Dev Builds » 20240217-1611

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 14:32:14 1138710 4994 2202 410 2382 +130.48 +/- 4.34 1 59 653 1715 69 +294.89 +/- 13.34
ncm-dbt-03 14:33:10 1199614 5000 2174 352 2474 +132.7 +/- 4.16 1 43 647 1751 58 +305.45 +/- 13.4
ncm-dbt-04 14:32:54 1189920 5008 2214 367 2427 +134.47 +/- 4.23 1 40 649 1739 75 +306.89 +/- 13.37
ncm-dbt-06 14:32:43 1190395 4998 2208 352 2438 +135.49 +/- 4.26 0 53 609 1765 72 +310.98 +/- 13.82
20000 8798 1481 9721 +133.28 +/- 2.12 3 195 2558 6970 274 +304.48 +/- 6.73

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
327255 ncm-dbt-03 1202728 0 0 0 0 -0.0 +/- 1199.83 0 0 0 0 0 -0.0 +/- 1199.83
327254 ncm-dbt-04 1197043 8 4 0 4 +190.56 +/- 13.88 0 0 0 4 0 +1199.83 +/- 250.99
327253 ncm-dbt-06 1186240 498 218 44 236 +126.73 +/- 14.32 0 11 58 175 5 +287.23 +/- 44.88
327252 ncm-dbt-03 1200961 500 213 22 265 +139.81 +/- 12.67 0 3 59 182 6 +330.23 +/- 44.94
327251 ncm-dbt-01 1141335 494 216 48 230 +123.06 +/- 14.61 1 9 63 169 5 +277.91 +/- 43.24
327250 ncm-dbt-04 1195916 500 218 35 247 +133.34 +/- 12.29 0 3 64 180 3 +315.35 +/- 43.03
327249 ncm-dbt-06 1189599 500 215 38 247 +128.55 +/- 12.63 0 3 71 172 4 +295.94 +/- 40.69
327248 ncm-dbt-03 1198751 500 221 35 244 +135.76 +/- 13.2 0 6 57 182 5 +318.25 +/- 45.73
327247 ncm-dbt-01 1134706 500 218 36 246 +132.54 +/- 13.47 0 4 68 170 8 +298.62 +/- 41.71
327246 ncm-dbt-04 1188566 500 225 39 236 +135.76 +/- 13.39 0 3 67 171 9 +306.84 +/- 41.99
327245 ncm-dbt-06 1192819 500 229 43 228 +135.76 +/- 13.76 0 6 60 176 8 +309.64 +/- 44.55
327244 ncm-dbt-03 1202081 500 215 41 244 +126.17 +/- 14.12 0 8 67 168 7 +280.42 +/- 42.05
327243 ncm-dbt-01 1125875 500 223 48 229 +126.97 +/- 14.61 0 8 69 163 10 +275.45 +/- 41.43
327242 ncm-dbt-04 1189407 500 229 38 233 +139.81 +/- 13.64 0 4 61 175 10 +318.25 +/- 44.18
327241 ncm-dbt-06 1192059 500 213 28 259 +134.95 +/- 13.78 0 7 58 178 7 +309.64 +/- 45.27
327240 ncm-dbt-01 1139453 500 221 37 242 +134.15 +/- 13.24 0 2 71 168 9 +301.33 +/- 40.6
327239 ncm-dbt-03 1200843 500 219 27 254 +140.62 +/- 13.04 0 4 57 182 7 +330.23 +/- 45.79
327238 ncm-dbt-04 1183246 500 207 35 258 +124.6 +/- 13.8 1 4 73 166 6 +280.42 +/- 40.21
327237 ncm-dbt-06 1186398 500 214 33 253 +131.74 +/- 12.54 0 1 73 170 6 +301.33 +/- 39.86
327236 ncm-dbt-03 1196176 500 216 31 253 +134.95 +/- 13.22 0 4 64 175 7 +309.64 +/- 43.08
327235 ncm-dbt-01 1143670 500 223 47 230 +127.76 +/- 15.09 0 11 62 167 10 +277.93 +/- 43.49
327234 ncm-dbt-04 1186553 500 226 41 233 +134.95 +/- 13.59 0 4 66 171 9 +304.07 +/- 42.38
327233 ncm-dbt-06 1193475 500 224 35 241 +138.18 +/- 14.06 0 4 65 169 12 +306.84 +/- 42.73
327232 ncm-dbt-01 1145935 500 214 36 250 +129.35 +/- 13.54 0 7 63 175 5 +295.94 +/- 43.42
327231 ncm-dbt-03 1194478 500 224 36 240 +137.37 +/- 12.76 0 3 62 179 6 +321.19 +/- 43.77
327230 ncm-dbt-04 1188961 500 227 41 232 +135.76 +/- 13.94 0 5 64 171 10 +304.07 +/- 43.1
327229 ncm-dbt-06 1193151 500 219 29 252 +138.99 +/- 13.1 0 7 50 189 4 +333.32 +/- 48.73
327228 ncm-dbt-03 1202577 500 214 38 248 +127.76 +/- 13.02 0 5 68 173 4 +293.29 +/- 41.75
327227 ncm-dbt-01 1143289 500 216 46 238 +123.02 +/- 14.16 0 8 71 164 7 +270.57 +/- 40.83
327226 ncm-dbt-04 1187310 500 218 39 243 +130.14 +/- 13.52 0 6 65 173 6 +295.94 +/- 42.75
327225 ncm-dbt-06 1189062 500 223 26 251 +144.71 +/- 13.08 0 5 50 188 7 +346.12 +/- 48.95
327224 ncm-dbt-03 1201084 500 210 41 249 +122.24 +/- 13.12 0 4 78 163 5 +273.0 +/- 38.74
327223 ncm-dbt-04 1188077 500 216 29 255 +136.56 +/- 12.59 0 2 65 177 6 +318.25 +/- 42.59
327222 ncm-dbt-01 1138528 500 222 35 243 +136.56 +/- 12.98 0 4 61 179 6 +318.25 +/- 44.18
327221 ncm-dbt-01 1130349 500 220 38 242 +132.54 +/- 12.32 0 2 68 176 4 +309.64 +/- 41.56
327220 ncm-dbt-06 1187439 500 223 39 238 +134.15 +/- 13.43 0 4 66 172 8 +304.07 +/- 42.38
327219 ncm-dbt-04 1194633 500 215 37 248 +129.35 +/- 14.24 0 7 67 167 9 +285.49 +/- 42.08
327218 ncm-dbt-03 1197803 500 217 42 241 +126.97 +/- 13.41 1 2 74 167 6 +288.06 +/- 39.79
327217 ncm-dbt-01 1143968 500 229 39 232 +138.99 +/- 12.7 0 4 57 184 5 +330.23 +/- 45.79
327216 ncm-dbt-06 1193714 500 230 37 233 +141.44 +/- 13.96 0 5 58 176 11 +321.19 +/- 45.36
327215 ncm-dbt-04 1189413 500 229 33 238 +143.89 +/- 12.51 0 2 57 184 7 +342.85 +/- 45.71
327214 ncm-dbt-03 1198274 500 225 39 236 +135.76 +/- 12.81 0 4 61 180 5 +318.25 +/- 44.18

