Sayuri https://github.com/CGLemon/Sayuri

"Let's ROCK!
Sayuri is a GTP-compliant go engine which supports variable komi and board size. Strongly inspired by Leela Zero and Kata Go. Based on Deep Convolutional Neural Network, Monte Carlo Tree Search and other powerful technics."

This is interesting, but I'd like some more information. Are those openings "trained"? When does the opening end? Are networks compatible with KG's? Can I train the network beyond the opening?

Thanks, take care

_________________
一碁一会

The program is only available on Ubuntu/MacOS. I have windows. I downloaded the network, looked at it, it's very different from the katago networks, I haven't tried it. as I understand it, the project is new, I wonder what will happen!

Release v0.1.0
https://github.com/CGLemon/Sayuri/releases

Features

Support Sabaki and GoGui analysis mode.
Support handicap game.
Support variable komi and board size (from 7x7 to 19x19).
Lock-free SMP MCTS.
Acceleration by multi-core processor and multi-Nvidia GPU.
Predict the current side winrate and draw-rate.
Predict the current side score lead and death strings.
Reuse the sub-tree.
Tromp-Taylor rules with forbidding suicide move.

Release v0.1.1
https://github.com/CGLemon/Sayuri/releases

Main Updating

Improve the MCTS performance.
Add prefix id for GTP command.
Fix ladder bug.
Add the new options, --symm-pruning, --kgs-hint.
Show the tree complexity and KL divergence.
Extend some GTP commands, sayuri-analyze, sayuri-genmove_analyze and others.
Change the default thinking time (to 10 sec per move).
Use a random sample input data read. This helps improve the spread of games in the shuffle buffer. The new network is stronger than before.
Clean some code and add more comments.

Sayuri(B) v.s. Pachi(W)
https://www.youtube.com/watch?v=82UclNrXGxg

Sayuri(B) v.s. Pachi(W)

Release v0.2.0
https://github.com/CGLemon/Sayuri/releases

Release v0.3.0
https://github.com/CGLemon/Sayuri/releases

windows version is expected next month
https://github.com/CGLemon/Sayuri/issues/2

windows version https://github.com/CGLemon/Sayuri/issues/2
Zenith Go 7 - Sayuri, Zen gave up on move 205 (Zenith Go 7 9dan, Sayuri -t 3 -w uce14-swa-1700k-v2.bin.txt), Zen spent a little more time

Release v0.5.0 (Furaregai Girl) https://github.com/CGLemon/Sayuri/releases/tag/v0.5.0

Network for 2023 CGF Open https://github.com/CGLemon/Sayuri/releases/

Release v0.6.0 (Hachi to Misemono) https://github.com/CGLemon/Sayuri/releases/

"This version is more efficiently than KataGo g104 in the early training phase! KataGo g170 is unknown but it looks like that Sayuri is still better than g170 in the early training phase. Thanks for Gumbel AlphaZero and other KataGo methods. We cat train a high amateur level network on a single GPU device in a just one week..."

Release v0.6.1 https://github.com/CGLemon/Sayuri/releases/tag/v0.6.1
The current version achieves superhuman level. Sayuri finally suppresses LZ157 and SAI project.

KataGo - Sayuri
KataGo b18-s770 maxVisits = 100
Sayuri Sayuri-v0.6.1-eigen-windows-x64, zero-swa-special-2035k.bin.txt, -t 10 -b 1 -p 10000 --cache-memory-mib 2048

leelaz - Sayuri
leelaz p 1000, b40_287
Sayuri p 1000

Sayuri - leelaz
leelaz p 1000, b40_287
Sayuri p 1000

if an error occurs when starting the Windows version, add the missing files .dll from here https://github.com/mstorsjo/llvm-mingw/ ... t-i686.zip

Sayuri - Zenith Go
Zenith Go 9 dan
Sayuri p 100