| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Guenther Demetz
Date: Fri Jul 19 08:32:16 2024 +0200 Timestamp: 1721370736 Simplify improving condition if we were in check at our previous move we look back until we weren't in check and take the staticEval of that position as reference. Passed STC: https://tests.stockfishchess.org/tests/view/668ba7b65034141ae5996665 LLR: 2.93 (-2.94,2.94) <-1.75,0.25> Total: 74784 W: 19454 L: 19274 D: 36056 Elo +0.84 Ptnml(0-2): 260, 8874, 18952, 9038, 268 Passted LTC: https://tests.stockfishchess.org/tests/view/668cb2db5034141ae599678b LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 241488 W: 61166 L: 61171 D: 119151 Elo -0.01 Ptnml(0-2): 190, 27154, 66062, 27147, 191 closes https://github.com/official-stockfish/Stockfish/pull/5492 bench: 1368313 see source |
| Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Linmiao Xu
Date: Mon Jul 15 13:38:14 2024 +0200 Timestamp: 1721043494 Separate eval params for smallnet and main net Values found with spsa around 80% of 120k games at 60+0.6: https://tests.stockfishchess.org/tests/view/669205dac6827afcdcee3ea4 Passed STC: https://tests.stockfishchess.org/tests/view/6692928b4ff211be9d4e98a9 LLR: 2.96 (-2.94,2.94) <0.00,2.00> Total: 313696 W: 81107 L: 80382 D: 152207 Elo +0.80 Ptnml(0-2): 934, 36942, 80363, 37683, 926 Passed LTC: https://tests.stockfishchess.org/tests/view/6692aab54ff211be9d4e9915 LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 228420 W: 57903 L: 57190 D: 113327 Elo +1.08 Ptnml(0-2): 131, 25003, 63243, 25688, 145 closes https://github.com/official-stockfish/Stockfish/pull/5486 bench 1319322 see source |
| Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Michael Chaly
Date: Mon Jul 15 13:35:06 2024 +0200 Timestamp: 1721043306 Remove quiet history pruning depth limit This patch removes lmrDepth limit for quiet moves history based pruning. Previously removal of this type of depth limits was considered bad because it was performing bad for matetrack - but with this pruning heuristic this shouldn't be that bad because it's "naturally" depth limited by history threshold and should be completely disabled at depth >= 15 or so. Also this heuristic in previous years was known to scale non-linearly - bigger lmrDepth thresholds were better at longer time controls and removing it completely probably should scale pretty well. Passed STC: https://tests.stockfishchess.org/tests/view/6692b89b4ff211be9d4eab21 LLR: 2.93 (-2.94,2.94) <-1.75,0.25> Total: 114464 W: 29675 L: 29545 D: 55244 Elo +0.39 Ptnml(0-2): 372, 12516, 31329, 12640, 375 Passed LTC: https://tests.stockfishchess.org/tests/view/6692c4554ff211be9d4eab3d LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 67746 W: 17182 L: 17014 D: 33550 Elo +0.86 Ptnml(0-2): 28, 6993, 19652, 7183, 17 closes https://github.com/official-stockfish/Stockfish/pull/5485 Bench: 1250388 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: MinetaS
Date: Mon Jul 15 13:30:47 2024 +0200 Timestamp: 1721043047 Use ValueList to represent searched moves array This PR replaces a pair of array and size with existing ValueList class. Removes two local variables in search and two parameters of update_all_stats. Passed non-regression STC: LLR: 2.93 (-2.94,2.94) <-1.75,0.25> Total: 227040 W: 58472 L: 58463 D: 110105 Elo +0.01 Ptnml(0-2): 495, 23572, 65427, 23481, 545 https://tests.stockfishchess.org/tests/view/669299204ff211be9d4e98dc closes https://github.com/official-stockfish/Stockfish/pull/5484 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: FauziAkram
