Kayra 1.1

[mehmet123]

Kayra 1.1 is ready:
https://www.mediafire.com/file/uwhfmsc0 ... .1.7z/file


1 Kayra 1.1 bmi2 : 2408 750 (+ 19,=729,- 2), 51.1 %
2 Kayra 1.0 bmi2 : 2400 750 (+ 2,=729,- 19), 48.9 %

Game Conditions: Cutechess Gui, 1 Core (Core-i7 9750h), 10 sec + 0.5 sec TC, Balsa 5 Moves Opening Book, 64 Mb Hash, Ponder Off
https://www.mediafire.com/file/6t8lrddt ... 1.pgn/file



1 Kayra 1.1 bmi2 : 2404 550 (+ 5,=545,- 0), 50.5 %
2 Kayra 1.0 bmi2 : 2400 550 (+ 0,=545,- 5), 49.5 %

Game Conditions: Cutechess Gui, 1 Core (Core-i7 9750h), 30 sec + 0.5 sec TC, Balsa 5 Moves Opening Book, 64 Mb Hash, Ponder Off
https://www.mediafire.com/file/py773qfz ... 2.pgn/file



1 Kayra 1.1 bmi2 (Non NNUE) : 2808 200 (+165,= 35,- 0), 91.2 %
2 Kayra 1.0 bmi2 (Non NNUE) : 2400 200 (+ 0,= 35,-165), 8.8 %

Game Conditions: Cutechess Gui, 1 Core (Core-i7 9750h), 10 sec + 0.5 sec TC, Balsa 5 Moves Opening Book, 64 Mb Hash, Ponder Off
Kayra 1.0 is very weak without NNUE but Kayra 1.1 is strong enough without NNUE.
https://www.mediafire.com/file/upc1qh19 ... 3.pgn/file

I focused on the search and evaluation part, mostly the search part at developing Kayra 1.1.

[DrEinstein]

psqt cpp now no more randomly initialized, just replaced by the original SF psqt.cpp. Thus some 100 ELO improvement to 1.0 version.

[Damir]

Here is alternative link to those that have problems loading with mediafire:

https://www.dropbox.com/s/9ruaj6xr1kyfy ... .1.7z?dl=0

PS: thanks a lot for this new updated version Mehmet.

[sarona]

It is definitely the psqt.cpp from latest Stockfish development version that I have (stockfish_21122211), but not a verbatim copy. He has made a few changes to the values. Seven modifications in total. No change in the bishop table.

I was skeptical at first - like many others. But my test results with Kayra 1.0 avx2 were very good. Even against Ceres and Lc0 (RTX 3080 Ti) Will definitely take a look at the new 1.1

[DrEinstein]

I can't compare files on my phone, so I just looked at the two files. These seven chances will probably worsen classical eval of v1.1 less than the 700 changes did in v1.0

[Dicaste]

It's really nice i don't really get how it's improved but it really felt like it so i tested with chessbase marathon suite.
It shares the first place with Fat Titz right now.

[Eduard]

I don't know what the author did, but on Playchess with 12 cores I achieve greater search depths than people with 32 to 64 cores. I know search depth isn't everything, but I've also tested the engine in analysis mode. There are also very good results.

[Sopel]

Looks strictly worse than stockfish. What's the point? Does it have any useful features?

[Eduard]

Can you explain this in simple terms to people like me?

I am testing different engines on servers. It is important to me that the engines play with Ponder ON and not less than 5m + 0s. Kayra certainly doesn't play any worse than Stockfish. I'm 800 games in a row undefeated with Kayra. Oh, I haven't even lost a game and I play a lot. Kayra had to play all kinds of openings, from Pirc to King's Gambit to Volga Gambit.

I saw a Kayra 1.0 BUG in a kibitz game. With me Kayra is flawless. Kayra 1.1 is better than 1.0 and I'm sure Stockfish dev. does not play better (I also tested that).

[Sopel]

Can you explain this in simple terms to people like me?

I am testing different engines on servers. It is important to me that the engines play with Ponder ON and not less than 5m + 0s. Kayra certainly doesn't play any worse than Stockfish. I'm 800 games in a row undefeated with Kayra. Oh, I haven't even lost a game and I play a lot. Kayra had to play all kinds of openings, from Pirc to King's Gambit to Volga Gambit.

I saw a Kayra 1.0 BUG in a kibitz game. With me Kayra is flawless. Kayra 1.1 is better than 1.0 and I'm sure Stockfish dev. does not play better (I also tested that).

Hmm? It obviously plays worse

Code: Select all

Score of stockfish_dev vs kayra_1_1: 1210 - 901 - 1889 [0.539]
...      stockfish_dev playing White: 1189 - 3 - 808  [0.796] 2000
...      stockfish_dev playing Black: 21 - 898 - 1081  [0.281] 2000
...      White vs Black: 2087 - 24 - 1889  [0.758] 4000
Elo difference: 26.9 +/- 7.8, LOS: 100.0 %, DrawRatio: 47.2 %
4000 of 10000 games finished.