Maybe right place this time...
Mayhem 3.0: https://github.com/SamuraiDangyo/mayhem ... s/tag/v3.0
- Tuning Mvv + bug fixes + speedups + last eval printed instead of 0
- Fast binaries for Windows + Unix. Compile yourself for extra kick.
- A little bit stronger. Maybe a few Elo. My estimations are always too positive.
Tuning Mvv should result better tactical eye.
Played a 5+5 against SF12. Mayhem managed to draw
[pgn][Event "Computer Chess Game"]
[Site "laptop"]
[Date "2021.01.28"]
[Round "-"]
[White "Mayhem 3.0"]
[Black "Stockfish 12"]
[Result "1/2-1/2"]
[TimeControl "300+5"]
1. d4 Nf6
{-0,21/25 16}
2. c4 e6
{-0,15/23 4}
3. Nf3 d5
{-0,30/24 10}
4. g3
{+0,09/12 15}
4... dxc4
{-0,24/22 5}
5. Bg2
{+0,09/1 0,1}
5... c6
{-0,14/25 32}
6. Ne5
{+0,09/1 0,1}
6... Bb4+
{+0,00/24 4}
7. Bd2
{+0,09/1 0,1}
7... Qxd4
{+0,00/26 5}
8. Bxb4
{+0,90/14 16}
8... Qxe5
{+0,00/29 8}
9. Na3
{+0,54/14 15}
9... b5
{+0,00/27 6}
10. f4
{+0,50/13 15}
10... Qc7
{+0,00/29 9}
11. Nxb5
{+0,30/13 15}
11... cxb5
{-0,07/29 6}
12. Bxa8
{+0,38/14 14}
12... Bb7
{+0,00/30 11}
13. Bxb7
{+0,46/13 14}
13... Qxb7
{+0,00/31 13}
14. Rf1
{+0,18/13 14}
14... Nc6
{+0,00/29 6}
15. Bc3
{+0,34/14 13}
15... O-O
{+0,00/32 8}
16. Bxf6
{+0,13/13 13}
16... gxf6
{+0,00/34 12}
17. Qd2
{+0,00/14 13}
17... Rd8
{+0,00/33 6}
18. Qc3
{-0,08/14 12}
18... Nd4
{+0,00/31 7}
19. Kf2
{+0,03/13 12}
19... Nxe2
{+0,00/35 15}
20. Qxf6
{+0,00/13 12}
20... Rd2
{+0,00/36 16}
21. Qg5+
{+0,17/12 12}
21... Kf8
{+0,00/31 4}
22. Ke1
{+0,00/12 12}
22... Rd3
{+0,00/31 9}
23. Rd1
{+0,00/12 11}
23... Nc1
{+0,00/34 7}
24. Qh6+
{-0,02/12 11}
24... Ke8
{+0,00/32 6}
25. Qxh7
{+0,00/13 11}
25... Qg2
{+0,00/37 6}
26. Qg8+
{-0,10/12 11}
26... Ke7
{+0,00/40 7}
27. Qg5+
{+0,00/11 11}
27... Kf8
{+0,00/42 7}
28. Qc5+
{+0,00/12 10}
28... Kg8
{+0,00/44 14}
29. Qg5+
{+0,00/13 10}
29... Kh7
{+0,00/41 6}
30. Qh5+
{+0,00/12 10}
30... Kg8
{+0,00/43 16}
31. Qg5+
{+0,00/15 10}
31... Kh7
{+0,00/46 13}
32. Qh4+
{+0,00/12 10}
32... Kg7
{+0,00/47 20}
33. Qg5+
{+0,00/14 9}
33... Kf8
{+0,00/48 1:41}
34. Qc5+
{+0,00/15 9}
34... Kg7
{+0,00/46 18}
35. Qg5+
{+0,00/15 9}
35... Kf8
{+0,00/67 29}
{Draw by repetition} 1/2-1/2
[/pgn]
Mayhem NNUE - New NN engine
Moderators: hgm, Rebel, chrisw
-
- Posts: 491
- Joined: Sat Mar 02, 2013 11:31 pm
-
- Posts: 491
- Joined: Sat Mar 02, 2013 11:31 pm
Re: Mayhem NNUE - New NN engine
Mayhem 3.1: https://github.com/SamuraiDangyo/mayhem ... s/tag/v3.1
Smallish update. See the release notes. Should be the best Mayhem. Fast binaries included.
Smallish update. See the release notes. Should be the best Mayhem. Fast binaries included.
-
- Posts: 491
- Joined: Sat Mar 02, 2013 11:31 pm
Re: Mayhem NNUE - New NN engine
Mayhem 3.2: https://github.com/SamuraiDangyo/mayhem ... s/tag/v3.2
See the release notes. Mainly speedups.
Nice game. Mayhem sacked a Q to 2 R in the EG.
Where Mayhem searched 26,000,000 NPS on 1 core.
[pgn][Event "Computer Chess Game"]
[Site "ThinkPad"]
[Date "2021.02.12"]
[Round "-"]
[White "Mayhem 3.2"]
[Black "Fruit 2.1"]
[Result "1-0"]
[TimeControl "40/60"]
[Annotator "6. +0,23 1... -0,02"]
1. c4 Nf6 {-0,02/12 2,1} 2. d4 e6 {+0,08/12 2,3} 3. Nc3 Bb4 {-0,02/12 3} 4.
Qc2 Nc6 {+0,23/11 1,5} 5. Nf3 O-O {+0,08/11 2,7} 6. Bd2 {+0,23/12 2,0} d5
{+0,08/10 2,1} 7. e3 {+0,22/12 1,9} Bd7 {-0,18/10 1,0} 8. a3 {+0,38/12 1,9}
Bd6 {-0,32/11 1,8} 9. c5 {+0,53/12 1,8} Be7 {-0,44/13 0,1} 10. b4
{+0,62/12 1,7} h6 {-0,52/11 2,5} 11. Bd3 {+0,67/12 1,7} a6 {-0,56/12 0,4}
12. O-O {+0,71/11 1,7} Qc8 {-0,56/12 0,1} 13. h3 {+0,78/10 1,7} Rd8
{-0,60/11 0,4} 14. Ne5 {+0,78/11 1,7} Nxe5 {-0,54/11 0,1} 15. dxe5
{+0,78/12 1,7} Ne8 {-0,53/12 0,1} 16. Ne2 {+0,76/12 1,7} Bb5 {-0,30/13 1,8}
17. Bxb5 {+0,74/13 1,7} axb5 {-0,05/10 0,4} 18. Nd4 {+0,77/12 1,7} Qd7
{-0,21/12 0,5} 19. f4 {+0,77/11 1,7} c6 {-0,10/11 1,3} 20. Qd3
{+0,64/10 1,7} g6 {-0,04/11 4} 21. g4 {+1,10/11 1,7} Ng7 {+0,00/11 1,6} 22.
