1 +71.89 ± 29.14
410262 ncm-dbt-02 582027 500 152 109 239 +29.95 ± 14.24 0 36 135 79 0 +60.36 ± 29.19
410261 ncm-dbt-04 572034 500 152 88 260 +44.72 ± 14.37 0 30 126 94 0 +90.97 ± 30.33
410260 ncm-dbt-01 583321 500 155 101 244 +37.67 ± 14.61 1 32 129 88 0 +77.71 ± 29.96
410259 ncm-dbt-03 586774 500 147 109 244 +26.46 ± 14.78 1 39 131 79 0 +54.65 ± 29.74
410258 ncm-dbt-05 586604 500 161 84 255 +53.94 ± 13.77 1 19 132 98 0 +112.14 ± 29.29
410257 ncm-dbt-02 584243 500 145 91 264 +37.67 ± 14.74 0 36 124 90 0 +76.25 ± 30.64
410256 ncm-dbt-04 569350 500 173 107 220 +46.13 ± 14.58 0 30 125 94 1 +92.46 ± 30.47
410255 ncm-dbt-01 583698 500 147 109 244 +26.46 ± 13.86 0 35 142 73 0 +53.22 ± 28.26
410254 ncm-dbt-03 585590 500 154 116 230 +26.46 ± 14.13 1 34 141 74 0 +54.65 ± 28.39
410253 ncm-dbt-05 580945 500 160 108 232 +36.26 ± 14.91 1 33 131 83 2 +71.89 ± 29.7
410252 ncm-dbt-02 583154 500 139 84 277 +38.37 ± 14.66 0 32 134 81 3 +73.34 ± 29.28
410251 ncm-dbt-04 569749 500 154 101 245 +36.97 ± 14.04 0 30 138 81 1 +73.34 ± 28.71
410250 ncm-dbt-01 583614 500 161 90 249 +49.67 ± 13.69 0 22 136 91 1 +99.95 ± 28.8
410249 ncm-dbt-03 586139 500 168 102 230 +46.13 ± 14.58 1 26 131 90 2 +92.46 ± 29.6
410248 ncm-dbt-05 584034 500 158 96 246 +43.3 ± 14.69 0 30 131 86 3 +83.57 ± 29.65
410247 ncm-dbt-02 585970 500 155 97 248 +40.48 ± 14.52 0 32 129 88 1 +80.63 ± 29.95
410246 ncm-dbt-04 570709 500 161 109 230 +36.26 ± 14.79 1 32 133 82 2 +71.89 ± 29.42
410245 ncm-dbt-03 585590 500 153 97 250 +39.08 ± 14.44 1 29 134 85 1 +79.17 ± 29.25
410244 ncm-dbt-01 584117 500 160 107 233 +36.97 ± 13.77 0 28 142 79 1 +73.34 ± 28.14
410243 ncm-dbt-05 583154 500 154 96 250 +40.48 ± 15.42 0 39 115 95 1 +80.63 ± 31.84
410242 ncm-dbt-03 587962 500 170 95 235 +52.51 ± 15.05 0 31 114 104 1 +106.01 ± 32.0
410241 ncm-dbt-02 584034 500 154 100 246 +37.67 ± 14.35 0 32 133 84 1 +74.79 ± 29.41
410240 ncm-dbt-04 570989 500 155 90 255 +45.42 ± 15.19 0 35 116 98 1 +90.97 ± 31.72
410239 ncm-dbt-01 583154 500 149 93 258 +39.08 ± 14.83 1 32 128 88 1 +79.17 ± 30.09

Commit

Commit ID 37160c4b1632245d46d86cec7bd22b76f5a87531
Author Linmiao Xu
Date 2023-03-29 19:37:52 UTC
Update default net to nn-dabb1ed23026.nnue Created by retraining the master net with these modifications: * New filtering methods for existing data from T80 sep+oct2022, T79 apr2022, T78 jun+jul+aug+sep2022, T77 dec2021 * Adding new filtered data from T80 aug2022 and T78 apr+may2022 * Increasing early-fen-skipping from 28 to 30 ``` python3 easy_train.py \ --experiment-name leela96-dfrc99-T80novT79mayT60novdec-v2-T80augsepoctT79aprT78aprtosep-v6-T77dec-v3-sk30 \ --training-dataset /data/leela96-dfrc99-T80novT79mayT60novdec-v2-T80augsepoctT79aprT78aprtosep-v6-T77dec-v3.binpack \ --nnue-pytorch-branch linrock/nnue-pytorch/misc-fixes \ --start-from-engine-test-net True \ --early-fen-skipping 30 \ --max_epoch 900 \ --start-lambda 1.0 \ --end-lambda 0.7 \ --lr 