New J.A Compiles

[Jim Ablett]

https://github.com/Nonlinear2/Bread-Engine
Bread 3.0.0 by Nonlinear

Bread 3.0.0 64 JA

Windows/Linux & Android builds

Proton link:
https://drive.proton.me/urls/R2SJ0J29YG#QRFGFPFn93fo

Smash link:
https://fromsmash.com/Bread300JA

Jim.

[Jim Ablett]

https://github.com/dan-homan/Leaf
Leaf dev.5382 by Dan Homan

Leaf dev.5382 64 JA

Windows & Linux profiled builds

Proton link:
https://drive.proton.me/urls/H2NN5Z6JJ4#Ok4SlLhm6v6F

Smash link:
https://fromsmash.com/LeafDev5382JA

Jim.

[dchoman]

Hi Jim,

Thanks for your interest in Leaf! Does the version you compile use the classical evaluation (EXchess's old evaluation) or the reference stockfish .nnue file (downloaded separately from them)? If it is a stockfish .nnue file, I really didn't intend that for competitive play... it is intended as a reference as I work to develop Leaf's TD learning from self-play starting with old EXchess psqt + randomized weights and biases.

If it is the the classic eval, no problem, but if it is the stockfish 15.1 nnue file, then I would ask people not to release results from play. It is not representative of what Leaf will be able to do with its own learned net... at least not until I've spent a lot more time on the TDleaf learning.

In terms of my TDleaf learning progress so far, I am still optimizing, but my best improvement after 30,000 games (a tiny amount) is a neural net that is within about 400 elo (below) the old EXchess eval. So I've got quite a ways to go.

- Dan

[Mark Tang]

Hi Jim, can you compile the Integral Dev again? The previous download link is outdated. (though it hadn't any improvement since 2026)

[Jim Ablett]

Hi Jim,

Thanks for your interest in Leaf! Does the version you compile use the classical evaluation (EXchess's old evaluation) or the reference stockfish .nnue file (downloaded separately from them)? If it is a stockfish .nnue file, I really didn't intend that for competitive play... it is intended as a reference as I work to develop Leaf's TD learning from self-play starting with old EXchess psqt + randomized weights and biases.

If it is the the classic eval, no problem, but if it is the stockfish 15.1 nnue file, then I would ask people not to release results from play. It is not representative of what Leaf will be able to do with its own learned net... at least not until I've spent a lot more time on the TDleaf learning.

In terms of my TDleaf learning progress so far, I am still optimizing, but my best improvement after 30,000 games (a tiny amount) is a neural net that is within about 400 elo (below) the old EXchess eval. So I've got quite a ways to go.

- Dan

Hi Dan,

It's using the Stockfish 15.1 nnue file with TD learning enabled. I will compile classic eval version and post updated links soon.

Jim.

[Jim Ablett]

Hi Jim, can you compile the Integral Dev again? The previous download link is outdated. (though it hadn't any improvement since 2026)

Hi Mark,

download from my repo >

https://drive.proton.me/urls/9TRAQG8SXC#3J2pemAZ0DpV

Jim.

[Sylwy]

THANK YOU, again !

viewtopic.php?p=990518#p990518

[Jim Ablett]

https://github.com/dan-homan/Leaf
Leaf dev.5382 by Dan Homan

Leaf dev.5382 64 JA

Windows & Linux profiled builds

Proton link:
https://drive.proton.me/urls/H2NN5Z6JJ4#Ok4SlLhm6v6F

Smash link:
https://fromsmash.com/LeafDev5382JA

Jim.

Updated links with classical eval builds added.

Proton link:
https://drive.proton.me/urls/H2NN5Z6JJ4#Ok4SlLhm6v6F

Smash link:
https://fromsmash.com/LeafDev538264JA

Jim.

[dchoman]

Thanks Jim!

I also just saw your github note about the fixes for Windows compiles. I'll check that out and try to integrate in the not to distant future. The TDleaf stuff is rapidly evolving as I test and tune, so the code base is changing rapidly right now.

- Dan

[Mark Tang]

Hi Jim, can you compile the Integral Dev again? The previous download link is outdated. (though it hadn't any improvement since 2026)

Hi Mark,

download from my repo >

https://drive.proton.me/urls/9TRAQG8SXC#3J2pemAZ0DpV

Jim.

I'm sorry, Jim.I can't visit the drive.proton.me site for some reason. Can you share another fromsmash link please.