My strongest NNUE net (yet)

[JohnS]

Trying to understand this but not much luck so far.

I run sf-nnue run (halfkp-256x2-32-32) x64-bmi2 from the command line and then used
go depth 20

The engine just stopped after some time without any output, is this expected.

Suggestion - allow nn.bin to be in same directory as the exe, this seems simpler to me for using the engine.

What happens if nn.bin is missing, is it just normal stockfish.

Put the net in "eval" folder under the folder of the engine, or change the EvalDir parameter.
Also you need to send an
isready
command
for the engine to load the eval file.

Thanks! Using isready made it work. nn.bin was in the eval folder, just suggested maybe easier for users to default location to engine directory.

[thunderstruck]

What does option SkipLoadingEval do?

Kind regards.

Antonio

[Raphexon]

What does option SkipLoadingEval do?

Kind regards.

Antonio

The binaries we use have the learning function still in them.
Skiploadingeval is used when we want to train a net from scratch.

If you use it to play a game it plays like Stockfish without an evaluation function.

You can also download binaries without the learn fuction:

https://github.com/nodchip/Stockfish/releases

Those won't have the "skiploadingeval" option.
They're smaller in size, but otherwise play identical.

[Raphexon]

Problably new best net, also from Gekkehenker:

Code: Select all

40+0.3 
   # PLAYER       : RATING  ERROR   POINTS  PLAYED    (%)
   1 stnnuegk2    : 2860.6    3.9    930.5    1814   51.3%
   2 stnnuegk1    : 2851.4    3.9    883.5    1814   48.7%

Here the net:
https://cdn.discordapp.com/attachments/ ... 24_june.7z

Thanks to Gekkehenker and also to the community that is helping to improve this engine bit a bit!

I've updated the net again:


Tested +25 +/-8 elo vs stnnuegk2 after 2000 games.
TC:10s+0.1s. 1 thread. Slowmover 50 for both

[cdani]

I've updated the net again:


Tested +25 +/-8 elo vs stnnuegk2 after 2000 games.
TC:10s+0.1s. 1 thread. Slowmover 50 for both

Thanks.
I have done two tests:

Code: Select all

40 seconds + 0.3 against previous version:
   # PLAYER       : RATING  ERROR   POINTS  PLAYED    (%)
   1 stnnuegk3    : 2863.5    4.9    650.5    1248   52.1%
   2 stnnuegk2    : 2848.5    4.9    597.5    1248   47.9%

200 seconds + 1 against same version of Stockfish that I used previously:
   # PLAYER       : RATING  ERROR   POINTS  PLAYED    (%)
   1 st200617     : 2882.2    6.7    377.0     657   57.4%
   2 stnnuegk3    : 2829.8    6.7    280.0     657   42.6%

So maybe is 50 to Stockfish.