Active : 9 Machines / 316 Threads / 250.21 MNPS
Priority 5
87Reckless-AVX512nnz_u8diff4.0+0.04
LLR: 0.52 (-2.25, 2.89) [0.00, 3.00]
Games: 48056 W: 12245 L: 12112 D: 23699
Ptnml(0-2): 195, 5120, 13234, 5315, 164
don't use packus
Priority -1
WorldRecklessbest-move-changes-ln1p-2diff8.0+0.08
LLR: 0.04 (-2.25, 2.89) [0.00, 3.00]
Games: 55904 W: 14029 L: 13911 D: 27964
Ptnml(0-2): 96, 5965, 15723, 6061, 107
Priority -5
Fauzi2Recklesstunediff40.0+0.40
Tuning 8 Parameters
6386/80000 Iterations
25546/320000 Games Played
[50% TP] [Threads=8 Hash=512] tune
Finished
87Recklessnnz-pext-2diff4.0+0.04
LLR: -2.73 (-2.25, 2.89) [0.00, 3.00]
Games: 15200 W: 3868 L: 3994 D: 7338
Ptnml(0-2): 70, 1721, 4131, 1621, 57
87Recklessnnz_u8-avx2-onlydiff4.0+0.04
LLR: 2.93 (-2.25, 2.89) [-3.00, 0.00]
Games: 118398 W: 30160 L: 30246 D: 57992
Ptnml(0-2): 466, 13022, 32277, 13000, 434
ensure AVX2/NEON impl is a non-regression
WorldRecklessexpt-26diffN=25000
Elo: -2.07 +- 2.15 (95%) [N=40000]
Games: 40342 W: 11946 L: 12186 D: 16210
Ptnml(0-2): 843, 4848, 9025, 4616, 839
SCALE=280
WorldRecklessexpt-26diff8.0+0.08
LLR: -2.33 (-2.25, 2.89) [0.00, 3.00]
Games: 9650 W: 2346 L: 2461 D: 4843
Ptnml(0-2): 18, 1203, 2500, 1084, 20
SCALE=280
WorldRecklessswiglu-swish-rebaseddiff8.0+0.08
LLR: -2.35 (-2.25, 2.89) [0.00, 3.00]
Games: 6172 W: 1507 L: 1636 D: 3029
Ptnml(0-2): 30, 807, 1532, 696, 21
cj-09-swiglu-swish
WorldRecklessswiglu-swish-rebaseddiffN=25000
Elo: -2.20 +- 2.15 (95%) [N=40000]
Games: 40200 W: 12032 L: 12286 D: 15882
Ptnml(0-2): 821, 4877, 8966, 4607, 829
cj-09-swiglu-swish
WorldRecklessmore-nodes-tmdiff40.0+0.40
LLR: -2.31 (-2.25, 2.89) [0.00, 3.00]
Games: 10274 W: 2474 L: 2574 D: 5226
Ptnml(0-2): 1, 1072, 3092, 970, 2
87Reckless-AVX512nnz_u8-3diff4.0+0.04
LLR: -2.26 (-2.25, 2.89) [0.00, 3.00]
Games: 6728 W: 1685 L: 1800 D: 3243
Ptnml(0-2): 32, 770, 1861, 683, 18
no packus, and redo the unrolling as well; uses all 8 mask registers
SwedisRecklessu8_nnzdiff4.0+0.04
LLR: -2.29 (-2.25, 2.89) [0.00, 3.00]
Games: 19270 W: 4944 L: 5035 D: 9291
Ptnml(0-2): 79, 2127, 5315, 2034, 80
faster NNZ using the fact that L1 < 4 * 2^8 so the indices can be 8 bits instead of 16 my idea but impl is almost copied verbatim from the one anematode wrote https://tests.stockfishchess.org/tests/view/69855f88c5ad5e83c2ca2871
WorldRecklessmore-nodes-tmdiff8.0+0.08
LLR: 2.96 (-2.25, 2.89) [0.00, 3.00]
Games: 97904 W: 24581 L: 24213 D: 49110
Ptnml(0-2): 183, 10409, 27399, 10779, 182
87Recklesssingular-tm2diff8.0+0.08
LLR: -1.25 (-2.25, 2.89) [0.00, 3.00]
Games: 20242 W: 5012 L: 5041 D: 10189
Ptnml(0-2): 45, 2409, 5233, 2398, 36
try multiplier reduction instead
PeregrRecklessrelative-to-absolutediff8.0+0.08
LLR: -2.27 (-2.25, 2.89) [0.00, 3.00]
Games: 43990 W: 10964 L: 11001 D: 22025
Ptnml(0-2): 94, 5274, 11320, 5189, 118
Maintain the same hit-rate when penalizing previous move in early ttcut by using the absolute history value instead of the relative move_count.
