Commercial Release of Ethereal 13.00 (NNUE) for AVX2 Systems

[AndrewGrant]

New Stockfish NNUE nets ara trained on Leela games. Isn't it time to give up a conservative stance on Ethereal NNUE nets.

"Ethereal 13 (NNUE) includes two NNUE Networks for evaluation. These Networks are the second of their kind, boasting themselves as the only other high level NNUEs not derived from, trained on, nor duplicated from the works of the Stockfish team."

I'll stick to that. I'm the last person doing it on my own.

And food for thought: Stockfish folks training on Leela data is far less concerning to me than other engines reusing the entire SF pipeline.

[mehmet123]

"Ethereal 13 (NNUE) includes two NNUE Networks for evaluation. These Networks are the second of their kind, boasting themselves as the only other high level NNUEs not derived from, trained on, nor duplicated from the works of the Stockfish team."
I'll stick to that. I'm the last person doing it on my own.
And food for thought: Stockfish folks training on Leela data is far less concerning to me than other engines reusing the entire SF pipeline.

We respect your decision. Is there any chance of us seeing an unofficial version with the feature to use large networks?

[AndrewGrant]

"Ethereal 13 (NNUE) includes two NNUE Networks for evaluation. These Networks are the second of their kind, boasting themselves as the only other high level NNUEs not derived from, trained on, nor duplicated from the works of the Stockfish team."
I'll stick to that. I'm the last person doing it on my own.
And food for thought: Stockfish folks training on Leela data is far less concerning to me than other engines reusing the entire SF pipeline.

We respect your decision. Is there any chance of us seeing an unofficial version with the feature to use large networks?

Large networks? Define large. Mine are the original halfkp size of 2x256 => 32 =>32 => 1. You cannot plug any one else's Network into Ethereal, since they are not compatible.

[willmorton]

"Ethereal 13 (NNUE) includes two NNUE Networks for evaluation. These Networks are the second of their kind, boasting themselves as the only other high level NNUEs not derived from, trained on, nor duplicated from the works of the Stockfish team."

I'll stick to that. I'm the last person doing it on my own.

so all the others who have said their nets were trained on their engines' games lied?

[AndrewGrant]

"Ethereal 13 (NNUE) includes two NNUE Networks for evaluation. These Networks are the second of their kind, boasting themselves as the only other high level NNUEs not derived from, trained on, nor duplicated from the works of the Stockfish team."

I'll stick to that. I'm the last person doing it on my own.

so now all the others who have said their nets were trained on their engines' games lied?

No no, not at all. Its all of the above, not just the data. So the only to NOT be trained on other data while also NOT using the Stockfish pipeline while also NOT using the Stockfish implementation while also BEING high level. I have no doubt that the other networks are trained on the data their authors claim. Sorry if there was any confusion there.

[AlexChess]

"Ethereal 13 (NNUE) includes two NNUE Networks for evaluation. These Networks are the second of their kind, boasting themselves as the only other high level NNUEs not derived from, trained on, nor duplicated from the works of the Stockfish team."

I'll stick to that. I'm the last person doing it on my own.

so now all the others who have said their nets were trained on their engines' games lied?

No no, not at all. Its all of the above, not just the data. So the only to NOT be trained on other data while also NOT using the Stockfish pipeline while also NOT using the Stockfish implementation while also BEING high level. I have no doubt that the other networks are trained on the data their authors claim. Sorry if there was any confusion there.

Hi!

Which is the size of the commercial NNUEs? Is it possible to train a custom net for the 13.00 github version, since it's different from Stockfish?

Regards,
AlexChess

[AndrewGrant]

"Ethereal 13 (NNUE) includes two NNUE Networks for evaluation. These Networks are the second of their kind, boasting themselves as the only other high level NNUEs not derived from, trained on, nor duplicated from the works of the Stockfish team."

I'll stick to that. I'm the last person doing it on my own.

so now all the others who have said their nets were trained on their engines' games lied?

No no, not at all. Its all of the above, not just the data. So the only to NOT be trained on other data while also NOT using the Stockfish pipeline while also NOT using the Stockfish implementation while also BEING high level. I have no doubt that the other networks are trained on the data their authors claim. Sorry if there was any confusion there.

Which is the size of the commercial NNUEs? Is it possible to train a custom net for the 13.00 github version, since it's different from Stockfish?

Regards,
AlexChess

Fat Fritz 2 was using 2x512 => 32 => 32 => 1 I believe? I cannot say for sure, but something like double the weights on the input layer.
Komodo ... I do not know the specifics. And if I did, I don't think they have said much publically, so I would not disclose that.

Yes, one could train their own Network for Ethereal. Such a person would need to take a peek at the NNUE code to figure out the format, as its different than Stockfish in order to exploit some AVX benefits for floats. You would then need to produce training code as well, as I have not released my training code written in C, nor do I plan to at any point. Although some other chess users do indeed have access to the training code. Halogen primarily, since Halogen was the guinea pig for my NN learning efforts.

[AlexChess]

Thank you!
I hope that other skilled programers will make their own NNUEs for it. I like to compare different versions

Regards, AlexChess

[kranium]

"Ethereal 13 (NNUE) includes two NNUE Networks for evaluation. These Networks are the second of their kind, boasting themselves as the only other high level NNUEs not derived from, trained on, nor duplicated from the works of the Stockfish team."
I'll stick to that. I'm the last person doing it on my own.
And food for thought: Stockfish folks training on Leela data is far less concerning to me than other engines reusing the entire SF pipeline.

We respect your decision. Is there any chance of us seeing an unofficial version with the feature to use large networks?

"We" ?
Mehmet, I don't believe you speak for everyone...certainly not for me.

Andrew, your posts are simply repugnant.
You and Ethereal owe a huge debt to Stockfish and all that came before., yet you act like you've invented all this technology by copying SF and rewriting a few python modules.

I I strongly suggest you step down from the fake ivory pedestal you've installed yourself on...
take a deep breath and inhale in large dose of much needed humility.

[booot]

No no, not at all. Its all of the above, not just the data. So the only to NOT be trained on other data while also NOT using the Stockfish pipeline while also NOT using the Stockfish implementation while also BEING high level.

I am trying to implement it in Booot just now. With Delphi+Python+Keras+Tensorflow. Not using any Stockfish pipeline. I am still not sure i can pass this way by my own, but today i am much closer to finish then i was 1 month ago before i started .