Commit

Commit ID 8e75548f2a10969c1c9211056999efbcebe63f9a
Author Linmiao Xu
Date 2024-02-17 16:11:46 UTC
Update default main net to nn-b1a57edbea57.nnue Created by retraining the previous main net `nn-baff1edbea57.nnue` with: - some of the same options as before: ranger21, more WDL skipping - the addition of T80 nov+dec 2023 data - increasing loss by 15% when prediction is too high, up from 10% - use of torch.compile to speed up training by over 25% ```yaml experiment-name: 2560--S9-514G-T80-augtodec2023-more-wdl-skip-15p-more-loss-high-q-sk28 training-dataset: # https://github.com/official-stockfish/Stockfish/pull/4782 - /data/S6-514G-1ee1aba5ed.binpack - /data/test80-aug2023-2tb7p.v6.min.binpack - /data/test80-sep2023-2tb7p.binpack - /data/test80-oct2023-2tb7p.binpack - /data/test80-nov2023-2tb7p.binpack - /data/test80-dec2023-2tb7p.binpack early-fen-skipping: 28 start-from-engine-test-net: True nnue-pytorch-branch: linrock/nnue-pytorch/r21-more-wdl-skip-15p-more-loss-high-q-torch-compile num-epochs: 1000 lr: 4.375e-4 gamma: 0.995 start-lambda: 1.0 end-lambda: 0.7 ``` Epoch 819 trained with the above config led to this PR. Use of torch.compile decorators in nnue-pytorch model.py was found to speed up training by at least 25% on Ampere gpus when using recent pytorch compiled with cuda 12: https://catalog.ngc.nvidia.com/orgs/nvidia/containers/pytorch See recent main net PRs for more info on - ranger21 and more WDL skipping: https://github.com/official-stockfish/Stockfish/pull/4942 - increasing loss when Q is too high: https://github.com/official-stockfish/Stockfish/pull/4972 Training data can be found at: https://robotmoon.com/nnue-training-data/ Passed STC: https://tests.stockfishchess.org/tests/view/65cd76151d8e83c78bfd2f52 LLR: 2.98 (-2.94,2.94) <0.00,2.00> Total: 78336 W: 20504 L: 20115 D: 37717 Ptnml(0-2): 317, 9225, 19721, 9562, 343 Passed LTC: https://tests.stockfishchess.org/tests/view/65ce5be61d8e83c78bfd43e9 LLR: 2.95 (-2.94,2.94) <0.50,2.50> Total: 41016 W: 10492 L: 10159 D: 20365 Ptnml(0-2): 22, 4533, 11071, 4854, 28 closes https://github.com/official-stockfish/Stockfish/pull/5056 Bench: 1351997
Copyright 2011–2024 Next Chess Move LLC