Ratsu - UCI chess engines
Author: Eetu Rantala
Ratsu 2.0.0 what's new?
This release elevates Ratsu to the next level by introducing an NNUE (Efficiently Updatable Neural Network) to replace the hand-crafted evaluation. Estimated CCRL blitz Elo 2643 +/- 40.
NNUE features
768 → 2x 128 → 1 network trained with PyTorch using lichess-big3-resolved dataset (7M FENs + game results)
incrementally updated accumulators with perspective
integer quantisation, int16 weights and accumulators
+331 Elo over Ratsu 1.2.0
Results of ratsu-200 vs ratsu-120 (8+0.08, NULL, NULL, 8moves_v3.pgn):
Elo: 331.80 +/- 27.29, nElo: 454.27 +/- 21.49
LOS: 100.00 %, DrawRatio: 13.75 %, PairsRatio: 53.12
Games: 1004, Wins: 807, Losses: 62, Draws: 135, Points: 874.5 (87.10 %)
Ptnml(0-2): [2, 6, 69, 95, 330], WL/DD Ratio: 3.06
Binaries for Windows and Linux are available. Use highest possible x86-64 v-level for best performance.
Comments
Post a Comment