Introducing Toga III chess engine

[dkappe]

A bunch more RL data has yielded a stronger net. Also have eliminated the need for libnnueprobe in Windows. Still training.

[pgn]

[Event "?"]
[Site "?"]
[Date "2021.05.08"]
[Round "1"]
[White "toga3.3.2"]
[Black "Defenchess"]
[Result "1-0"]
[ECO "E10"]
[GameDuration "00:04:32"]
[GameEndTime "2021-05-08T07:17:19.138 CDT"]
[GameStartTime "2021-05-08T07:12:46.763 CDT"]
[Opening "Queen's pawn game"]
[PlyCount "179"]
[Termination "adjudication"]
[TimeControl "60+1"]

1. d4 {book} Nf6 {book} 2. c4 {book} e6 {book} 3. Nf3 {book} d5 {book}
4. Nc3 {book} dxc4 {book} 5. e3 {book} c5 {book} 6. Bxc4 {book}
a6 {-0.21/19 2.5s} 7. O-O {+0.25/14 2.8s} b5 {-0.05/18 2.8s}
8. Be2 {+0.10/14 2.8s} Bb7 {+0.05/20 3.1s} 9. dxc5 {+0.11/14 3.6s}
Qxd1 {+0.01/20 1.3s} 10. Rxd1 {+0.26/16 2.1s} Bxc5 {-0.05/21 2.5s}
11. Nd4 {+0.13/16 3.1s} Nbd7 {+0.03/19 1.9s} 12. Nb3 {+0.54/16 2.5s}
Bb6 {-0.02/20 1.6s} 13. a4 {+0.63/17 2.2s} b4 {-0.19/22 1.8s}
14. a5 {+0.55/17 1.9s} Bc7 {-0.46/19 3.2s} 15. Na2 {+0.38/16 3.2s}
Bd5 {-0.60/20 2.7s} 16. Nd4 {+0.45/17 2.4s} Bxa5 {-0.45/22 5.4s}
17. Bd2 {+0.62/17 2.3s} Rb8 {-0.67/22 2.6s} 18. Bxa6 {+0.59/16 3.2s}
O-O {-0.58/22 5.0s} 19. Nc1 {+0.70/16 1.5s} Bd8 {-0.59/23 6.7s}
20. Nd3 {+0.74/16 2.2s} b3 {-0.49/20 2.5s} 21. Nb4 {+0.80/15 2.2s}
Nc5 {-0.59/20 1.8s} 22. Bf1 {+0.93/14 2.0s} Rc8 {-0.58/18 3.9s}
23. Bb5 {+0.85/14 1.9s} Bb6 {-0.71/20 2.1s} 24. Nxd5 {+0.78/13 1.5s}
Nxd5 {-0.67/21 1.8s} 25. Ra3 {+0.86/14 1.6s} Ne4 {-0.57/19 2.6s}
26. Nxb3 {+0.89/14 2.0s} Nxd2 {-0.55/19 2.7s} 27. Nxd2 {+0.95/14 1.2s}
Bc5 {-0.65/20 2.5s} 28. Rb3 {+0.83/15 2.0s} Be7 {-0.69/19 2.3s}
29. g3 {+0.89/14 1.8s} Rb8 {-0.64/19 3.6s} 30. Nf3 {+1.02/14 1.5s}
Rfc8 {-0.62/20 1.0s} 31. Kg2 {+0.90/15 1.9s} g6 {-0.58/19 1.9s}
32. Ra1 {+0.80/14 3.9s} Nf6 {-0.57/19 2.1s} 33. Ra7 {+0.91/14 1.6s}
Bc5 {-0.28/20 1.3s} 34. Ra4 {+0.81/14 1.7s} Bf8 {-0.48/19 2.4s}
35. h3 {+0.76/13 2.3s} Kg7 {-0.44/19 2.3s} 36. g4 {+0.74/14 2.4s}
h6 {-0.35/18 1.8s} 37. Ra7 {+0.65/14 1.9s} Bc5 {-0.35/17 2.2s}
38. Ra2 {+0.71/15 1.5s} Ne4 {-0.20/20 1.1s} 39. Ba4 {+0.70/14 1.1s}
Ra8 {-0.24/19 1.2s} 40. Rb7 {+0.65/15 1.6s} Bd6 {0.00/21 1.4s}
41. b3 {+0.40/14 0.97s} Rc1 {-0.13/20 1.8s} 42. h4 {+0.42/13 1.6s}