Kh1 {+1,01/11 1,7} Ra4 {+0,00/12 0,2} 23. Bc3 {+1,04/11 1,7} Rda8
{-0,02/11 1,8} 24. Bb2 {+1,09/11 1,7} R4a6 {-0,16/10 1,0} 25. Qc2
{+1,29/11 1,7} b6 {-0,13/10 2,3} 26. cxb6 {+1,24/12 1,7} Rxb6
{-0,13/11 0,2} 27. f5 {+1,70/12 1,7} gxf5 {-0,38/11 2,0} 28. gxf5
{+1,67/12 1,7} exf5 {-0,75/11 0,5} 29. Rg1 {+1,67/11 1,7} Bg5
{-1,22/10 1,5} 30. h4 {+1,95/12 1,7} Bxh4 {-1,93/11 0,6} 31. Rxg7+
{+2,22/12 1,7} Kxg7 {-4,48/12 9} 32. e6 {+2,12/12 1,7} fxe6 {-4,69/12 1,0}
33. Nxf5+ {+2,06/11 1,7} Kf8 {-4,81/13 0,1} 34. Nxh4 {+1,98/12 1,7} Qf7
{-5,18/12 2,1} 35. Bd4 {+2,23/11 1,7} Rba6 {-1,37/11 1,5} 36. Rg1
{+1,84/11 1,7} e5 {-2,08/11 0,1} 37. Qe2 {+1,95/11 1,7} exd4 {-1,71/11 2,6}
38. Rf1 {+2,12/14 1,7} Rxa3 {-1,94/11 0,1} 39. Rxf7+ {+2,11/14 1,7} Kxf7
{-1,90/11 0,1} 40. Qh5+ {+2,08/14 1,7} Ke7 {-1,88/12 0,9} 41. Nf5+
{+2,28/14 2,0} Kf6 {-1,92/13 0,4} 42. Nxd4 {+2,43/14 1,9} Ra1+
{-1,96/13 1,2} 43. Kg2 {+2,40/13 1,9} R8a2+ {-1,96/13 0,5} 44. Ne2
{+2,46/13 1,8} Kg7 {-1,96/13 0,6} 45. Kg3 {+2,53/13 1,7} Re1 {-1,88/11 1,5}
46. Qg4+ {+2,53/13 1,7} Kf6 {-2,53/13 2,7} 47. Qf4+ {+2,49/13 1,6} Kg6
{-2,75/13 2,0} 48. Qd6+ {+2,85/14 1,6} Kh7 {-2,79/13 0,8} 49. Qe7+
{+2,87/13 1,5} Kh8 {-2,79/13 0,5} 50. Nd4 {+2,89/12 1,5} Rg1+
{-2,38/13 2,5} 51. Kf4 {+2,99/13 1,4} Rf2+ {-2,27/14 0,1} 52. Ke5
{+2,61/13 1,4} Rg7 {-2,65/14 1,7} 53. Qe6 {+3,62/13 1,4} Rg5+
{-2,94/13 0,4} 54. Kd6 {+3,35/13 1,4} Kh7 {-3,41/13 2,1} 55. Kc5
{+3,32/14 1,4} Rg6 {-3,39/12 0,1} 56. Qd7+ {+3,74/13 1,4} Kg8
{-3,63/14 0,1} 57. Nf5 {+3,92/13 1,4} Rc2+ {-4,68/14 20} 58. Kb6
{+4,51/13 1,4} Rcg2 {-4,68/14 0,1} 59. Ne7+ {+4,52/13 1,4} Kf8
{-4,70/14 0,1} 60. Qd8+ {+4,78/13 1,4} Kf7 {-4,03/12 0,9} 61. Qc7
{+4,19/13 1,4} Kf8 {-3,96/12 0,1} 62. Nxc6 {+4,23/13 1,4} Rc2 {-4,09/15 11}
63. Qd8+ {+4,23/15 1,4} Kg7 {-5,04/14 0,7} 64. Qd7+ {+3,36/15 1,4} Kh8
{-5,04/14 0,1} 65. Qc8+ {+3,15/15 1,4} Kh7 {-1,77/12 0,7} 66. Qb7+
{+1,45/15 1,4} Kg8 {-1,77/12 0,9} 67. Qa8+ {+1,01/15 1,4} Kh7
{-1,11/13 0,9} 68. Qa7+ {+0,58/14 1,4} Kg8 {-1,15/14 0,1} 69. Kxb5
{+0,45/14 1,4} Rcxc6 {-1,12/16 0,1} 70. Qa8+ {+0,87/15 1,4} Kf7
{-1,13/17 0,1} 71. Qxc6 {+0,59/16 1,4} Rxc6 {-1,68/21 0,1} 72. Kxc6
{+0,87/20 1,4} h5 {-1,12/18 0,1} 73. Kxd5 {+1,14/19 1,4} h4 {-6,28/24 0,1}
74. Ke4 {+5,00/19 1,4} h3 {-10,14/26 0,1} 75. Kf3 {+5,00/20 1,4} Ke6
{-6,44/19 0,4} 76. Kg3 {+4,95/21 1,4} Kd5 {-10,33/26 0,1} 77. Kxh3
{+5,00/21 1,4} Ke4 {-10,33/20 0,9} 78. Kg4 {+6,73/20 1,4} Kd5 {-10,48/20 3}
79. Kf5 {+14,06/20 1,4} Kc4 {-11,00/24 1,1} 80. Ke6 {+14,51/19 1,4} Kxb4
{-10,95/18 2,2} 81. Kd6 {+14,68/18 2,0} Kc4 {-10,95/17 0,1} 82. e4
{+15,12/18 1,9} Kd4 {-11,07/16 0,4} 83. e5 {+15,51/18 1,9} Ke4
{-11,05/14 0,5} 84. e6 {+15,67/17 1,8} Kf5 {-11,05/12 0,6} 85. e7
{+15,85/16 1,7} Ke4 {-1000,08/12 11} 86. e8=Q+ {+16,18/15 1,7} Kd3
{-10,85/9 0,1} 87. Kc5 {+17,45/13 1,6} Kc2 {-1000,05/51 0,1} 88. Kc4
{+104,85/11 0,5} Kd2 {-1000,04/50 1,5} 89. Qe4 {+104,85/9 0,1} Kc1
{-1000,03/48 1,4} 90. Qe2 {+104,85/7 0,1} Kb1 {-1000,02/38 0,1} 91. Kb3
{+104,85/5 0,1} Ka1 {-1000,01/44 1,5} 92. Qe1# {+104,85/3 0,1}
{Xboard adjudication: Checkmate} 1-0[/pgn]
See the release notes. Mainly speedups.
