Page 2 of 8
Re: The new Tafl engine (AI)
Posted: Sat Apr 25, 2026 3:22 pm
by sovaz1997
Draganov wrote: Sat Apr 25, 2026 9:30 am
I found out another weak point in its play with the whites. When playing with the whites, the engine is playing very sharply trying to set up guillotine positions or attacking the corners in a strong way, mobilizing the king early and putting a lot of pressure on the corners. However, I found out that in some games, the engine managed to clear 5-6 squares on the edge and instead of trying to build an exit fort, which will be impossible for me to block, decided to continue penetrating the corners. I think it put much more weight on attacking the corners and neglecting the exit fort options. In some of the games, I managed to block its king on the edge. The engine had a good position but missed my moves that blocked the king on the edge and it lost the game.
Yeah! I think because white strategy is easier to learn.
It's really difficult for engine to make the fort, because it hard to get in train process. But at some point, a "snowball" might happen and it will quickly see new strategies, as it did with corners.
Re: The new Tafl engine (AI)
Posted: Sat Apr 25, 2026 3:24 pm
by sovaz1997
Draganov wrote: Sat Apr 25, 2026 10:15 am
Is it possible to add an option where a player can upload a game notation and the engine to analyze how well both players played. I want to analyze the championship games of the world champions and see how well Schachus played in 2014 or Herjan in 2016. It will answer a lot of interesting questions like whether the best version of Herjan played better than the champion version of Adam? Was Casshern better in his champion years than Plantagenet in 2017? How strong was I in 2021?
For example in chess.com we have a similar function to analyze the games of other players. And some chess enthusiasts made projects of comparing the world champions and their accuracy.
Yeah, I plan to add game uploading.
Also in local gui I plan to add analyze things, etc. Because local analysis is much better anyway (on any gpu).
Re: The new Tafl engine (AI)
Posted: Sat Apr 25, 2026 3:35 pm
by sovaz1997
I launched a match between a small 6x64 network and a new 10x128 network that's currently training (it's currently 50 ELO behind the small network).
I runned 15 seconds per move. It's like if we run it in browser for maybe 5-10 minutes for every move, because locally engine plays 2000 nodes per second
No opening book