Nc5 {0.00/20 1.2s} 43. Rb6 {+0.36/14 1.4s} Be7 {0.00/20 0.99s}
44. Rc6 {+0.35/14 1.1s} Rd1 {0.00/22 1.1s} 45. Ne5 {+0.14/14 1.5s}
Rd5 {0.00/19 0.83s} 46. Nc4 {+0.40/14 0.96s} Rb8 {0.00/20 1.5s}
47. Rc7 {+0.70/12 1.3s} Bd8 {-0.23/21 1.4s} 48. Ra7 {+0.66/13 1.1s}
Nxa4 {0.00/21 1.1s} 49. R2xa4 {+0.25/11 1.6s} Bxh4 {-0.14/19 1.2s}
50. R4a6 {+0.46/14 2.2s} Kg8 {0.00/22 1.2s} 51. Rc6 {+0.67/13 1.0s}
Bd8 {0.00/22 0.92s} 52. b4 {+0.23/12 2.0s} Rxb4 {-0.30/19 1.2s}
53. Rc8 {+0.16/14 1.5s} Kg7 {-0.49/19 1.1s} 54. e4 {+1.13/16 1.4s}
Rxc4 {-0.99/19 1.1s} 55. Rxc4 {+0.98/15 0.95s} Rd2 {-1.34/19 0.92s}
56. g5 {+1.26/15 1.3s} hxg5 {-1.43/21 1.0s} 57. Rb4 {+1.47/16 1.1s}
Kf6 {-1.41/20 1.1s} 58. Rb5 {+1.67/15 1.6s} Rd4 {-1.57/20 1.1s}
59. e5+ {+2.45/15 0.97s} Kf5 {-1.66/21 1.1s} 60. Rxf7+ {+2.41/16 0.91s}
Ke4 {-2.22/20 1.0s} 61. Kg3 {+2.51/16 1.4s} Rd1 {-2.24/22 1.0s}
62. f3+ {+2.50/14 0.90s} Ke3 {-2.43/22 1.0s} 63. Rb3+ {+2.52/15 0.83s}
Ke2 {-2.81/18 1.0s} 64. Ra7 {+2.62/15 1.3s} Rd2 {-2.95/19 1.0s}
65. Rb4 {+3.60/15 0.91s} Rd5 {-4.00/17 1.0s} 66. Re4+ {+3.93/15 1.4s}
Kd3 {-5.00/19 3.3s} 67. Ra3+ {+4.10/15 1.1s} Kc2 {-4.96/17 1.6s}
68. Ra6 {+4.08/14 0.40s} g4 {-5.30/16 1.2s} 69. Kxg4 {+4.59/13 1.1s}
Rd1 {-5.35/18 0.99s} 70. Rxe6 {+4.58/11 0.36s} Rg1+ {-5.58/17 1.0s}
71. Kh3 {+4.72/12 0.41s} Kb3 {-5.74/17 1.0s} 72. Rg4 {+4.81/13 1.4s}
Rh1+ {-5.10/13 0.99s} 73. Kg2 {+4.98/13 1.1s} Re1 {-5.44/14 1.0s}
74. Rgxg6 {+5.03/12 0.90s} Re2+ {-5.80/14 0.99s} 75. Kg3 {+5.12/14 1.4s}
Ba5 {-6.42/17 1.0s} 76. f4 {+5.23/13 1.0s} Be1+ {-6.48/16 0.99s}
77. Kf3 {+5.39/14 1.4s} Rf2+ {-6.94/17 1.0s} 78. Ke4 {+5.51/15 1.4s}
Re2+ {-7.37/16 1.0s} 79. Kd3 {+5.67/14 1.4s} Rd2+ {-7.65/17 1.00s}
80. Ke3 {+5.73/13 0.66s} Ra2 {-7.32/17 1.00s} 81. Ra6 {+6.00/13 1.4s}
Rh2 {-7.61/17 1.0s} 82. f5 {+6.16/12 1.4s} Kc4 {-7.41/13 1.00s}
83. f6 {+6.44/13 1.3s} Bf2+ {-7.91/13 1.00s} 84. Kf3 {+6.60/14 1.3s}
Kb5 {-8.28/13 1.00s} 85. Ra2 {+9.37/18 1.1s} Rh7 {-209.57/15 1.0s}
86. Rxf2 {+11.61/17 0.90s} Rh3+ {-209.60/24 0.90s} 87. Kg2 {+16.39/27 0.83s}
Rd3 {-209.63/29 1.0s} 88. f7 {+18.01/16 0.32s} Ra3 {-209.65/36 0.75s}
89. f8=Q {+23.97/20 0.96s} Rg3+ {-209.66/128 0.21s}
90. Kxg3 {+M29/63 0.18s, White wins by adjudication: SyzygyTB} 1-0
[/pgn]