Nice game. Mayhem sacked a Q to 2 R in the EG.
Where Mayhem searched 26,000,000 NPS on 1 core.
[pgn][Event "Computer Chess Game"]
[Site "ThinkPad"]
[Date "2021.02.12"]
[Round "-"]
[White "Mayhem 3.2"]
[Black "Fruit 2.1"]
[Result "1-0"]
[TimeControl "40/60"]
[Annotator "6. +0,23 1... -0,02"]
1. c4 Nf6 {-0,02/12 2,1} 2. d4 e6 {+0,08/12 2,3} 3. Nc3 Bb4 {-0,02/12 3} 4.
Qc2 Nc6 {+0,23/11 1,5} 5. Nf3 O-O {+0,08/11 2,7} 6. Bd2 {+0,23/12 2,0} d5
{+0,08/10 2,1} 7. e3 {+0,22/12 1,9} Bd7 {-0,18/10 1,0} 8. a3 {+0,38/12 1,9}
Bd6 {-0,32/11 1,8} 9. c5 {+0,53/12 1,8} Be7 {-0,44/13 0,1} 10. b4
{+0,62/12 1,7} h6 {-0,52/11 2,5} 11. Bd3 {+0,67/12 1,7} a6 {-0,56/12 0,4}
12. O-O {+0,71/11 1,7} Qc8 {-0,56/12 0,1} 13. h3 {+0,78/10 1,7} Rd8
{-0,60/11 0,4} 14. Ne5 {+0,78/11 1,7} Nxe5 {-0,54/11 0,1} 15. dxe5
{+0,78/12 1,7} Ne8 {-0,53/12 0,1} 16. Ne2 {+0,76/12 1,7} Bb5 {-0,30/13 1,8}
17. Bxb5 {+0,74/13 1,7} axb5 {-0,05/10 0,4} 18. Nd4 {+0,77/12 1,7} Qd7
{-0,21/12 0,5} 19. f4 {+0,77/11 1,7} c6 {-0,10/11 1,3} 20. Qd3
{+0,64/10 1,7} g6 {-0,04/11 4} 21. g4 {+1,10/11 1,7} Ng7 {+0,00/11 1,6} 22.
Kh1 {+1,01/11 1,7} Ra4 {+0,00/12 0,2} 23. Bc3 {+1,04/11 1,7} Rda8
{-0,02/11 1,8} 24. Bb2 {+1,09/11 1,7} R4a6 {-0,16/10 1,0} 25. Qc2
{+1,29/11 1,7} b6 {-0,13/10 2,3} 26. cxb6 {+1,24/12 1,7} Rxb6
{-0,13/11 0,2} 27. f5 {+1,70/12 1,7} gxf5 {-0,38/11 2,0} 28. gxf5
{+1,67/12 1,7} exf5 {-0,75/11 0,5} 29. Rg1 {+1,67/11 1,7} Bg5
{-1,22/10 1,5} 30. h4 {+1,95/12 1,7} Bxh4 {-1,93/11 0,6} 31. Rxg7+
{+2,22/12 1,7} Kxg7 {-4,48/12 9} 32. e6 {+2,12/12 1,7} fxe6 {-4,69/12 1,0}
33. Nxf5+ {+2,06/11 1,7} Kf8 {-4,81/13 0,1} 34. Nxh4 {+1,98/12 1,7} Qf7
{-5,18/12 2,1} 35. Bd4 {+2,23/11 1,7} Rba6 {-1,37/11 1,5} 36. Rg1
{+1,84/11 1,7} e5 {-2,08/11 0,1} 37. Qe2 {+1,95/11 1,7} exd4 {-1,71/11 2,6}
38. Rf1 {+2,12/14 1,7} Rxa3 {-1,94/11 0,1} 39. Rxf7+ {+2,11/14 1,7} Kxf7
{-1,90/11 0,1} 40. Qh5+ {+2,08/14 1,7} Ke7 {-1,88/12 0,9} 41. Nf5+
{+2,28/14 2,0} Kf6 {-1,92/13 0,4} 42. Nxd4 {+2,43/14 1,9} Ra1+
{-1,96/13 1,2} 43. Kg2 {+2,40/13 1,9} R8a2+ {-1,96/13 0,5} 44. Ne2
{+2,46/13 1,8} Kg7 {-1,96/13 0,6} 45. Kg3 {+2,53/13 1,7} Re1 {-1,88/11 1,5}
46. Qg4+ {+2,53/13 1,7} Kf6 {-2,53/13 2,7} 47. Qf4+ {+2,49/13 1,6} Kg6
{-2,75/13 2,0} 48. Qd6+ {+2,85/14 1,6} Kh7 {-2,79/13 0,8} 49. Qe7+
{+2,87/13 1,5} Kh8 {-2,79/13 0,5} 50. Nd4 {+2,89/12 1,5} Rg1+
{-2,38/13 2,5} 51. Kf4 {+2,99/13 1,4} Rf2+ {-2,27/14 0,1} 52. Ke5
{+2,61/13 1,4} Rg7 {-2,65/14 1,7} 53. Qe6 {+3,62/13 1,4} Rg5+
{-2,94/13 0,4} 54. Kd6 {+3,35/13 1,4} Kh7 {-3,41/13 2,1} 55. Kc5
{+3,32/14 1,4} Rg6 {-3,39/12 0,1} 56. Qd7+ {+3,74/13 1,4} Kg8
{-3,63/14 0,1} 57. Nf5 {+3,92/13 1,4} Rc2+ {-4,68/14 20} 58. Kb6
{+4,51/13 1,4} Rcg2 {-4,68/14 0,1} 59. Ne7+ {+4,52/13 1,4} Kf8
{-4,70/14 0,1} 60. Qd8+ {+4,78/13 1,4} Kf7 {-4,03/12 0,9} 61. Qc7
{+4,19/13 1,4} Kf8 {-3,96/12 0,1} 62. Nxc6 {+4,23/13 1,4} Rc2 {-4,09/15 11}
63. Qd8+ {+4,23/15 1,4} Kg7 {-5,04/14 0,7} 64. Qd7+ {+3,36/15 1,4} Kh8
{-5,04/14 0,1} 65. Qc8+ {+3,15/15 1,4} Kh7 {-1,77/12 0,7} 66. Qb7+
{+1,45/15 1,4} Kg8 {-1,77/12 0,9} 67. Qa8+ {+1,01/15 1,4} Kh7
{-1,11/13 0,9} 68. Qa7+ {+0,58/14 1,4} Kg8 {-1,15/14 0,1} 69. Kxb5
{+0,45/14 1,4} Rcxc6 {-1,12/16 0,1} 70. Qa8+ {+0,87/15 1,4} Kf7
{-1,13/17 0,1} 71. Qxc6 {+0,59/16 1,4} Rxc6 {-1,68/21 0,1} 72. Kxc6
{+0,87/20 1,4} h5 {-1,12/18 0,1} 73. Kxd5 {+1,14/19 1,4} h4 {-6,28/24 0,1}
74. Ke4 {+5,00/19 1,4} h3 {-10,14/26 0,1} 75. Kf3 {+5,00/20 1,4} Ke6
{-6,44/19 0,4} 76. Kg3 {+4,95/21 1,4} Kd5 {-10,33/26 0,1} 77. Kxh3
{+5,00/21 1,4} Ke4 {-10,33/20 0,9} 78. Kg4 {+6,73/20 1,4} Kd5 {-10,48/20 3}
79. Kf5 {+14,06/20 1,4} Kc4 {-11,00/24 1,1} 80. Ke6 {+14,51/19 1,4} Kxb4
{-10,95/18 2,2} 81. Kd6 {+14,68/18 2,0} Kc4 {-10,95/17 0,1} 82. e4
{+15,12/18 1,9} Kd4 {-11,07/16 0,4} 83. e5 {+15,51/18 1,9} Ke4
{-11,05/14 0,5} 84. e6 {+15,67/17 1,8} Kf5 {-11,05/12 0,6} 85. e7
{+15,85/16 1,7} Ke4 {-1000,08/12 11} 86. e8=Q+ {+16,18/15 1,7} Kd3
{-10,85/9 0,1} 87. Kc5 {+17,45/13 1,6} Kc2 {-1000,05/51 0,1} 88. Kc4
{+104,85/11 0,5} Kd2 {-1000,04/50 1,5} 89. Qe4 {+104,85/9 0,1} Kc1
