Moderators: hgm, Rebel, chrisw
There is an updated version of FF2 (called update1) (source code on github), where the net is a separate file. Because the first release with the nnue-net being part of the binaries was a gpl license violation.Gabor Szots wrote: ↑Sat Feb 13, 2021 2:04 pmHow did you do that? The official release has no separate net, it's embedded.
No. Only the commercial version of FF2 has the strong FF2 net. The ratinglists (CEGT, CCRL and me (SPCC)) got this version for testing. The net, which can be downloaded on the FF2-github website is weaker. Of course. Otherwise nobody would buy FF2 anymore.Sylwy wrote: ↑Sat Feb 13, 2021 3:41 pmThanks ! This means that the net in the archive uploaded by Mr. Michael Taktikos is the correct one. My test is evolving similarly to the version tested by CCRL.
On what test have you confirmed it is "weaker"?... I seriously doubt they would produce a "false" additional net just to publish on github.. That would get them into more problems GPLv3 Wise if they try murky stuff like that.
http://talkchess.com/forum3/viewtopic.p ... 09#p882009Werewolf wrote: ↑Sat Feb 13, 2021 5:12 pm Can someone comment on this:
At 31:30 Albert S clearly says Fat Fritz 2 is "genuinely stronger than what's out there right now, without a question and clearly so" And SF Dev is referenced saying Fat Fritz 2 is measurably stronger.
I'm just not seeing this at all on my tests, and I used SF Dev which was released the same day as FF2 came out.
https://en.chessbase.com/post/interview ... -fritz-2-0
I can only think of 4 possible explanations:
a) My test results are inaccurate (Test run ongoing, only 500 games so far, but they are in line with others.)
b) the test criteria is very specific on a large number of cores (I'm testing single-threaded on Cutechess for time's sake)
c) the video is old, perhaps around Christmas time, and doesn't allow for SF Dev improvement
d) a more sinister conclusion.
Can someone comment as I'd prefer to give the benefit of the doubt if possible, I just can't replicate this result.
He's not providing any data to sustain his claim, that should give you a good idea as to where the wind is blowing.Werewolf wrote: ↑Sat Feb 13, 2021 5:12 pm Can someone comment on this:
At 31:30 Albert S clearly says Fat Fritz 2 is "genuinely stronger than what's out there right now, without a question and clearly so" And SF Dev is referenced saying Fat Fritz 2 is measurably stronger.
I'm just not seeing this at all on my tests, and I used SF Dev which was released the same day as FF2 came out.
https://en.chessbase.com/post/interview ... -fritz-2-0
I can only think of 4 possible explanations:
a) My test results are inaccurate (Test run ongoing, only 500 games so far, but they are in line with others.)
b) the test criteria is very specific on a large number of cores (I'm testing single-threaded on Cutechess for time's sake)
c) the video is old, perhaps around Christmas time, and doesn't allow for SF Dev improvement
d) a more sinister conclusion.
Can someone comment as I'd prefer to give the benefit of the doubt if possible, I just can't replicate this result.
The amount of BS in these two paragraphs alone are enough for a whole thread, and he didn't even have any other stake in the matter other than having the article published.A new paradigm
On December 5 the DeepMind group published a new paper at the site of Cornell University called "Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm", and the results were nothing short of staggering. AlphaZero had done more than just master the game, it had attained new heights in ways considered inconceivable. The test is in the pudding of course, so before going into some of the fascinating nitty-gritty details, let’s cut to the chase. It played a match against the latest and greatest version of Stockfish, and won by an incredible score of 64 : 36, and not only that, AlphaZero had zero losses (28 wins and 72 draws)!
Stockfish needs no introduction to ChessBase readers, but it's worth noting that the program was on a computer that was running nearly 900 times faster! Indeed, AlphaZero was calculating roughly 80 thousand positions per second, while Stockfish, running on a PC with 64 threads (likely a 32-core machine) was running at 70 million positions per second. To better understand how big a deficit that is, if another version of Stockfish were to run 900 times slower, this would be equivalent to roughly 8 moves less deep. How is this possible?
A.Silver said, that the github net is weaker. And it is obvious, that this must be true. If not, nobody needed to buy FF2, so it would be crazy, if Chessbase would provide the strong commercial net on github for free.twobeer wrote: ↑Sat Feb 13, 2021 4:53 pmOn what test have you confirmed it is "weaker"?... I seriously doubt they would produce a "false" additional net just to publish on github.. That would get them into more problems GPLv3 Wise if they try murky stuff like that.
Can you include test with this net for refernce? or is it just an assumption on your part that it is weaker and not equal...
Do you have both bin files? can you post the MD5 checksums? whould be good refrence with the checksum.. How do we even now that SPCC got the same net as CEGT, CCRL etc.. If those are "hand-picked" by mr. A.Silver..
This means that CCRL also tested an old version of the 20x512 net! 