Join here https://lichess.org/team/qcp-workmans
Tournament points: 148,880
Team leader: duropo
* To satisfy those players that are more interested on close to perfect play
* To allow players to test their newly created opening ideas at best possible resistance
* To challenge the skill of engine developers
* To challenge the skill of opening book creators
* To generate high quality games that could possibly be used in Lichess opening explorer
* Probably to minimize the cheating practices in rated games by providing a playground suitable for players who are in the first place are probably just interested in playing as a Cyborg
An important parameter, which HEAVILY influence leelas search process and move selection is --softmax_temp value. Default value is 1.0. Higher values -> broader search but less depth. Lower values -
My Leela ID 531 still likes to blunder: I used cutechess-cli to generate game: https://lichess.org/xAun43D5#38 Check in terminal with all infos: jo@Deneb:~/lczero_new/build$ ./lczero Usi
#6 yep, confirmed: Chris Whittington (by the way, i am pretty sure he is the developer of 'chess system tal') said in lczero forum: """... LCZero requires a move history, you can't just send it posit
Yes, that could be the reason.
Thanks for info, very impotant to know, i guess she is not 100% compatible with my old scid gui (default version in ubuntu 16.04), because scid sends only fen to engine and not the move history?
I tried fen and I have the same funny result, too. Yes, Leela is trained with move history (last 8 plies or smth like this) and is used to having this additional info. Without it she gives weird r
I think i found the reason for this blunders, my engine lczero v0.10 does not like fen input. With fen input method: position fen rnq1kbnr/ppp1pppp/3p4/8/2P5/5NPb/PP1PPPBP/RNBQK2R w KQkq - 3 5 id 5
Id 527 just castles, at least until about 3000 nodes.