{-1000,03/48 1,4} 90. Qe2 {+104,85/7 0,1} Kb1 {-1000,02/38 0,1} 91. Kb3
{+104,85/5 0,1} Ka1 {-1000,01/44 1,5} 92. Qe1# {+104,85/3 0,1}
{Xboard adjudication: Checkmate} 1-0[/pgn]
-
- Posts: 2056
- Joined: Mon Mar 13, 2006 2:31 am
- Location: North Carolina, USA
Re: Mayhem NNUE - New NN engine
Wow, after 15+ years somebody agrees with my sentiment. That is exactly why I created Ares the way I did. It has human knowledge and tuning to make it a human like partner for play and analysis. I use it to analyze my games instead of all the clones/derivatives. Ares' analysis suggestions make more sense to anybody rated 2300 or below. I've put it on ICC in the past and watched it beat lots of IMs.carldaman wrote: ↑Fri Oct 23, 2020 9:00 am Yeah, in many ways it's a sad state of affairs when you just plug in SF-NNUE as a shortcut. The worst part is that it's a black box that doesn't even let you set contempt, let alone other settings. It takes the fun out for guys like Brendan and me, and also for some conscientious programmers like Andrew G of Ethereal fame, who take pride in fine tuning an engine.
My advice to developers that know they are not chasing after Stockfish is to program for style, and to create an enjoyable opponent or analysis partner. That would be something to be proud of, regardless of actual Elo rating.
If you are weaker than FM, I suggest doing what I do. Play it with the search depth fixed. It will play fast, weak and human like. Set the depth to 1 and give your self plenty of clock time - play a reasonably slow TC. If that is not strong enough move the fixed depth up to 2 and so on.
On playing style of Ares: the GB versions are more king hunters and the strongest of the nonGB versions is more balanced.
-
- Posts: 2283
- Joined: Sat Jun 02, 2012 2:13 am
Re: Mayhem NNUE - New NN engine
Well, thanks for dropping by, Charles, I'm sure glad you agree - this is really something I've stated on other occasions, too.CRoberson wrote: ↑Wed Feb 17, 2021 8:34 pmWow, after 15+ years somebody agrees with my sentiment. That is exactly why I created Ares the way I did. It has human knowledge and tuning to make it a human like partner for play and analysis. I use it to analyze my games instead of all the clones/derivatives. Ares' analysis suggestions make more sense to anybody rated 2300 or below. I've put it on ICC in the past and watched it beat lots of IMs.carldaman wrote: ↑Fri Oct 23, 2020 9:00 am Yeah, in many ways it's a sad state of affairs when you just plug in SF-NNUE as a shortcut. The worst part is that it's a black box that doesn't even let you set contempt, let alone other settings. It takes the fun out for guys like Brendan and me, and also for some conscientious programmers like Andrew G of Ethereal fame, who take pride in fine tuning an engine.
My advice to developers that know they are not chasing after Stockfish is to program for style, and to create an enjoyable opponent or analysis partner. That would be something to be proud of, regardless of actual Elo rating.
If you are weaker than FM, I suggest doing what I do. Play it with the search depth fixed. It will play fast, weak and human like. Set the depth to 1 and give your self plenty of clock time - play a reasonably slow TC. If that is not strong enough move the fixed depth up to 2 and so on.
On playing style of Ares: the GB versions are more king hunters and the strongest of the nonGB versions is more balanced.
I really like good sparring partners like Ares-GB and RomiChess that allow me to sometimes get playable positions against them. Such 'coffeehouse' and king hunter styled engines are not the norm in computer chess, but they are a special and welcome treat when they surface.
Ares-GB not being the strongest Ares version while being a better opponent is a case in point. I wish more developers would do this. Pawel Koziol has done some marvelous things with Rodent's various personalities, but such programmers are a small minority. OpenTal 1.0 deserves a special mention.