Date: Mon Jul 15 13:27:42 2024 +0200 Timestamp: 1721042862 bonus calculation for prior countermoves Introduce a new term to the bonus calculation for prior countermoves Passed STC: LLR: 2.93 (-2.94,2.94) <0.00,2.00> Total: 140896 W: 36545 L: 36079 D: 68272 Elo +1.15 Ptnml(0-2): 383, 16505, 36217, 16949, 394 https://tests.stockfishchess.org/tests/view/6691c73cc6827afcdcee1816 Passed LTC: LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 126660 W: 32089 L: 31587 D: 62984 Elo +1.38 Ptnml(0-2): 63, 13774, 35154, 14276, 63 https://tests.stockfishchess.org/tests/view/6691cdc4c6827afcdcee1930 closes https://github.com/official-stockfish/Stockfish/pull/5483 bench: 1250388 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Linmiao Xu
Date: Mon Jul 15 13:24:14 2024 +0200 Timestamp: 1721042654 Set best value to futility value after pruned quiet move Passed non-regression STC: https://tests.stockfishchess.org/tests/view/6691592f5034141ae599c68d LLR: 2.93 (-2.94,2.94) <-1.75,0.25> Total: 278496 W: 71818 L: 71865 D: 134813 Elo -0.06 Ptnml(0-2): 865, 33311, 70978, 33194, 900 Passed non-regression LTC: https://tests.stockfishchess.org/tests/view/66918fca5034141ae599e761 LLR: 2.95 (-2.94,2.94) <-1.75,0.25> Total: 202986 W: 51048 L: 51013 D: 100925 Elo +0.06 Ptnml(0-2): 107, 22552, 56133, 22601, 100 closes https://github.com/official-stockfish/Stockfish/pull/5480 bench 1715206 see source |
| Windows x64 for modern computers + AVX2 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Michael Chaly
Date: Mon Jul 15 13:22:31 2024 +0200 Timestamp: 1721042551 Decrease delta Decrease delta in aspiration windows - both initial value and quadratic function of previous best value. Passed STC: https://tests.stockfishchess.org/tests/view/6691a52ec6827afcdcee1569 LLR: 2.93 (-2.94,2.94) <0.00,2.00> Total: 55456 W: 14449 L: 14107 D: 26900 Elo +2.14 Ptnml(0-2): 174, 6416, 14193, 6784, 161 Passed LTC: https://tests.stockfishchess.org/tests/view/6691aac1c6827afcdcee1625 LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 107940 W: 27530 L: 27065 D: 53345 Elo +1.50 Ptnml(0-2): 52, 11787, 29840, 12226, 65 closes https://github.com/official-stockfish/Stockfish/pull/5479 bench 1547707 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: MinetaS
Date: Mon Jul 15 13:19:46 2024 +0200 Timestamp: 1721042386 Simplify futility_move_count This patch reverts changes from #4032 which was introduced as a speedup. Modern compilers no longer use DIV/IDIV instructions, potentially making the explicit branch perform worse. Since evaluations spend significantly more time now, the impact of the speedup in search diminishes with old compilers as well. GCC 14.1.0 profile-build, x86-64-vnni512 ``` .text:000000014010FEA9 mov ecx, [rsp+3FB8h+var_3F5C] ... .text:000000014010FEBD mov r10d, ecx .text:000000014010FEC0 imul r10d, ecx .text:000000014010FEC4 mov ecx, dword ptr [rsp+3FB8h+var_3F44+4] .text:000000014010FEC8 add r10d, 3 .text:000000014010FECC mov r11d, r10d .text:000000014010FECF sar r11d, 1 .text:000000014010FED2 cmp [rsp+3FB8h+var_3EE7], 0 .text:000000014010FEDA cmovnz r11d, r10d ``` LLVM 18.1.18 profile-build, x86-64-vnni512 ``` .text:0000000140001EDC mov [rsp+40h+arg_E0], r13 .text:0000000140001EE4 movsxd rcx, r13d .text:0000000140001EE7 mov rax, rcx .text:0000000140001EEA mov [rsp+40h+arg_B8], rcx .text:0000000140001EF2 imul eax, eax .text:0000000140001EF5 add eax, 3 .text:0000000140001EF8 mov ecx, [rsp+40h+arg_8C] .text:0000000140001EFF shrx eax, eax, ecx .text:0000000140001F04 mov [rsp+40h+arg_190], rax ``` Passed non-regression STC: LLR: 2.96 (-2.94,2.94) <-1.75,0.25> Total: 109504 W: 28420 L: 28280 D: 52804 Elo +0.44 Ptnml(0-2): 355, 12326, 29273, 12420, 378 https://tests.stockfishchess.org/tests/view/6690dc095034141ae599c5fe closes https://github.com/official-stockfish/Stockfish/pull/5478 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Shawn Xu
