CEGT Forum
Würden Sie gerne auf diese Nachricht reagieren? Erstellen Sie einen Account in wenigen Klicks oder loggen Sie sich ein, um fortzufahren.

LCZero Testmatch

Nach unten

LCZero Testmatch Empty LCZero Testmatch

Beitrag von Werner Mo Apr 23, 2018 9:12 pm

Testmatches under CEGT conditions 40/20: - running on this pc 40/10 real.
(startposition 87kns) (166 = 5169.70)
1   LCZero 0.7/166 1CPU   2057      011½10½0½0½½111½10½0  10.5/20
2   Anechka 0.08          2040     100½01½1½1½½000½01½1    9.5/20

Lczero 187= 5297.69 (startposition 87kns)
1   LCZero 0.7 1CPU  2166  +126   100½1111111010½1½½½1  13.5/20
2   Anechka 0.08     2040  -126  011½0000000101½0½½½0    6.5/20

Lczero 181= 5267.33 (startposition 87kns)                         
1   EveAnn 1.72 2071      +108   0½1110½110111010½11½  13.0/20
2   LCZero 0.7/181 1CPU 1963  -108  1½0001½001000101½00½    7.0/20

Lczero 187 = 5297.69 and GPU Intel Graphics 630 (startposition 145kns)
1   LCZero GPU v0.7/187 1CPU 2124   +53   ½010½01½111½1100101½  11.5/20
2   EveAnn 1.72 2071             -53  ½101½10½000½0011010½    8.5/20

Lczero 217 = 5405.12 1CPU (startposition 87kns)                        
1   LCZero/217 0.7 1CPU  2197  +126   1010100111111101½½½1  13.5/20
2   EveAnn 1.72      2071  -126  0101011000000010½½½0    6.5/20

Lczero 187 = 5297.69  (startposition 87kns)
1   LCZero 0.7 1CPU 2123   010½001101100½111½0½  10.0/20  100.00
2   Popochin 4.0 2123     101½110010011½000½1½   10.0/20  100.00

Lczero 217 = 5405.12 1CPU (startposition 87kns)
1   LCZero 0.7 1CPU  2504  +381   1111111111½111111½10  18.0/20 !?
2   Popochin 4.0     2123  -381  0000000000½000000½01    2.0/20


Lczero 187 = 5297.69  (startposition 365 kns)
1   The Baron 3.42 x64 1CPU  2669  +190   11110½101111110½0111  15.0/20
2   LCZero 0.7 CPU 8Threads  2479  -190  00001½010000001½1000    5.0/20

Lczero 187 = 5297.69  (startposition 572 kns)
1   lczero v0.7 4CPU 2598   +89   ½1010½1111000101½111  12.5/20
2   Djinn 1.021 x64 2509    -89  ½0101½0000111010½000    7.5/20

Lczero 217 = 5405.12 1CPU (startposition 87kns) (+100 in selfplay = +50 against other opponents)
1   Djinn 1.021 x64  2509  +190   ½½11½1½101110½11111½  15.0/20
2   LCZero 0.7 1CPU  2319  -190  ½½00½0½010001½00000½    5.0/20

Werner
Admin

Anzahl der Beiträge : 877
Anmeldedatum : 22.10.13

https://cegt.forumieren.com

Nach oben Nach unten

LCZero Testmatch Empty Re: LCZero Testmatch

Beitrag von Werner Di Mai 01, 2018 10:36 am

Now testing new Network  15x192! CEGT Conditions 40/20
PC Benchmarks: t=1 Startposition depth 20 CPU Version 31 nps (before I had 87)

Network 227 = 5501.53   (I did not await such a jump from 2319 before)                    
1   Djinn 1.021 x64  2509   +70   ½½11½10100111½100½1½  12.0/20
2   LCZero 0.7 1CPU  2439   -70  ½½00½01011000½011½0½    8.0/20

Network 227 = 5501.53
1   The Baron 3.42 x64 1CPU  2669  +240   11111½11½10½111½1011  16.0/20
2   LCZero 0.7 1CPU          2429  -240  00000½00½01½000½0100    4.0/20

Network 227 = 5501.53                                
1   Counter 2.6 x64 1CPU 2435  +145  +58/=29/-17 69.71%   72.5/104
2   LCZero 0.7 1CPU 2290       -145  +17/=29/-58 30.29%   31.5/104

