Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

change NNUE file ? #1

Open
tissatussa opened this issue Dec 27, 2024 · 4 comments
Open

change NNUE file ? #1

tissatussa opened this issue Dec 27, 2024 · 4 comments

Comments

@tissatussa
Copy link

I can compile your source on Xubuntu 22.04 and i wonder : can i change the NN file and compile ? Will your search / eval logic work with any another NN ? (being the same format)

@bitArtisan1
Copy link
Owner

Yes, it should work with another .nnue file, make sure to change the file name to your new .nnue in my init.cpp file.

@tissatussa
Copy link
Author

@bitArtisan1 tnx, i will experiment with different compiles .. in README you state the "NNUE Evaluation" is "Features=HalfKP(Friend)[41024->256x2],Network=AffineTransform1<-32" which means the good-old Stockfish 20 Mb ones .. lots of them exist, i once found "red neuronal", from some "Fischermann" or so, he trained many NN and maybe still does .. is your NNUE file unique ?

@bitArtisan1
Copy link
Owner

bitArtisan1 commented Dec 27, 2024

No problem! No, it's not unique—it's a pre-trained network from the Stockfish repositories, and I haven't made any modifications to it. You can find it in the Stockfish NNUE database here.
Good luck with your experiments!

@tissatussa
Copy link
Author

@bitArtisan1 yes, that SF db is well-known .. in the meantime i compiled Wize with several such 20 Mb NNs and let those engines play against eachother and also other NN-based engines of equal strength or having a (own) small NN (say upto 5 Mb) and see how they perform .. i used this position :

e4-c6-d4-d6

  1. e4 c6 2. d4 d6!

Black plays the Caro-Kann Defense and should do 2...d5 but i'm experimenting with this d6-setup in my own games. SF often plays d6-d5! soon, though wasting a tempo .. pawns can not go back ! White should be able to win such game, but with many pieces on the board it's all about good moves!

soon i will post the game list here - have to prepare - but here's one nice game : your default NN wins with Black against Wize-with-Mida-NN (see https://github.com/GiacomoPorpiglia/Mida ) .. but Mida itself beats default-Wize, so i keep wondering what happens to the strength of any NN, all trained differently, in your Wize ?

wize-demo-ApronusDiagram1735598274

[Event "engine vs engine"]
[Site "Holland"]
[Date "2024.12.27"]
[Round "?"]
[White "Wize v1.0 NNUE mida"]
[Black "Wize v1.0 NNUE default"]
[Result "0-1"]
[FEN "rnbqkbnr/pp2pppp/2pp4/8/3PP3/8/PPP2PPP/RNBQKBNR w KQkq - 0 1"]
[GameDuration "00:13:34"]
[PlyCount "114"]
[SetUp "1"]
[TimeControl "300+3"]

