Dev Builds » 20220704-1342

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 06:55:24 584104 4000 1102 952 1946 +13.03 ± 5.14 7 357 1122 507 7 +26.11 ± 10.08
ncm-dbt-02 06:55:58 585612 4010 1098 970 1942 +11.09 ± 5.14 3 382 1111 502 7 +21.51 ± 10.15
ncm-dbt-03 06:55:01 584243 4000 1075 967 1958 +9.38 ± 5.25 5 404 1074 512 5 +18.78 ± 10.36
ncm-dbt-04 06:56:17 568746 3996 1073 943 1980 +11.31 ± 5.38 9 406 1033 546 4 +23.51 ± 10.59
ncm-dbt-05 06:56:44 581229 3994 1119 968 1907 +13.14 ± 5.19 3 377 1089 522 6 +25.8 ± 10.27
20000 5467 4800 9733 +11.59 ± 2.34 27 1926 5429 2589 29 +23.14 ± 4.6

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
432421 ncm-dbt-02 584916 10 3 2 5 +34.85 ± 119.9 0 1 2 2 0 +70.44 ± 307.49
432420 ncm-dbt-05 577889 494 149 116 229 +23.24 ± 14.7 0 40 136 69 2 +43.84 ± 29.06
432419 ncm-dbt-04 566336 496 126 127 243 -0.7 ± 15.24 2 56 131 59 0 +1.4 ± 29.77
432418 ncm-dbt-01 583782 500 143 117 240 +18.08 ± 14.02 0 40 145 64 1 +34.86 ± 27.92
432417 ncm-dbt-03 584453 500 150 117 233 +22.96 ± 14.53 0 41 136 72 1 +44.72 ± 29.1
432416 ncm-dbt-02 587028 500 134 130 236 +2.78 ± 14.55 0 55 136 59 0 +5.56 ± 29.15
432415 ncm-dbt-05 581610 500 137 126 237 +7.64 ± 14.84 1 50 137 61 1 +15.3 ± 29.02
432414 ncm-dbt-04 569949 500 129 119 252 +6.95 ± 15.4 0 59 122 69 0 +13.9 ± 30.9
432413 ncm-dbt-01 583950 500 138 128 234 +6.95 ± 14.66 2 47 140 61 0 +16.69 ± 28.62
432412 ncm-dbt-03 582694 500 131 116 253 +10.43 ± 14.95 1 49 135 64 1 +20.87 ± 29.27
432411 ncm-dbt-02 584201 500 143 126 231 +11.82 ± 14.81 2 44 140 63 1 +25.06 ± 28.61
432410 ncm-dbt-05 583531 500 140 133 227 +4.86 ± 14.6 0 54 135 61 0 +9.73 ± 29.28
432409 ncm-dbt-04 570749 500 139 119 242 +13.9 ± 15.22 0 52 127 70 1 +26.46 ± 30.28
432408 ncm-dbt-01 581860 500 142 127 231 +10.43 ± 14.7 1 48 136 65 0 +22.27 ± 29.14
432407 ncm-dbt-03 584159 500 137 114 249 +15.99 ± 14.12 0 43 141 66 0 +32.05 ± 28.47
432406 ncm-dbt-02 586562 500 139 122 239 +11.82 ± 14.43 0 48 137 65 0 +23.66 ± 29.01
432405 ncm-dbt-04 568912 500 132 111 257 +14.6 ± 15.4 2 48 127 73 0 +32.05 ± 30.28
432404 ncm-dbt-01 584622 500 129 121 250 +5.56 ± 14.66 2 47 143 57 1 +12.51 ± 28.23
432403 ncm-dbt-05 578465 500 127 112 261 +10.43 ± 13.78 0 44 147 59 0 +20.87 ± 27.68
432402 ncm-dbt-03 585253 500 130 128 242 +1.39 ± 14.68 0 56 137 56 1 +1.39 ± 29.02
432401 ncm-dbt-02 583531 500 129 123 248 +4.17 ± 14.41 1 50 141 58 0 +9.73 ± 28.5
432400 ncm-dbt-04 568832 500 127 105 268 +15.3 ± 15.57 0 53 124 71 2 +27.85 ± 30.66
432399 ncm-dbt-01 585885 500 144 108 248 +25.06 ± 14.68 0 38 142 66 4 +44.72 ± 28.29
432398 ncm-dbt-05 579909 500 148 113 239 +24.36 ± 13.84 0 35 146 68 1 +47.55 ± 27.73
432397 ncm-dbt-03 584369 500 132 128 240 +2.78 ± 15.17 1 56 132 60 1 +5.56 ± 29.66