It seems very hard to get a NNUE-based engine to vary its style to the point of taking serious risks and aiming for murky imbalances where the opponent can go wrong. With a traditional evaluation it is possible to tweak it to favor a certain style, or load it with knowledge regardless of any related slowdowns, provided that Elo is not the ultimate goal, of course.
For several years I've had the opportunity to be a tester for and advise on a couple of private projects, CyberNezh (which is sometimes available as a bot on lichess) and SultanKhan. What got me hooked on these projects was what transpired when I first went sparring against Nezh. It was the only time ever I felt that I didn't beat myself with a stupid blunder, or a series of bad moves, but instead I was blown off the board with some mind-boggling violent sacs that I didn't see coming - being used, up to that point, to more 'normal' engines which did not have such weapons in their arsenals. It was an amazing feeling, witnessing how the engine actively and creatively 'came out' and made the win happen, instead of sitting back and cashing in on some bad moves, which is how most engine-human games are decided.
The secret behind such engines is not caring that much about Elo in the first place, and being mindful of having an attractive style, as you must well know, I'm sure. I hope we can see more good things coming from Ares!
Cheers,
-
- Posts: 491
- Joined: Sat Mar 02, 2013 11:31 pm
Re: Mayhem NNUE - New NN engine
Mayhem 3.3: https://github.com/SamuraiDangyo/mayhem ... s/tag/v3.3
Very minor update. But time management is better + a few tricks
I didn't compile more EG knowledge because poseidon is already 1MB. It contains mostly KRPvKR / KRRPvKRR (rook + pawn) endgames anyway.
Mayhem has a level parameter. 0 means random mover. 10 means full strength. But those levels 1-9 blunder too much. Not very good but I didn't to slow down search for such a "feature".
Strong human style isn't very different to engines. 8x8 is too simple for creativity. Engines don't get tired/emotional/etc they just keep going the same ad nauseam. When I play I try to not care if the fact I have lost 20 games in a row and just try to win next one. Because engines surely don't.
Very minor update. But time management is better + a few tricks
I didn't compile more EG knowledge because poseidon is already 1MB. It contains mostly KRPvKR / KRRPvKRR (rook + pawn) endgames anyway.
Hello.carldaman wrote: ↑Thu Feb 18, 2021 7:16 amWell, thanks for dropping by, Charles, I'm sure glad you agree - this is really something I've stated on other occasions, too.CRoberson wrote: ↑Wed Feb 17, 2021 8:34 pmWow, after 15+ years somebody agrees with my sentiment. That is exactly why I created Ares the way I did. It has human knowledge and tuning to make it a human like partner for play and analysis. I use it to analyze my games instead of all the clones/derivatives. Ares' analysis suggestions make more sense to anybody rated 2300 or below. I've put it on ICC in the past and watched it beat lots of IMs.carldaman wrote: ↑Fri Oct 23, 2020 9:00 am Yeah, in many ways it's a sad state of affairs when you just plug in SF-NNUE as a shortcut. The worst part is that it's a black box that doesn't even let you set contempt, let alone other settings. It takes the fun out for guys like Brendan and me, and also for some conscientious programmers like Andrew G of Ethereal fame, who take pride in fine tuning an engine.
My advice to developers that know they are not chasing after Stockfish is to program for style, and to create an enjoyable opponent or analysis partner. That would be something to be proud of, regardless of actual Elo rating.
If you are weaker than FM, I suggest doing what I do. Play it with the search depth fixed. It will play fast, weak and human like. Set the depth to 1 and give your self plenty of clock time - play a reasonably slow TC. If that is not strong enough move the fixed depth up to 2 and so on.
On playing style of Ares: the GB versions are more king hunters and the strongest of the nonGB versions is more balanced.
I really like good sparring partners like Ares-GB and RomiChess that allow me to sometimes get playable positions against them. Such 'coffeehouse' and king hunter styled engines are not the norm in computer chess, but they are a special and welcome treat when they surface.
Ares-GB not being the strongest Ares version while being a better opponent is a case in point. I wish more developers would do this. Pawel Koziol has done some marvelous things with Rodent's various personalities, but such programmers are a small minority. OpenTal 1.0 deserves a special mention.
It seems very hard to get a NNUE-based engine to vary its style to the point of taking serious risks and aiming for murky imbalances where the opponent can go wrong. With a traditional evaluation it is possible to tweak it to favor a certain style, or load it with knowledge regardless of any related slowdowns, provided that Elo is not the ultimate goal, of course.
For several years I've had the opportunity to be a tester for and advise on a couple of private projects, CyberNezh (which is sometimes available as a bot on lichess) and SultanKhan. What got me hooked on these projects was what transpired when I first went sparring against Nezh. It was the only time ever I felt that I didn't beat myself with a stupid blunder, or a series of bad moves, but instead I was blown off the board with some mind-boggling violent sacs that I didn't see coming - being used, up to that point, to more 'normal' engines which did not have such weapons in their arsenals. It was an amazing feeling, witnessing how the engine actively and creatively 'came out' and made the win happen, instead of sitting back and cashing in on some bad moves, which is how most engine-human games are decided.
The secret behind such engines is not caring that much about Elo in the first place, and being mindful of having an attractive style, as you must well know, I'm sure. I hope we can see more good things coming from Ares!
Cheers,
Mayhem has a level parameter. 0 means random mover. 10 means full strength. But those levels 1-9 blunder too much. Not very good but I didn't to slow down search for such a "feature".
Strong human style isn't very different to engines. 8x8 is too simple for creativity. Engines don't get tired/emotional/etc they just keep going the same ad nauseam. When I play I try to not care if the fact I have lost 20 games in a row and just try to win next one. Because engines surely don't.
-
- Posts: 491
- Joined: Sat Mar 02, 2013 11:31 pm
Re: Mayhem NNUE - New NN engine
Mayhem 3.4 unleashed !
Built the binary and they will come.
Release: https://github.com/SamuraiDangyo/mayhem ... s/tag/v3.4
Massive increase in strength as I spent time in Xiphos source code looking for LMR / Null move formulas.
Mayhem is now searching super deep. In some positions up to 30 depth. Thanks to better LMR / Null move.
Pesto is included too. Mayhem is propably 2100 Elo w/o NNUE + opening book.
C++17 is supported too.
In 1 minute game. Which is really bad tc for Mayhem. Mayhem demolished Hakkapeliitta 3.0!