[Chessqueen]

Unfortunately nothing could. My processor is http://talkchess.com/forum3/viewtopic.p ... 1&start=10

We know your processor. You are doing something wrong. Try to read carefully again what I wrote.
Ofc I have tested it myself on even older hardware than yours and even here the sse version runs
after doing what I wrote.

Code: Select all

[Event "CuteChessDevGS"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.24"]
[Round "1"]
[White "Berserk_330-64"]
[Black "FruitTogaIII-NNUE_03-64"]
[Result "0-1"]
[ECO "E00"]
[GameDuration "00:07:31"]
[GameEndTime "2021-04-24T20:52:28.329 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-24T20:44:56.990 Mitteleuropõische Sommerzeit"]
[Opening "Queen's pawn game"]
[PlyCount "164"]
[TimeControl "40/120"]

1. d4 {book} Nf6 {book}
2. c4 {book} e6 {book}
3. e3 {book} Bb4+ {book}
4. Nd2 {+0.22/16 3.21} Be7 {-0.19/13 2.38}
5. Ngf3 {+0.28/16 3.23} d6 {-0.21/14 2.91}
6. Bd3 {+0.36/16 3.23} O-O {-0.18/14 2.73}
7. O-O {+0.29/16 3.23} c5 {-0.19/14 3.53}
8. b3 {+0.22/17 3.22} b6 {-0.11/14 3.18}
9. Bb2 {+0.23/15 3.23} Nbd7 {-0.11/14 2.98}
10. Bc2 {+0.24/14 3.22} Qc7 {-0.09/13 3.24}
11. e4 {+0.33/16 3.22} cxd4 {0.00/14 2.51}
12. Nxd4 {+0.23/15 3.23} a6 {-0.03/16 3.15}
13. h3 {+0.28/14 3.22} Bb7 {+0.03/13 2.32}
14. a4 {+0.26/15 3.22} Rfc8 {+0.07/12 2.00}
15. Qe2 {+0.22/16 3.22} g6 {+0.08/14 2.47}
16. N4f3 {0.00/16 3.23} Nh5 {+0.25/15 3.14}
17. Rfe1 {0.00/17 3.22} Nf4 {+0.32/15 2.82}
18. Qe3 {0.00/18 3.22} e5 {+0.28/14 3.30}
19. g3 {-0.20/14 3.23} Ne6 {+0.30/15 3.31}
20. a5 {-0.21/16 3.23} b5 {+0.38/15 3.31}
21. Bc3 {-0.18/15 3.24} Qc5 {+0.58/12 4.10}
22. Rac1 {-0.22/15 3.24} Qxe3 {+0.61/14 2.03}
23. Rxe3 {-0.10/17 3.23} Bf8 {+0.63/15 3.48}
24. Rd1 {-0.16/15 3.25} Rc7 {+0.57/14 3.37}
25. Kg2 {0.00/17 3.24} Rac8 {+0.72/13 2.89}
26. Bb4 {-0.07/16 3.24} Nb8 {+0.81/13 3.44}
27. Bc3 {-0.51/16 3.24} Nc6 {+0.97/14 2.35}
28. Nf1 {-0.72/16 3.24} b4 {+1.18/14 3.09}
29. Bb2 {-0.74/17 3.24} Nxa5 {+1.22/15 5.91}
30. Ra1 {-0.91/18 3.26} Nc6 {+1.30/16 3.38}
31. Ree1 {-0.90/17 3.25} Ncd4 {+1.39/16 3.38}
32. Bxd4 {-0.94/17 3.26} exd4 {+1.63/16 2.19}
33. Red1 {-0.71/18 3.25} Ra8 {+1.64/18 4.28}
34. Ra5 {-1.27/20 3.25} Bg7 {+1.88/16 2.61}
