Dev Builds » 20230329-1937

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 09:39:55 1948911 3329 2840 4 485 +438.83 +/- 15.46
ncm-et-4 09:39:45 1956874 3344 2855 6 483 +438.92 +/- 15.5
ncm-et-9 09:39:56 1950278 3307 2818 4 485 +437.59 +/- 15.46
ncm-et-10 09:39:30 1950333 3319 2812 8 499 +430.06 +/- 15.25
ncm-et-13 09:39:39 1959423 3378 2889 6 483 +440.82 +/- 15.5
ncm-et-15 09:39:46 1952824 3323 2844 4 475 +442.34 +/- 15.63
20000 17058 32 2910 +438.07 +/- 6.3

Test Detail

ID Host Started (UTC) Duration Base NPS Games Wins Losses Draws Elo CLI PGN
181596 ncm-et-9 2023-03-31 14:04 00:54:03 1949639 307 266 0 41 +458.14 +/- 54.45
181595 ncm-et-10 2023-03-31 14:02 00:55:34 1947112 319 263 0 56 +406.69 +/- 46.09
181594 ncm-et-3 2023-03-31 14:01 00:56:11 1958335 329 282 0 47 +445.57 +/- 50.64
181593 ncm-et-15 2023-03-31 14:01 00:56:58 1958487 323 271 0 52 +423.11 +/- 47.96
181592 ncm-et-4 2023-03-31 13:59 00:59:02 1959438 344 302 0 42 +474.79 +/- 53.81
181591 ncm-et-13 2023-03-31 13:54 01:04:02 1959432 378 320 1 57 +428.95 +/- 45.87
181585 ncm-et-9 2023-03-31 12:36 01:26:45 1956972 500 413 0 87 +408.38 +/- 36.75
181584 ncm-et-3 2023-03-31 12:34 01:26:45 1951152 500 433 0 67 +457.52 +/- 42.15
181583 ncm-et-10 2023-03-31 12:34 01:27:38 1944362 500 426 1 73 +436.43 +/- 40.36
181582 ncm-et-15 2023-03-31 12:33 01:27:02 1932363 500 437 0 63 +468.95 +/- 43.54
181581 ncm-et-4 2023-03-31 12:30 01:27:44 1949376 500 429 0 71 +446.7 +/- 40.89
181580 ncm-et-13 2023-03-31 12:28 01:25:16 1970174 500 436 1 63 +463.16 +/- 43.6
181574 ncm-et-9 2023-03-31 11:08 01:27:37 1952548 500 431 1 68 +449.35 +/- 41.89
181573 ncm-et-3 2023-03-31 11:07 01:26:03 1956057 500 419 0 81 +421.93 +/- 38.15
181572 ncm-et-10 2023-03-31 11:07 01:26:01 1964672 500 422 2 76 +424.28 +/- 39.56
181571 ncm-et-15 2023-03-31 11:05 01:27:36 1947897 500 427 0 73 +441.5 +/- 40.29
181570 ncm-et-4 2023-03-31 11:04 01:25:41 1967544 500 435 3 62 +454.76 +/- 43.98
181569 ncm-et-13 2023-03-31 11:01 01:25:42 1971732 500 430 1 69 +446.7 +/- 41.57
181560 ncm-et-9 2023-03-31 09:39 01:27:43 1957889 500 430 0 70 +449.35 +/- 41.19
181559 ncm-et-3 2023-03-31 09:39 01:27:25 1951461 500 441 1 58 +477.99 +/- 45.54
181558 ncm-et-10 2023-03-31 09:38 01:28:20 1954505 500 420 1 79 +421.93 +/- 38.73
181557 ncm-et-4 2023-03-31 09:37 01:26:24 1954371 500 419 0 81 +421.93 +/- 38.15
181556 ncm-et-15 2023-03-31 09:36 01:27:42 1956647 500 437 0 63 +468.95 +/- 43.54
181555 ncm-et-13 2023-03-31 09:34 01:26:08 1948748 500 422 0 78 +429.05 +/- 38.92
181549 ncm-et-9 2023-03-31 08:12 01:26:37 1956660 500 435 2 63 +457.52 +/- 43.63
181548 ncm-et-3 2023-03-31 08:11 01:27:11 1947538 500 424 0 76 +433.94 +/- 39.45
181547 ncm-et-10 2023-03-31 08:10 01:27:39 1946004 500 439 0 61 +474.93 +/- 44.28
181546 ncm-et-4 2023-03-31 08:09 01:27:29 1950102 500 414 1 85 +408.38 +/- 37.27
181545 ncm-et-15 2023-03-31 08:09 01:27:12 1951018 500 421 2 77 +421.93 +/- 39.29
181544 ncm-et-13 2023-03-31 08:07 01:26:54 1951154 500 413 3 84 +401.92 +/- 37.57
181538 ncm-et-9 2023-03-31 06:44 01:27:58 1938077 500 416 0 84 +415.04 +/- 37.43
181537 ncm-et-3 2023-03-31 06:42 01:28:51 1936697 500 419 0 81 +421.93 +/- 38.15
181536 ncm-et-10 2023-03-31 06:41 01:27:46 1951608 500 428 2 70 +438.95 +/- 41.29
181535 ncm-et-4 2023-03-31 06:41 01:26:55 1958964 500 434 2 64 +454.76 +/- 43.27
181534 ncm-et-15 2023-03-31 06:41 01:27:04 1957514 500 423 2 75 +426.65 +/- 39.83
181533 ncm-et-13 2023-03-31 06:39 01:27:03 1953487 500 440 0 60 +477.98 +/- 44.67
181527 ncm-et-10 2023-03-31 05:14 01:26:32 1944069 500 414 2 84 +406.2 +/- 37.55
181526 ncm-et-13 2023-03-31 05:14 01:24:34 1961239 500 428 0 72 +444.08 +/- 40.59
181525 ncm-et-15 2023-03-31 05:14 01:26:12 1965844 500 428 0 72 +444.08 +/- 40.59
181524 ncm-et-4 2023-03-31 05:14 01:26:30 1958326 500 422 0 78 +429.05 +/- 38.92
181523 ncm-et-3 2023-03-31 05:14 01:27:29 1941139 500 422 3 75 +421.93 +/- 39.85
181522 ncm-et-9 2023-03-31 05:14 01:29:13 1940167 500 427 1 72 +438.95 +/- 40.66

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 2011–2024 Next Chess Move LLC