4.375e-4 \ --gamma 0.995 \ --tui False \ --gpus "0," \ --seed $RANDOM ``` The v3 filtering used for data from T77dec 2021 differs from v2 filtering in that: * To improve binpack compression, positions after ply 28 were skipped during training by setting position scores to VALUE_NONE (32002) instead of removing them entirely * All early-game positions with ply <= 28 were removed to maximize binpack compression * Only bestmove captures at d6pv2 search were skipped, not 2nd bestmove captures * Binpack compression was repaired for the remaining positions by effectively replacing bestmoves with "played moves" to maintain contiguous sequences of positions in the training game data After improving binpack compression, The T77 dec2021 data size was reduced from 95G to 19G. The v6 filtering used for data from T80augsepoctT79aprT78aprtosep 2022 differs from v2 in that: * All positions with only one legal move were removed * Tighter score differences at d6pv2 search were used to remove more positions with only one good move than before * d6pv2 search was not used to remove positions where the best 2 moves were captures ``` python3 interleave_binpacks.py \ nn-547-dataset/leela96-eval-filt-v2.binpack \ nn-547-dataset/dfrc99-eval-filt-v2.binpack \ nn-547-dataset/test80-nov2022-12tb7p-eval-filt-v2-d6.binpack \ nn-547-dataset/T79-may2022-12tb7p-eval-filt-v2.binpack \ nn-547-dataset/T60-nov2021-12tb7p-eval-filt-v2.binpack \ nn-547-dataset/T60-dec2021-12tb7p-eval-filt-v2.binpack \ filt-v6/test80-aug2022-16tb7p-filter-v6.binpack \ filt-v6/test80-sep2022-16tb7p-filter-v6.binpack \ filt-v6/test80-oct2022-16tb7p-filter-v6.binpack \ filt-v6/test79-apr2022-16tb7p-filter-v6.binpack \ filt-v6/test78-aprmay2022-16tb7p-filter-v6.binpack \ filt-v6/test78-junjulaug2022-16tb7p-filter-v6.binpack \ filt-v6/test78-sep2022-16tb7p-filter-v6.binpack \ filt-v3/test77-dec2021-16tb7p-filt-v3.binpack \ /data/leela96-dfrc99-T80novT79mayT60novdec-v2-T80augsepoctT79aprT78aprtosep-v6-T77dec-v3.binpack ``` The code for the new data filtering methods is available at: https://github.com/linrock/Stockfish/tree/nnue-data-v3/nnue-data The code for giving hexword names to .nnue files is at: https://github.com/linrock/nnue-namer Links for downloading the training data components can be found at: https://robotmoon.com/nnue-training-data/ Local elo at 25k nodes per move: nn-epoch779.nnue : 0.6 +/- 3.1 Passed STC: https://tests.stockfishchess.org/tests/view/64212412db43ab2ba6f8efb0 LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 82256 W: 22185 L: 21809 D: 38262 Ptnml(0-2): 286, 9065, 22067, 9407, 303 Passed LTC: https://tests.stockfishchess.org/tests/view/64223726db43ab2ba6f91d6c LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 30840 W: 8437 L: 8149 D: 14254 Ptnml(0-2): 14, 2891, 9323, 3177, 15 closes https://github.com/official-stockfish/Stockfish/pull/4465 bench 5101970
Copyright 2011–2025 Next Chess Move LLC