Dev Builds » 20240402-0649

Use this dev build

NCM plays each Stockfish dev build 20,000 times against Stockfish 14. 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 11:50:45 1198337 4024 1735 326 1963 +127.03 +/- 4.82 0 48 560 1351 53 +284.19 +/- 14.4
ncm-dbt-02 11:42:54 1227249 3968 1733 316 1919 +129.79 +/- 4.85 0 47 528 1354 55 +292.29 +/- 14.84
ncm-dbt-03 11:54:25 1232917 4040 1773 315 1952 +131.3 +/- 4.79 1 38 545 1374 62 +295.66 +/- 14.6
ncm-dbt-05 11:41:33 1222045 3950 1735 290 1925 +133.27 +/- 5.0 2 56 471 1387 59 +303.24 +/- 15.69
ncm-dbt-06 11:53:05 1233233 4018 1759 307 1952 +131.49 +/- 4.82 1 45 520 1387 56 +298.2 +/- 14.95
20000 8735 1554 9711 +130.56 +/- 2.17 4 234 2624 6853 285 +294.61 +/- 6.65

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
339712 ncm-dbt-06 1254725 18 7 0 11 +142.54 +/- 56.5 0 0 2 7 0 +361.06 +/- 503.16
339711 ncm-dbt-01 1194362 24 10 3 11 +104.32 +/- 83.57 0 1 4 6 1 +190.85 +/- 204.66
339710 ncm-dbt-03 1237637 40 15 3 22 +107.51 +/- 41.27 0 0 8 12 0 +240.78 +/- 127.62
339709 ncm-dbt-05 1218170 450 208 40 202 +136.29 +/- 13.92 0 3 58 157 7 +312.16 +/- 45.28
339708 ncm-dbt-02 1222223 468 213 34 221 +140.0 +/- 14.68 0 4 59 159 12 +310.83 +/- 44.93
339707 ncm-dbt-06 1233104 500 217 47 236 +123.02 +/- 13.82 0 7 72 165 6 +273.0 +/- 40.54
339706 ncm-dbt-01 1212308 500 215 40 245 +126.97 +/- 12.66 0 5 67 176 2 +295.94 +/- 42.08
339705 ncm-dbt-03 1226668 500 223 32 245 +139.81 +/- 13.26 0 4 59 179 8 +324.17 +/- 44.97
339704 ncm-dbt-05 1218853 500 226 33 241 +141.44 +/- 15.03 1 8 49 181 11 +324.17 +/- 48.88
339703 ncm-dbt-02 1229107 500 223 45 232 +129.35 +/- 14.75 0 8 67 164 11 +280.42 +/- 42.05
339702 ncm-dbt-06 1237051 500 223 37 240 +135.76 +/- 14.47 0 7 61 171 11 +301.33 +/- 44.13
339701 ncm-dbt-01 1211092 500 216 41 243 +126.97 +/- 13.22 0 5 70 170 5 +288.06 +/- 41.11
339700 ncm-dbt-03 1218280 500 223 37 240 +135.76 +/- 13.76 0 3 69 167 11 +301.33 +/- 41.33
339699 ncm-dbt-05 1226129 500 223 36 241 +136.56 +/- 13.55 0 5 61 176 8 +312.48 +/- 44.19
339698 ncm-dbt-02 1243014 500 214 35 251 +130.14 +/- 12.39 0 3 68 176 3 +304.07 +/- 41.65
339697 ncm-dbt-06 1237426 500 215 37 248 +129.35 +/- 14.41 1 8 59 176 6 +295.94 +/- 44.72
339696 ncm-dbt-01 1198438 500 207 43 250 +118.33 +/- 14.36 0 11 69 165 5 +261.07 +/- 41.31
339695 ncm-dbt-03 1239318 500 212 33 255 +130.14 +/- 12.39 0 3 68 176 3 +304.07 +/- 41.65
339694 ncm-dbt-05 1224999 500 214 29 257 +134.95 +/- 13.96 0 8 56 179 7 +309.64 +/- 45.98
339693 ncm-dbt-02 1235941 500 223 41 236 +132.54 +/- 13.47 0 6 62 176 6 +304.07 +/- 43.81
339692 ncm-dbt-01 1194126 500 225 38 237 +136.56 +/- 12.79 0 1 69 172 8 +312.48 +/- 41.11
339691 ncm-dbt-06 1232786 500 223 36 241 +136.56 +/- 13.36 0 4 63 175 8 +312.48 +/- 43.44
339690 ncm-dbt-03 1223588 500 217 40 243 +128.55 +/- 14.26 0 6 71 163 10 +280.42 +/- 40.83
339689 ncm-dbt-05 1215474 500 219 35 246 +134.15 +/- 13.61 0 5 64 173 8 +304.07 +/- 43.1
339688 ncm-dbt-02 1208571 500 212 35 253 +128.55 +/- 13.19 0 7 62 178 3 +298.62 +/- 43.77
339687 ncm-dbt-01 1197835 500 212 52 236 +115.22 +/- 14.37 0 10 76 158 6 +249.64 +/- 39.41
339686 ncm-dbt-06 1224860 500 223 33 244 +138.99 +/- 12.7 0 2 63 178 7 +324.17 +/- 43.32
339685 ncm-dbt-03 1240068 500 218 50 232 +121.46 +/- 14.17 1 6 73 164 6 +270.57 +/- 40.25
339684 ncm-dbt-05 1236009 500 211 29 260 +132.54 +/- 13.83 1 7 55 183 4 +312.48 +/- 46.38
339683 ncm-dbt-02 1235004 500 219 43 238 +127.76 +/- 13.57 0 5 71 167 7 +285.49 +/- 40.81
339682 ncm-dbt-01 1184198 500 225 43 232 +132.54 +/- 14.36 0 6 67 166 11 +290.66 +/- 42.08
339681 ncm-dbt-03 1249916 500 219 47 234 +124.6 +/- 13.62 0 6 72 166 6 +277.93 +/- 40.53
339680 ncm-dbt-06 1226877 500 220 42 238 +129.35 +/- 12.61 0 6 61 182 1 +306.84 +/- 44.17
339679 ncm-dbt-05 1214894 500 218 44 238 +126.17 +/- 14.62 0 10 64 168 8 +277.93 +/- 42.91
339678 ncm-dbt-02 1222306 500 213 40 247 +125.38 +/- 13.79 0 6 72 165 7 +277.93 +/- 40.53
339677 ncm-dbt-01 1210944 500 218 26 256 +140.62 +/- 13.04 0 2 63 176 9 +324.17 +/- 43.32
339676 ncm-dbt-06 1236576 500 223 37 240 +135.76 +/- 13.01 0 2 68 172 8 +309.64 +/- 41.56
339675 ncm-dbt-03 1217366 500 225 42 233 +133.34 +/- 14.85 0 7 66 164 13 +288.06 +/- 42.4
339674 ncm-dbt-05 1221832 500 216 44 240 +124.6 +/- 14.31 0 10 64 170 6 +277.93 +/- 42.91
339673 ncm-dbt-02 1221826 500 216 43 241 +125.38 +/- 13.96 0 8 67 169 6 +280.42 +/- 42.05
339672 ncm-dbt-01 1181730 500 207 40 253 +120.67 +/- 13.84 0 7 75 162 6 +265.78 +/- 39.69
339671 ncm-dbt-06 1215698 500 208 38 254 +123.02 +/- 14.65 0 9 71 161 9 +265.78 +/- 40.81
339670 ncm-dbt-03 1243413 500 221 31 248 +138.99 +/- 12.5 0 3 59 183 5 +330.23 +/- 44.94