Network 227 = 5501.53                              
1   NanoSzachy 4.1 x64 2357  +165  +30/=15/-7 72.12%   37.5/52
2   LCZero 0.7 1CPU 2192     -165  +7/=15/-30 27.88%   14.5/52  Sad

Network 227 = 5501.53    (the other 52 games are included)                          
1   NanoSzachy 4.1 x64 2356  +169  +62/=27/-15 72.60%   75.5/104
2   LCZero 0.7 1CPU 2188     -169  +15/=27/-62 27.40%   28.5/104

Network 246 = 5599.02 (+100 and ~+50 Elo against other engines?)
1   NanoSzachy 4.1 x64 2356  +125   ½01100111½11½0101110111½1½  17.5/26
2   LCZero 0.8 1CPU 2231     -125  ½10011000½00½1010001000½0½    8.5/26

Network 253 = 5641.391; nps: 213 in startposition
The Baron 3.42 x64 2669 111½½001½1½0½00100½11½½½0½ 13.0/26 169.00
2 lczero 0.8 4CPU 2669! 000½½110½0½1½11011½00½½½1½ 13.0/26 169.00


Werner
Admin

Anzahl der Beiträge : 877
Anmeldedatum : 22.10.13

https://cegt.forumieren.com

Nach oben Nach unten

LCZero Testmatch Empty Re: LCZero Testmatch

Beitrag von Werner Mi Mai 09, 2018 7:11 pm

Now testing new Network  15x192! CEGT Conditions 40/20
LCZero 0.9 avx
PC Benchmarks: t=1 Startposition depth 20 CPU Version 66 nps (before I had 31)

Network 263 = 5609.13
1   The Baron 3.42 x64 1CPU 2669  +125   0½½01½111½½½1½101100111111  17.5/26
2   lczero v0.9 1CPU avx 2544          -125  1½½10½000½½½0½010011000000    8.5/26

Network 270 = 5563.32 something is going terrible wrong at the moment!
1   The Baron 3.42 x64 1CPU 2669  +323   11011111110111½11111110111  22.5/26
2   lczero v0.9 avx 1CPU 2346         -323  00100000001000½00000001000    3.5/26

Network 285 = 5462
1   The Baron 3.42 x64 1CPU  +354   10111111111111111110111110  23.0/26
2   LCZero2 0.10avx 1CPU     -354  01000000000000000001000001    3.0/26

Network 270 = 5563
1  The Baron 3.42 x64 1CPU  +229   ½1½111½101½1111½0111½1111½  20.5/26
2   LCZero3 0.10avx 1CPU     -229  ½0½000½010½0000½1000½0000½    5.5/26

Network 263 = 5609
1   The Baron 3.42 x64 1CPU  +125   01½½1½½11½0½½½1111011110½1  17.5/26
2   LCZero4 0.10avx 1CPU     -125  10½½0½½00½1½½½0000100001½0    8.5/26

Djinn 1.021 x64    2200 - LCZero 0.10avx303 1CPU     2200   13.5 - 12.5    +9/=9/-8    51.92%
Djinn 1.021 x64    2200 - Lczero2 0.10avx253 1CPU    2200   21.0 - 5.0    +17/=8/-1    80.77%

104 games with weight 303 - 5710                                  
1   Djinn 1.021 x64         2510   +78  +49/=29/-26 61.06%   63.5/104
2   LCZero 0.10avx303 1CPU  2432   -78  +26/=29/-49 38.94%   40.5/104

104 games with weight 331 - 5900                                  
1   Djinn 1.021 x64         2510   +26  +39/=34/-31 53.85%   56.0/104
2   LCZero 0.10avx331 1CPU  2484   -26  +31/=34/-39 46.15%   48.0/104

104 games with weight 349 - 5901
1   GreKo 2018.02 x64       2507   +23  +43/=25/-36 53.37%   55.5/104
2   LCZero 0.10avx349 1CPU  2484   -23  +36/=25/-43 46.63%   48.5/104

104 games with weight 349 - 5901  
1   LCZero 0.10avx349 1CPU  2487   +23  +40/=31/-33 53.37%   55.5/104
2   Nebula 2.0 x64 1CPU     2464   -23  +33/=31/-40 46.63%   48.5/104

104 games with weight 431 - 5906                                
1   DeepGandalf 7 x64 1CPU  2464   +64  +45/=33/-26 59.13%   61.5/104
2   LCZero 0.10avx431 1CPU  2400   -64  +26/=33/-45 40.87%   42.5/104