87Recklesssingular-tmdiff8.0+0.08
LLR: -0.86 (-2.25, 2.89) [0.00, 3.00]
Games: 2936 W: 708 L: 751 D: 1477
Ptnml(0-2): 9, 357, 772, 328, 2
87Recklessprefetch-threat-index2diff4.0+0.04
LLR: -0.33 (-2.25, 2.89) [0.00, 3.00]
Games: 8910 W: 2321 L: 2321 D: 4268
Ptnml(0-2): 26, 978, 2456, 960, 35
87Recklessprefetch-threat-indexdiff4.0+0.04
LLR: -0.22 (-2.25, 2.89) [0.00, 3.00]
Games: 26870 W: 6823 L: 6778 D: 13269
Ptnml(0-2): 99, 2915, 7363, 2958, 100
Cj5716Recklessquad-extdiff8.0+0.08
LLR: 0.17 (-2.25, 2.89) [0.00, 3.00]
Games: 26638 W: 6767 L: 6698 D: 13173
Ptnml(0-2): 74, 3142, 6812, 3223, 68
Cj5716Recklessquad-extdiff40.0+0.40
LLR: -0.60 (-2.25, 2.89) [0.00, 3.00]
Games: 40382 W: 9674 L: 9624 D: 21084
Ptnml(0-2): 12, 4543, 11027, 4601, 8
scaling ltc
WorldRecklessexpt-26diff8.0+0.08
LLR: -2.25 (-2.25, 2.89) [0.00, 3.00]
Games: 6792 W: 1669 L: 1788 D: 3335
Ptnml(0-2): 22, 880, 1701, 781, 12
x.concat(-x).sqrrelu() on L2/L3
WorldRecklessexpt-26diffN=25000
Elo: -1.78 +- 2.16 (95%) [N=40000]
Games: 40034 W: 11964 L: 12169 D: 15901
Ptnml(0-2): 845, 4807, 8903, 4632, 830
x.concat(-x).sqrrelu() on L2/L3
WorldRecklessbest-move-changes-ln1pdiff40.0+0.40
LLR: -2.04 (-2.25, 2.89) [0.00, 3.00]
Games: 51536 W: 12386 L: 12392 D: 26758
Ptnml(0-2): 5, 4996, 15776, 4982, 9
PeregrRecklessunbias-ttmdiff8.0+0.08
LLR: -2.28 (-2.25, 2.89) [0.00, 3.00]
Games: 56026 W: 14018 L: 14029 D: 27979
Ptnml(0-2): 108, 6606, 14589, 6609, 101
tt_moves may already have poor history, so exclude them from the penalty
PeregrRecklesslate-bestdiff40.0+0.40
LLR: -2.30 (-2.25, 2.89) [0.00, 3.00]
Games: 33788 W: 8248 L: 8305 D: 17235
Ptnml(0-2): 10, 3802, 9327, 3745, 10
[LTC] account for lateness in bonus
WorldRecklessself-swigludiff8.0+0.08
LLR: -2.25 (-2.25, 2.89) [0.00, 3.00]
Games: 14584 W: 3593 L: 3695 D: 7296
Ptnml(0-2): 40, 1826, 3662, 1724, 40
cj-08-self-swiglu
WorldRecklessself-swigludiffN=25000
Elo: -0.95 +- 2.14 (95%) [N=40000]
Games: 40204 W: 11964 L: 12074 D: 16166
Ptnml(0-2): 812, 4788, 9004, 4694, 804
cj-08-self-swiglu
WorldRecklessexpt-25diff8.0+0.08
LLR: -1.55 (-2.25, 2.89) [0.00, 3.00]
Games: 29378 W: 7319 L: 7346 D: 14713
Ptnml(0-2): 85, 3552, 7444, 3521, 87
increase l1 reg 0.005 -> 0.01