Stockfish NN release (NNUE)

[Raphexon]

This is my current net:

Trained on depth 8 data.
Validated with depth 19 data.


For use rename it to nn.bin and replace the original nn.bin in the "eval" folder.

For use with:

stockfish.nnue-learn-use-blas.halfkp_256x2-32-32.exe
And
stockfish.nnue.halfkp_256x2-32-32.exe
https://github.com/nodchip/Stockfish/re ... 2020-06-09

I'm currently generating new training data with it, will be able to release something new in a few days I hope.

[marsell]

May I ask how to use these engines? They don't run in the gui arena.
ty

[Raphexon]

May I ask how to use these engines? They don't run in the gui arena.
ty

What CPU do you have.
The binaries I linked require BMI2 support (Haswell or newer, Ryzen or newer if AMD)
Are you installing the halfkp binaries, and did you put the 20mb nn in the eval folder? (And renamed it to nn.bin)

[marsell]

my cpu is a intel i7 4790k running on windows 10 Pro
I tried to install the halfkp binaries with the 20 mb nn, renamed as nn.bin.
But it looks like it's still not working

[RubiChess]

May I ask how to use these engines? They don't run in the gui arena.
ty

What CPU do you have.
The binaries I linked require BMI2 support (Haswell or newer, Ryzen or newer if AMD)
Are you installing the halfkp binaries, and did you put the 20mb nn in the eval folder? (And renamed it to nn.bin)

I also tried your halfkp binary stockfish.nnue.halfkp_256x2-32-32.exe
Have replaced eval\nn.bin with the candidate one.
My system: Intel(R) Core(TM) i5-5257U CPU @ 2.70GHz
Crashes at "go" command with 0xc0000005 at offset 00000000000365d4 (if that helps).

[Raphexon]

my cpu is a intel i7 4790k running on windows 10 Pro
I tried to install the halfkp binaries with the 20 mb nn, renamed as nn.bin.
But it looks like it's still not working

Does this work?

[marsell]

great, thanks, now it works.

[Joerg Oster]

This is my current net:

Trained on depth 8 data.
Validated with depth 19 data.


For use rename it to nn.bin and replace the original nn.bin in the "eval" folder.

For use with:

stockfish.nnue-learn-use-blas.halfkp_256x2-32-32.exe
And
stockfish.nnue.halfkp_256x2-32-32.exe
https://github.com/nodchip/Stockfish/re ... 2020-06-09

I'm currently generating new training data with it, will be able to release something new in a few days I hope.

Are you still training without draws?
Probably not a good idea, imho.

In my copy of SF-NNUE https://github.com/joergoster/Stockfish-NNUE I enabled to use draw results, as well.
Hopefully, I did everything right.
When compiling from source it will use halfkp_256x2-32-32 architecture by default.

I also think that the initial learning rate (eta = 1.0) might be too high.

[cucumber]

This is my current net:

Trained on depth 8 data.
Validated with depth 19 data.


For use rename it to nn.bin and replace the original nn.bin in the "eval" folder.

For use with:

stockfish.nnue-learn-use-blas.halfkp_256x2-32-32.exe
And
stockfish.nnue.halfkp_256x2-32-32.exe
https://github.com/nodchip/Stockfish/re ... 2020-06-09

I'm currently generating new training data with it, will be able to release something new in a few days I hope.

Are you still training without draws?
Probably not a good idea, imho.

In my copy of SF-NNUE https://github.com/joergoster/Stockfish-NNUE I enabled to use draw results, as well.
Hopefully, I did everything right.
When compiling from source it will use halfkp_256x2-32-32 architecture by default.

I also think that the initial learning rate (eta = 1.0) might be too high.

Can you share all of your binaries (gensfen, half-kp, half-kp learn, etc.,) on Google Drive or as GitHub releases? I know it's a lot to ask for, but I'd really appreciate it.

[Joerg Oster]

This is my current net:

Trained on depth 8 data.
Validated with depth 19 data.


For use rename it to nn.bin and replace the original nn.bin in the "eval" folder.

For use with:

stockfish.nnue-learn-use-blas.halfkp_256x2-32-32.exe
And
stockfish.nnue.halfkp_256x2-32-32.exe
https://github.com/nodchip/Stockfish/re ... 2020-06-09

I'm currently generating new training data with it, will be able to release something new in a few days I hope.

Are you still training without draws?
Probably not a good idea, imho.

In my copy of SF-NNUE https://github.com/joergoster/Stockfish-NNUE I enabled to use draw results, as well.
Hopefully, I did everything right.
When compiling from source it will use halfkp_256x2-32-32 architecture by default.

I also think that the initial learning rate (eta = 1.0) might be too high.

Can you share all of your binaries (gensfen, half-kp, half-kp learn, etc.,) on Google Drive or as GitHub releases? I know it's a lot to ask for, but I'd really appreciate it.

Try these: https://www.file-upload.net/download-14 ... 20.7z.html