{ https://www.lichess.org/r4UEkWWe } 1. Nf3 {+1.33/11 10s} Nf6 {-1.29/12 14s} 2. Nc3 {+1.65/11 8.3s}
g6 {-1.30/12 13s} 3. h3 {+1.70/11 9.0s} Bg7 {-1.16/12 22s} 4. Bd3 {+1.53/10 17s}
Nbd7 {-0.99/11 13s} 5. Be3 {+1.23/9 8.8s} e5 {-1.00/10 10s}
6. dxe5 {+1.12/10 15s} Nxe5 {-0.94/11 10s} 7. Nxe5 {+1.15/10 11s}
dxe5 {-0.94/11 11s} 8. a4 {+1.15/9 23s} Be6 {-0.94/10 21s} 9. a5 {+0.83/10 13s}
O-O {-0.84/11 17s} 10. Be2 {+0.92/9 7.5s} Qc7 {-0.44/10 7.5s}
11. O-O {+0.63/9 6.8s} b5 {-0.30/11 13s} 12. b4 {+0.26/9 13s} Rad8 {+0.50/9 14s}
13. Qc1 {+0.39/8 7.5s} Rd4 {+0.63/8 6.8s} 14. Rb1 {+0.19/8 15s}
Rfd8 {+0.35/8 6.5s} 15. a6 {+0.31/8 11s} Bf8 {+0.85/8 7.1s}
16. Bxb5 {-0.63/8 9.1s} cxb5 {+2.95/10 9.1s} 17. Nxb5 {-1.49/11 7.4s}
Qe7 {+2.72/10 5.6s} 18. Nxa7 {-1.68/10 6.8s} Qxa7 {+4.62/11 5.5s}
19. c3 {-5.64/10 11s} Nxe4 {+6.11/11 4.6s} 20. cxd4 {-6.17/10 4.8s}
exd4 {+7.37/12 11s} 21. Bh6 {-7.01/10 15s} Nc3 {+8.57/11 6.2s}
22. Rb2 {-8.63/11 13s} Qxa6 {+9.08/11 6.0s} 23. Bxf8 {-9.06/10 5.4s}
Rxf8 {+9.37/11 6.0s} 24. Qd2 {-8.31/9 4.4s} Bc4 {+10.02/12 6.5s}
25. Re1 {-8.73/11 4.2s} Qf6 {+10.95/12 6.6s} 26. Rc2 {-9.68/11 7.0s}
Kg7 {+11.05/11 13s} 27. f4 {-9.49/11 4.5s} Ra8 {+11.00/11 4.5s}
28. Kh2 {-9.01/10 5.0s} Rd8 {+10.69/11 5.0s} 29. Re5 {-9.16/11 5.9s}
Rd5 {+11.65/12 6.9s} 30. Re8 {-10.54/10 3.9s} Rf5 {+11.15/11 5.4s}
31. g3 {-9.91/11 3.9s} h5 {+11.39/11 6.8s} 32. h4 {-10.67/10 4.3s}
Rb5 {+12.07/10 11s} 33. Rc1 {-10.33/10 9.9s} Rxb4 {+11.63/10 3.6s}
34. f5 {-10.70/10 8.7s} gxf5 {+12.20/10 4.2s} 35. Rce1 {-11.94/9 4.3s}
f4 {+14.67/10 4.2s} 36. Qf2 {-13.49/9 3.7s} Nd5 {+18.45/10 7.0s}
37. gxf4 {-14.85/9 5.6s} Qxf4+ {+20.74/11 3.5s} 38. Qxf4 {-14.77/11 3.2s}
Nxf4 {+20.88/13 4.0s} 39. Rd8 {-14.82/12 4.5s} Ne6 {+20.58/12 3.2s}
40. Rd6 {-14.80/11 4.0s} Kf6 {+21.16/12 4.7s} 41. Kg3 {-15.65/12 5.5s}
Kf5 {+21.97/12 2.9s} 42. Rd7 {-16.49/12 4.5s} Rb3+ {+21.53/12 3.1s}
43. Kh2 {-16.71/12 9.8s} f6 {+22.34/12 4.6s} 44. Rc1 {-16.37/11 4.4s}
Rc3 {+22.33/12 3.4s} 45. Re1 {-18.68/11 3.3s} Kf4 {+22.95/13 5.8s}
46. Kh1 {-19.40/12 5.2s} f5 {+24.21/12 3.8s} 47. Rd1 {-19.61/12 3.2s}
Be2 {+26.48/11 2.9s} 48. Ra1 {-18.76/11 3.0s} Kg3 {+26.48/11 2.8s}
49. Kg1 {-21.45/11 2.8s} Nf4 {+27.01/11 3.9s} 50. Rg7+ {-24.58/12 3.4s}
Bg4 {+27.46/10 2.7s} 51. Rga7 {-26.97/13 6.4s} Rc2 {+27.76/13 6.4s}
52. R7a2 {-M12/14 5.0s} Be2 {+M11/17 2.7s} 53. Kh1 {-M10/20 2.8s}
Bf3+ {+M9/39 2.7s} 54. Kg1 {-M8/65 8.5s} Bg2 {+M7/65 8.8s}
55. Ra3+ {-M6/24 2.8s} Rc3 {+M5/24 3.9s} 56. Rxc3+ {-M4/44 8.1s}
dxc3 {+M3/128 0.029s} 57. Rd1 {-M2/128 0.027s}
Ne2# {+M1/128 0.015s, Black mates} 0-1

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants