worthjas.blogg.se

Moving forward, I will spend my efforts working on the CudAD trainer alongside Finn to improve it for both Berserk and Koivisto. This speedup is a large motivator for me to continue developing Berserk and its networks. His CUDA based trainer improved Berserk network training time from a week to four hours per network. First, Finn Eggers (Koivisto Author), for helping getting Berserk setup on the Koivisto Trainer. I'd like to personally thank two community members during this release. Self play results vs Berserk 8.5.1, self play results do not reflect exactly when playing vs other engines. Fix hashmove reference in code by in #414.Adjust LMR to use a new formula + legal moves. Improve NNUE auto-vectorization and update logic by in #408.Don't reduce more for previous PV moves by in #407.Reduce more in NMP when there are no opponent threats by in #405.Accumulator update via Move by in #404.Separate promotions in tactical history on captures by in #403.Return to 512 Hidden Neurons by in #402.Remove Counter History Pruning by in #396.Reverse Futility Pruning Margin Tweak by in #393.Corrent singular search results by in #392.Disable late move pruning at root + improve ordering by in #390.Split Good and Bad captures with history and SEE by in #467.Utilize LMR Depth when move pruning quiets by in #465.Add WDL Output and Normalize CP to 50% at 100cp by in #463.Identify upcoming repetitions by in #462.