Chess engine: Lizard 11

-

 


Lizard - UCI chess engine and NNUE  Author: Liam McGuire
Rating Chess Engines Diary CEDR=3658

v.11:
This release is the first one to use self-play data to train networks. The network from version 10.5 was used to train ~200m positions, and subsequent networks used that data to train and produce more data. The milestone I was targeting here was to be nearly equal to 10.5 since the elo hit from removing Lc0 data was significant.

For the sake of my sanity (and because .NET can't/doesn't autovectorize NNUE code), this requires at least AVX2 to be performant, but should still function without it.

Picking a binary:
Start with the non-aot binary, and pick ...-512 if your processor has Avx512.

If you want a slightly slower but more compact binary, start with ...-aot-v4 and use the first one that doesn't crash:

v4 enables Avx512,
v3 enables Avx256,
v2 is a fallback for older CPU's
8+0.08:
Lizard 11.0 vs Lizard 10.5
Elo   | 18.54 +- 6.98 (95%)
Conf  | 8.0+0.08s Threads=1 Hash=32MB
Games | N: 3002 W: 873 L: 713 D: 1416
Penta | [13, 311, 728, 401, 48]
http://somelizard.pythonanywhere.com/test/1730/
60+0.6:
Lizard 11.0 vs Lizard 10.5
Elo   | 2.43 +- 10.92 (95%)
Conf  | 60.0+0.60s Threads=1 Hash=256MB
Games | N: 1000 W: 242 L: 235 D: 523
Penta | [1, 121, 249, 128, 1]
http://somelizard.pythonanywhere.com/test/1728/
DFRC:
Lizard 11.0 vs Lizard 10.5
Elo   | 22.42 +- 14.20 (95%)
Conf  | 8.0+0.08s Threads=1 Hash=32MB
Games | N: 2002 W: 770 L: 641 D: 591
Penta | [166, 128, 295, 235, 177]
http://somelizard.pythonanywhere.com/test/1732/

 Lizard 10.5 - results:
OpponentScore+-Games
Obsidian 13.027.5/64-964 Games
Alexandria 7.0.025.5/53-253 Games
Chess-System-Tal-225.5/47+447 Games
PlentyChess 3.0.019.5/47-847 Games
Berserk 1316.5/38-538 Games
RubiChess 2024011215.5/35-435 Games
Ethereal 14.2515/35-535 Games
Peacekeeper 3.0119.5/33+633 Games
Caissa 1.20 JA14/29-129 Games
Uralochka 3.41a JA14/28+028 Games
Caissa 1.1912/26-226 Games
Motor 0.7.011/26-426 Games
JigSaw 5.97/26-1226 Games
Viridithas 14.0.014/25+325 Games
Clover 7.1 JA13.5/25+225 Games
Yuliana 5.09/23-523 Games
Stockfish 16.17.5/20-520 Games
Nalwald 1917/18+1618 Games
Patricia 316/18+1418 Games
Patricia 3.116/18+1418 Games
Viridithas 13.0.011.5/18+518 Games



Comments

Post a Comment

Popular posts from this blog

Torch-4 UCI v 1.01 - new version chess engine

-
Image
Torch-4 UCI v 1.01 what's new? Bugfix regarding relative node.exe and engine path. If torch.exe previously did not work in some GUIs, it now should. Download the source code as a zip file, extract it and use either torch.exe or torch.bat as the engine. github:https://github.com/Felixk45/Torch-4-UCI-chess-engine/releases Torch-4 UCI v 1.01 download from page
Read more

Reckless 0.10.0 - new version super strong chess engine

-
Image
  Reckless , a Rust-based UCI chess engine. Author: Arseniy Surkov    Rating Chess Engines Diary CEDR=3743 Reckless is the new star among chess engines. On the new Chess Engines Diary CEDR ranking list, Reckless is ranked 2th with a rating of 3743 . Testing of the new engine version is coming soon. Reckless 0.9.0 what's new? Reckless 0.9.0 is ready! You can download pre-built binaries for Windows, Linux, and macOS from the GitHub Releases page. Thanks to everyone who contributed through code, testing, hardware support, or any other effort. This release came together because of your help. Update highlights NNUE The neuron network has been updated to a modern threat-based architecture. It models the interactions between pieces, where one piece attacks or defends another, and extends the existing king-bucketed piece-square-based inputs with additional features. Multithreading and NUMA support NUMA (Non-Uniform Memory Access) support has been added to Reckless. Work...
Read more

Reckless 0.9.0 - new version super strong chess engine

-
Image
  Reckless , a Rust-based UCI chess engine. Author: Arseniy Surkov    Rating Chess Engines Diary CEDR=3743 Reckless is the new star among chess engines. On the new Chess Engines Diary CEDR ranking list, Reckless is ranked 2th with a rating of 3743 . Testing of the new engine version is coming soon. Reckless 0.9.0 what's new? Reckless 0.9.0 is ready! You can download pre-built binaries for Windows, Linux, and macOS from the GitHub Releases page. Thanks to everyone who contributed through code, testing, hardware support, or any other effort. This release came together because of your help. Update highlights NNUE The neuron network has been updated to a modern threat-based architecture. It models the interactions between pieces, where one piece attacks or defends another, and extends the existing king-bucketed piece-square-based inputs with additional features. Multithreading and NUMA support NUMA (Non-Uniform Memory Access) support has been added to Reckless. Work...
Read more

We already have the new Fritz 20 program - what does it look like and how does it work?

-
Image
The purchase process for the download version on the ChessBase website took only 5 minutes. Downloading the files from the cloud was even quicker. The installation itself, which involved five large files, took maybe 7 minutes, including restarting the system and entering the serial number I received upon purchase. The start image for Fritz 20 has a strange mosaic of colors; perhaps I'll get used to it. As you can see in the second screenshot, the program interface hasn't changed much; only the icons are different (we've been using Fritz 17 until now). It looks nice, but it's no major revolution. The next two screenshots show engine management and the installation of the UCI chess engine (here, Cataphract 1.1). As you can see, there are no changes. Screenshot no. 5 shows the analysis of some position by Fritz 20 engines. The changes are minimal - it is still very legible. Finally, on screen 6 we have the engine tournament table generated by Fritz 20 . I think it is sti...
Read more