Can somebody here run a tournament of all the NN and NNUE engines Besides SF NNUE, Komodo Dragon, Lela, which are obviously the strongest one. and post it here or create a Link on Youtube.com
Tornado News
Moderators: hgm, Rebel, chrisw
-
- Posts: 5589
- Joined: Wed Sep 05, 2018 2:16 am
- Location: Moving
- Full name: Jorge Picado
Re: Tornado News
Who is 17 years old GM Gukesh 2nd at the Candidate in Toronto?
https://indianexpress.com/article/sport ... t-9281394/
https://indianexpress.com/article/sport ... t-9281394/
-
- Posts: 1242
- Joined: Sat Jul 05, 2014 7:54 am
- Location: Southwest USA
Re: Tornado News
Was going to run a new tourney featuring Tornado but this single exe compile must be popcnt compile...(not working in my machine......need w64 npopcnt or x32 Thx AREngin wrote: ↑Wed Dec 09, 2020 11:41 am Hello Computer and Chessfriends,
i created a new Tornado site here: https://sites.google.com/view/tornadochess/start
you can download the last version 8 or just new Tornado with can be NNUE use it in the evalution function now.
it's amazing increased in strength with just only NNUE!
i did not really well test it, but seems it is now over 3000 elo or more
later i will do my own neural network but that's need many time .... :-/
enjoy,
Engin
-
- Posts: 918
- Joined: Mon Jan 05, 2009 7:40 pm
- Location: Germany
- Full name: Engin Üstün
Re: Tornado News
sorry, yes this is only 64 bit popcount and SSE support version !supersharp77 wrote: ↑Wed Dec 16, 2020 11:38 pmWas going to run a new tourney featuring Tornado but this single exe compile must be popcnt compile...(not working in my machine......need w64 npopcnt or x32 Thx AREngin wrote: ↑Wed Dec 09, 2020 11:41 am Hello Computer and Chessfriends,
i created a new Tornado site here: https://sites.google.com/view/tornadochess/start
you can download the last version 8 or just new Tornado with can be NNUE use it in the evalution function now.
it's amazing increased in strength with just only NNUE!
i did not really well test it, but seems it is now over 3000 elo or more
later i will do my own neural network but that's need many time .... :-/
enjoy,
Engin
-
- Posts: 319
- Joined: Sat Oct 31, 2020 1:04 am
- Full name: Aleksey Glebov
Re: Tornado News
Interesting idea. Ptobably, I can organize such a tourney after my next planned tournament. But I will include only engines using their own nets (at least not SF and not Leela nets). So Tornado, BBC, Mayhem and some others engines using SF NNUE will not be included. Instead, I will include Ethereal, Slow, Nemorino, RubiChess, Minic, Orion, Seer, Halogen (may be Igel) and others engines using original nets. Also Leela CPU may be included.Chessqueen wrote: ↑Tue Dec 15, 2020 10:57 pmCan somebody here run a tournament of all the NN and NNUE engines Besides SF NNUE, Komodo Dragon, Lela, which are obviously the strongest one. and post it here or create a Link on Youtube.com
Incredibly fast systems miscount incredibly fast.
-
- Posts: 5589
- Joined: Wed Sep 05, 2018 2:16 am
- Location: Moving
- Full name: Jorge Picado
Re: Tornado News
Pedone 2.1 now has its own NNUE, not simply implemented by replacing SF NNUEAngle wrote: ↑Thu Dec 17, 2020 1:37 amInteresting idea. Probably, I can organize such a tourney after my next planned tournament. But I will include only engines using their own nets (at least not SF and not Leela nets). So Tornado, BBC, Mayhem, and some others engines using SF NNUE will not be included. Instead, I will include Ethereal, Slow, Nemorino, RubiChess, Minic, Orion, Seer, Halogen (may be Igel) and others engines using original nets. Also Leela CPU may be included.Chessqueen wrote: ↑Tue Dec 15, 2020 10:57 pmCan somebody here run a tournament of all the NN and NNUE engines Besides SF NNUE, Komodo Dragon, Lela, which are obviously the strongest one. and post it here or create a Link on Youtube.com
"Yes Pedone is now using NNUE. I have implemented my own network with included castle rights and with independent weights for the both sides.
For the training I have used 14 depth search score of Pedone with traditional evaluation and I have used a simple SGD to train the net.