35. N1d2 {-1.40/19 3.27} Nc5 {+1.90/16 2.43}
36. Ne1 {-1.05/15 3.28} f5 {+1.95/17 3.96}
37. f3 {-1.56/17 3.27} Bh6 {+2.04/15 4.02}
38. exf5 {-1.42/15 3.28} gxf5 {+2.17/15 2.61}
39. Ra2 {-1.89/15 3.30} Rf8 {+2.28/15 3.90}
40. Bb1 {-1.99/16 3.33} Re7 {+2.58/15 3.63}
41. Kf1 {-2.18/17 2.98} Kh8 {+2.86/14 2.98}
42. Ng2 {-1.99/17 2.97} Bxd2 {+2.92/14 5.25}
43. Rdxd2 {-2.31/19 2.97} Nxb3 {+2.97/14 2.52}
44. Rdb2 {-2.06/17 2.98} Nc5 {+2.44/14 2.45}
45. Rxb4 {-2.33/17 2.97} Bxf3 {+2.26/13 3.42}
46. Rb6 {-2.39/16 2.97} Rg7 {+2.67/13 3.38}
47. Rxd6 {-2.40/17 2.97} Rxg3 {+2.60/13 3.27}
48. Ne1 {-2.37/16 2.98} Ba8 {+2.59/13 3.00}
49. Rh2 {-2.14/20 2.97} Be4 {+2.46/12 2.21}
50. Bxe4 {-2.06/18 2.98} fxe4+ {+3.07/14 1.85}
51. Rf2 {-2.13/17 3.00} Rfg8 {+2.90/15 2.94}
52. Rxd4 {-1.79/17 2.98} Rxh3 {+2.65/13 2.09}
53. Ke2 {-2.55/17 2.99} Rb8 {+2.81/13 2.43}
54. Rg2 {-2.30/18 2.99} a5 {+3.28/11 2.27}
55. Kd1 {-2.94/18 3.00} a4 {+3.18/14 7.37}
56. Rd5 {-2.96/18 3.00} Rc8 {+3.19/14 2.88}
57. Rgg5 {-3.18/18 2.99} Nb3 {+3.60/15 2.88}
58. Rg2 {-3.46/17 2.99} a3 {+3.73/13 2.89}
59. Ra2 {-3.21/16 2.99} Rxc4 {+4.49/12 2.88}
60. Rxa3 {-4.54/21 2.99} Rc1+ {+4.92/13 0.92}
61. Ke2 {-4.58/17 2.99} Rh2+ {+4.78/13 0.76}
62. Kf1 {-4.53/21 2.99} Nd2+ {+4.85/15 3.36}
63. Rxd2 {-4.51/19 2.99} Rxd2 {+4.84/16 0.84}
64. Re3 {-5.20/20 2.99} Rc4 {+5.09/18 2.03}
65. Re2 {-5.16/20 2.99} Rxe2 {+8.90/31 2.38}
66. Kxe2 {-8.85/22 3.00} h5 {+10.04/19 0.75}
67. Ng2 {-11.17/22 3.00} Rc2+ {+10.64/16 0.73}
68. Kf1 {-11.13/24 3.01} Rxg2 {+10.76/18 3.79}
69. Kxg2 {-11.89/26 3.00} h4 {+10.48/12 1.20}
70. Kf2 {-11.51/23 3.01} Kg7 {+10.76/13 1.15}
71. Ke3 {-11.61/23 3.01} h3 {+10.77/11 2.70}
72. Kxe4 {-M112/21 3.02} h2 {+10.85/13 4.56}
73. Kd5 {-M22/25 3.02} h1=Q+ {+297.44/16 4.57}
74. Kd4 {-M16/27 3.02} Qf3 {+297.44/13 3.00}
75. Kc4 {-M16/27 3.01} Kf7 {+M45/63 0.13}
76. Kd4 {-M14/30 3.03} Ke6 {+M13/25 3.57}
77. Kc4 {-M12/35 3.04} Qe3 {+M11/63 0.081}
78. Kb5 {-M10/58 3.05} Qc3 {+M9/63 1.52}
79. Ka4 {-M8/79 3.06} Kd6 {+M7/63 0.071}
80. Kb5 {-M6/127 0.25} Qb3+ {+M5/63 0.045}
81. Ka6 {-M4/127 0.12} Kc6 {+M3/63 0.039}
82. Ka7 {-M2/127 0.092} Qb7# {+M1/63 0.054}
{Black mates}
0-1

