Dark Horse Update

[purechess]

Having a CC BY-ND 4.0 license prevents distribution and getting a wider audience for your nets in the GPL v3 chess world

[dkappe]

Having a CC BY-ND 4.0 license prevents distribution and getting a wider audience for your nets in the GPL v3 chess world

That doesn’t concern me.

[MikeB]

I’ve had some requests to update Dark Horse to the new larger HalfKA architecture. I haven’t done any testing beyond making sure the net is the strongest of the training run. Same data as DH 0.3 (less than 1b positions).

https://www.patreon.com/posts/dark-horse-0-3-52384228

Good stuff Dietrich!

[dkappe]

I plugged the new model (after a little surgery) and data loader into a different training framework that I use for some non-chess work and ran it on the Dark Horse data. More bells and whistles. Seems much stronger at STC than the vanilla one. Dark Horse 0.3b XL.

https://www.patreon.com/posts/53624852

P.S. I find it interesting and gratifying that people are using Dark Horse XL for correspondence chess to counter the Stockfish groupthink.

[dkappe]

[pgn]

[Event "?"]
[Site "?"]
[Date "2021.07.12"]
[Round "16"]
[White "SF14"]
[Black "dh0.3b xl"]
[Result "0-1"]
[ECO "C02"]
[GameDuration "00:01:12"]
[GameEndTime "2021-07-12T22:42:05.927 CDT"]
[GameStartTime "2021-07-12T22:40:53.843 CDT"]
[Opening "French"]
[PlyCount "176"]
[Termination "adjudication"]
[TimeControl "20+0.2"]
[Variation "Advance, Paulsen attack"]