I'm not an expert on NN but I have found on the net a lot of material that have helped the development.
I also have a discussion with Marco Belli and we have helped each other in implementing NNUE."
Who is 17 years old GM Gukesh 2nd at the Candidate in Toronto?
https://indianexpress.com/article/sport ... t-9281394/
https://indianexpress.com/article/sport ... t-9281394/
-
- Posts: 512
- Joined: Tue Sep 29, 2020 4:29 pm
- Location: Dublin, Ireland
- Full name: Madeleine Birchfield
Re: Tornado News
Pedone 2.1 does not have its own NNUE, as stated in the first post of the Pedone 2.1 thread; instead it is the last version of Pedone that does not have NNUE, all future versions of Pedone have NNUE, such as the dev verson of Pedone playing at TCEC.Chessqueen wrote: ↑Thu Dec 17, 2020 8:58 pm Pedone 2.1 now has its own NNUE, not simply implemented by replacing SF NNUE
"Yes Pedone is now using NNUE. I have implemented my own network with included castle rights and with independent weights for the both sides.
For the training I have used 14 depth search score of Pedone with traditional evaluation and I have used a simple SGD to train the net.
I'm not an expert on NN but I have found on the net a lot of material that have helped the development.
I also have a discussion with Marco Belli and we have helped each other in implementing NNUE."
-
- Posts: 918
- Joined: Mon Jan 05, 2009 7:40 pm
- Location: Germany
- Full name: Engin Üstün
Re: Tornado News
Why not? What is the problem with SF NNUE? Well every engine is playing diffently even it's using same NNUE in the eval function, for example BBC is very weaker then my engine if they using same NNUE file.Angle wrote: ↑Thu Dec 17, 2020 1:37 amInteresting idea. Ptobably, I can organize such a tourney after my next planned tournament. But I will include only engines using their own nets (at least not SF and not Leela nets). So Tornado, BBC, Mayhem and some others engines using SF NNUE will not be included. Instead, I will include Ethereal, Slow, Nemorino, RubiChess, Minic, Orion, Seer, Halogen (may be Igel) and others engines using original nets. Also Leela CPU may be included.Chessqueen wrote: ↑Tue Dec 15, 2020 10:57 pmCan somebody here run a tournament of all the NN and NNUE engines Besides SF NNUE, Komodo Dragon, Lela, which are obviously the strongest one. and post it here or create a Link on Youtube.com
I will using my own Network in future but thats takes a lot of time to training
-
- Posts: 319
- Joined: Sat Oct 31, 2020 1:04 am
- Full name: Aleksey Glebov
Re: Tornado News
The problem is that I am not excited by a competition of eval-clones. Observing a bunch of engines with the same eval (which is not native for them) in one tourney is annoying. It's like a film festival, where all the participants offer the same movie, but in different quality (quality of search). However, the ''plot of movie'' is strongly defined by the eval rather than by search.
Incredibly fast systems miscount incredibly fast.
-
- Posts: 319
- Joined: Sat Oct 31, 2020 1:04 am
- Full name: Aleksey Glebov
Re: Tornado News
It would be nice to have Pedone NNUE with the original net in this tourney. Fabio says that the new release may be available in January.Chessqueen wrote: ↑Thu Dec 17, 2020 8:58 pm Pedone 2.1 now has its own NNUE, not simply implemented by replacing SF NNUE
"Yes Pedone is now using NNUE. I have implemented my own network with included castle rights and with independent weights for the both sides.
For the training I have used 14 depth search score of Pedone with traditional evaluation and I have used a simple SGD to train the net.
I'm not an expert on NN but I have found on the net a lot of material that have helped the development.
I also have a discussion with Marco Belli and we have helped each other in implementing NNUE."
Incredibly fast systems miscount incredibly fast.
-
- Posts: 918
- Joined: Mon Jan 05, 2009 7:40 pm
- Location: Germany
- Full name: Engin Üstün
Re: Tornado News
then you also not use same books for all engines tooAngle wrote: ↑Thu Dec 17, 2020 11:27 pmThe problem is that I am not excited by a competition of eval-clones. Observing a bunch of engines with the same eval (which is not native for them) in one tourney is annoying. It's like a film festival, where all the participants offer the same movie, but in different quality (quality of search). However, the ''plot of movie'' is strongly defined by the eval rather than by search.