Repository of networks for the Berserk chess engine
Jay Honnold
dae8f890ce
Training from bexp42-e1000, bexp44-e1000 was re-trained using old Berserk 9 data of DFRC fens. The resulting network was the sorted for sparse matmul chunking. Elo | 91.44 +- 12.79 (95%) SPRT | 10.0+0.10s Threads=1 Hash=8MB LLR | 2.91 (-2.25, 2.89) [0.00, 2.00] Games | N: 1446 W: 550 L: 178 D: 718 Penta | [2, 84, 264, 286, 87] http://chess.grantnet.us/test/34159/ Elo | 58.38 +- 10.28 (95%) SPRT | 60.0+0.60s Threads=1 Hash=64MB LLR | 2.90 (-2.25, 2.89) [0.00, 2.25] Games | N: 1604 W: 431 L: 164 D: 1009 Penta | [2, 73, 412, 286, 29] http://chess.grantnet.us/test/34160/ Elo | -0.81 +- 1.59 (95%) SPRT | 10.0+0.10s Threads=1 Hash=8MB LLR | -0.70 (-2.25, 2.89) [-2.00, 0.00] Games | N: 85600 W: 19865 L: 20065 D: 45670 Penta | [597, 10290, 21245, 10052, 616] http://chess.grantnet.us/test/34158/ |
||
|---|---|---|
| berserk-fb675dad41b4.nn | ||