Dev Builds » 20230114-0712

You are viewing an old NCM Stockfish dev build test. You may find the most recent dev build tests using Stockfish 15 as the baseline here.

Use this dev build

NCM plays each Stockfish dev build 20,000 times against Stockfish 7. This yields an approximate Elo difference and establishes confidence in the strength of the dev builds.

Summary

Host Duration Avg Base NPS Games Wins Losses Draws Elo
ncm-et-3 11:16:44 1950103 3331 2850 5 476 +441.63 ± 15.61
ncm-et-4 11:16:12 1951952 3310 2873 4 433 +458.59 ± 16.38
ncm-et-9 11:16:50 1955317 3337 2864 5 468 +445.07 ± 15.75
ncm-et-10 11:16:34 1956109 3336 2851 0 485 +442.29 ± 15.45
ncm-et-13 11:16:23 1955824 3341 2912 6 423 +462.87 ± 16.58
ncm-et-15 11:16:59 1957122 3346 2909 1 436 +461.87 ± 16.31
20001 17259 21 2721 +451.85 ± 6.52

Test Detail

ID Host Started (UTC) Duration Base NPS Games Wins Losses Draws Elo CLI PGN
164124 ncm-et-4 2023-01-15 08:14 01:01:55 1953023 310 262 1 47 +426.58 ± 50.72
164123 ncm-et-10 2023-01-15 08:09 01:07:07 1959852 336 286 0 50 +437.92 ± 49.0
164122 ncm-et-3 2023-01-15 08:08 01:07:50 1951991 331 278 0 53 +424.13 ± 47.49
164121 ncm-et-15 2023-01-15 08:07 01:08:57 1965225 346 303 0 43 +471.5 ± 53.14
164120 ncm-et-9 2023-01-15 08:07 01:09:06 1953910 337 287 1 49 +434.77 ± 49.65
164119 ncm-et-13 2023-01-15 08:06 01:10:20 1952347 341 303 0 38 +491.63 ± 56.79
164118 ncm-et-4 2023-01-15 06:31 01:42:33 1954198 500 430 0 70 +449.35 ± 41.19
164117 ncm-et-3 2023-01-15 06:26 01:41:27 1945598 500 434 2 64 +454.76 ± 43.27
164116 ncm-et-10 2023-01-15 06:25 01:42:42 1958957 500 426 0 74 +438.95 ± 40.01
164115 ncm-et-9 2023-01-15 06:24 01:41:40 1953144 500 427 1 72 +438.95 ± 40.66
164114 ncm-et-15 2023-01-15 06:24 01:42:22 1944795 500 439 1 60 +471.92 ± 44.73
164113 ncm-et-13 2023-01-15 06:23 01:42:02 1957110 500 445 0 55 +494.02 ± 46.77
164112 ncm-et-4 2023-01-15 04:48 01:42:29 1955729 500 432 0 68 +454.76 ± 41.82
164111 ncm-et-10 2023-01-15 04:44 01:40:22 1957117 500 430 0 70 +449.35 ± 41.19
164110 ncm-et-15 2023-01-15 04:43 01:39:50 1967081 500 437 0 63 +468.95 ± 43.54
164109 ncm-et-3 2023-01-15 04:43 01:41:54 1929928 500 430 2 68 +444.09 ± 41.92
164108 ncm-et-9 2023-01-15 04:42 01:41:54 1947278 500 427 0 73 +441.5 ± 40.29
164107 ncm-et-13 2023-01-15 04:41 01:41:14 1948178 500 428 1 71 +441.5 ± 40.95
164106 ncm-et-4 2023-01-15 03:05 01:41:43 1954689 500 435 0 65 +463.15 ± 42.83
164105 ncm-et-10 2023-01-15 03:03 01:40:58 1953033 500 438 0 62 +471.92 ± 43.9
164104 ncm-et-3 2023-01-15 03:01 01:41:33 1955807 500 427 1 72 +438.95 ± 40.66
164103 ncm-et-15 2023-01-15 03:00 01:42:46 1956066 500 429 0 71 +446.7 ± 40.89
164102 ncm-et-9 2023-01-15 02:59 01:41:45 1954680 500 435 1 64 +460.32 ± 43.24
164101 ncm-et-13 2023-01-15 02:59 01:41:22 1956485 500 438 2 60 +466.04 ± 44.75
164100 ncm-et-4 2023-01-15 01:22 01:42:54 1958019 500 435 0 65 +463.15 ± 42.83
164099 ncm-et-10 2023-01-15 01:21 01:41:21 1951504 500 422 0 78 +429.05 ± 38.92
164098 ncm-et-9 2023-01-15 01:20 01:39:16 1962389 500 438 1 61 +468.96 ± 44.35
164097 ncm-et-15 2023-01-15 01:19 01:40:25 1962803 500 426 0 74 +438.95 ± 40.01
164096 ncm-et-3 2023-01-15 01:18 01:42:23 1946350 500 422 0 78 +429.05 ± 38.92
164095 ncm-et-13 2023-01-15 01:17 01:41:16 1952945 500 438 0 62 +471.92 ± 43.9
164094 ncm-et-4 2023-01-14 23:39 01:42:23 1952120 500 442 2 56 +477.99 ± 46.39
164093 ncm-et-10 2023-01-14 23:38 01:42:28 1954705 500 421 0 79 +426.65 ± 38.66
164092 ncm-et-3 2023-01-14 23:38 01:39:44 1962351 500 437 0 63 +468.95 ± 43.54
164091 ncm-et-9 2023-01-14 23:36 01:42:44 1951815 500 412 1 87 +404.05 ± 36.82
164090 ncm-et-15 2023-01-14 23:36 01:42:52 1944649 500 449 0 51 +507.87 ± 48.67
164089 ncm-et-13 2023-01-14 23:35 01:41:19 1960629 500 430 2 68 +444.09 ± 41.92
164088 ncm-et-4 2023-01-14 21:56 01:42:15 1935890 500 437 1 62 +466.04 ± 43.97
164087 ncm-et-13 2023-01-14 21:56 01:38:50 1963076 500 430 1 69 +446.7 ± 41.57
164086 ncm-et-10 2023-01-14 21:56 01:41:36 1957595 500 428 0 72 +444.08 ± 40.59
164085 ncm-et-3 2023-01-14 21:55 01:41:53 1958700 500 422 0 78 +429.05 ± 38.92
164084 ncm-et-15 2023-01-14 21:55 01:39:47 1959237 500 426 0 74 +438.95 ± 40.01
164083 ncm-et-9 2023-01-14 21:55 01:40:25 1964009 500 438 0 62 +471.92 ± 43.9

