Senpai 3.0 by Fabien LETOUZEY is available!

[Pablo Vazquez]

Hi Fabien,

I can add each time a new version on my site.

Jim like it to compare the own compiles with the compiles by the programmers.

In my opinion better ...
To add your own compiles in the Senpai download file and Jim can add his own compiles
on his Repro directory, if Jim do that.

Important is to create a new version number, like Senpai 3.0.1 for an example (of course
your decision).

Your own compiles for the RC1 = release version are 10% slower only.

Hint to Senpai 2.0:
Jim's Senpai 2.0 compiles are 40% faster but I didn't put his Senpai 2 compilations with
the other Senpai files.

Best
Frank

Yes Frank 3.0.1 and compile myself, I will do that.

Fabien.

Have you considered using a public github repo? It would help with the releases and also to track changes between versions/commits. For example, I've noticed that you removed 7th rank pawn push extension in 3.0.1 in addition to the other two changes.

[Pvt. Ryan]

Hi. Does this engine have a build that uses HCE instead of NNUE? Cheers

[Xann]

Have you considered using a public github repo? It would help with the releases and also to track changes between versions/commits. For example, I've noticed that you removed 7th rank pawn push extension in 3.0.1 in addition to the other two changes.

Hi Pablo,

Public commits are not at all how I roll; there are multiple aspects.

The development version has a lot more code, and is organised in 'lab' modules that generate game examples, learn evaluation weights, test changes etc; they use the engine module directly. There is also a lot of unused chess code for future experiments. I consider all of this unfit for a UCI release.

Then, there is another big aspect. To me, GitHub and other popular places are more an 'AI-training farm' than a public utility (hence the value). You know how it works, they take your upload as 'consent' to use it for their models (which are not free to use). When I want to distribute something, I look for enthusiasts.

Specifically for the extension you mentioned, Ryan proposed to remove it this week. Since I had to make a change (Move Overhead) that affects Elo somewhat, I applied his proposal (after testing of course). I was focused on the release, and wasn't testing chess changes at the same time.

Fabien.

[Xann]

Hi. Does this engine have a build that uses HCE instead of NNUE? Cheers

First of all, Senpai does not use a neural network.

Secondly, I question the HCE vs NN categories. I am claiming earlier in this thread, and the link below that there is a 'grey area' inbetween. I hope that experts will give their opinion.

viewtopic.php?p=981661#p981661
See my answer to 'Will it use NN evaluation or hand-tuned eval?'

To come back to your question, I think that the community already considers Senpai's evaluation as HCE, apparently following this definition: it's not a neural network.

Fabien.

[Pvt. Ryan]

Thanks for that. I just thought the 'eval' file had something to do with NN. The engine won't work if 'eval' is not in the same folder.

[Xann]

I just thought the 'eval' file had something to do with NN.

Yeah, exactly. I think you felt that grey area already!

[Ryan Benitez]

Hi. Does this engine have a build that uses HCE instead of NNUE? Cheers

First of all, Senpai does not use a neural network.

Secondly, I question the HCE vs NN categories. I am claiming earlier in this thread, and the link below that there is a 'grey area' inbetween. I hope that experts will give their opinion.

viewtopic.php?p=981661#p981661
See my answer to 'Will it use NN evaluation or hand-tuned eval?'

To come back to your question, I think that the community already considers Senpai's evaluation as HCE, apparently following this definition: it's not a neural network.

Fabien.

I think you have shown that there are multiple things between HCE and NN.