Trained on Reinforcement Learn and fine tuned using CCRL games:
https://www.dropbox.com/s/71qy3j8gtgch2 ... 5.zip?dl=0
An amazing net for SF-NNUE.
Moderators: hgm, Rebel, chrisw
-
- Posts: 4468
- Joined: Fri Apr 21, 2006 4:19 pm
- Location: IASI - the historical capital of MOLDOVA
- Full name: SilvianR
-
- Posts: 919
- Joined: Sat May 31, 2014 8:28 am
Re: An amazing net for SF-NNUE.
How does this net fair against some of Sergio's more recent nets?Sylwy wrote: ↑Mon Aug 03, 2020 8:49 pm Trained on Reinforcement Learn and fine tuned using CCRL games:
https://www.dropbox.com/s/71qy3j8gtgch2 ... 5.zip?dl=0
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.
-
- Posts: 10948
- Joined: Wed Jul 26, 2006 10:21 pm
- Full name: Kai Laskos
Re: An amazing net for SF-NNUE.
Hmmm, there seem to be chunks of identical hex strings with Sergio nets. Strength-wise, also almost identical to latest Sergio nets.Sylwy wrote: ↑Mon Aug 03, 2020 8:49 pm Trained on Reinforcement Learn and fine tuned using CCRL games:
https://www.dropbox.com/s/71qy3j8gtgch2 ... 5.zip?dl=0
-
- Posts: 1439
- Joined: Sat Oct 27, 2018 12:58 am
- Location: Germany
- Full name: N.N.
Re: An amazing net for SF-NNUE.
There are differences to the Sergio network. Here are some examples (german):
https://forum.computerschach.de/cgi-bin ... ?tid=11372
https://forum.computerschach.de/cgi-bin ... ?tid=11372
-
- Posts: 10948
- Joined: Wed Jul 26, 2006 10:21 pm
- Full name: Kai Laskos
Re: An amazing net for SF-NNUE.
These are a couple positions. On similarity Sim 8300 positions, this net is behaving as a late SV net among SV nets:Eduard wrote: ↑Mon Aug 03, 2020 11:05 pm There are differences to the Sergio network. Here are some examples (german):
https://forum.computerschach.de/cgi-bin ... ?tid=11372
Code: Select all
sim version 3
Key:
1) Lc0 96x8 (time: 100 ms scale: 1.0)
2) Lc0 SV_3010 (time: 100 ms scale: 1.0)
3) Lc0 T40 (time: 100 ms scale: 1.0)
4) Lc0 T70 (time: 100 ms scale: 1.0)
5) SF_FiNN02 (time: 100 ms scale: 1)
6) SF_NNUE_GK (time: 100 ms scale: 1)
7) SF_NNUE_nightnurse (time: 100 ms scale: 1)
8) SF_NNUE_qhapaq (time: 100 ms scale: 1)
9) SF_NNUE_SV_1515 (time: 100 ms scale: 1)
10) SF_NNUE_SV_1515 Bis (time: 100 ms scale: 1)
11) SF_NNUE_SV_2141 (time: 100 ms scale: 1)
12) SF_NNUE_SV_2257 (time: 100 ms scale: 1)
13) Stockfish 10 (time: 100 ms scale: 1)
14) Stockfish 11 (time: 100 ms scale: 1)
15) Stockfish 6 (time: 100 ms scale: 1)
16) Stockfish 8 (time: 100 ms scale: 1)
17) Stockfish dev (time: 100 ms scale: 1)
18) Stockfish dev Bis (time: 100 ms scale: 1)
19) Andscacs (time: 100 ms scale: 1)
20) Komodo 14 (time: 100 ms scale: 1)
21) Ethereal (time: 100 ms scale: 1)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
1. ----- 63.60 64.23 67.24 46.61 53.10 51.52 53.65 53.14 53.58 53.24 53.02 50.62 50.30 47.90 49.45 50.57 50.42 47.98 49.50 49.73