Date: Mon Jul 15 13:18:37 2024 +0200 Timestamp: 1721042317 Unify Movepick Initializer Passed Non-regression STC: LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 168704 W: 43524 L: 43455 D: 81725 Elo +0.14 Ptnml(0-2): 414, 17173, 49076, 17308, 381 https://tests.stockfishchess.org/tests/view/66904b7b5034141ae599a197 Passed Non-regression LTC: LLR: 2.95 (-2.94,2.94) <-1.75,0.25> Total: 120294 W: 30473 L: 30364 D: 59457 Elo +0.31 Ptnml(0-2): 40, 10974, 38032, 11039, 62 https://tests.stockfishchess.org/tests/view/66905b235034141ae599a223 closes https://github.com/official-stockfish/Stockfish/pull/5477 bench 1459677 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Shawn Xu
Date: Mon Jul 15 13:15:35 2024 +0200 Timestamp: 1721042135 Simplify Away Refutation Stage Simplify away killer stage to a constant bonus given to the killer move during quiet move scoring. Passed Non-regression STC (Against then-pending PR #5472): LLR: 2.93 (-2.94,2.94) <-1.75,0.25> Total: 106176 W: 27685 L: 27539 D: 50952 Elo +0.48 Ptnml(0-2): 410, 12765, 26637, 12821, 455 https://tests.stockfishchess.org/tests/view/668dd0835034141ae5999e8f Passed Non-regression LTC: LLR: 2.95 (-2.94,2.94) <-1.75,0.25> Total: 92472 W: 23426 L: 23276 D: 45770 Elo +0.56 Ptnml(0-2): 55, 10376, 25215, 10544, 46 https://tests.stockfishchess.org/tests/view/669019e45034141ae5999fd2 closes https://github.com/official-stockfish/Stockfish/pull/5476 Bench 1459677 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Andyson007
Date: Mon Jul 15 13:14:57 2024 +0200 Timestamp: 1721042097 Fixed non UCI compliance print `<empty>` and accept `<empty>` for UCI string options, accepting empty strings as well. Internally use empty strings (`""`). closes https://github.com/official-stockfish/Stockfish/pull/5474 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: yl25946
Date: Thu Jul 11 07:29:49 2024 +0200 Timestamp: 1720675789 removed second killer move STC with movepicker rewrite: LLR: 2.93 (-2.94,2.94) <-1.75,0.25> Total: 46656 W: 12208 L: 11995 D: 22453 Elo +1.59 Ptnml(0-2): 203, 5461, 11777, 5694, 193 https://tests.stockfishchess.org/tests/view/668d98a15034141ae5999e68 Earlier version passed STC: LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 468896 W: 120999 L: 120054 D: 227843 Elo +0.70 Ptnml(0-2): 1207, 55209, 120639, 56218, 1175 https://tests.stockfishchess.org/tests/view/668b17d2cf91c430fca58630 Earlier version passed LTC: LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 550524 W: 139553 L: 139877 D: 271094 Elo -0.20 Ptnml(0-2): 333, 61646, 151616, 61346, 321 https://tests.stockfishchess.org/tests/view/668b2e04cf91c430fca586b1 closes https://github.com/official-stockfish/Stockfish/pull/5472 bench 1234309 Co-authored-by: rn5f107s2 <> see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Thu Jul 11 07:29:44 2024 +0200 Timestamp: 1720675784 Provide more info on found TB files now uses the following format: `info string Found 510 WDL and 510 DTZ tablebase files (up to 6-man).` this clarifies exactly what has been found, as the difference matters, e.g. for the PV extension of TB scores. closes https://github.com/official-stockfish/Stockfish/pull/5471 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Stéphane Nicolet
