Stockfish: Our lawsuit against ChessBase

[ChickenLogic]

Albert, you do realize that
a) nodchip ported nnue to SF as a a way of giving back
and
b) Gekkehenker and me were the first ever to train NNUE nets, push it into this forum and were also the first to beat SF dev?

If it wasn't for Gekke and me NNUE probably would potentially have been buried for years since the original claim of the "mysterious Japanese programmer" of being able to gain > 90 elo for SF was ridiculed. Nodchip didn't show that NNUE would be so awesome for chess. It was Gekke, Jjosh, Sergio and me.

Yes, the computer chess community owes a lot to nodchip and the original creator of NNUE - I NEVER disputed this. But without those proving its worth (which are exclusively people doing it for and with SF) it would've been buried in open sight for potentially years.

Btw, it was also me who asked Cscuile to open the NNUE channel in the then unofficial SF discord. It was us who spread the curiosity about NNUE. And it was Sopel who actually innovated upon it. And many other's who optimized it beyond anything I'd ever hoped for.

Get in line man.

[connor_mcmonigle]

And you know what? If it wasn't for SF your beloved clones wouldn't exist at all. Komodo, Fire, Rubi, Ethereal and basically any other engine now using NNUE would be more than 150 elo weaker.

NNUE didn't originate with SF nor was it even created for it.

I think it is fair to state that NNUE originated with the nodchip implementation using Stockfish in the chess-engine /clone content. I am also sure you understood the intention of the quote perfectly well and I truly wonder what reaction you expect from pointing out that NNUE existed outside the chess world before it got implemented in Chess-engines (just like NNs and deep learning existed before it was used in the stronger chess engines).

... but their implementations are not taken from SF.

Outside of Komodo, they most definitely are.

[dkappe]

I think it is fair to state that NNUE originated with the nodchip implementation using Stockfish in the chess-engine /clone content. I am also sure you understood the intention of the quote perfectly well and I truly wonder what reaction you expect from pointing out that NNUE existed outside the chess world before it got implemented in Chess-engines (just like NNs and deep learning existed before it was used in the stronger chess engines).

Hmmm, nnue was first used in a shogi engine as referenced in the original paper. It’s source is available on github. https://github.com/tttak/tnk-

It was only later ported to a stockfish search based shogi engine and from there used in a fork of stockfish for chess.

Now I’m sure you’ll be able to use this in your moral arithmetic spreadsheet to figure out who owes what to whom. Please report back when you’ve figured it out.

[connor_mcmonigle]

I think it is fair to state that NNUE originated with the nodchip implementation using Stockfish in the chess-engine /clone content. I am also sure you understood the intention of the quote perfectly well and I truly wonder what reaction you expect from pointing out that NNUE existed outside the chess world before it got implemented in Chess-engines (just like NNs and deep learning existed before it was used in the stronger chess engines).

Hmmm, nnue was first used in a shogi engine as referenced in the original paper. It’s source is available on github. https://github.com/tttak/tnk-

It was only later ported to a stockfish search based shogi engine and from there used in a fork of stockfish for chess.

Now I’m sure you’ll be able to use this in your moral arithmetic spreadsheet to figure out who owes what to whom. Please report back when you’ve figured it out.

I really fail to see how this relates to the original discussion...

The original point was that "openness" is good as openly sharing ideas has resulted in the incredible progress in computer chess we observe. It's entirely irrelevant whether Stockfish is responsible for NNUE or not. Had the original Shogi developers kept their observations closed, chess engine development would be far behind where it is today. To argue otherwise is ridiculous.

Were it not for individuals sharing their ideas openly, we'd be far behind where we are today. It doesn't matter who those individuals are. That's why the Eman clone's failure to comply with the GPL is incredibly disrespectful to all those who have contributed to Stockfish and worked to advance computer chess. And now we have fools in this forum trying to paint the situation as if Stockfish contributors are evil for requesting that the creators of illegal clones respect Stockfish's license.

[AndrewGrant]

Note: Using customized NNUe like dkappe's Night Nurse CF and Dark Horse CF or Zeus.nn on all supported clones is perfectly allowed by GPLv3. It's not clear if a GPLv3 engine can lock NNUE offering it only in the commercial version.

How does GPL apply to neural network models?
https://opensource.stackexchange.com/qu ... ork-models

I read this as a (weak) jab at Ethereal, since I feel like you have no issue with FF2 based on your use of illegal clones? Or am I paranoid.

[AlexChess]

Note: Using customized NNUe like dkappe's Night Nurse CF and Dark Horse CF or Zeus.nn on all supported clones is perfectly allowed by GPLv3. It's not clear if a GPLv3 engine can lock NNUE offering it only in the commercial version.

How does GPL apply to neural network models?
https://opensource.stackexchange.com/qu ... ork-models

I read this as a (weak) jab at Ethereal, since I feel like you have no issue with FF2 based on your use of illegal clones? Or am I paranoid.

You are paranoid. I was talking about the use of NNUE FatFritz2_vX.bin (they still sell the Fat Fritz 2.0 SE version and I'd like to test the new NNUE with FF on Stockfish 14 GPLv3 and other compatible engines) since the private use without sharing them is perfectly allowed. But sorry I really don't like to interact with you.

Regards, Alex

[Raphexon]

I think it is fair to state that NNUE originated with the nodchip implementation using Stockfish in the chess-engine /clone content. I am also sure you understood the intention of the quote perfectly well and I truly wonder what reaction you expect from pointing out that NNUE existed outside the chess world before it got implemented in Chess-engines (just like NNs and deep learning existed before it was used in the stronger chess engines).

Hmmm, nnue was first used in a shogi engine as referenced in the original paper. It’s source is available on github. https://github.com/tttak/tnk-

It was only later ported to a stockfish search based shogi engine and from there used in a fork of stockfish for chess.

Now I’m sure you’ll be able to use this in your moral arithmetic spreadsheet to figure out who owes what to whom. Please report back when you’ve figured it out.

Did you even look at the github repo you linked?

[dkappe]

Did you even look at the github repo you linked?

Ah, my bad. Thanks for the heads up. Apparently this version is derived from https://github.com/yaneurao/YaneuraOu which was in turn derived from stockfish search. That’s what I get for misreading the Chess Programming Wiki. It’s clear from the original paper that the earliest nnue engines were stockfish derived. https://dev.exherbo.org/~alip/doc/nnue_en.pdf

[AlexChess]

Did you even look at the github repo you linked?

Ah, my bad. Thanks for the heads up. Apparently this version is derived from https://github.com/yaneurao/YaneuraOu which was in turn derived from stockfish search. That’s what I get for misreading the Chess Programming Wiki. It’s clear from the original paper that the earliest nnue engines were stockfish derived. https://dev.exherbo.org/~alip/doc/nnue_en.pdf

Hi dkappe, just a little OT question: which is stronger between Toga III and Toga 1.1 + Dark Rose NNUE? I get controversial results using Arena or BanksiaGui. Night Nurse CF and Dark Horse CF are much stronger, anyway. Thank you for these strong alternatives to SF test NNUEs!

Note my Night Nurse customized logo against the myth MikeB's Black Diamond.



DIVERSITY from an already strong open source base is the right way to improve all engines for free, in the interest of the chess community! Fork and multiply engines, but release the modified source code!

[dkappe]

Hi dkappe, just a little OT question: which is stronger between Toga III and Toga 1.1 + Dark Rose NNUE? I get controversial results using Arena or BanksiaGui. Night Nurse CF and Dark Horse CF are much stronger, anyway. Thank you for these strong alternatives to SF test NNUEs!

Dark Toga 1.1 is much stronger, though the next version of both should be closer.