[pgn][Event "Computer Chess Game"]
[Site "pc"]
[Date "2021.03.05"]
[Round "-"]
[White "Mayhem 3.4"]
[Black "Hakkapeliitta 3.0"]
[Result "1-0"]
[TimeControl "60"]
[Annotator "7. +0,07 1... -0,01"]
1. c4 e5 {-0,01/17 2,4} 2. Nc3 Nc6 {-0,04/19 2,3} 3. g3 Nf6 {-0,02/19 3} 4.
Bg2 Bc5 {+0,03/19 2,0} 5. a3 d6 {+0,04/19 4} 6. e3 Bb6 {+0,16/17 1,8} 7. b4
{+0,07/12 2,0} Be6 {+0,21/17 1,7} 8. d3 {+0,11/12 1,9} O-O {+0,21/18 1,6}
9. Bb2 {+0,17/12 1,9} a6 {+0,26/17 1,8} 10. Nge2 {+0,33/11 1,8} Qd7
{+0,17/17 1,5} 11. Nd5 {+0,25/11 1,7} Bxd5 {+0,20/17 1,4} 12. cxd5
{+0,24/13 1,7} Ne7 {+0,16/17 1,4} 13. Nc3 {+0,10/12 1,6} a5 {+0,25/16 1,5}
14. b5 {+0,27/11 1,6} h5 {+0,22/16 1,3} 15. h3 {+0,08/10 1,5} h4
{+0,33/17 1,9} 16. g4 {+0,14/12 1,5} Nh7 {+0,25/17 1,1} 17. O-O
{+0,33/11 1,4} f5 {+0,39/17 1,1} 18. Kh2 {+0,20/11 1,4} fxg4 {+0,53/16 1,3}
19. Qxg4 {+0,18/11 1,3} Qxg4 {+0,73/18 1,0} 20. hxg4 {+0,15/13 1,3} a4
{+0,67/16 1,0} 21. Rac1 {+0,14/10 1,2} Ng5 {+1,17/17 0,9} 22. f4
{+0,59/12 1,2} h3 {+1,75/18 0,9} 23. Bh1 {-0,91/12 1,2} Bxe3 {+2,37/18 1,7}
24. f5 {-0,93/12 1,1} Ra5 {+2,26/17 0,8} 25. Rc2 {-0,33/11 1,1} Bf4+
{+2,41/17 0,8} 26. Rxf4 {-0,04/13 1,0} exf4 {+1,71/19 0,7} 27. Bc1
{+0,15/12 1,0} Nxf5 {+1,57/20 0,8} 28. gxf5 {+0,55/12 1,0} Rxf5
{+1,52/20 0,7} 29. Rf2 {+0,51/12 0,9} Ra8 {+1,03/20 0,7} 30. Ne4
{+0,69/13 0,9} Raf8 {+1,56/17 0,7} 31. b6 {+0,43/13 0,9} cxb6
{+1,08/19 0,6} 32. Nxd6 {+1,16/13 0,8} Re5 {+0,77/18 0,8} 33. Nc4
{+1,15/13 0,8} Re1 {+0,81/16 0,5} 34. Bd2 {+1,60/12 0,8} Ra1 {+0,72/16 1,1}
35. d6 {+2,03/12 0,8} Ra2 {+0,24/15 0,5} 36. Bd5+ {+2,07/12 0,7} Kh7
{-0,10/16 0,5} 37. d7 {+2,19/12 0,7} Rc2 {-0,38/16 0,8} 38. Ne5
{+2,34/11 0,7} g6 {-1,56/16 1,0} 39. Be1 {+2,55/11 0,7} Rc5 {-1,05/17 0,3}
40. Rxf4 {+2,18/11 0,6} Rd8 {-1,10/18 0,4} 41. Rd4 {+2,66/11 0,6} Rc2+
{-1,23/15 0,3} 42. Kg3 {+2,32/12 0,6} Kg7 {-1,75/16 0,6} 43. Bb4
{+2,32/10 0,6} Kf6 {-2,64/15 0,9} 44. Ng4+ {+3,67/10 0,6} Kg7
{-3,23/17 0,3} 45. Be7 {+3,71/12 0,5} Rxd7 {-3,91/17 0,3} 46. Bf6+
{+4,25/11 0,5} Kf8 {-3,78/17 0,2} 47. Bxg5 {+3,17/10 0,5} Rc5
{-4,02/16 0,2} 48. Rf4+ {+4,70/9 0,5} Kg7 {-4,41/17 0,2} 49. Bf6+
{+5,33/10 0,5} Kf8 {-4,75/17 0,2} 50. Bd4+ {+6,07/9 0,5} Ke7 {-5,48/18 0,2}
51. Rf7+ {+6,37/10 0,4} Kd8 {-6,22/20 0,3} 52. Rxd7+ {+6,11/10 0,4} Kxd7
{-6,39/19 0,2} 53. Bxc5 {+5,60/14 0,4} bxc5 {-6,42/18 0,2} 54. Bxb7
{+5,99/15 0,4} Kd6 {-6,56/19 0,2} 55. Be4 {+6,30/15 0,4} g5 {-6,56/17 0,2}
56. Kxh3 {+6,58/16 0,4} Ke6 {-7,48/17 0,1} 57. Bc6 {+6,73/17 0,4} Kf5
{-8,15/18 0,2} 58. Kg3 {+7,39/15 0,4} Ke6 {-8,75/18 0,1} 59. Bxa4
{+7,58/14 0,3} Kd5 {-9,15/17 0,2} 60. Nf6+ {+8,81/14 0,3} Ke5
{-9,40/18 0,2} 61. Nd7+ {+9,29/14 0,3} Kd5 {-10,79/18 0,2} 62. Kg4
{+9,46/15 0,3} c4 {-8,49/18 0,1} 63. dxc4+ {+11,41/15 0,3} Kxc4
{-8,89/18 0,1} 64. Kxg5 {+11,42/15 0,3} Kc3 {-9,01/18 0,1} 65. Bb5
{+12,17/13 0,3} Kb3 {-9,01/19 0,1} 66. a4 {+12,48/13 0,3} Kb4
{-9,01/18 0,1} 67. Kf4 {+13,04/12 0,3} Ka5 {-10,38/17 0,1} 68. Ke4
{+13,46/12 0,2} Kb4 {-8,49/7 0,1} 69. Kd4 {+13,84/13 0,2} Ka5
{-16,54/20 0,1} 70. Bc6 {+15,84/13 0,2} Ka6 {-13,62/18 0,1} 71. a5
{+16,70/12 0,2} Kxa5 {-7,71/15 0,1} 72. Kc5 {+16,00/11 0,2} Ka6
{-7,66/7 0,1} 73. Kb4 {+16,74/11 0,2} Ka7 {-7,34/7 0,1} 74. Ka4
{+16,70/10 0,2} Ka6 {-7,23/7 0,1} 75. Bg2 {+16,74/11 0,2} Ka7 {-7,15/7 0,1}
76. Kb3 {+16,60/10 0,2} Ka6 {-7,23/7 0,1} 77. Kb4 {+16,74/11 0,2} Ka7
