Fritz 17

[Fulvio]

As of now, it runs in four that I know of.

If the reference is to the various derivatives of lc0, they cannot be counted as different engines.

The links you cited do not apply here either.

The links explain why something that is a PART of a GPL program is considered derivative work (regardless of how it is connected or who the author is).
The copyright covers the result of human creativity.
In the case of a chess engine software that result is how the engines plays.
In the case of lc0 how it plays depend on many PARTS: the NN, the implementation of the MCTS search, of the multi-threading, how optimized is the code, ...
Modifying, replacing or adding a part (search's implementation, NN, etc...) must also be released under the GPL.
The viral nature of the GPL is well known and that reasoning cannot be dismissed without producing valid arguments (as you have already done twice).

Also, it might be worth pointing out the NN weights is not code, it is data.

It might be worth pointing out that the copyright is not related to the physical format. I can write an article in Word, convert it in PDF or print it on paper: still the same copyright.
However the thesis that a neural network is like a mathematical formula and cannot be copyrighted is also worth consideration.
I repeat: it is complex and I do not want to enter into endless polemics.
The point and the reason why I replayed to the previous post is that no one can say: shut up, I am obviously right and you are wrong.

[Albert Silver]

As of now, it runs in four that I know of.

If the reference is to the various derivatives of lc0, they cannot be counted as different engines.

The links you cited do not apply here either.

The links explain why something that is a PART of a GPL program is considered derivative work (regardless of how it is connected or who the author is).
The copyright covers the result of human creativity.
In the case of a chess engine software that result is how the engines plays.
In the case of lc0 how it plays depend on many PARTS: the NN, the implementation of the MCTS search, of the multi-threading, how optimized is the code, ...
Modifying, replacing or adding a part (search's implementation, NN, etc...) must also be released under the GPL.
The viral nature of the GPL is well known and that reasoning cannot be dismissed without producing valid arguments (as you have already done twice).

Also, it might be worth pointing out the NN weights is not code, it is data.

It might be worth pointing out that the copyright is not related to the physical format. I can write an article in Word, convert it in PDF or print it on paper: still the same copyright.
However the thesis that a neural network is like a mathematical formula and cannot be copyrighted is also worth consideration.
I repeat: it is complex and I do not want to enter into endless polemics.
The point and the reason why I replayed to the previous post is that no one can say: shut up, I am obviously right and you are wrong.

I am pretty sure no one said shut up, and that is entirely in your imagination. In the case of Lc0, the weights have nothing to do with the GPL, nor are they a part of lc0, they merely run in it. Just as you can use the freely offered weights of others in lc0, you can use privately developed and commercial ones as well.

It is interesting that you consider Scorpio to be a derivative of Lc0. I strongly suspect that Daniel Shawul would take exception to this description.

[AndrewGrant]

It is interesting that you consider Scorpio to be a derivative of Lc0. I strongly suspect that Daniel Shawul would take exception to this description.

He is not alone in his thoughts.

[Dann Corbit]

There are some among the GPL crowd that believe if they give you a pan, then you must give away everything you cook in it.
Even if you eat the food, you belong to the GPL, as does your house and eventually the entire planet.

They eventually gave up on that model, since (early on) they tried to apply it as a license to things like bison (if bison generates a grammar, then the things using the grammar are also GPL was attempted or something along those lines).
But then they discovered that nobody wanted to use their stuff that had *THAT* viral of an impact, so they stopped trying to do that.

I think it could be argued that the weights are an output of the tool, given a collection of user input.
But I do not think it reasonable that the weights belong to the GPL, or all outputs of all GPL tools belong to the GPL.
People who use MySQL might be unhappy about that, since all of their customer data would become GPL.

[dkappe]

It is interesting that you consider Scorpio to be a derivative of Lc0. I strongly suspect that Daniel Shawul would take exception to this description.

He is not alone in his thoughts.

Daniel or the original poster?

