2700+ rated Leela is available online now!
*1. Follow the account. 2. Wait until LeelaChess Follows you 3. Send a challenge.
https://lichess.org/@/LeelaChess
2700+ rated Leela is available online now!
*1. Follow the account. 2. Wait until LeelaChess Follows you 3. Send a challenge.
https://lichess.org/@/LeelaChess
Now we have three Leela bots
1. https://lichess.org/@/LeelaChess (strongest I think)
2. https://kontrachess.com/ (middle)
3. http://play.lczero.org/ (weakest)
Cool that we can play the strongest Leela. I dont think Leela is known much, so more exposure will help.
Assuming Leela learns to play at 3400+, does that mean it will require even a better GPU chip to operate, or is it going to be like some massive file to download for personal use?
Current weight ( her knowledge library)= 38M, which is 10x128. They may increase network layers and size in the future, so possibly larger.
Current weight ( her knowledge library)= 38M, which is 10x128. They may increase network layers and size in the future, so possibly larger.
Does that mean 10 layers, with 128 "neurons" per layer? And, if so, is each layer connected only to the adjacent layers?
It doesnt work like that.I cant explain in detail. I am not a Leela programmer.
I realize that Leela's neural net is emulated in digital software, as opposed to direct implementation in hardware (like our brains). But nevertheless, every neural net has multiple layers, and many neurons within each layer. I'd like to find out more about the design of the neural net itself, not the software used to emulate it. I'll try asking at the TCEC folks.
Maybe someone already posted this, but here is an interesting short article about Leela Chess Zero by cyberpunk author Bruce Stirling in WIRED magazine:
wired.com/beyond-the-beyond/2018/04/leela-zero-open-source-chess-playing-neural-network/
Some of that IM's play against Leela was shockingly bad. I found good moves in all of the positions identified as blunders or mistakes by Lichess in this game.
I realize that Leela's neural net is emulated in digital software, as opposed to direct implementation in hardware (like our brains). But nevertheless, every neural net has multiple layers, and many neurons within each layer. I'd like to find out more about the design of the neural net itself, not the software used to emulate it. I'll try asking at the TCEC folks.
Why ask anyone at all? In particular TCEC people who have very little to do with it? LeelaZeroChessis open source so defaut wa tofind the answer is look for it from the code
That assumes I know the programming language it is written in. But I know only PHP and some Python. In any case, the info at the link provided by drmrboss was what I needed.
I understand basic C++ programming, ( I attended that 3 months course in 20 years ago in my teenage assuming that I could write my own chess engine).A kind of unrealistic kid's dream.
What's the easiest way to set up Leela to play herself, or are there videos of people streaming this match up?
Just go to Lczero.org main page. You will see the instruction for set up. If you are stacked, you can ask me.
As Leela totally crushed 2400 blitz IM with score of 9-1, this week will be featuring one of top Bullet player, GrandMaster Andrew Tang.
Andrew Tang is very strong in bullet with 2900 rating and won 2nd prize in 4th Lichess Title arena recently(1st was Magnus Carlsen).
Tang has at times been the highest-rated bullet chess player on both lichess and chess.com and is the highest-rated bullet chess player ever on lichess, with a peak lichess bullet rating of 3196 and a peak chess.com bullet rating of 3250.
https://lichess.org/blog/WtNG7CcAAFMTTHPj/gm-andrew-tang-vs-leela-chess-zero
https://lichess.org/blog/WtzZAyoAALvE8ZSQ/gm-andrew-tang-defends-humanity-against-leela-chess-zero
Score 40-4 ( leela won), Andrew won one game though.
I told you several times, that play.lczero.orgis for demonstration only. She think only 200 positions in normal mode and 2000 in hard mode. In that match it was full mode, she might think about 20,000-30,000 positions per turn.
Still blundered very badly https://lichess.org/3Bl93u7n/black#84
Try Leela on that website and tell me stats vs leela! @AIM-AceMove
https://kontrachess.com/