{-7,34/7 0,1} 78. Bc6 {+16,98/11 0,2} Ka6 {-7,43/7 0,1} 79. Ka4
{+16,74/11 0,2} Ka7 {-7,17/7 0,1} 80. Bh1 {+15,71/10 0,2} Ka6 {-7,17/7 0,1}
81. Kb4 {+16,28/11 0,2} Ka7 {-7,23/7 0,1} 82. Bg2 {+16,37/11 0,2} Ka6
{-7,43/7 0,1} 83. Kc5 {+16,37/11 0,2} Ka5 {-7,70/13 0,1} 84. Nb6
{+15,95/12 0,1} Ka6 {-7,69/7 0,1} 85. Nc4 {+15,91/12 0,1} Ka7 {-7,69/7 0,1}
86. Kc6 {+16,24/11 0,1} Ka8 {-1000,07/12 0,1} 87. Kb6+ {+104,85/9 0,1} Kb8
{-1000,08/7 0,1} 88. Bh3 {+104,85/8 0,1} Ka8 {-1000,07/7 0,1} 89. Ne5
{+104,85/7 0,1} Kb8 {-7,98/7 0,1} 90. Nc6+ {+104,85/6 0,1} Ka8
{-1000,04/7 0,1} 91. Nb4 {+104,85/5 0,1} Kb8 {-1000,02/7 0,1} 92. Na6+
{+104,85/4 0,1} Ka8 {-1000,01/7 0,1} 93. Bg2# {+104,85/3 0,1}
{Xboard adjudication: Checkmate} 1-0[/pgn]
Built the binary and they will come.
Release: https://github.com/SamuraiDangyo/mayhem ... s/tag/v3.4
Massive increase in strength as I spent time in Xiphos source code looking for LMR / Null move formulas.
Mayhem is now searching super deep. In some positions up to 30 depth. Thanks to better LMR / Null move.
Pesto is included too. Mayhem is propably 2100 Elo w/o NNUE + opening book.
C++17 is supported too.
In 1 minute game. Which is really bad tc for Mayhem. Mayhem demolished Hakkapeliitta 3.0!
[pgn][Event "Computer Chess Game"]
[Site "pc"]
[Date "2021.03.05"]
[Round "-"]
[White "Mayhem 3.4"]
[Black "Hakkapeliitta 3.0"]
[Result "1-0"]
[TimeControl "60"]
[Annotator "7. +0,07 1... -0,01"]
1. c4 e5 {-0,01/17 2,4} 2. Nc3 Nc6 {-0,04/19 2,3} 3. g3 Nf6 {-0,02/19 3} 4.
Bg2 Bc5 {+0,03/19 2,0} 5. a3 d6 {+0,04/19 4} 6. e3 Bb6 {+0,16/17 1,8} 7. b4
{+0,07/12 2,0} Be6 {+0,21/17 1,7} 8. d3 {+0,11/12 1,9} O-O {+0,21/18 1,6}
9. Bb2 {+0,17/12 1,9} a6 {+0,26/17 1,8} 10. Nge2 {+0,33/11 1,8} Qd7
{+0,17/17 1,5} 11. Nd5 {+0,25/11 1,7} Bxd5 {+0,20/17 1,4} 12. cxd5
{+0,24/13 1,7} Ne7 {+0,16/17 1,4} 13. Nc3 {+0,10/12 1,6} a5 {+0,25/16 1,5}
14. b5 {+0,27/11 1,6} h5 {+0,22/16 1,3} 15. h3 {+0,08/10 1,5} h4
{+0,33/17 1,9} 16. g4 {+0,14/12 1,5} Nh7 {+0,25/17 1,1} 17. O-O
{+0,33/11 1,4} f5 {+0,39/17 1,1} 18. Kh2 {+0,20/11 1,4} fxg4 {+0,53/16 1,3}
19. Qxg4 {+0,18/11 1,3} Qxg4 {+0,73/18 1,0} 20. hxg4 {+0,15/13 1,3} a4
{+0,67/16 1,0} 21. Rac1 {+0,14/10 1,2} Ng5 {+1,17/17 0,9} 22. f4
{+0,59/12 1,2} h3 {+1,75/18 0,9} 23. Bh1 {-0,91/12 1,2} Bxe3 {+2,37/18 1,7}
24. f5 {-0,93/12 1,1} Ra5 {+2,26/17 0,8} 25. Rc2 {-0,33/11 1,1} Bf4+
{+2,41/17 0,8} 26. Rxf4 {-0,04/13 1,0} exf4 {+1,71/19 0,7} 27. Bc1
{+0,15/12 1,0} Nxf5 {+1,57/20 0,8} 28. gxf5 {+0,55/12 1,0} Rxf5
{+1,52/20 0,7} 29. Rf2 {+0,51/12 0,9} Ra8 {+1,03/20 0,7} 30. Ne4
{+0,69/13 0,9} Raf8 {+1,56/17 0,7} 31. b6 {+0,43/13 0,9} cxb6
{+1,08/19 0,6} 32. Nxd6 {+1,16/13 0,8} Re5 {+0,77/18 0,8} 33. Nc4
{+1,15/13 0,8} Re1 {+0,81/16 0,5} 34. Bd2 {+1,60/12 0,8} Ra1 {+0,72/16 1,1}
35. d6 {+2,03/12 0,8} Ra2 {+0,24/15 0,5} 36. Bd5+ {+2,07/12 0,7} Kh7
{-0,10/16 0,5} 37. d7 {+2,19/12 0,7} Rc2 {-0,38/16 0,8} 38. Ne5
{+2,34/11 0,7} g6 {-1,56/16 1,0} 39. Be1 {+2,55/11 0,7} Rc5 {-1,05/17 0,3}
40. Rxf4 {+2,18/11 0,6} Rd8 {-1,10/18 0,4} 41. Rd4 {+2,66/11 0,6} Rc2+
{-1,23/15 0,3} 42. Kg3 {+2,32/12 0,6} Kg7 {-1,75/16 0,6} 43. Bb4
{+2,32/10 0,6} Kf6 {-2,64/15 0,9} 44. Ng4+ {+3,67/10 0,6} Kg7
{-3,23/17 0,3} 45. Be7 {+3,71/12 0,5} Rxd7 {-3,91/17 0,3} 46. Bf6+
{+4,25/11 0,5} Kf8 {-3,78/17 0,2} 47. Bxg5 {+3,17/10 0,5} Rc5
