Author - Martin Sedlak
v.4.43:
NN based evaluation (topo 736-192-4-1), goodbye HCE
no incremental inference updates, so a bit slower
custom inference with layer 1 zero skip trick
inverse sigmoid to convert win probability to cp, oscillates a lot with big evals
trained on 1.7M cheng self-play fixed node games, yielding ~125M filtered positions
labeled on 6-ply cheng search and game outcome
includes custom hacky libtorch-based tuner
bench 11152861
dropped support for Windows XP
Cheng 4.41 engine results with other engines:
| Tomcat 1.0 | 6.5/10 | +3 | 10 Partii |
| Amoeba 3.4 | 5/10 | +0 | 10 Partii |
| Monolith 2 | 5/9 | +1 | 9 Partii |
| BBC 1.4 | 6.5/8 | +5 | 8 Partii |
| Topple 0.8.1 | 4.5/8 | +1 | 8 Partii |
| Demolito 2021-10-21 | 2.5/8 | -3 | 8 Partii |
| Beef 0.3.6 | 0/8 | -8 | 8 Partii |
| FabChess 1.16 | 3.5/7 | +0 | 7 Partii |
| Godel 7.0 | 4/6 | +2 | 6 Partii |
| Nirvanachess 2.5 | 3/6 | +0 | 6 Partii |
| Deuterium 2019.2.37.73 | 3/6 | +0 | 6 Partii |
| Invictus r382 | 2/6 | -2 | 6 Partii |
| Combusken 2.0.0 | 0.5/6 | -5 | 6 Partii |
| Coiled 1.1 | 0/6 | -6 | 6 Partii |
| Francesca 0.31a | 3.5/5 | +2 | 5 Partii |
| Tantabus 2.0.0 | 3.5/4 | +3 | 4 Partii |
| Gargantua 1.0 | 1.5/4 | -1 | 4 Partii |
| Devre 4.0 | 0/4 | -4 | 4 Partii |
| Rodent NNUE 1.0 | 2.5/3 | +2 | 3 Partii |
| MadChess 3.2 beta | 2/3 | +1 | 3 Partii |
| Mittens 0.9 | 1.5/3 | +0 | 3 Partii |
Comments
Post a Comment