New version of Allie v0.6

[Jamal Bubker]

Thank you Adam !!

[MMarco]

Hi,

I just released a new version of Allie: https://github.com/manyoso/allie/releases/tag/v0.6

You can find windows exe on Allie discord channel under #releases. This version will be playing in the upcoming TCEC Cup and is substantially stronger and faster than v0.5. Major parts of the code were rewritten and perft tests implemented along with NPS increases due to lockless threading redesign and lots of changes to TM.

Cheers!

Wow, Allie is blazingly fast!!

It achieves better results than Lc0 24.1 in test suite with the same net (LS 14.2), i.e. 123/200 on arasan20.epd on my 1660ti at 10s/move (Lc0 24.1 gets 115/200). I'm eager to see how that transposes into game play.

Thank you for the release!

[Master Om]

Allie 0.6

Thanks.
But it is only the source code of allie.
Where is the executable out of discord??

Thats the exe file. Source is on github.

[gonzochess75]

Hi,

I just released a new version of Allie: https://github.com/manyoso/allie/releases/tag/v0.6

You can find windows exe on Allie discord channel under #releases. This version will be playing in the upcoming TCEC Cup and is substantially stronger and faster than v0.5. Major parts of the code were rewritten and perft tests implemented along with NPS increases due to lockless threading redesign and lots of changes to TM.

Cheers!

Wow, Allie is blazingly fast!!

It achieves better results than Lc0 24.1 in test suite with the same net (LS 14.2), i.e. 123/200 on arasan20.epd on my 1660ti at 10s/move (Lc0 24.1 gets 115/200). I'm eager to see how that transposes into game play.

Thank you for the release!

Be careful here. Allie counts minimaxed nodes to compute NPS so it is apples to oranges comparison. To see which one is really faster you'd be better off looking at nvidia-smi to measure the GPU utilization.

[corres]

Allie 0.6

Thanks.
But it is only the source code of allie.
Where is the executable out of discord??

Thats the exe file. Source is on github.

Did you download it from this site?
Obviously not.
Or you do not know what is an executable.

[MMarco]

Hi,

I just released a new version of Allie: https://github.com/manyoso/allie/releases/tag/v0.6

You can find windows exe on Allie discord channel under #releases. This version will be playing in the upcoming TCEC Cup and is substantially stronger and faster than v0.5. Major parts of the code were rewritten and perft tests implemented along with NPS increases due to lockless threading redesign and lots of changes to TM.

Cheers!

Wow, Allie is blazingly fast!!

It achieves better results than Lc0 24.1 in test suite with the same net (LS 14.2), i.e. 123/200 on arasan20.epd on my 1660ti at 10s/move (Lc0 24.1 gets 115/200). I'm eager to see how that transposes into game play.

Thank you for the release!

Be careful here. Allie counts minimaxed nodes to compute NPS so it is apples to oranges comparison. To see which one is really faster you'd be better off looking at nvidia-smi to measure the GPU utilization.

Yes I came to the conclusion something wasn't the same as Lc0 with the node count!

I ran a 100 match game at 40/2min repeating + 1s/move against Stockfish 11 4CPU (on my laptop, ryzen 7 + gtx 1660ti. The Leela-ratio measure is around 8.1 for Stockfish 8 on one thread et Lc0 24.1 with LS 14.2, and around 1.65 with SF8 on 4 threads). Allie + LS 14.2 won the match +20, -13, =67:

Code: Select all

 
   # PLAYER                    : RATING    POINTS  PLAYED    (%)
   1 Allie v0.6 (LS 14.2)      : 3612.3      53.5     100   53.5%
   2 Stockfish 11 POPCNT 4CPU  : 3587.7      46.5     100   46.5%

White advantage = 24.69
Draw rate (equal opponents) = 50.00 %

Here's an interesting game in which Allie's queen seemed locked-in, and SF eventually captured it on move 49... but Allie nicely played and won... rather strangely in the end... in the final position (the GUI awarded the game to white at this point...) Allie needs a long time before clearly seeing a win after having had scores over +10 a few moves earlier. Interestingly, Komodo 10 is also having difficulties in the final position.

