We have released Dragon 2.6 at komodochess.com. Dragon uses NNUE (Neural Network Updated Efficiently) technology, originally developed for the game of shogi. Komodo has a great deal of chess knowledge in its evaluation. Training an NNUE network based on this evaluation was both an advantage and a challenge, requiring experimentation with architectures and data generation of billions of positions. The reinforcement learning phase for Dragon is in its infancy, but is already showing great promise.
Dragon 2.6 is a huge strength improvement over Dragon 2, roughly 70 elo at blitz on four threads and 83 elo on one, more than 100 elo in MCTS mode and for playing Fischerandom (chess960). It is a significant improvement over Dragon 2.5, about 18 elo on one thread blitz, about 25 elo in MCTS mode or playing chess960. The gains over Komodo 14.1, the last pre-dragon release (Nov 2020), are in the 250 to 350 Elo range depending on threads, mode, and game type, all at CCRL blitz time control (2' + 1"). The improvement over Dragon 2 is due to a larger, "smarter", and better-trained net, to deeper search due to search enhancements, and to parameter tuning. The improvement from Komodo to Dragon is due to the embedded neural network providing much more accurate evaluation and also in some sense gaining an extra ply or so of search by seeing some tactics that a normal eval can not see. Our testing also shows that Dragon is especially strong in Fischerandom (960) chess, which was confirmed when Dragon won the TCEC FRC 2021 championship, and again in our latest testing that shows Dragon 2.6 convincingly defeated Stockfish 14 in a 2000 game blitz Fischerandom match. Dragon also won the 2021 TCEC Swiss and Swiss2 championships (normal chess). We believe that Dragon will play in a more human-like style than standard Komodo since it relies on learning what actually works in games rather than just on pre-assigned values for eval terms.
The Skill levels have been replaced by Elo values, made much more human-like, and are now intended to correspond to human Fide ratings in Rapid chess, so for example an 1800 FIDE rated player should be closely matched with UCI Elo 1800 at 15' + 10". Also the MultiPV play of standard mode is dramatically improved from the initial dragon release, on top of the actual strength gain. Other features added since Komodo 13 include Armageddon mode, MCTS Optimism, Personalities, and Auto-Skill. Standard mode is usually the stronger mode objectively, while MCTS mode is usually better at setting problems for human opponents as it does not assume perfect defense, especially with more than a few seconds to consider, and is also stronger than standard mode when looking at several lines of play at once (MultiPV).
ChessSirs;;;;
ReplyDeleteHow did Fisherov do against Dragon 2.6 ?
He won without a problem: https://chessengines.blogspot.com/2022/01/match-fisherov-098j-dragon-26-265-235.html
Delete