1. e4 {book} e6 {book} 2. d4 {book} d5 {book} 3. e5 {book} c5 {book}
4. c3 {book} Nc6 {book} 5. Nf3 {book} Qb6 {book} 6. a3 {book}
f6 {-0.24/18 0.70s} 7. Bd3 {+0.81/20 1.6s} fxe5 {-0.23/17 0.30s}
8. dxe5 {+0.64/17 0.46s} c4 {-0.23/17 0.32s} 9. Bc2 {+0.61/17 0.31s}
Nh6 {-0.23/16 0.42s} 10. O-O {+0.73/18 1.3s} Bd7 {-0.32/19 2.0s}
11. Nbd2 {+0.72/17 0.24s} Ng4 {-0.15/18 0.49s} 12. b3 {+0.56/20 1.6s}
O-O-O {-0.13/16 0.49s} 13. bxc4 {+0.73/16 0.22s} Ngxe5 {-0.20/15 0.38s}
14. Nxe5 {+0.63/21 3.0s} Nxe5 {-0.24/16 0.34s} 15. cxd5 {+0.51/17 0.39s}
exd5 {-0.11/17 0.58s} 16. Rb1 {+0.65/21 1.7s} Qc7 {-0.21/16 0.84s}
17. Nb3 {+0.86/17 0.27s} Bd6 {-0.36/19 1.5s} 18. Qxd5 {+0.87/19 0.43s}
Bc6 {-0.17/16 0.28s} 19. Qe6+ {+0.60/22 1.6s} Bd7 {-0.46/17 1.1s}
20. Qd5 {+0.71/19 0.25s} Bc6 {-0.28/17 0.48s} 21. Qa5 {+0.64/18 0.36s}
b6 {-0.29/16 0.44s} 22. Qa6+ {+1.22/19 0.72s} Bb7 {-0.21/15 0.33s}
23. Qa4 {+0.75/19 0.69s} Nd3 {0.00/15 0.43s} 24. g3 {+0.74/18 0.34s}
Qxc3 {-0.04/17 0.77s} 25. Na5 {+0.45/17 1.0s} Bd5 {-0.23/14 0.36s}
26. Be3 {+0.57/17 0.98s} Nc5 {-0.03/14 0.33s} 27. Bf5+ {+0.84/17 0.29s}
Kc7 {+0.06/15 0.37s} 28. Qb4 {+1.79/17 0.58s} Qe5 {-0.18/18 1.3s}
29. Bh3 {+0.19/18 1.4s} Qe4 {-0.13/17 0.41s} 30. Qxe4 {+0.50/17 0.17s}
Bxe4 {-0.21/18 0.52s} 31. Rbc1 {+0.46/19 0.42s} Kb8 {-0.07/14 0.33s}
32. Bg2 {+0.38/23 2.3s} Bxg2 {-0.18/18 0.56s} 33. Kxg2 {+0.60/17 0.14s}
Rc8 {-0.19/18 0.47s} 34. Nc4 {+0.36/17 0.44s} Be7 {-0.20/18 0.56s}
35. Bd4 {+0.35/14 0.21s} Bf6 {-0.18/19 0.69s} 36. Bxf6 {+0.50/17 0.39s}
gxf6 {-0.26/18 0.31s} 37. Ne3 {+0.46/16 0.23s} Rhd8 {-0.15/18 0.45s}
38. Rc4 {+0.50/15 0.21s} Rd3 {-0.23/18 1.2s} 39. Rh4 {+0.78/14 0.16s}
Rc7 {-0.19/16 0.34s} 40. Ra1 {+0.70/14 0.20s} Kb7 {-0.14/17 0.56s}
41. Rh5 {+0.83/13 0.17s} Rd2 {-0.25/16 0.45s} 42. Ng4 {+0.72/18 1.4s}
Nd3 {-0.21/18 0.78s} 43. Rf5 {+0.81/13 0.064s} a5 {-0.26/17 1.0s}
44. Rxf6 {+0.50/16 0.50s} Rcc2 {-0.15/17 0.26s} 45. Rf1 {+0.21/14 0.11s}
b5 {-0.08/16 0.38s} 46. Ne3 {+0.46/14 0.13s} Ra2 {0.00/18 0.64s}
47. h4 {+0.46/13 0.087s} a4 {0.00/17 0.54s} 48. h5 {+0.36/16 0.51s}
Re2 {0.00/18 0.70s} 49. Rf7+ {+0.71/13 0.036s} Kc6 {0.00/15 0.26s}
50. Ng4 {+0.54/14 0.11s} Rxa3 {0.00/16 0.57s} 51. Rxh7 {+0.67/15 0.11s}
Ne1+ {+0.08/18 0.67s} 52. Kh3 {+0.72/15 0.14s} Rf3 {0.00/18 0.49s}
53. Rg7 {+0.70/14 0.10s} a3 {0.00/15 0.14s} 54. h6 {-0.01/17 0.29s}
Rf8 {+0.55/13 0.18s} 55. f4 {0.00/15 0.091s} b4 {+0.52/16 0.54s}
56. h7 {0.00/15 0.12s} Rh8 {+0.75/14 0.11s} 57. Ne5+ {-0.13/15 0.20s}
Kb6 {+1.00/14 0.15s} 58. Nc4+ {-2.42/15 0.90s} Kb5 {+0.83/17 0.54s}
59. Nd2 {-0.95/14 0.11s} a2 {+1.09/15 0.11s} 60. Nb3 {-1.30/14 0.12s}
Kc4 {+0.28/16 0.38s} 61. Na1 {-1.23/13 0.063s} Kc3 {+0.47/15 0.12s}
62. Rb7 {-0.74/14 0.096s} Nc2 {+0.83/14 0.15s} 63. Nxc2 {-1.54/17 0.67s}
Rxc2 {+0.77/13 0.16s} 64. f5 {-1.46/13 0.049s} b3 {+0.93/17 0.89s}
65. Ra7 {-1.53/13 0.062s} Kb2 {+1.12/13 0.040s} 66. Kg4 {-1.30/12 0.059s}
Rh2 {+1.19/17 0.54s} 67. f6 {-1.52/14 0.18s} R2xh7 {+1.24/12 0.044s}
68. Rf2+ {-1.40/14 0.14s} Kb1 {+1.22/13 0.10s} 69. Rf1+ {-1.59/12 0.18s}
Kc2 {+1.25/11 0.064s} 70. Rf2+ {-1.63/12 0.15s} Kd1 {+1.29/13 0.15s}
71. Ra6 {-3.17/16 0.76s} Rb7 {+1.29/11 0.12s} 72. Kf3 {-2.86/12 0.054s}
Rc8 {+1.86/13 0.23s} 73. g4 {-2.62/11 0.034s} Rc4 {+2.38/14 0.57s}
74. Kg3 {-4.57/13 0.56s} Rbb4 {+3.42/14 0.22s} 75. Rf1+ {-5.86/12 0.23s}
Kc2 {+4.72/12 0.10s} 76. Rf2+ {-5.86/12 0.091s} Kb1 {+5.61/11 0.081s}
77. Rf1+ {-5.44/13 0.041s} Rc1 {+5.94/14 0.12s} 78. Rxc1+ {-5.55/13 0.44s}
Kxc1 {+7.08/15 0.36s} 79. f7 {-7.46/12 0.22s} Rb8 {+7.15/12 0.10s}
80. g5 {-7.30/12 0.23s} b2 {+9.28/14 0.087s} 81. Rc6+ {-7.23/11 0.21s}
Kd1 {+10.46/12 0.080s} 82. Rd6+ {-12.68/11 0.22s} Ke2 {+15.07/16 0.84s}
83. Re6+ {-13.23/13 0.20s} Kd3 {+15.67/12 0.071s} 84. Rd6+ {-14.90/11 0.20s}
Kc4 {+15.70/16 0.43s} 85. Ra6 {-15.76/12 0.20s} b1=Q {+78.47/12 0.035s}
86. Rc6+ {-152.53/14 0.20s} Kd5 {+152.63/15 0.27s} 87. f8=R {-152.62/15 0.16s}
Rxf8 {+M11/15 0.033s} 88. Rf6 {-M10/17 0.069s}
Rxf6 {+M9/22 0.050s, Black wins by adjudication: SyzygyTB} 0-1

[/pgn]

[peter]

More bells and whistles. Seems much stronger at STC than the vanilla one. Dark Horse 0.3b XL.

https://www.patreon.com/posts/53624852

Thanks a lot for this one again!

New leader in my personal MultiPV-SMP-list at Vincent Lejeune's Hard Talkchess small set: Crystal with Dark Horse 0.3bxl.nnue with 92/114.
As always 24 threads of 12x3GHz Xeon CPU (SSE4.1 pocnt), 60"/pos., 8G hash, 6men Syzygys, MultiPV=4:

Code: Select all

Crystal 0907210
Bisher gelöst: 92 von 114  ;  37:43m

         1   2   3   4   5   6   7   8   9  10  11  12  13  14  15  16  17  18  19  20
 -------------------------------------------------------------------------------------
   0 |  55   0   -   -   -   1   7  31   4   2  25   -   6   3   9   0  53   5  15   6
  20 |  17   -   -   4   2   0   0   9  57   0   0  42   3  11   1   0   0   -   -   8
  40 |  11   0  20   0   0   -   0   0  13   5   -   3   -  60   3  30   0   1   0   0
  60 |   1   -   2   -   -   -   3  42  50   0   -   0   0   0   -  14  60   0  60  29
  80 |   0   0   -   1   -   0   0  13   -   0  17  10   -   5   0   5  19   1   1   3
 100 |  13   4   0   5   3   -   0   0   0   0   0   1   0   2