Date: Thu Jul 11 07:29:33 2024 +0200 Timestamp: 1720675773 Clean up comments in code - Capitalize comments - Reformat multi-lines comments to equalize the widths of the lines - Try to keep the width of comments around 85 characters - Remove periods at the end of single-line comments closes https://github.com/official-stockfish/Stockfish/pull/5469 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Disservin
Date: Thu Jul 11 07:21:04 2024 +0200 Timestamp: 1720675264 CI give correct permissions for the clang-format action closes https://github.com/official-stockfish/Stockfish/pull/5470 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: yl25946
Date: Tue Jul 9 18:54:12 2024 +0200 Timestamp: 1720544052 Move Loop Consistency in Probcut In probcut move loop, everything is enclosed within a large if statement. I've changed it to guard clauses to stay consistent with other move loops. closes https://github.com/official-stockfish/Stockfish/pull/5463 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Linmiao Xu
Date: Tue Jul 9 18:49:28 2024 +0200 Timestamp: 1720543768 Update default main net to nn-e8bac1c07a5a.nnue Created by modifying L2 weights from the previous main net (nn-74f1d263ae9a.nnue) with params found by spsa around 9k / 120k games at 120+1.2. 370 spsa params - L2 weights in nn-74f1d263ae9a.nnue where |val| >= 50 A: 6000, alpha: 0.602, gamma: 0.101 weights: [-127, 127], c_end = 6 To print the spsa params with nnue-pytorch: ``` import features from serialize import NNUEReader feature_set = features.get_feature_set_from_name("HalfKAv2_hm") with open("nn-74f1d263ae9a.nnue", "rb") as f: model = NNUEReader(f, feature_set).model c_end = 6 for i in range(8): for j in range(32): for k in range(30): value = int(model.layer_stacks.l2.weight[32 * i + j, k] * 64) if abs(value) >= 50: print(f"twoW[{i}][{j}][{k}],{value},-127,127,{c_end},0.0020") ``` Among the 370 params, 229 weights were changed. avg change: 0.0961 ± 1.67 range: [-4, 3] The number of weights changed, grouped by layer stack index, shows more weights were modified in the lower piece count buckets: [54, 52, 29, 23, 22, 18, 14, 17] Found with the same method described in: https://github.com/official-stockfish/Stockfish/pull/5459 Passed STC: https://tests.stockfishchess.org/tests/view/668aec9a58083e5fd88239e7 LLR: 3.00 (-2.94,2.94) <0.00,2.00> Total: 52384 W: 13569 L: 13226 D: 25589 Elo +2.27 Ptnml(0-2): 127, 6141, 13335, 6440, 149 Passed LTC: https://tests.stockfishchess.org/tests/view/668af50658083e5fd8823a0b LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 46974 W: 12006 L: 11668 D: 23300 Elo +2.50 Ptnml(0-2): 25, 4992, 13121, 5318, 31 closes https://github.com/official-stockfish/Stockfish/pull/5466 bench 1300471 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Shawn Xu
Date: Tue Jul 9 18:44:39 2024 +0200 Timestamp: 1720543479 Simplify Probcut Malus Passed Non-regression STC: LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 74880 W: 19261 L: 19083 D: 36536 Elo +0.83 Ptnml(0-2): 202, 8861, 19120, 9071, 186 https://tests.stockfishchess.org/tests/view/668a0966eca84f4d25864cba Passed Non-regression LTC: LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 263916 W: 66690 L: 66718 D: 130508 Elo -0.04 Ptnml(0-2): 125, 29348, 73040, 29320, 125 https://tests.stockfishchess.org/tests/view/668a17e3eca84f4d25864e91 closes https://github.com/official-stockfish/Stockfish/pull/5464 bench 1331408 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Stéphane Nicolet
