Rebel 14

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

Vernon Crawford
Posts: 73
Joined: Wed Sep 01, 2021 2:05 am
Location: London, England
Full name: Vernon Crawford

Re: Rebel 14

Post by Vernon Crawford »

Rebel wrote: Wed Jan 12, 2022 8:46 pm
Gabor Szots wrote: Wed Jan 12, 2022 7:56 pm
Rebel wrote: Wed Jan 12, 2022 7:48 pm
Gabor Szots wrote: Wed Jan 12, 2022 5:53 pm 2. Is what you have done the same what Albert Silver did with Fat Fritz 2, that is, take the search of an other engine and replace its evaluation with your own?
Brrrr.... what a comparison, which engine did Albert use, Fritz, Stockfish? I used my own and the discount is 500%.
What am I missing? Don't you write you have taken Fruit 2.1 as your starting point and later replaced its evaluation with that of Benjamin? I thought the search of Fruit has been kept.
Yep, that's what the page states.
Seems like calling it 'Rebel' is a real stretch.
The engine is Fruit with a Benjamin NNUE.

It's Fruit NNUE!
wickedpotus
Posts: 153
Joined: Sun May 16, 2021 5:33 pm
Full name: Aron Rodgriges

Re: Rebel 14

Post by wickedpotus »

JohnW wrote: Wed Jan 12, 2022 10:29 pm Anyone try this in the Arena gui? I can't get the engine to make a move. Seems to work fine in Fritz 17 though..
When you install engine in arena it defaults to winboard.. just switch to UCI under ENGINES - Manage - Details for the engine... it will work..

I am not sure if there is some additional settings needed as it don't seem to use all 10-cores 20 threads of the CPU i tested.. only uses like 20% of the CPU..

Cheers,
Potus...
Frank Quisinsky
Posts: 6831
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: Rebel 14

Post by Frank Quisinsky »

Hi there,

after all the discusses in the past about programs used Fruit code.

Very sure after all the contacts I had with Fabien ...
Fabien will be happy if an other programmer wrote very open that Fruit sources are used.

That is absolutly genial if a person like Ed Schröder used a code and give his own knowledge with the important NN part and all the work on Benjamin.
In German I would like to say ... GEIL.

For me to 100% perfect and again, I hope, wish me ... that Ed will have a long time fun with it to make the first version stronger and stronger. A big order for Ed with all the programming knowledge Ed have.

I am not a fan of open-source for different reasons I wrote that often in the past.
Important for me is at first: Which information gave a person ____at first___, used a work by an other.

What Ed do is the best example for it.

Really good news after the new Hiarcs last week.
I wish me more of the "older" computer chess heros with such nice things.

Again ... good luck Ed.
Very interesting ... very strong idea and to 100% with the meaning of Fabien (after the opinion about it I have from him).
2 super heros are working in combination together!

Go Rebel 14 go ...

Best
Frank
connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Rebel 14

Post by connor_mcmonigle »

I have to agree with Vernon (which probably isn't a good sign) in that I'd have much preferred if a different name was used/the name indicated more clearly that this is a Fruit derivative. In any case, after some effort due to the excessive reliance on nonstandard windows specific crap in the NNUE implementation, I was able to produce a Linux compile. It seems the source is slightly incomplete as implementations for "read_u16_le" and "read_u32_le" are missing, though these were trivial to implement myself. It seems to play about as expected in my very brief testing:

[pgn]
[Event "?"]
[Site "?"]
[Date "2022.01.12"]
[Round "?"]
[White "rebel-14"]
[Black "mr.bob-dev"]
[Result "1-0"]
[ECO "C07"]
[GameDuration "00:04:19"]
[GameEndTime "2022-01-12T14:12:06.447 PST"]
[GameStartTime "2022-01-12T14:07:46.622 PST"]
[Opening "French"]
[PlyCount "79"]
[TimeControl "120+1"]
[Variation "Tarrasch, Open Variation"]