Commit

Commit ID 0716b845fdef8a20102b07eaec074b8da8162523
Author Viren6
Date 2024-04-02 06:49:48 UTC
Update NNUE architecture to SFNNv9 and net nn-ae6a388e4a1a.nnue Part 1: PyTorch Training, linrock Trained with a 10-stage sequence from scratch, starting in May 2023: https://github.com/linrock/nnue-tools/blob/master/exp-sequences/3072-10stage-SFNNv9.yml While the training methods were similar to the L1-2560 training sequence, the last two stages introduced min-v2 binpacks, where bestmove capture and in-check position scores were not zeroed during minimization, for compatibility with skipping SEE >= 0 positions and future research. Training data can be found at: https://robotmoon.com/nnue-training-data This net was tested at epoch 679 of the 10th training stage: https://tests.stockfishchess.org/tests/view/65f32e460ec64f0526c48dbc Part 2: SPSA Training, Viren6 The net was then SPSA tuned. This consisted of the output weights (32 * 8) and biases (8) as well as the L3 biases (32 * 8) and L2 biases (16 * 8), totalling 648 params in total. The SPSA tune can be found here: https://tests.stockfishchess.org/tests/view/65fc33ba0ec64f0526c512e3 With the help of Disservin , the initial weights were extracted with: https://github.com/Viren6/Stockfish/tree/new228 The net was saved with the tuned weights using: https://github.com/Viren6/Stockfish/tree/new241 Earlier nets of the SPSA failed STC compared to the base 3072 net of part 1: https://tests.stockfishchess.org/tests/view/65ff356e0ec64f0526c53c98 Therefore it is suspected that the SPSA at VVLTC has added extra scaling on top of the scaling of increasing the L1 size. Passed VVLTC 1: https://tests.stockfishchess.org/tests/view/6604a9020ec64f0526c583da LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 53042 W: 13554 L: 13256 D: 26232 Ptnml(0-2): 12, 5147, 15903, 5449, 10 Passed VVLTC 2: https://tests.stockfishchess.org/tests/view/660ad1b60ec64f0526c5dd23 LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 17506 W: 4574 L: 4315 D: 8617 Ptnml(0-2): 1, 1567, 5362, 1818, 5 STC Elo estimate: https://tests.stockfishchess.org/tests/view/660b834d01aaec5069f87cb0 Elo: -7.66 ± 3.8 (95%) LOS: 0.0% Total: 9618 W: 2440 L: 2652 D: 4526 Ptnml(0-2): 80, 1281, 2261, 1145, 42 nElo: -13.94 ± 6.9 (95%) PairsRatio: 0.87 closes https://tests.stockfishchess.org/tests/view/660b834d01aaec5069f87cb0 bench 1823302 Co-Authored-By: Linmiao Xu <lin@robotmoon.com>
Copyright 2011–2024 Next Chess Move LLC