A hybrid Stockfish NNUE?

[Milos]

BTW, is there a parameter to change "eval\nn.bin" ?

Reply to self, yes there is you stupid.

https://github.com/nodchip/Stockfish/bl ... on.cpp#L86

[Rebel]

NN : sergio-1907 TC 40m/20s

Code: Select all

Score of henk vs sf11: 726 - 369 - 905  [0.589] 2000
...      henk playing White: 447 - 121 - 432  [0.663] 1000
...      henk playing Black: 279 - 248 - 473  [0.515] 1000
...      White vs Black: 695 - 400 - 905  [0.574] 2000
Elo difference: 62.7 +/- 11.3, LOS: 100.0 %, DrawRatio: 45.3 %
Finished match

So far:

Code: Select all

https://github.com/nodchip/Stockfish
https://www.comp.nus.edu.sg/~sergio-v/nnue/

Stockfish 11        Nodes/second   : 1.664.010
SF-NNUE (popcount)  Nodes/second   :   796.674

henk.bin      54.7%   2020-07-19
sergio-1432   56.1%   2020-07-21 15:56
sergio-1907   58.9%   2020-07-21 19:14

Where does this end?

[Werner]

hi Ed,
is it possible to test sfnnue with MEA?
I think this engine needs ucinewgame to start the net?

[Rebel]

hi Ed,
is it possible to test sfnnue with MEA?
I think this engine needs ucinewgame to start the net?

I had no problem with the popcount version.

[Chessqueen]

I wonder if it would be possible to make a hybrid version of Stockfish NNUE, with the search running on CPU and the evaluation on GPU. If so, would it be possible to use larger networks?

Here it is, and you can use a Medium NET to Catch Him