If the latter, then you’ve produced a real head scratcher. Is leela chess, as the first to colonize DeepMind’s intellectual territory with c++ (there were earlier alpha zero implementations in python), the only original engine, and all others — whether derived from the alpha zero pseudo code or some other source — are henceforth clones? By that logic, all ab chess engines are just clones of Shannon’s original endgame solver. Whether that’s ridiculous is left as an exercise for the reader.

[dkappe]

They eventually gave up on that model, since (early on) they tried to apply it as a license to things like bison (if bison generates a grammar, then the things using the grammar are also GPL was attempted or something along those lines).
But then they discovered that nobody wanted to use their stuff that had *THAT* viral of an impact, so they stopped trying to do that.

I think there was some boilerplate code included in the Bison output which was GPL’d. Yeah, why use bison when yacc was available?

Anyhow, the difference between the output of a program and the program is a long established distinction. As I mentioned previously, even where the output is itself an executable program (see gcc and FSF’s own discussion on the topic), the output is not covered by the GPL (that would have to be a different kind of license).

[AndrewGrant]

It is interesting that you consider Scorpio to be a derivative of Lc0. I strongly suspect that Daniel Shawul would take exception to this description.

He is not alone in his thoughts.

Daniel or the original poster?

If the latter, then you’ve produced a real head scratcher. Is leela chess, as the first to colonize DeepMind’s intellectual territory with c++ (there were earlier alpha zero implementations in python), the only original engine, and all others — whether derived from the alpha zero pseudo code or some other source — are henceforth clones? By that logic, all ab chess engines are just clones of Shannon’s original endgame solver. Whether that’s ridiculous is left as an exercise for the reader.

Any engine using in full or in part the code written by the Leela team is a derivative.

If Scorpio is not a Leela clone, what other explanation do you have for a mid-tier engine with an unreadable code base and a rampant history of crashes, bugs, and other play limiting or preventing issues, suddenly shooting up almost overnight to become one of the most powerful engines featured in TCEC / CCC. I won't make you guess the explanation -- the explanation is that anyone can piggy back off the work of the Leela team and with minimal effort build an engine that is a couple hundred elo weaker than Leela.

[John Conway]

Any engine using in full or in part the code written by the Leela team is a derivative.

If Scorpio is not a Leela clone, what other explanation do you have for a mid-tier engine with an unreadable code base and a rampant history of crashes, bugs, and other play limiting or preventing issues, suddenly shooting up almost overnight to become one of the most powerful engines featured in TCEC / CCC. I won't make you guess the explanation -- the explanation is that anyone can piggy back off the work of the Leela team and with minimal effort build an engine that is a couple hundred elo weaker than Leela.

Are you saying Scorpio is a derivative or are you saying it is a clone or are you saying a derivative is a clone?

[dkappe]

If Scorpio is not a Leela clone, what other explanation do you have for a mid-tier engine with an unreadable code base and a rampant history of crashes, bugs, and other play limiting or preventing issues, suddenly shooting up almost overnight to become one of the most powerful engines featured in TCEC / CCC. I won't make you guess the explanation -- the explanation is that anyone can piggy back off the work of the Leela team and with minimal effort build an engine that is a couple hundred elo weaker than Leela.

I hate to hurt your feelings, but I can beat ethereal with 1 core on a dual 2080 with my python engine. It's puct implementation is only 85 lines of python. I'm sure you'll argue its a clone.

Face it, the hardest thing about the nn engines is the training of the nets. The MCTS code is dead simple, and if you use a highly optimized library like TensorRT, the gpu code isn't hard at all.

Green jealousy is such an unbecoming color for you. I think you owe Daniel an apology for your unfounded insinuations.

[dannyb]

I hate to hurt your feelings, but I can beat ethereal with 1 core on a dual 2080 with my python engine. It's puct implementation is only 85 lines of python. I'm sure you'll argue its a clone.

Face it, the hardest thing about the nn engines is the training of the nets. The MCTS code is dead simple, and if you use a highly optimized library like TensorRT, the gpu code isn't hard at all.

Green jealousy is such an unbecoming color for you. I think you owe Daniel an apology for your unfounded insinuations.

is your engine available for testing?