If you're interested, I can run them after on a specific opening. But for statistics, a lot of games are needed anyway.
Re: The new Tafl engine (AI)
Posted: Sat Apr 25, 2026 4:05 pm
by sovaz1997
@Draganov I'm added import & export games feature!
Re: The new Tafl engine (AI)
Posted: Sat Apr 25, 2026 7:59 pm
by sovaz1997
Some games played locally 15s/move on GPU:
gen0146: previous small NN
10x128_gen0042: new bigger net, slightly weaker than 0146
Games could be imported to
https://playtafl.org/
https://playtafl.org/?game=RwAzYWFhYWEz ... 2NKA%3D%3D
Code: Select all
[Event "Zevratafl testing"]
[Site "local"]
[Date "2026.04.25"]
[Round "Match 3 - Game A"]
[White "TaflZero NN: ./gen0146.onnx"]
[Black "TaflZero NN: ./10x128_gen0042.onnx"]
[Result "0-1"]
[SetUp "1"]
[FEN "3aaaaa3/5a5/11/a4d4a/a3ddd3a/aa1ddkdd1aa/a3ddd3a/a4d4a/11/5a5/3aaaaa3 a"]
[TimeControl "15000ms"]
1. a8c8 {+0.55} h6h9 {-0.45} 2. e11e9 {+0.71} g5j5 {-0.62} 3. k4j4 {+0.75} g6g2 {-0.67} 4. h1h2 {+0.76} f6i6 {-0.75} 5. k5i5 {+0.79} f4i4 {-0.74} 6. f10i10 {+0.81} f5k5 {-0.75} 7. j4k4 {+0.81} i4i1 {-0.77} 8. a4i4 {+0.81} i6i5 {-0.79} 9. g1g3 {+0.85} i1i3 {-0.78} 10. j6j3 {+0.88} i3i1 {-0.78} 11. h2j2 {+0.89} i5i9 {-0.21} 12. k6i6 {+0.89} i9j9 {-0.34} 13. i10j10 {+0.92} j9k9 {-0.40} 14. j10k10 {+0.93} k9i9 {-0.40} 15. k10i10 {+0.94} i9j9 {-0.47} 16. i10j10 {+0.97} j9k9 {-0.54} 17. j10k10 {+1.00} g7i7 {-0.55} 18. h11i11 {+0.95} k9i9 {-0.58} 19. k10i10 {+1.00} i9k9 {-0.55} 20. i10k10 {+0.94} e5i5 {-0.62} 21. d11d10 {+0.99} e7b7 {-0.64} 22. a5h5 {+1.11} k9j9 {-0.72} 23. k10j10 {+1.09} j9k9 {-0.80} 24. j10k10 {+1.14} i7i10 {-0.80} 25. k7j7 {+1.26} k9j9 {-0.85} 26. k10j10 {+1.25} j9k9 {-0.93} 27. j10k10 {+1.31} e6e5 {-0.96} 28. c8b8 {+1.47} k9j9 {-1.13} 29. k10j10 {+1.40} j9k9 {-1.22} 30. j10k10 {+1.46} e5g5 {-1.28} 31. g3i3 {+1.58} k9j9 {-1.13} 32. k10j10 {+1.50} j9k9 {-1.24} 33. j10k10 {+1.61} i10g10 {-1.28} 34. a7c7 {+1.66} k9j9 {-1.47} 35. k10j10 {+1.73} j9k9 {-1.55} 36. j10k10 {+1.84} h9f9 {-1.67} 37. b8d8 {+1.94} k9j9 {-1.74} 38. k10j10 {+2.06} j9k9 {-1.86} 39. j10k10 {+2.19} i5j5 {-1.92} 40. j7j10 {+3.42} f7k7 {-2.77} 41. a6a3 {+3.80} g5g9 {-3.10} 42. f11f10 {+4.52} i1h1 {-3.52} 43. g11h11 {+5.89} h1h10 {-4.50} 44. f2b2 {+5.54} d6i6 {-4.19} 45. e1e8 {+7.35} g9g1 {-4.11} 46. i3i1 {+8.50} g1g9 {-4.17} 47. f1g1 {+11.73} g9g3 {-5.16} 48. g1h1 {+11.34} g3h3 {-5.14} 49. d1c1 {+11.41} h3h9 {-5.19} 50. h1g1 {+13.23} i6g6 {-6.49} 51. i11i10 {+13.01} h9h6 {-6.83} 52. h11i11 {+19.20} h6h4 {-8.94} 53. f10f11 {+19.38} h4i4 {-9.06} 54. f11h11 {+19.74} g6g3 {-9.44} 55. g1h1 {+19.29} f9h9 {-9.86} 56. h11g11 {+18.99} h9g9 {-9.92} 57. g11h11 {+18.42} g9h9 {-9.76} 58. h11g11 {+19.10} h9g9 {-9.86} 59. g11h11 {+18.93} g3h3 {-10.37} 60. h1g1 {+19.85} f8f11 {-10.15} 61. g1f1 {+17.66} f11f8 {-8.90} 62. f1g1 {+19.92} f8f11 {-10.19} 63. g1f1 {+17.76} k7k6 {-9.13} 64. f1f9 {+18.77} g9h9 {-12.40} 65. e9e11 {+25.18} h9g9 {-14.79} 66. h11g11 {+31.29} h10h8 {-22.82} 67. e11e9 {+31.37} h8f8 {-21.53} 68. c7f7 {+38.39} k9j9 {-28.70} 69. k10k9 {+49.76} j5b5 {-44.45} 70. b2b4 {+52.72} h3b3 {-36.61} 71. c1c3 {+60.55} i4c4 {-41.34} 72. d8d4 {+63.98} g10e10 {-44.67} 73. b6h6 {+63.91} k6k5 {-44.50} 74. h6h9 {+72.39} k5c5 {-50.08} 75. j3j8 {+107.60} j9i9 {-42.39} 76. i1i8 {+109.38} i9j9 {-60.87} 77. i8i9 {+125.74} 0-1
Code: Select all
[Event "Zevratafl testing"]
[Site "local"]
[Date "2026.04.25"]
[Round "Match 2 - Game A"]
[White "TaflZero NN: ./gen0146.onnx"]
[Black "TaflZero NN: ./10x128_gen0042.onnx"]
[Result "0-1"]
[SetUp "1"]
[FEN "3aaaaa3/5a5/11/a4d4a/a3ddd3a/aa1ddkdd1aa/a3ddd3a/a4d4a/11/5a5/3aaaaa3 a"]
[TimeControl "15000ms"]
1. a8c8 {+0.55} h6h9 {-0.45} 2. e11e9 {+0.71} g5j5 {-0.62} 3. k4j4 {+0.74} g6g2 {-0.67} 4. h1h2 {+0.76} f6i6 {-0.75} 5. k5i5 {+0.78} f4i4 {-0.76} 6. f10i10 {+0.80} f5k5 {-0.75} 7. j4k4 {+0.82} i4i1 {-0.76} 8. a4i4 {+0.81} i6i5 {-0.79} 9. g1g3 {+0.84} i1i3 {-0.78} 10. j6j3 {+0.87} i3i1 {-0.78} 11. h2j2 {+0.88} i5i9 {-0.22} 12. k6i6 {+0.89} i9j9 {-0.33} 13. i10j10 {+0.91} j9k9 {-0.39} 14. j10k10 {+0.93} k9i9 {-0.41} 15. k10i10 {+0.94} i9j9 {-0.47} 16. i10j10 {+0.95} j9k9 {-0.53} 17. j10k10 {+1.01} g7i7 {-0.56} 18. h11i11 {+0.96} k9i9 {-0.54} 19. k10i10 {+1.01} i9k9 {-0.55} 20. i10k10 {+0.95} e5i5 {-0.62} 21. d11d10 {+0.99} e7b7 {-0.64} 22. a5h5 {+1.12} k9j9 {-0.72} 23. k10j10 {+1.07} j9k9 {-0.80} 24. j10k10 {+1.13} i7i10 {-0.80} 25. k7j7 {+1.27} k9j9 {-0.84} 26. k10j10 {+1.24} j9k9 {-0.93} 27. j10k10 {+1.31} e6e5 {-0.96} 28. c8b8 {+1.43} k9j9 {-1.13} 29. k10j10 {+1.38} j9k9 {-1.23} 30. j10k10 {+1.45} e5g5 {-1.27} 31. g3i3 {+1.56} k9j9 {-1.12} 32. k10j10 {+1.49} j9k9 {-1.25} 33. j10k10 {+1.61} i10g10 {-1.30} 34. a7c7 {+1.67} g5g9 {-1.48} 35. i11i10 {+2.24} k9j9 {-2.09} 36. k10j10 {+2.22} j9k9 {-2.33} 37. j10k10 {+2.43} g9f9 {-2.43} 38. g11i11 {+3.16} k9j9 {-2.74} 39. k10j10 {+3.03} j9k9 {-2.83} 40. j10k10 {+2.94} h9h2 {-2.95} 41. e9c9 {+3.29} d6k6 {-3.19} 42. j7j10 {+4.05} f7k7 {-3.91} 43. e1e3 {+4.76} k7e7 {-4.18} 44. f1g1 {+5.73} k6d6 {-4.64} 45. b6b2 {+6.19} d6c6 {-5.03} 46. c7b7 {+7.54} f8d8 {-5.61} 47. d1c1 {+9.70} f9e9 {-6.42} 48. f11c11 {+12.96} e9f9 {-7.94} 