Date: Tue Jul 9 18:42:21 2024 +0200 Timestamp: 1720543341 Fix compilation on Apple Always use the posix function posix_memalign() as aligned memory allocator on Apple computers. This should allow to compile Stockfish out of the box on all versions of Mac OS X. Patch tested on the following systems (apart from the CI) : • Mac OS 10.9.6 (arch x86-64-sse41-popcnt) with gcc-10 • Mac OS 10.13.6 (arch x86-64-bmi2) with gcc-10, gcc-14 and clang-11 • Mac OS 14.1.1 (arch apple-silicon) with clang-15 closes https://github.com/official-stockfish/Stockfish/pull/5462 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Michael Chaly
Date: Tue Jul 9 18:41:36 2024 +0200 Timestamp: 1720543296 Further simplify stat bonuses Based on recent simplification by linrock Since it completely removed any special bonuses based on values difference and made it flat stat_bonus(depth + 1) I got an idea that we might as well remove all (depth + 1) bonuses and make them usual depth bonuses. Also removes weird negative bonus for depth 1 as a side effect. Passed STC: https://tests.stockfishchess.org/tests/view/6689d817eca84f4d25863746 LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 18080 W: 4789 L: 4552 D: 8739 Elo +4.55 Ptnml(0-2): 46, 1987, 4727, 2244, 36 Passed LTC: https://tests.stockfishchess.org/tests/view/6689daa4eca84f4d258639d7 LLR: 2.95 (-2.94,2.94) <-1.75,0.25> Total: 109062 W: 27548 L: 27417 D: 54097 Elo +0.42 Ptnml(0-2): 58, 11983, 30293, 12164, 33 Passed direct LTC vs linrock patch: https://tests.stockfishchess.org/tests/view/668a46f8eca84f4d25866fe9 LLR: 2.95 (-2.94,2.94) <-1.75,0.25> Total: 100002 W: 25351 L: 25209 D: 49442 Elo +0.49 Ptnml(0-2): 54, 11119, 27529, 11229, 70 closes https://github.com/official-stockfish/Stockfish/pull/5461 Bench 1175744 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Joost VandeVondele
Date: Tue Jul 9 18:36:46 2024 +0200 Timestamp: 1720543006 Fix output for GUI Fritz 19 can hang with the current way to provide output, i.e. too much output in a short time for a mate / depth 245 found quickly. fallout from 25361e514bffb81284d4311601a9f7a4a7ced79b closes https://github.com/official-stockfish/Stockfish/pull/5460 No functional change see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Linmiao Xu
Date: Tue Jul 9 18:35:23 2024 +0200 Timestamp: 1720542923 Update default main net to nn-74f1d263ae9a.nnue Created by setting output weights (256) and biases (8) of the previous main net nn-ddcfb9224cdb.nnue to values found around 12k / 120k spsa games at 120+1.2 This used modified fishtest dev workers to construct .nnue files from spsa params, then load them with EvalFile when running tests: https://github.com/linrock/fishtest/tree/spsa-file-modified-nnue/worker Inspired by researching loading spsa params from files: https://github.com/official-stockfish/fishtest/pull/1926 Scripts for modifying nnue files and preparing params: https://github.com/linrock/nnue-pytorch/tree/no-gpu-modify-nnue spsa params: weights: [-127, 127], c_end = 6 biases: [-8192, 8192], c_end = 64 Example of reading output weights and biases from the previous main net using nnue-pytorch and printing spsa params in a format compatible with fishtest: ``` import features