1. e4 {+0.59/16 3.4s} e6 {-0.19/20 6.0s} 2. d4 {+0.46/16 4.6s}
d5 {-0.50/22 5.8s} 3. Nd2 {+0.39/16 4.6s} c5 {-0.07/21 5.5s}
4. Ngf3 {+0.38/16 8.4s} Nf6 {-0.19/22 5.4s} 5. e5 {+0.71/18 3.9s}
Nfd7 {-0.12/24 5.2s} 6. c3 {+0.60/18 4.1s} Nc6 {0.00/22 5.5s}
7. Bd3 {+0.75/17 4.0s} Qb6 {0.00/25 4.6s} 8. O-O {+0.96/17 3.9s}
cxd4 {0.00/25 5.3s} 9. cxd4 {+0.74/18 2.9s} Nxd4 {0.00/26 3.9s}
10. Nxd4 {+0.87/18 2.9s} Qxd4 {0.00/27 4.0s} 11. Nf3 {+0.92/19 3.5s}
Qb6 {-0.06/23 4.1s} 12. h3 {+0.93/16 2.7s} Nc5 {+0.22/21 4.0s}
13. Be3 {+1.50/17 5.3s} Bd7 {0.00/22 3.7s} 14. Rc1 {+1.62/17 3.6s}
Rc8 {0.00/21 3.5s} 15. Ng5 {+1.35/17 6.6s} Be7 {-0.71/22 8.9s}
16. Qh5 {+2.15/17 4.3s} g6 {-1.10/22 7.0s} 17. Qh6 {+3.81/17 3.1s}
d4 {-1.24/21 3.6s} 18. Bxd4 {+2.56/17 13s} Bf8 {-1.08/23 2.5s}
19. Qh4 {+2.54/17 2.7s} Be7 {-1.14/23 3.9s} 20. Qg3 {+2.01/16 6.2s}
O-O {-0.30/18 2.2s} 21. Bb1 {+2.51/14 1.6s} Bb5 {-0.21/17 3.2s}
22. Rfd1 {+2.42/15 2.4s} Be2 {-0.20/21 4.3s} 23. Rd2 {+2.31/14 1.8s}
Qb5 {-0.22/21 2.7s} 24. Re1 {+2.16/14 2.3s} Bc4 {-0.49/18 2.3s}
25. h4 {+3.79/13 1.7s} Qa5 {-0.80/15 3.8s} 26. Red1 {+5.41/15 3.3s}
Bd5 {-1.21/15 1.9s} 27. Be3 {+5.40/14 1.3s} Qd8 {-1.51/16 2.2s}
28. h5 {+6.86/15 2.2s} Bxg5 {-2.06/18 2.3s} 29. Bxg5 {+7.03/15 1.3s}
Qc7 {-1.99/18 1.7s} 30. Rc2 {+7.12/15 4.5s} Qb6 {-1.50/15 1.6s}
31. Bh6 {+7.39/15 1.6s} Qb5 {-2.50/14 1.6s} 32. Qg5 {+18.33/14 1.7s}
Nd7 {-3.23/14 1.4s} 33. hxg6 {+33.98/13 2.0s} hxg6 {-9.83/15 1.4s}
34. Rxc8 {+M13/14 1.4s} Qe2 {-11.73/17 3.8s} 35. Bxg6 {+M11/18 1.5s}
Qxd1+ {-M10/42 1.2s} 36. Kh2 {+M9/4 0.001s} Qh1+ {-M8/99 0.54s}
37. Kxh1 {+M7/43 0.42s} Bxg2+ {-M6/99 0.035s} 38. Qxg2 {+M5/45 1.2s}
Kh8 {-M4/99 0.099s} 39. Bxf7 {+M3/46 1.5s} Rxc8 {-M2/99 0.007s}
40. Qg7# {+M1/44 1.5s, White mates} 1-0
[/pgn]

It does seem the inference code is original and I'd be inclined to guess the training code is also original given the model's parameters are stored in a 35 MiB txt file and the quantization scheme relies only on 16 bit and 32 bit integers afaic... Chris or Ed can hopefully provide more details there.
Frank Quisinsky
Posts: 6831
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: Rebel 14

Post by Frank Quisinsky »

Hi Conner,

Rebel 14 (Fruit) ...

On the other hand ...
The name of Dark Toga is Dark Toga and not Dark Toga (Fruit).

But maybe it's better to set (Fruit).

Most important information at first by Ed:

Technical
Rebel 14 consists of 4 parts,

1. The famous Fruit 2.1 engine from 2004 by Fabien Letouzey, elo ~2700
2. Growing Fruit by Pawel Koziol improving the Fruit search, elo ~2800 making it about equal in strength with ProDeo 3.1
3. Replacing the Fruit evaluation with Benjamin 1.1 NNUE evaluation.
4. NNUE code to excecute the Benjamin NN evaluation by Chris Whittington.