[pgn][Event "Allie-Stein 14.2 vs SF 11 4CPU"]
[Date "2020.04.22"]
[Round "55"]
[White "Allie v0.6"]
[Black "Stockfish 11 64 POPCNT"]
[Result "1-0"]
[ECO "A08"]
[Annotator "0.11;-0.33"]
[PlyCount "169"]
[TimeControl "40/120+1:40/120+1:0+4"]

{AMD Ryzen 7 3750H with Radeon Vega Mobile Gfx 2296 MHz W=12.4 plies; 31kN/s;
40 970 TBAs B=27.2 plies; 5 829kN/s; 2 516 208 TBAs} 1. Nf3 Nf6 2. g3 d5 3. Bg2
e6 4. d3 c5 {Both last book move} 5. O-O {0.11/12 9} Be7 {-0.33/28 11} 6. e4 {
0.11/13 5} Nc6 {-0.41/25 5 (0-0)} 7. Bf4 {0.15/16 5 (Na3)} O-O {-0.12/22 5} 8.
Ne5 {0.15/17 3} Nxe5 {-0.09/25 2} 9. Bxe5 {0.14/14 2} Nd7 {-0.24/24 2} 10. Bf4
{0.12/14 4} e5 {-0.09/26 8 (Bf6)} 11. Bc1 {0.19/12 10 (Bd2)} d4 {-0.28/25 1}
12. Nd2 {0.16/12 5 (f4)} Nb8 {-0.44/23 2 (b5)} 13. f4 {0.26/9 5} Nc6 {-0.47/23
2 (Bd7)} 14. f5 {0.31/10 6 (Qh5)} Bd7 {-0.36/25 8 (Bg5)} 15. Nf3 {0.36/8 7 (h4)
} f6 {-0.40/22 4 (b5)} 16. Rf2 {0.34/8 5 (g4)} Be8 {-0.53/21 1 (b5)} 17. Bf1 {
0.44/8 5 (g4)} Kh8 {0.00/25 18 (a5)} 18. Nh4 {0.53/7 6 (g4)} Bf7 {-0.31/20 4
(Rc8)} 19. Qf3 {0.57/8 6 (Be2)} Nb8 {0.43/23 21 (b5)} 20. g4 {1.37/7 4 (Qg4)}
c4 {0.45/23 7 (Nd7)} 21. Qh3 {1.97/9 6} Kg8 {0.61/22 2} 22. dxc4 {1.98/11 5}
Nd7 {1.27/22 7} 23. Bd3 {1.99/11 4 (Nf3)} Rc8 {1.18/22 2 (Qc7)} 24. b3 {
2.38/11 3} Nc5 {0.94/23 4} 25. Bd2 {2.34/13 4 (Rg2)} Qe8 {1.15/23 5 (b5)} 26.
Rg2 {2.73/9 5 (Nf3)} h6 {1.50/22 3 (b5)} 27. Nf3 {6.57/8 4 (b4)} Qc6 {1.75/21 8
} 28. Re1 {6.22/11 3} Rfd8 {1.76/20 3} 29. Qg3 {5.47/11 3} Kf8 {1.99/21 1 (b5)}
30. g5 {5.40/10 4} hxg5 {1.70/24 1} 31. h4 {5.22/12 3} Bh5 {1.81/25 3 (Ke8)}
32. hxg5 {5.70/11 3} Bxf3 {1.76/25 3} 33. Qxf3 {6.44/13 3} Ke8 {1.91/25 1} 34.
Qh5+ {6.33/13 3 (Rh2)} Kd7 {2.01/26 2} 35. Qf7 {7.17/12 2} Rh8 {2.15/28 6 (Rg8)
} 36. g6 {40.67/9 3 (Qxg7)} Nxd3 {0.00/25 1 (Rcf8)} 37. cxd3 {138.36/7 3} Rcf8
{0.00/27 1 (Qd6)} 38. Qxg7 {3.27/11 3 (Qd5+)} Rhg8 {0.00/31 3} 39. Qh6 {