49. a6a3 {+15.39} d8d4 {-9.13} 50. j2i2 {+15.32} h2h8 {-10.37} 51. f2h2 {+21.12} e7j7 {-9.90} 52. b7a7 {+22.37} i5e5 {-8.96} 53. d10b10 {+25.00} e5i5 {-9.37} 54. a7a9 {+27.29} j7e7 {-10.47} 55. e3f3 {+28.98} c6j6 {-11.82} 56. f3f4 {+29.72} h8h4 {-11.56} 57. f4f3 {+24.03} d4d8 {-10.88} 58. f3h3 {+25.39} e7c7 {-10.25} 59. j3j4 {+24.05} h4h8 {-10.20} 60. k4k5 {+30.03} j6j5 {-10.60} 61. h3h6 {+34.10} j5j6 {-11.05} 62. h6h4 {+29.34} j6j5 {-11.27} 63. h4i4 {+31.30} h8h6 {-13.21} 64. i10i9 {+22.66} d8i8 {-12.01} 65. i9i10 {+25.34} f9e9 {-11.94} 66. i2j2 {+29.95} h6h8 {-11.75} 67. h2h1 {+38.51} c7k7 {-16.99} 68. j2j1 {+47.67} k7d7 {-21.88} 69. j1j2 {+48.84} e9f9 {-21.08} 70. h1h6 {+55.53} i8i6 {-21.40} 71. h6h2 {+47.12} i6i8 {-21.57} 72. k5k7 {+41.92} i5i6 {-20.12} 73. i3k3 {+45.51} i8k8 {-28.00} 74. k3k6 {+54.45} k9j9 {-62.65} 75. k10k9 {+67.93} h8h11 {-59.33} 76. k6j6 {+77.57} f9i9 {-60.33} 77. h2h6 {+82.65} g10g2 {-57.24} 78. j6j8 {+86.64} g2g10 {-50.22} 79. j4j7 {+101.66} d7d4 {-50.79} 80. i4i8 {+102.40} j9i9 {-64.42} 81. h6h9 {+103.81} g10g9 {-60.91} 82. j7h7 {+90.94} d4i4 {-60.16} 83. k7i7 {+93.44} i4a4 {-65.61} 84. h7h9 {+115.70} i9j9 {-73.78} 85. i10i9 {+125.74} 0-1
Code: Select all
[Event "Zevratafl testing"]
[Site "local"]
[Date "2026.04.25"]
[Round "Match 3 - Game B"]
[White "TaflZero NN: ./10x128_gen0042.onnx"]
[Black "TaflZero NN: ./gen0146.onnx"]
[Result "0-1"]
[SetUp "1"]
[FEN "3aaaaa3/5a5/11/a4d4a/a3ddd3a/aa1ddkdd1aa/a3ddd3a/a4d4a/11/5a5/3aaaaa3 a"]
[TimeControl "15000ms"]
1. a7b7 {+0.49} g5i5 {-0.51} 2. f2i2 {+0.54} g6g3 {-0.52} 3. j6j3 {+0.50} g3b3 {-0.53} 4. e1e2 {+0.58} i5i10 {-0.50} 5. k6j6 {+0.69} f8c8 {-0.59} 6. e2b2 {+0.83} c8c1 {-0.72} 7. d1d3 {+0.93} c1b1 {-0.81} 8. f1c1 {+1.01} e5e2 {-0.85} 9. a4c4 {+1.25} e2b2 {-0.95} 10. a5a3 {+1.34} e6e1 {-1.02} 11. c4c3 {+1.40} e7e2 {-1.06} 12. c3c4 {+1.47} f7c7 {-1.09} 13. h11h9 {+1.74} e2d2 {-1.18} 14. c1d1 {+1.90} f6e6 {-1.56} 15. g1f1 {+2.10} e6e2 {-1.74} 16. a3a2 {+2.44} g7g9 {-1.93} 17. k7d7 {+2.87} i10i9 {-2.38} 18. e11e9 {+3.19} e2c2 {-2.67} 19. d1c1 {+4.28} f5a5 {-2.81} 20. d3a3 {+4.90} d6d5 {-2.94} 21. g11i11 {+6.04} i9j9 {-3.24} 22. b6b4 {+6.87} h6h10 {-3.69} 23. f11h11 {+8.18} h10i10 {-3.94} 24. k8k10 {+10.84} j9k9 {-5.53} 25. j6j10 {+12.75} k9k10 {-6.73} 26. k5k9 {+15.34} i10j10 {-7.14} 27. f1d1 {+15.47} d5d6 {-7.31} 28. d1d5 {+24.19} g9g3 {-8.77} 29. d5d3 {+31.03} a5f5 {-10.03} 30. a8j8 {+38.58} g3g9 {-11.84} 31. b7b10 {+45.92} f5e5 {-12.89} 32. d11c11 {+53.62} e5e1 {-15.79} 33. d3d1 {+55.66} e1e3 {-15.12} 34. d7d3 {+55.75} e3e7 {-16.79} 35. d1d2 {+60.08} e7e2 {-18.75} 36. d3d2 {+53.47} e2e1 {-16.43} 37. d2d1 {+54.38} e1e7 {-13.49} 38. a6a9 {+53.09} e7b7 {-16.24} 39. d1d2 {+59.26} f4e4 {-19.09} 40. i2e2 {+49.43} c2c3 {-14.15} 41. e2i2 {+29.80} c3b3 {-12.39} 42. a9a4 {+19.74} g9g2 {-8.08} 43. b4b5 {+40.55} b7b6 {-10.33} 44. b10b7 {+43.80} g2f2 {-14.39} 45. c4c3 {+43.77} e4e3 {-19.20} 46. d2d3 {+46.10} e3e1 {-24.76} 47. d3d1 {+42.83} e1e3 {-20.02} 48. d1d3 {+41.58} e3e1 {-24.77} 49. d3d1 {+44.08} f2f3 {-18.07} 50. e9e3 {+53.75} f3f4 {-26.72} 51. b5b4 {+53.04} f4d4 {-29.35} 52. b7b5 {+41.77} d4d3 {-34.01} 53. c11c3 {+43.09} e1e2 {-49.21} 54. k4d4 {+51.77} e2d2 {-38.22} 55. e3d3 {+28.89} j10i10 {-56.34} 56. k9i9 {+117.66} b2d2 {-18.46} 57. a2b2 {+125.74} 0-1
Re: The new Tafl engine (AI)
Posted: Sat Apr 25, 2026 10:04 pm
by Draganov
Sovaz can you make adjustments in importing the games to allow importing games in the notation format that we use in our game archive here on this website?
Re: The new Tafl engine (AI)
Posted: Sat Apr 25, 2026 10:26 pm
by sovaz1997
Draganov wrote: Sat Apr 25, 2026 10:04 pm
Sovaz can you make adjustments in importing the games to allow importing games in the notation format that we use in our game archive here on this website?
Will do it!
I'll support bnoth formats I think, because use my format in testing for now
Do you mean this format?
Code: Select all
1. d1-d3 e5-e2
2. d3-d2xe2 e6-e2
3. f2-i2 f4-f2
4. a5-c5 f6-e6
5. d2-b2 e6-e3
6. j6-j3
Re: The new Tafl engine (AI)
Posted: Sat Apr 25, 2026 11:58 pm
by Draganov
Exactly, this is the format.
Re: The new Tafl engine (AI)
Posted: Sun Apr 26, 2026 3:24 am
by sovaz1997
Draganov wrote: Sat Apr 25, 2026 11:58 pm
Exactly, this is the format.
Now works! (Classic format)
Re: The new Tafl engine (AI)
Posted: Sun Apr 26, 2026 6:36 am
by Draganov
On a smartphone it is difficult to review a game. The move history is bellow and to see the next move we need to scroll down and click on the exact move. Can you add back and forward buttons that are close to the board for easier navigation in a game?
I managed to upload a game in our format. However, to see the move evaluation I need to press the 'Go back' button and then the button 'Make engine move'. Then, the move history is rewritten and the game can't be reviewed forward until I import the moves again.
Is it possible to make the move evaluation visible on each move while reviewing a game? If this is done, we can analyze the games of the world champions and see how well they played.