Commit

Commit ID 3d2381d76d7bf9686ef0e0671f60c3b885a7058a
Author Linmiao Xu
Date 2023-01-14 07:12:11 UTC
Update default net to nn-1e7ca356472e.nnue Created by retraining the master net on a dataset composed of: * The Leela-dfrc_n5000.binpack dataset filtered with depth6 multipv2 search to remove positions with only one good move, in addition to removing positions where either of the two best moves are captures * The same Leela T80 oct+nov 2022 training data used in recent best datasets * Additional Leela training data from T60 nov+dec 2021 and T79 apr+may 2022 Trained with end lambda 0.7 and started with max epoch 800. All positions with ply <= 28 were skipped: ``` python easy_train.py \ --experiment-name leela95-dfrc96-mpv-eval-fonly-T80octnov-T79aprmayT60novdec-12tb7p-sk28-lambda7 \ --training-dataset /data/leela95-dfrc96-mpv-eval-fonly-T80octnov-T79aprmayT60novdec-12tb7p.binpack \ --nnue-pytorch-branch linrock/nnue-pytorch/misc-fixes-skip-ply-lteq-28 \ --start-from-engine-test-net True \ --gpus "0," \ --start-lambda 1.0 \ --end-lambda 0.7 \ --gamma 0.995 \ --lr 4.375e-4 \ --tui False \ --seed $RANDOM \ --max_epoch 800 ``` Around epoch 780, training was manually paused and max epoch increased to 920 before resuming. During depth6 multipv2 data filtering, positions were considered to have only one good move if the score of the best move was significantly better than the 2nd best move in a way that changes the outcome of the game: * the best move leads to a significant advantage while the 2nd best move equalizes or loses * the best move is about equal while the 2nd best move loses The modified stockfish branch and exact score thresholds used for filtering are at: https://github.com/linrock/Stockfish/tree/tools-filter-multipv2-eval-diff/src/filter About 95% of the Leela portion and 96% of the DFRC portion of the Leela-dfrc_n5000.binpack dataset was filtered. Unfiltered parts of the dataset were left out. The additional Leela training data from T60 nov+dec 2021 and T79 apr+may 2022 was WDL-rescored with about 12TB of syzygy 7-piece tablebases where the material difference is less than around 6 pawns. Best moves were exported to .plain data files during data conversion with the lc0 rescorer. The exact training data can be found at: https://robotmoon.com/nnue-training-data/ Local elo at 25k nodes per move experiment_leela95-dfrc96-mpv-eval-fonly-T80octnov-T79aprmayT60novdec-12tb7p-sk28-lambda7 run_0/nn-epoch899.nnue : 3.8 +/- 1.6 Passed STC https://tests.stockfishchess.org/tests/view/63bed1f540aa064159b9c89b LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 103344 W: 27392 L: 26991 D: 48961 Ptnml(0-2): 333, 11223, 28099, 11744, 273 Passed LTC https://tests.stockfishchess.org/tests/view/63c010415705810de2deb3ec LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 21712 W: 5891 L: 5619 D: 10202 Ptnml(0-2): 12, 2022, 6511, 2304, 7 closes https://github.com/official-stockfish/Stockfish/pull/4338 bench 4106793
Copyright 2011–2024 Next Chess Move LLC