1.94/14 3 (Qh7)} Rh8 {0.00/34 1} 40. Qg7 {2.10/15 1} Rhg8 {0.00/34 1} 41. Qh7 {
2.18/14 12} Rh8 {0.00/36 5} 42. Kf2 {2.39/17 6} Qa6 {0.00/36 5} 43. a4 {
2.47/14 7} Qb6 {0.00/37 4} 44. b4 {2.78/13 6 (Rb1)} Qd8 {1.28/32 12 (Qa6)} 45.
Qg7 {2.09/12 13 (b5)} Rhg8 {0.57/32 11} 46. Qh6 {2.09/15 5 (Qh7)} Rh8 {0.15/33
3} 47. Qg7 {2.17/14 4} Rhg8 {1.05/33 8} 48. Qh7 {2.21/15 4} Rh8 {1.24/34 3} 49.
b5 {2.41/14 3 (Rh1)} Rxh7 {0.08/30 3} 50. gxh7 {2.37/14 2} Rh8 {0.00/30 3} 51.
Rg7 {2.45/19 5} Kc8 {0.00/32 3} 52. Rh1 {2.51/13 4} b6 {0.00/34 3 (Qe8)} 53.
Kf3 {3.48/8 11 (Bh6)} Kb8 {1.54/32 8} 54. Rh6 {2.95/7 7 (Ke2)} Kc8 {1.60/32 3}
55. Rf7 {2.73/10 5 (Rh1)} Qe8 {1.61/32 6} 56. Rg7 {2.71/14 4} Qd8 {2.53/34 11}
57. Rh4 {2.82/10 3 (Rh1)} Qe8 {2.15/33 7} 58. Kf2 {2.82/10 8} Bd6 {2.00/33 4
(Qd8)} 59. Rxa7 {3.65/12 5 (Kf3)} Bc7 {3.12/29 3} 60. a5 {3.49/16 4} bxa5 {
2.91/30 3} 61. Bxa5 {3.45/15 2} Bxa5 {3.06/30 2} 62. Ra8+ {3.37/16 1} Kd7 {
2.95/33 4} 63. Rxe8 {3.27/14 1} Kxe8 {4.17/32 7} 64. c5 {3.36/12 1} Kd7 {
5.38/32 14 (Bb4)} 65. Rh6 {5.89/16 5} Bb4 {2.63/29 1} 66. c6+ {6.88/14 2} Kc7 {
5.29/30 2} 67. Kg3 {6.68/16 2} Bd2 {3.65/26 1} 68. Rxf6 {7.03/15 2} Rxh7 {
5.73/26 3} 69. Re6 {7.64/13 2} Bb4 {5.75/25 1} 70. f6 {6.76/12 6 (Kg4)} Kb6 {
2.69/22 2} 71. Kg4 {6.53/12 0} Kxb5 {3.69/23 2} 72. Kf5 {6.26/14 2} Ba3 {
2.77/24 4} 73. Kxe5 {6.07/12 2} Rh5+ {4.23/24 4} 74. Kf4 {5.95/10 3} Rc5 {
5.47/25 3 (Kb6)} 75. e5 {8.73/10 3} Rxc6 {5.80/23 1} 76. Re8 {11.94/10 2} Rc1 {
3.27/22 0} 77. e6 {23.58/9 2 (Ke4)} Re1 {2.32/23 1} 78. e7 {16.68/12 2 (Kf5)}
Bd6+ {2.30/26 1} 79. Kg4 {5.17/13 2 (Kg5)} Kb4 {9.55/28 2} 80. f7 {4.60/13 1}
Rg1+ {12.84/22 1} 81. Kh4 {2.98/16 11 (Kh3)} Rh1+ {2.25/24 1} 82. Kg5 {2.83/16
0 (Kg4)} Rg1+ {30.84/25 7} 83. Kh4 {2.44/18 8} Rh1+ {20.62/31 4} 84. Kg5 {
2.30/19 0 (Kg4)} Rg1+ {37.20/35 4} 85. Kh5 {1.80/18 9} 1-0[/pgn]