104 games with weight 526 - 6167                                    
1   Deep Shredder 12 x64 1CPU  2800  +144  +63/=19/-22 69.71%   72.5/104
2   LCZero 0.10avx526 1CPU     2656  -144  +22/=19/-63 30.29%   31.5/104

104 games with weight 526 - 6167 and 4CPUs for LCZero
1 Deep Shredder 12 x64 1CPU 2800 +67 +45/=34/-25 59.62% 62.0/104
2 LCZero 0.10avx 526 4CPU 2733 -67 +25/=34/-45 40.38% 42.0/104


Werner
Admin

Anzahl der Beiträge : 877
Anmeldedatum : 22.10.13

https://cegt.forumieren.com

Nach oben Nach unten

LCZero Testmatch Empty LCZero testmatches with longer time controls

Beitrag von Werner Mo Dez 10, 2018 7:32 am

1   Deep Shredder 12 x64 1CPU  +272   111½1½111111½½01½111½½1111  21.5/26
2   LC0 0.19.1 w31918 4CPU     -272  000½0½000000½½10½000½½0000    4.5/26
and now the surprise, same openings but longer time oontrol and only 1CPU:

40 Züge in 40 min  2018 weights_run2_31969.pb.gz; 9141
1   Deep Shredder 12 x64 1CPU   +82   0½11½1½011½½½011½1½½½11100  16.0/26
2   LC0 0.19.1 1CPU             -82  1½00½0½100½½½100½0½½½00011   10.0/26 = 2718
seems 4CPU do still not work correct and endgames are Long....

40 Züge in 40 min  2018 weights_32247_9999.pb.gz; 9999
1   Lc0 0.20.0-rc2 1CPU        +110   1111½1½½½10½1½½0½½½11½½½½1  17.0/26 = 2910
2   Deep Shredder 12 x64 1CPU  -110  0000½0½½½01½0½½1½½½00½½½½0    9.0/26
this is a jump with about +200 points from Networks with 9000 points to networks with 10000 points! (on longer time controls).

40 Züge in 40 min  2019
1   RofChade 2.0 x64 1CPU 2992        +157   0½1½11½011½½1½11111½110½1½  18.5/26
2   LC0 0.20.1 32494-10810 1CPU 2835  -157  1½0½00½100½½0½00000½001½0½    7.5/26 not the awaited result!

40 Züge in 40 min  2019
1   RofChade 2.0 x64 1CPU 2992           ½½1½½10½100½½½½10½½½½½½01½  13.0/26  169.00
2   Lc0 0.20.1 11258-96x8-se 1CPU 2992  ½½0½½01½011½½½½01½½½½½½10½   13.0/26  169.00
this distilled network is very strong! (https://github.com/dkappe/leela-chess-weights/wiki/Distilled-Networks)

40 Züge in 10 min  2019
1   RofChade 2.0 x64 1CPU          +70   01½½1½½0½1½½½1½½½101½1½½101½½1½11½0  21.0/35
2   Lc0 0.20.1a 11258-112x9 1CPU   -70  10½½0½½1½0½½½0½½½010½0½½010½½0½00½1   14.0/35

40 Züge in 40 min  2019
1   RofChade 2.0 x64 1CPU          +40   ½½½11½0½0½10½11101½10½½1½½0½½½01½½1  19.5/35
2   Lc0 0.20.1a 11258-112x9 1CPU   -40  ½½½00½1½1½01½00010½01½½0½½1½½½10½½0   15.5/35
I am not longer sure that longer time controls give better results….

40 Züge in 40 min 2019
1 Lc0 0.20.1b 11258-112x9 1CPU +10 0½½½½01½0111½1½½0101½1½½½½1½0½000½1 18.0/35
2 Arasan 21.2 x64 1CPU -10 1½½½½10½1000½0½½1010½0½½½½0½1½111½0 17.0/35
result not ok - too less hash used


Werner
Admin

Anzahl der Beiträge : 877
Anmeldedatum : 22.10.13

https://cegt.forumieren.com

Nach oben Nach unten

LCZero Testmatch Empty Re: LCZero Testmatch

Beitrag von Gesponserte Inhalte


Gesponserte Inhalte


Nach oben Nach unten

Nach oben


 
Befugnisse in diesem Forum
Sie können in diesem Forum nicht antworten