Code: Select all

Toga III 0.3 UCI based on Fruit 2.1 by Dietrich Kappe, Jerry Donald Watson, Thomas Gaksch and Fabien Letouzey.
Embedded NNUE loaded !
uci
id name Toga III 0.2
id author Dietrich Kappe, Jerry Donald Watson, Thomas Gaksch and Fabien Letouzey
option name Hash type spin default 64 min 4 max 16384
option name Search Time type spin default 0 min 0 max 3600
option name Search Depth type spin default 0 min 0 max 20
option name Ponder type check default false
option name OwnBook type check default true
option name BookFile type string default performance.bin
option name MultiPV type spin default 1 min 1 max 10
option name Hash Pruning type check default true
option name Hash Pruning Depth type spin default 1 min 0 max 6
option name Hash Pruning Margin type spin default 100 min 0 max 1000
option name NullMove Pruning type combo default Always var Always var Fail High var Never
option name NullMove Reduction type spin default 3 min 1 max 4
option name Razoring type check default true
option name Razoring Depth type spin default 4 min 0 max 6
option name Razoring Margin type spin default 300 min 0 max 1000
option name Evaluation Pruning type check default true
option name Evaluation Pruning Depth type spin default 5 min 0 max 6
option name History Pruning type check default true
option name History Threshold type spin default 70 min 0 max 100
option name Delta Pruning type check default true
option name Delta Margin type spin default 50 min 0 max 500
option name Quiescence Check Plies type spin default 1 min 0 max 2
option name Material type spin default 100 min 0 max 400
option name Piece Activity type spin default 100 min 0 max 400
option name Piece Square Activity type spin default 100 min 0 max 400
option name King Safety type spin default 100 min 0 max 400
option name Pawn Structure type spin default 100 min 0 max 400
option name Passed Pawns type spin default 100 min 0 max 400
option name Toga Lazy Eval type check default true
option name Toga Lazy Eval Margin type spin default 200 min 0 max 900
option name Toga Lazy Eval Mobility Margin type spin default 125 min 0 max 900
option name Toga King Pawn Endgame Bonus type spin default 30 min 0 max 100
option name Toga Rook Pawn Endgame Penalty type spin default 10 min 0 max 100
option name Number of Threads type spin default 1 min 1 max 64
uciok
go infinite
info depth 1
info multipv 1 depth 1 seldepth 1 score cp -8 time 0 nodes 2 pv b1a3
info multipv 1 depth 1 seldepth 1 score cp 20 time 15 nodes 3 pv b1c3
info multipv 1 depth 1 seldepth 1 score cp 21 time 15 nodes 4 pv g1f3
info multipv 1 depth 1 seldepth 1 score cp 28 time 31 nodes 15 pv e2e4
info depth 1 seldepth 1 time 46 nodes 21 nps 0
...
info depth 14 seldepth 32 time 3759 nodes 865039 nps 230125
...
info depth 17 seldepth 35 time 15350 nodes 4304800 nps 280443

[dkappe]

After generating some more RL data, I have a stronger net and Windows binaries that don’t need a library.

https://github.com/dkappe/TogaIII/relea ... ga3-0.3.12

At a guess it’s 20 elo stronger than Toga III 0.2. I also provide some BMI compiles, but do test to see if it is stronger than avx2.

[Gabor Szots]

After generating some more RL data, I have a stronger net and Windows binaries that don’t need a library.

https://github.com/dkappe/TogaIII/relea ... ga3-0.3.12

At a guess it’s 20 elo stronger than Toga III 0.2. I also provide some BMI compiles, but do test to see if it is stronger than avx2.

Is the toganet.bin in the src the newest net?

[dkappe]

After generating some more RL data, I have a stronger net and Windows binaries that don’t need a library.

https://github.com/dkappe/TogaIII/relea ... ga3-0.3.12

At a guess it’s 20 elo stronger than Toga III 0.2. I also provide some BMI compiles, but do test to see if it is stronger than avx2.

Is the toganet.bin in the src the newest net?

Yes.

[dkappe]

I’ve uploaded some, I hope, better compiles, especially for Windows. https://www.patreon.com/posts/51269439

[Scally]

Thanks,

It compiles and runs fine on my Raspberry Pi4 for the DGT Pi & PicoChess.


Al.

[dkappe]

Thanks,

It compiles and runs fine on my Raspberry Pi4 for the DGT Pi & PicoChess.


Al.

The dark branch contains the 70 elo stronger Dark Toga.

[Scally]

Thanks

Al.

[Rebel]

Dietrich, I totally overlooked the new Toga.

https://prodeo.actieforum.com/t358p25-g ... sults#4064

Enjoy.