432396 ncm-dbt-02 583908 500 143 120 237 +15.99 ± 14.38 0 43 143 62 2 +29.25 ± 28.2
432395 ncm-dbt-04 569789 500 132 126 242 +4.17 ± 15.89 3 56 123 68 0 +12.51 ± 30.78
432394 ncm-dbt-01 584579 500 143 130 227 +9.04 ± 14.71 2 46 139 63 0 +20.87 ± 28.75
432393 ncm-dbt-05 583824 500 143 124 233 +13.21 ± 14.28 1 43 142 64 0 +27.85 ± 28.34
432392 ncm-dbt-03 581361 500 115 120 265 -3.48 ± 14.99 0 62 132 55 1 -8.34 ± 29.66
432391 ncm-dbt-02 587579 500 143 99 258 +30.65 ± 14.81 0 40 126 84 0 +61.79 ± 30.39
432390 ncm-dbt-01 583154 500 137 114 249 +15.99 ± 13.99 0 42 143 65 0 +32.05 ± 28.2
432389 ncm-dbt-04 568434 500 140 124 236 +11.12 ± 14.88 2 46 136 66 0 +25.06 ± 29.14
432388 ncm-dbt-05 584076 500 144 135 221 +6.26 ± 15.7 1 59 120 70 0 +13.9 ± 31.14
432387 ncm-dbt-02 587070 500 136 131 233 +3.47 ± 14.09 0 49 149 50 2 +4.17 ± 27.43
432386 ncm-dbt-03 583950 500 122 127 251 -3.47 ± 15.71 3 60 126 61 0 -2.78 ± 30.41
432385 ncm-dbt-04 566967 500 148 112 240 +25.06 ± 14.02 0 36 143 70 1 +48.96 ± 28.14
432384 ncm-dbt-01 585000 500 126 107 267 +13.21 ± 14.8 0 49 134 66 1 +25.06 ± 29.4
432383 ncm-dbt-02 585717 500 128 117 255 +7.64 ± 14.84 0 52 137 59 2 +12.51 ± 29.02
432382 ncm-dbt-03 587707 500 158 117 225 +28.55 ± 14.28 0 37 135 78 0 +57.5 ± 29.2
432381 ncm-dbt-05 580530 500 131 109 260 +15.3 ± 15.45 0 52 126 70 2 +27.85 ± 30.41

Commit

Commit ID 85f8ee6199f8578fbc082fc0f37e1985813e637a
Author Joost VandeVondele
Date 2022-07-04 13:42:34 UTC
Update default net to nn-3c0054ea9860.nnu First things first... this PR is being made from court. Today, Tord and Stéphane, with broad support of the developer community are defending their complaint, filed in Munich, against ChessBase. With their products Houdini 6 and Fat Fritz 2, both Stockfish derivatives, ChessBase violated repeatedly the Stockfish GPLv3 license. Tord and Stéphane have terminated their license with ChessBase permanently. Today we have the opportunity to present our evidence to the judge and enforce that termination. To read up, have a look at our blog post https://stockfishchess.org/blog/2022/public-court-hearing-soon/ and https://stockfishchess.org/blog/2021/our-lawsuit-against-chessbase/ This PR introduces a net trained with an enhanced data set and a modified loss function in the trainer. A slight adjustment for the scaling was needed to get a pass on standard chess. passed STC: https://tests.stockfishchess.org/tests/view/62c0527a49b62510394bd610 LLR: 2.94 (-2.94,2.94) <0.00,2.50> Total: 135008 W: 36614 L: 36152 D: 62242 Ptnml(0-2): 640, 15184, 35407, 15620, 653 passed LTC: https://tests.stockfishchess.org/tests/view/62c17e459e7d9997a12d458e LLR: 2.94 (-2.94,2.94) <0.50,3.00> Total: 28864 W: 8007 L: 7749 D: 13108 Ptnml(0-2): 47, 2810, 8466, 3056, 53 Local testing at a