New J.A Compiles

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

User avatar
Jim Ablett
Posts: 2350
Joined: Fri Jul 14, 2006 7:56 am
Location: London, England
Full name: Jim Ablett

Re: New J.A Compiles

Post by Jim Ablett »

Image
https://github.com/olithink/OliThink/tree/nnue-ja
Olithink 5.11.9 NNUE-Hybrid by Oliver Brausch

An update to my experimental nnue branch.
Add hybrid nnue mode.
Added UCI configuration option to address The NNUE "Halo Effect". The problem is Networks might look at a
wild position and see a massive, non-material advantage (like a powerful attack) and completely ignore the fact that
it has to give up a lot of pieces to get it. It believes its attack is so strong that the sacrifice is worth it, even
if, deep down, it might not be, spotting subtle tactical and positional advantages, but can get a little too aggressive
or "speculative." So I want to mix the material score with the NNUE score to create a good hybrid solution.

This modification implements a hybrid evaluation that mixes the NNUE score with the material score to mitigate the NNUE
"Halo Effect". The material score is blended in at a ratio of 1/8, meaning the final score is (7/8 * Score_NNUE) + (1/8 * Score_Material)
Olithink 5.11.9 NNUE-Hybrid 64 JA

Windows/Linux & Android builds

Proton link:
https://drive.proton.me/urls/BAJDYHFWPM#LMLTNQu1bRwe

Smash link:
https://fromsmash.com/Olithink5119NNUEHYBRIDJA

Jim.
User avatar
Jim Ablett
Posts: 2350
Joined: Fri Jul 14, 2006 7:56 am
Location: London, England
Full name: Jim Ablett

Re: New J.A Compiles

Post by Jim Ablett »

Image
https://github.com/matthiaslang/jackych ... /tag/25.09
JackyChess 25.09 by Matthias Lang

JackyChess 25.09 64 JA

Compiled from Java to profiled native Windows executable.

Proton link:
https://drive.proton.me/urls/182MPPT630#HUrmY0oDEVJ2

Smash link:
https://fromsmash.com/JackyChess250964JA

Jim.
User avatar
Jim Ablett
Posts: 2350
Joined: Fri Jul 14, 2006 7:56 am
Location: London, England
Full name: Jim Ablett

Re: New J.A Compiles

Post by Jim Ablett »

Image
https://github.com/Bobingstern/Tarnished
Tarnished 4.0 Dev.1137 by Anik Patel

Tarnished 4.0 Dev.1137 64 JA

Windows/Linux & Android builds

Proton link:
https://drive.proton.me/urls/TSXEF29018#1eIuXIacIia2

Smash link:
https://fromsmash.com/Tarnished40Dev1137JA

Jim.
User avatar
Jim Ablett
Posts: 2350
Joined: Fri Jul 14, 2006 7:56 am
Location: London, England
Full name: Jim Ablett

Re: New J.A Compiles

Post by Jim Ablett »

Jim Ablett wrote: Wed Nov 19, 2025 3:42 pm Image
https://github.com/olithink/OliThink/tree/nnue-ja
Olithink 5.11.9 NNUE-Hybrid by Oliver Brausch

An update to my experimental nnue branch.
Add hybrid nnue mode.
Added UCI configuration option to address The NNUE "Halo Effect". The problem is Networks might look at a
wild position and see a massive, non-material advantage (like a powerful attack) and completely ignore the fact that
it has to give up a lot of pieces to get it. It believes its attack is so strong that the sacrifice is worth it, even
if, deep down, it might not be, spotting subtle tactical and positional advantages, but can get a little too aggressive
or "speculative." So I want to mix the material score with the NNUE score to create a good hybrid solution.

This modification implements a hybrid evaluation that mixes the NNUE score with the material score to mitigate the NNUE
"Halo Effect". The material score is blended in at a ratio of 1/8, meaning the final score is (7/8 * Score_NNUE) + (1/8 * Score_Material)
Olithink 5.11.9 NNUE-Hybrid 64 JA

Windows/Linux & Android builds

Proton link:
https://drive.proton.me/urls/BAJDYHFWPM#LMLTNQu1bRwe

Smash link:
https://fromsmash.com/Olithink5119NNUEHYBRIDJA

Jim.
Small update that fixes analysis mode.

Proton link: (same as above)
https://drive.proton.me/urls/BAJDYHFWPM#LMLTNQu1bRwe

Smash link:
https://fromsmash.com/Olithink-51119NNUEHYBRIDJA

Jim.
Frank Quisinsky
Posts: 7114
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: New J.A Compiles

Post by Frank Quisinsky »

Hi Jim,

thank you!
Your Tarnished is newer as the official dev from the programmer (251106 dev).

:-)

My overview is more complicated than I anticipated before starting this file.
I like that, means to create from a small chaos a bigger chaos.

My chaos file:
https://www.amateurschach.de/main/_engines.htm

354 engine *.eng files added now in my Shredder 12 GUI.
Today I added: Donna, Rodin xb and Drosophila ... I can't get enough from this stuff.
I must be completely crazy, but that is not new for me since I started with a ChessChallenger 7 for around 48 years.

Soon, I will start with a new test idea. Tomorrow I will test the following hardware:
CTONE M2 MINI PC, Core i9 13900HK (14K/20T, bis zu 5.4GHz), 32GB RAM+1TB PCIe 4.0 SSD Micro Desktop Computer mit Iris Xe 96EU| WiFi 6E| BT 5.2|2.5GbE LAN Support Tripe Display|8K

Performance Cores = 6 (max. 5.4 Ghz)
Efficient Cores = 8 (max. 4.1 Ghz)

The idea is to use the 8 Efficent Cores only on two on such machines for engine testing with a significantly lower power consumption per machine, compare to my 16 cores AMD systems. In the morning I give the first system electricity. Not more, to many work on my engine overview. Hope the hardware idea is good. Perhaps I can use all 14 cores with 4.0 Ghz per core. I don't know at the moment.

If to complicated I will go back to my ChessChallenger 7 for the rest of my life.
The topic is CL4 vs. CL6 ... Level 4 ~20 minutes per move vs. Level 6 ~6 minutes per move.
Possible with DM PB keys in combination after every move.
A secret feature and a good tipp for the others in the same situation.
You have to know that not only level 5 (analyze mode) can use a mate in 3 moves.
Also very rarely the level 4 ... very important I think.

:-)

Best
Frank