Black goes on with 49...Rxh7
[d]3q1r1r/pp1kb2Q/5pP1/1P2pP2/P1PpP3/3P4/3B1KR1/4R3 b - - 0 49

The final position:
[d]4R3/4PP2/3b4/7K/1k1p4/3P4/8/6r1 b - - 0 85

Code: Select all

Analysis by Komodo 10 64-bit:

85...Rh1+ 86.Kg6 Rg1+ 87.Kf5 Rf1+ 88.Ke4 Rf4+ 89.Kd5 Bxe7 90.Rxe7 Kc3 91.Ke5 Rf1 92.Ke4 Re1+ 93.Kf5 Rf1+ 94.Kg6 Kxd3 95.Kh6 Kc4 96.Rc7+ Kd5 97.Kg7 Rxf7+ 
  White is clearly better: +- (1.72)  Depth: 10   00:00:00
  The position is equal: = (0.00)  Depth: 56   00:02:26  1019MN, tb=31237385

Code: Select all

Analysis by Allie v0.6 (LS 14.2):

85...Rh1+ 86.Kg6 Rg1+ 87.Kh7 Rh1+ 88.Kg7 Rg1+ 89.Kh6 Rh1+ 90.Kg7 Rg1+ 91.Kh6 Rh1+ 92.Kg6 Rg1+ 93.Kh5 Rh1+ 94.Kg4 Rg1+ 95.Kh3 Bxe7 96.Rxe7 Rf1 97.Rc7 Kb5 98.Kh4 Kb6 99.Rd7 Kc6 100.Ra7 Kb6 101.Rd7 Kc6 102.Ra7 Kd5 103.Kh5 Ke6 104.Kg6 Rf3 105.Rc7 Rg3+ 106.Kh7 Rh3+ 
  The position is equal: = (0.08)   00:00:04  31N
  White is better: +/- (1.07)  Depth: 22/54   00:01:55  8799kN, tb=93488
  ...
  White is clearly better: +- (2.40)  Depth: 23/54   00:05:03  22802kN, tb=306214

Code: Select all

Analysis by Stockfish 11 64 POPCNT:

85...Rh1+ 86.Kg4 Rg1+ 87.Kh3 Bxe7 88.Rxe7 Rf1 89.Rc7 Kb5 90.Kg4 Kb6 91.Re7 Kc6 92.Kg5 Kd6 93.Ra7 Rf2 94.Kg6 Rg2+ 95.Kf6 Rf2+ 96.Kg7 Rg2+ 97.Kf8 Rg1 98.Ke8 Re1+ 99.Kd8 Rf1 100.Ra6+ Kc5 101.Ke7 Kb5 102.Ra8 Re1+ 103.Kf6 Rf1+ 104.Ke6 Re1+ 105.Kd5 Rf1 106.Rb8+ Ka4 107.Kxd4 Rf4+ 108.Kd5 Rf5+ 109.Ke6 Ra5 110.f8Q Rc5 
  White is clearly better: +- (2.28)  Depth: 8/14   00:00:01  26kN, tb=63
  White is clearly winning: +- (147.92 --)  Depth: 38/43   00:00:18  200MN, tb=1547667

[the_real_greco]

Allie Hype

[Master Om]

Allie 0.6

Thanks.
But it is only the source code of allie.
Where is the executable out of discord??

Thats the exe file. Source is on github.

Did you download it from this site?
Obviously not.
Or you do not know what is an executable.

Thats the link of executable from discord.

[corres]

Allie 0.6

Thanks.
But it is only the source code of allie.
Where is the executable out of discord??

Thats the exe file. Source is on github.

Did you download it from this site?
Obviously not.
Or you do not know what is an executable.

Thats the link of executable from discord.

Thanks. It is.

[corres]

Hi,
I just released a new version of Allie: https://github.com/manyoso/allie/releases/tag/v0.6
You can find windows exe on Allie discord channel under #releases. This version will be playing in the upcoming TCEC Cup and is substantially stronger and faster than v0.5. Major parts of the code were rewritten and perft tests implemented along with NPS increases due to lockless threading redesign and lots of changes to TM.
Cheers!

What is your propose what net fit well in Elo according to the new allie?