2. 63.60 ----- 74.82 70.39 46.82 53.96 51.30 55.46 55.33 55.20 55.22 55.88 51.30 51.23 48.35 50.07 51.76 51.92 48.46 49.84 49.00
3. 64.23 74.82 ----- 70.25 46.08 53.28 51.01 53.92 54.05 54.30 53.87 54.18 50.23 50.10 47.41 49.22 50.44 50.45 48.06 49.15 48.14
4. 67.24 70.39 70.25 ----- 47.63 55.83 52.67 56.60 56.41 56.05 56.71 57.10 52.62 52.61 49.78 51.07 52.62 52.18 49.73 50.44 50.30
5. 46.61 46.82 46.08 47.63 ----- 50.76 48.09 49.48 49.13 49.10 50.46 49.39 47.71 48.15 45.04 47.51 47.84 48.07 45.74 46.16 45.39
6. 53.10 53.96 53.28 55.83 50.76 ----- 54.60 58.07 58.19 57.67 60.96 57.55 55.12 55.38 51.52 54.33 55.80 55.24 52.05 52.11 51.76
7. 51.52 51.30 51.01 52.67 48.09 54.60 ----- 55.40 55.66 55.49 55.07 55.09 50.79 50.53 47.80 50.33 50.39 50.34 48.82 48.97 48.64
8. 53.65 55.46 53.92 56.60 49.48 58.07 55.40 ----- 63.57 62.61 60.33 62.44 52.20 52.06 48.18 50.89 51.44 51.51 49.14 50.01 49.67
9. 53.14 55.33 54.05 56.41 49.13 58.19 55.66 63.57 ----- 64.01 59.95 63.05 51.81 52.10 48.91 51.25 52.18 51.63 48.66 49.77 49.43
10. 53.58 55.20 54.30 56.05 49.10 57.67 55.49 62.61 64.01 ----- 59.88 62.73 51.68 51.70 48.17 49.99 51.43 51.51 48.24 49.72 48.79
11. 53.24 55.22 53.87 56.71 50.46 60.96 55.07 60.33 59.95 59.88 ----- 60.06 52.65 53.02 49.65 52.02 53.03 52.61 49.87 50.51 50.12
12. 53.02 55.88 54.18 57.10 49.39 57.55 55.09 62.44 63.05 62.73 60.06 ----- 51.78 52.20 48.34 51.32 51.75 51.70 49.17 49.81 49.04
13. 50.62 51.30 50.23 52.62 47.71 55.12 50.79 52.20 51.81 51.68 52.65 51.78 ----- 64.79 56.43 61.67 63.57 63.51 56.05 55.83 56.25
14. 50.30 51.23 50.10 52.61 48.15 55.38 50.53 52.06 52.10 51.70 53.02 52.20 64.79 ----- 55.52 60.31 66.04 65.53 55.55 55.90 54.95
15. 47.90 48.35 47.41 49.78 45.04 51.52 47.80 48.18 48.91 48.17 49.65 48.34 56.43 55.52 ----- 60.96 54.79 54.86 56.02 54.90 54.05
16. 49.45 50.07 49.22 51.07 47.51 54.33 50.33 50.89 51.25 49.99 52.02 51.32 61.67 60.31 60.96 ----- 59.03 58.62 58.55 56.19 55.45
17. 50.57 51.76 50.44 52.62 47.84 55.80 50.39 51.44 52.18 51.43 53.03 51.75 63.57 66.04 54.79 59.03 ----- 66.06 54.88 54.60 53.96
18. 50.42 51.92 50.45 52.18 48.07 55.24 50.34 51.51 51.63 51.51 52.61 51.70 63.51 65.53 54.86 58.62 66.06 ----- 54.22 54.90 54.47
19. 47.98 48.46 48.06 49.73 45.74 52.05 48.82 49.14 48.66 48.24 49.87 49.17 56.05 55.55 56.02 58.55 54.88 54.22 ----- 53.91 53.44
20. 49.50 49.84 49.15 50.44 46.16 52.11 48.97 50.01 49.77 49.72 50.51 49.81 55.83 55.90 54.90 56.19 54.60 54.90 53.91 ----- 53.86
21. 49.73 49.00 48.14 50.30 45.39 51.76 48.64 49.67 49.43 48.79 50.12 49.04 56.25 54.95 54.05 55.45 53.96 54.47 53.44 53.86 -----
Interesting also to note than self-similarity of SF dev on 4 threads (66%) is a bit higher than self-similarity of SF_NNUE_SV on 4 threads (64%).
-
- Posts: 2438
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl
Re: An amazing net for SF-NNUE.