from serialize import NNUEReader feature_set = features.get_feature_set_from_name("HalfKAv2_hm") with open("nn-ddcfb9224cdb.nnue", "rb") as f: model = NNUEReader(f, feature_set).model c_end_weights = 6 c_end_biases = 64 for i in range(8): for j in range(32): value = round(int(model.layer_stacks.output.weight[i, j] * 600 * 16) / 127) print(f"oW[{i}][{j}],{value},-127,127,{c_end_weights},0.0020") for i in range(8): value = int(model.layer_stacks.output.bias[i] * 600 * 16) print(f"oB[{i}],{value},-8192,8192,{c_end_biases},0.0020") ``` For more info on spsa tuning params in nets: https://github.com/official-stockfish/Stockfish/pull/5149 https://github.com/official-stockfish/Stockfish/pull/5254 Passed STC: https://tests.stockfishchess.org/tests/view/66894d64e59d990b103f8a37 LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 32000 W: 8443 L: 8137 D: 15420 Elo +3.32 Ptnml(0-2): 80, 3627, 8309, 3875, 109 Passed LTC: https://tests.stockfishchess.org/tests/view/6689668ce59d990b103f8b8b LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 172176 W: 43822 L: 43225 D: 85129 Elo +1.20 Ptnml(0-2): 97, 18821, 47633, 19462, 75 closes https://github.com/official-stockfish/Stockfish/pull/5459 bench 1120091 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: MinetaS
Date: Tue Jul 9 18:34:00 2024 +0200 Timestamp: 1720542840 Clean up refutations array in MovePicker This is a follow-up cleanup to a45c2bc34ae03dd35402e6cf26d515bae1425517. closes https://github.com/official-stockfish/Stockfish/pull/5458 No functional change see source |
| Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 + SSSE3 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Linmiao Xu
Date: Tue Jul 9 18:31:02 2024 +0200 Timestamp: 1720542662 Larger bonus when updating quiet stats Also removes unused arguments to update_all_stats to fix compiler warnings about unused parameters. Passed non-regression STC: https://tests.stockfishchess.org/tests/view/6689a79a0fdd852d63cf52e9 LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 26496 W: 6901 L: 6669 D: 12926 Elo +3.04 Ptnml(0-2): 62, 3094, 6715, 3304, 73 Passed non-regression LTC: https://tests.stockfishchess.org/tests/view/6689a9960fdd852d63cf532d LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 41214 W: 10373 L: 10173 D: 20668 Elo +1.69 Ptnml(0-2): 11, 4491, 11412, 4673, 20 closes https://github.com/official-stockfish/Stockfish/pull/5456 bench 1169958 see source |
| Windows x64 for Haswell CPUs Windows x64 for modern computers + AVX2 Windows x64 for modern computers Windows x64 Linux x64 for Haswell CPUs Linux x64 for modern computers + AVX2 Linux x64 for modern computers Linux x64 + SSSE3 Linux x64 | Author: Michael Chaly
Date: Tue Jul 9 18:29:09 2024 +0200 Timestamp: 1720542549 Adjust usage of previous statscore in bonus assignments This patch adjusts usage of previous statscore for bonus assginments - allowing it for any statscores and clamping it to wider range. Passed STC: https://tests.stockfishchess.org/tests/view/66892e76e59d990b103f6a91 LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 431520 W: 111767 L: 110872 D: 208881 Elo +0.72 Ptnml(0-2): 1180, 51165, 110133, 52144, 1138 Passed LTC: https://tests.stockfishchess.org/tests/view/66897176e59d990b103f9605 LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 143184 W: 36463 L: 35929 D: 70792 Elo +1.30 Ptnml(0-2): 55, 15540, 39863, 16084, 50 closes https://github.com/official-stockfish/Stockfish/pull/5455 bench 1330556 see source |