Not logged inCSS-Forum
Forum CSS-Online Help Search Login
CSS-Shop Impressum Datenschutz
Up Topic Hauptforen / CSS-Forum / LC0 Parameter Optimierung
- By Lothar Jung Date 2023-12-25 13:57 Upvotes 1
**Hardware:** AMD Ryzen 9 5900X CPU @ 3.70GHz, 12 cores, 64GB, gpu RTX 3080, 10GB
**Gauntlet:** SF-dev vs. LC0 192x15b nets
**Network IDs:** 791556
**LC0-version:** v0.30.0
**LC0-options:** MiniBatchSize=92 MoveLeftThresold=0.0 or 0.8 default and Contempt=0 or 70, ContemptMode=play, WDLCalibrationElo=3200, WDLContemptAttenuation=1, WDLEvalObjectivity=0, WDLDrawRateReference=0.47, ScoreType=WDL_mu
**SF-version:** SF-dev 22011322
**SF-options:** 8 Threads, 1024MB Hash
**Software:** Cutechess-CLI
**Time control:** LC0: 5,000 nodes/move, SF-dev: 125,000 nodes/move
**Book:** Chad's 8-ply book, book-ply8-unifen-Q-0.0-0.25.pgn and book-ply8-unifen-Q-0.25-0.40.pgn (10K openings in each book), order=sequential
**Tablebases:** 6-men
**Adjudication:** -draw movenumber=50 movecount=5 score=8 -resign movecount=5 score=1000
**Ordo:** -a 0  -A "" -W -D -U "0,1,2,3,4,5,6,7,8,9,10" -s 2000 -n 1
**Comment:** The aim of the test is to compare v0.30.0 default, MLT=0.8, with MLT=0.0 and Contempt usage. The order is considered sure with 99% CFS (a bit more than 3.5 Elo). CFS matrix and detailed results in 7z file.
Up Topic Hauptforen / CSS-Forum / LC0 Parameter Optimierung

Powered by mwForum 2.29.3 © 1999-2014 Markus Wichitill