fixed 25k nodes resulted in Test run1026/easy_train_data/experiments/experiment_2/training/run_0/nn-epoch799.nnue localElo: 4.2 +- 1.6 The real strength of the net is in FRC and DFRC chess where it gains significantly. Tested at STC with slightly different scaling: FRC: https://tests.stockfishchess.org/tests/view/62c13a4002ba5d0a774d20d4 Elo: 29.78 +-3.4 (95%) LOS: 100.0% Total: 10000 W: 2007 L: 1152 D: 6841 Ptnml(0-2): 31, 686, 2804, 1355, 124 nElo: 59.24 +-6.9 (95%) PairsRatio: 2.06 DFRC: https://tests.stockfishchess.org/tests/view/62c13a5702ba5d0a774d20d9 Elo: 55.25 +-3.9 (95%) LOS: 100.0% Total: 10000 W: 2984 L: 1407 D: 5609 Ptnml(0-2): 51, 636, 2266, 1779, 268 nElo: 96.95 +-7.2 (95%) PairsRatio: 2.98 Tested at LTC with identical scaling: FRC: https://tests.stockfishchess.org/tests/view/62c26a3c9e7d9997a12d6caf Elo: 16.20 +-2.5 (95%) LOS: 100.0% Total: 10000 W: 1192 L: 726 D: 8082 Ptnml(0-2): 10, 403, 3727, 831, 29 nElo: 44.12 +-6.7 (95%) PairsRatio: 2.08 DFRC: https://tests.stockfishchess.org/tests/view/62c26a539e7d9997a12d6cb2 Elo: 40.94 +-3.0 (95%) LOS: 100.0% Total: 10000 W: 2215 L: 1042 D: 6743 Ptnml(0-2): 10, 410, 3053, 1451, 76 nElo: 92.77 +-6.9 (95%) PairsRatio: 3.64 This is due to the mixing in a significant fraction of DFRC training data in the final training round. The net is trained using the easy_train.py script in the following way: ``` python easy_train.py \ --training-dataset=../Leela-dfrc_n5000.binpack \ --experiment-name=2 \ --nnue-pytorch-branch=vondele/nnue-pytorch/lossScan4 \ --additional-training-arg=--param-index=2 \ --start-lambda=1.0 \ --end-lambda=0.75 \ --gamma=0.995 \ --lr=4.375e-4 \ --start-from-engine-test-net True \ --tui=False \ --seed=$RANDOM \ --max_epoch=800 \ --auto-exit-timeout-on-training-finished=900 \ --network-testing-threads 8 \ --num-workers 12 ``` where the data set used (Leela-dfrc_n5000.binpack) is a combination of our previous best data set (mix of Leela and some SF data) and DFRC data, interleaved to form: The data is available in https://drive.google.com/drive/folders/1S9-ZiQa_3ApmjBtl2e8SyHxj4zG4V8gG?usp=sharing Leela mix: https://drive.google.com/file/d/1JUkMhHSfgIYCjfDNKZUMYZt6L5I7Ra6G/view?usp=sharing DFRC: https://drive.google.com/file/d/17vDaff9LAsVo_1OfsgWAIYqJtqR8aHlm/view?usp=sharing The training branch used is https://github.com/vondele/nnue-pytorch/commits/lossScan4 A PR to the main trainer repo will be made later. This contains a revised loss function, now computing the loss from the score based on the win rate model, which is a more accurate representation than what we had before. Scaling constants are tweaked there as well. closes https://github.com/official-stockfish/Stockfish/pull/4100 Bench: 5186781
Copyright 2011–2025 Next Chess Move LLC