{-4,02/16 0,2} 48. Rf4+ {+4,70/9 0,5} Kg7 {-4,41/17 0,2} 49. Bf6+
{+5,33/10 0,5} Kf8 {-4,75/17 0,2} 50. Bd4+ {+6,07/9 0,5} Ke7 {-5,48/18 0,2}
51. Rf7+ {+6,37/10 0,4} Kd8 {-6,22/20 0,3} 52. Rxd7+ {+6,11/10 0,4} Kxd7
{-6,39/19 0,2} 53. Bxc5 {+5,60/14 0,4} bxc5 {-6,42/18 0,2} 54. Bxb7
{+5,99/15 0,4} Kd6 {-6,56/19 0,2} 55. Be4 {+6,30/15 0,4} g5 {-6,56/17 0,2}
56. Kxh3 {+6,58/16 0,4} Ke6 {-7,48/17 0,1} 57. Bc6 {+6,73/17 0,4} Kf5
{-8,15/18 0,2} 58. Kg3 {+7,39/15 0,4} Ke6 {-8,75/18 0,1} 59. Bxa4
{+7,58/14 0,3} Kd5 {-9,15/17 0,2} 60. Nf6+ {+8,81/14 0,3} Ke5
{-9,40/18 0,2} 61. Nd7+ {+9,29/14 0,3} Kd5 {-10,79/18 0,2} 62. Kg4
{+9,46/15 0,3} c4 {-8,49/18 0,1} 63. dxc4+ {+11,41/15 0,3} Kxc4
{-8,89/18 0,1} 64. Kxg5 {+11,42/15 0,3} Kc3 {-9,01/18 0,1} 65. Bb5
{+12,17/13 0,3} Kb3 {-9,01/19 0,1} 66. a4 {+12,48/13 0,3} Kb4
{-9,01/18 0,1} 67. Kf4 {+13,04/12 0,3} Ka5 {-10,38/17 0,1} 68. Ke4
{+13,46/12 0,2} Kb4 {-8,49/7 0,1} 69. Kd4 {+13,84/13 0,2} Ka5
{-16,54/20 0,1} 70. Bc6 {+15,84/13 0,2} Ka6 {-13,62/18 0,1} 71. a5
{+16,70/12 0,2} Kxa5 {-7,71/15 0,1} 72. Kc5 {+16,00/11 0,2} Ka6
{-7,66/7 0,1} 73. Kb4 {+16,74/11 0,2} Ka7 {-7,34/7 0,1} 74. Ka4
{+16,70/10 0,2} Ka6 {-7,23/7 0,1} 75. Bg2 {+16,74/11 0,2} Ka7 {-7,15/7 0,1}
76. Kb3 {+16,60/10 0,2} Ka6 {-7,23/7 0,1} 77. Kb4 {+16,74/11 0,2} Ka7
{-7,34/7 0,1} 78. Bc6 {+16,98/11 0,2} Ka6 {-7,43/7 0,1} 79. Ka4
{+16,74/11 0,2} Ka7 {-7,17/7 0,1} 80. Bh1 {+15,71/10 0,2} Ka6 {-7,17/7 0,1}
81. Kb4 {+16,28/11 0,2} Ka7 {-7,23/7 0,1} 82. Bg2 {+16,37/11 0,2} Ka6
{-7,43/7 0,1} 83. Kc5 {+16,37/11 0,2} Ka5 {-7,70/13 0,1} 84. Nb6
{+15,95/12 0,1} Ka6 {-7,69/7 0,1} 85. Nc4 {+15,91/12 0,1} Ka7 {-7,69/7 0,1}
86. Kc6 {+16,24/11 0,1} Ka8 {-1000,07/12 0,1} 87. Kb6+ {+104,85/9 0,1} Kb8
{-1000,08/7 0,1} 88. Bh3 {+104,85/8 0,1} Ka8 {-1000,07/7 0,1} 89. Ne5
{+104,85/7 0,1} Kb8 {-7,98/7 0,1} 90. Nc6+ {+104,85/6 0,1} Ka8
{-1000,04/7 0,1} 91. Nb4 {+104,85/5 0,1} Kb8 {-1000,02/7 0,1} 92. Na6+
{+104,85/4 0,1} Ka8 {-1000,01/7 0,1} 93. Bg2# {+104,85/3 0,1}
{Xboard adjudication: Checkmate} 1-0[/pgn]
-
- Posts: 1525
- Joined: Tue Jul 15, 2014 12:47 pm
Re: Mayhem NNUE - New NN engine
How you are doing new versions of program please to make it available these are versions - mayhem-windows-slow-64bit.exe
-
- Posts: 491
- Joined: Sat Mar 02, 2013 11:31 pm
Re: Mayhem NNUE - New NN engine
I guess you need a Ryzen for these builds. Intel CPUs run too hot anyway.Krzysztof Grzelak wrote: ↑Sun Mar 07, 2021 12:19 am How you are doing new versions of program please to make it available these are versions - mayhem-windows-slow-64bit.exe
I'll add "slow" builds when I have time.
They are normal builds but without =native. A little bit slower but maximum compatibility.
I'll provide only those in the future.
-
- Posts: 491
- Joined: Sat Mar 02, 2013 11:31 pm
Re: Mayhem NNUE - New NN engine
Mayhem 3.5 unleashed !
Build the binary and they will come.
Release: https://github.com/SamuraiDangyo/mayhem ... s/tag/v3.5
Added only slow binaries.
Don't use superslow. Because it's super slow
Classical evaluation + other improvements.
Should be around 3000 ELO. At least in my testing.
Build the binary and they will come.
Release: https://github.com/SamuraiDangyo/mayhem ... s/tag/v3.5
Added only slow binaries.
Don't use superslow. Because it's super slow
Classical evaluation + other improvements.
Should be around 3000 ELO. At least in my testing.