That is strong!!
Thinking on all the programs never we can find such an information.

Houdini programmer, Houdini 1.0.
All is from myself ... and in reality 99,5 Robolito, the work Norman Schmidt do.

I am very happy about it and added the information directly in my Engine Link selection with ... derived from Fruit.

Best
Frank
Vernon Crawford
Posts: 73
Joined: Wed Sep 01, 2021 2:05 am
Location: London, England
Full name: Vernon Crawford

Re: Rebel 14

Post by Vernon Crawford »

Frank, it's clearly disingenuous and misleading to include himself as an author and name it under his well-known commercial trademark 'Rebel', especially when it appears he hasn't written any of the code.
connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Rebel 14

Post by connor_mcmonigle »

Vernon Crawford wrote: Wed Jan 12, 2022 11:40 pm Frank, it's clearly disingenuous and misleading to include himself as an author and name it under his well-known commercial trademark 'Rebel', especially when it appears he hasn't written any of the code.
Frank listed it as "derived from Fruit" which is sufficient in my opinion. It does appear Chris/Ed are responsible for the inference code so the latter part of your comment is inaccurate. In any case, I am disappointed to see Winter and Halogen removed (and "ShashChess" included :P)
Vernon Crawford
Posts: 73
Joined: Wed Sep 01, 2021 2:05 am
Location: London, England
Full name: Vernon Crawford

Re: Rebel 14

Post by Vernon Crawford »

connor_mcmonigle wrote: Wed Jan 12, 2022 11:46 pm
Vernon Crawford wrote: Wed Jan 12, 2022 11:40 pm Frank, it's clearly disingenuous and misleading to include himself as an author and name it under his well-known commercial trademark 'Rebel', especially when it appears he hasn't written any of the code.
Frank listed it as "derived from Fruit" which is sufficient in my opinion. It does appear Chris/Ed are responsible for the inference code so the latter part of your comment is inaccurate.
The info on website is as follows:

"4. NNUE code to excecute the Benjamin NN evaluation by Chris Whittington."

How do you know Ed wrote inference code?
Seems you're eager to spin things positively.
connor_mcmonigle
Posts: 544
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Rebel 14

Post by connor_mcmonigle »

Vernon Crawford wrote: Wed Jan 12, 2022 11:50 pm
connor_mcmonigle wrote: Wed Jan 12, 2022 11:46 pm
Vernon Crawford wrote: Wed Jan 12, 2022 11:40 pm Frank, it's clearly disingenuous and misleading to include himself as an author and name it under his well-known commercial trademark 'Rebel', especially when it appears he hasn't written any of the code.
Frank listed it as "derived from Fruit" which is sufficient in my opinion. It does appear Chris/Ed are responsible for the inference code so the latter part of your comment is inaccurate.
The info on website is as follows:

"4. NNUE code to excecute the Benjamin NN evaluation by Chris Whittington."

How do you know Ed wrote inference code?
Seems you're eager to spin things positively.
Rebel 14 is a collaboration between Chris and Ed allegedly (Chris's NNUE code wasn't public previously afaik and Chris is listed as an author). I'm pretty ambivalent on Rebel 14 on the whole and would have much preferred a different name was used as mentioned previously. I think we mostly are in agreeance.
Frank Quisinsky
Posts: 6831
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: Rebel 14

Post by Frank Quisinsky »

Hi Vernon,

from this point of view I can follow Vernon.

But Rebel released under GPL v3.

Trademark "Rebel" can be a problem ... OK!
Could be right, I don't know.

But the name is "Rebel 14" and not "Rebel" or "Rebel x" ... version number from a commercial version.

Hm ...
For me is the idea very strong with most important information by Ed.
On all the other things Ed can work ...

But honestly ... without to say you are not right ... Rebel stands for Ed.
And the fans of the work by Ed await Rebel.
The good name should be free for using for Ed himself.

Again, not sure here.
Maybe you are right.

Best
Frank

If I am in the position ... I believe I like to make the same and give the same name Ed do here.
Last edited by Frank Quisinsky on Thu Jan 13, 2022 12:00 am, edited 1 time in total.