Kayra 1.2:

[mehmet123]

Kayra 1.3:
https://www.mediafire.com/file/7a1z0ppz ... .3.7z/file

Kayra 1.3 sse3:
https://www.mediafire.com/file/pz239484 ... e3.7z/file


Kayra 1.3 compiled by Alex Morales for MacOS M1
Binary for Apple Silicon M1:
https://onedrive.live.com/?cid=B0F79706 ... 27&o=OneUp
Source and NNUE (Not embedded):
https://www.mediafire.com/file/7a1z0ppz ... .3.7z/file


Kayra 1.3 compiled by "Archimedes" for Android:
https://pixeldrain.com/u/PckZ82g1

[Eduard]

What exactly is new in Kayra 1.3? What is different from Stockfish dev.? Unfortunately I don't see anything - I compared it with SF 280322 with my test set.

[mehmet123]

What exactly is new in Kayra 1.3? What is different from Stockfish dev.? Unfortunately I don't see anything - I compared it with SF 280322 with my test set.

I talked about this in the 2nd post of this topic.

[mehmet123]

Kayra 1.4:
https://www.mediafire.com/file/rfz05eox ... .4.7z/file

[mehmet123]

Kayra 1.5:
https://www.mediafire.com/file/e49qs3ta ... .5.7z/file

[Eduard]

Kayra 1.5:
https://www.mediafire.com/file/e49qs3ta ... .5.7z/file

What is new and better?
It seems that the engine uses the old scaling 1092 with the new network.

[mehmet123]

Kayra 1.5:
https://www.mediafire.com/file/e49qs3ta ... .5.7z/file

What is new and better?
It seems that the engine uses the old scaling 1092 with the new network.

There isn't any radical changes only minor changes for Kayra 1.5. I'm also working on a Stockfish derivative chess engine with radical changes especially at search codes, but this chess engine isn't strong enough to beat Stockfish yet.

[Eduard]

The new networks at Stockfish are now working with a new scale:

https://talkchess.com/forum3/viewtopic. ... 1f#p929968

Why does Kayra 1.5 use the old scaling?

[mehmet123]

The new networks at Stockfish are now working with a new scale:

https://talkchess.com/forum3/viewtopic. ... 1f#p929968

Why does Kayra 1.5 use the old scaling?

Actually, I was using a different valuation from these two in the development process of Kayra 1.5 with new NNUE net. But since I couldn't get the result I wanted in one of my tests, I canceled a few changes in the evaluation part. I choose the previous formula as I did not have the opportunity to test whether the new formula had a positive effect on Kayra.

[Eduard]

I see that you are only deleting things, but not implementing anything new. What you say today, you said half a year ago. Like I said, I don't see anything different than the Stockfish. I don't see any other style. At least a Blue Marlin has a style of its own.

In the analysis mode I saw that some positions were solved a little faster, but a few seconds or not is nothing. In fact, there are a few positions where Kayra rates higher (wrongly) than Stockfish, but it may just be different because of scaling. There is no reason for me to use Kayra instead of Stockfish.