I tested it vs. SF 200717, like all S.Vieri nets. Look here:Zenmastur wrote: ↑Mon Aug 03, 2020 9:33 pmHow does this net fair against some of Sergio's more recent nets?Sylwy wrote: ↑Mon Aug 03, 2020 8:49 pm Trained on Reinforcement Learn and fine tuned using CCRL games:
https://www.dropbox.com/s/71qy3j8gtgch2 ... 5.zip?dl=0
https://www.sp-cc.de/files/ergebnisse_nnue_schnell.txt
The net is very strong, but not the best in that tests. The best is still sv200730_1934
-
- Posts: 3657
- Joined: Wed Nov 18, 2015 11:41 am
- Location: hungary
Re: An amazing net for SF-NNUE.
Sylwy wrote: ↑Mon Aug 03, 2020 8:49 pm Trained on Reinforcement Learn and fine tuned using CCRL games:
https://www.dropbox.com/s/71qy3j8gtgch2 ... 5.zip?dl=0
Is there another site to download it?
Dropbox is not a good place.
-
- Posts: 1296
- Joined: Sun Mar 12, 2006 6:46 pm
- Location: Kelowna
- Full name: Tony Mokonen
Re: An amazing net for SF-NNUE.
corres wrote: ↑Tue Aug 04, 2020 3:52 pmSylwy wrote: ↑Mon Aug 03, 2020 8:49 pm Trained on Reinforcement Learn and fine tuned using CCRL games:
https://www.dropbox.com/s/71qy3j8gtgch2 ... 5.zip?dl=0
Is there another site to download it?
Dropbox is not a good place.
There's nothing wrong with DropBox. Bizarre comment.
Also, any time there's a DropBox link that ends in ?dl=0, you can change the 0 to a 1 so that it's a direct download.
https://www.dropbox.com/s/71qy3j8gtgch2 ... 5.zip?dl=1
-
- Posts: 4607
- Joined: Wed Oct 01, 2008 6:33 am
- Location: Regensburg, Germany
- Full name: Guenther Simon
Re: An amazing net for SF-NNUE.
Good judgement...Laskos wrote: ↑Mon Aug 03, 2020 10:08 pmHmmm, there seem to be chunks of identical hex strings with Sergio nets. Strength-wise, also almost identical to latest Sergio nets.Sylwy wrote: ↑Mon Aug 03, 2020 8:49 pm Trained on Reinforcement Learn and fine tuned using CCRL games:
https://www.dropbox.com/s/71qy3j8gtgch2 ... 5.zip?dl=0
As expected the cloning on NNUE nets has begun.
People can now test clones of engines with clones of nets w/o knowing they play A+A vs. A+A.
This so called qh-fine-0803-5 net is 99.99999% identical to Sergio Vieris net 20200729-0109.
Only at the very end of millions of hex entries 'someone' changed a few hex values ;-)
UltraEdit compare mode:
Code: Select all
Comparing: C:\Engines\NNUE-SV\qh-fine-0803-5.bin
To: C:\Engines\NNUE-SV\20200729-0109.bin
====
====
166883 <! üúþúþû/öÿÿ=
!> üúþúþûAöÿÿ+
166893 <! ùéööñÆØ°íÐúûúçûîëäÙ÷Ìýÿÿí\GÔ"Bñ&æ ë8ñ-#Ùîåà˼ðç
!> ùéööñÆØ°íÐúûúçûîëäÙ÷¸ýÿÿí\GÔ"Bñ&æ ë8ñ-#Ùîåà˼ðç
https://rwbc-chess.de
trollwatch:
Talkchess nowadays is a joke - it is full of trolls/idiots/people stuck in the pleistocene > 80% of the posts fall into this category...
trollwatch:
Talkchess nowadays is a joke - it is full of trolls/idiots/people stuck in the pleistocene > 80% of the posts fall into this category...
-
- Posts: 3657
- Joined: Wed Nov 18, 2015 11:41 am
- Location: hungary
Re: An amazing net for SF-NNUE.
Thanks for the advice. I did not know this trick. Is this bizarre?tmokonen wrote: ↑Tue Aug 04, 2020 4:37 pmThere's nothing wrong with DropBox. Bizarre comment.corres wrote: ↑Tue Aug 04, 2020 3:52 pmSylwy wrote: ↑Mon Aug 03, 2020 8:49 pm Trained on Reinforcement Learn and fine tuned using CCRL games:
https://www.dropbox.com/s/71qy3j8gtgch2 ... 5.zip?dl=0
Is there another site to download it?
Dropbox is not a good place.
Also, any time there's a DropBox link that ends in ?dl=0, you can change the 0 to a 1 so that it's a direct download.
https://www.dropbox.com/s/71qy3j8gtgch2 ... 5.zip?dl=1