Try   HackMD

Run 2024-12-25

You could get the weights from here.

  • 2024-12-25-16:46
    • Start the training.
    • The version is v0.8.1.
    • learning rate = 5e-3
    • batch size = 256
    • The network size is 6bx96c.
    • Set the visits as 16.

  • 2024-12-26-03:58
    • Played 155k games.
    • Accumulate around
      0.0808×108       20bx256c eval queries.
    • Set the visits as 50 (from 16).

  • 2024-12-27-14:57
    • Played 365k games.
    • Accumulate around
      0.4184×108       20bx256c eval queries.
    • The strengh is better than Leela Zero with 91 weights. Elo different is +62.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w 091.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 635 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          203             171          374  (58.90%)
​​​​​​​​Leela Zero 0.17        146             115          261  (41.10%)

  • 2024-12-28-14:48
    • Played 500k games.
    • Accumulate around
      0.6396×108       20bx256c eval queries.
    • The strengh is better than Leela Zero with 92 weights. Elo different is +89.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w 092.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 729 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          213             243          456  (62.55%)
​​​​​​​​Leela Zero 0.17        122             151          273  (37.45%)

  • 2024-12-29-19:46
    • Played 665k games.
    • Accumulate around
      0.9077×108       20bx256c eval queries.
    • The strengh is better than Leela Zero with 105 weights. Elo different is +19.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w 105.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 535 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          131             151          282  (52.71%)
​​​​​​​​Leela Zero 0.17        116             137          253  (47.29%)

  • 2024-12-30-16:04
    • Played 780k games.
    • Accumulate around
      1.0931x10×108       20bx256c eval queries.
    • Halt 6bx96c training.
    • The strengh is better than Leela Zero with 110 weights. Elo different is +14.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w 110.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 590 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          161             146          307  (52.03%)
​​​​​​​​Leela Zero 0.17        149             134          283  (47.97%)
    • The strengh is as same as Leela Zero with 116 weights. Elo different is +2.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w 110.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 590 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          202             241          443  (50.23%)
​​​​​​​​Leela Zero 0.17        201             238          439  (49.77%)

  • 2025-1-2-03:46
    • Start 10bx128c training
    • Soft weights = 1.0

  • 2025-1-3-00:01
    • Played 865k games.
    • Accumulate around
      1.4977×108       20bx256c eval queries.
    • The strengh is as same as Leela Zero with 117 weights. Elo different is +13.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w 117.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 632 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          160             168          328  (51.90%)
​​​​​​​​Leela Zero 0.17        148             156          304  (48.10%)

  • 2025-1-4-20:42
    • Played 1055k games.
    • learning rate = 2.5e-3
    • Set the visits as 150 (from 50).

  • 2025-1-5-14:49
    • Played 1095k games.
    • Accumulate around
      2.8773×108       20bx256c eval queries.
    • The strengh is as same as Leela Zero with 130 weights. Elo different is +14.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w 130.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 600 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          136             176          312  (52.00%)
​​​​​​​​Leela Zero 0.17        124             164          288  (48.00%)

  • 2025-1-8-03:02
    • Played 1230k games.
    • Accumulate around
      5.0774×108       20bx256c eval queries.
    • The strengh is as same as Leela Zero with 140 weights. Elo different is +5.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w 140.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 527 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          125             142          267  (50.66%)
​​​​​​​​Leela Zero 0.17        120             140          260  (49.34%)

  • 2025-1-13-16:04
    • Played 1530k games.
    • Accumulate around
      7.9209x10×108       20bx256c eval queries.
    • Halt 10bx128c training.
    • The strengh is better than Leela Zero with 145 weights. Elo different is +77.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w 145.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 681 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          192             223          415  (60.94%)
​​​​​​​​Leela Zero 0.17        117             149          266  (39.06%)
    • The strengh is worse than Leela Zero with 150 weights. Elo different is -25.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w 150.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 656 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          147             157          304  (60.94%)
​​​​​​​​Leela Zero 0.17        171             181          352  (39.06%)

  • 2025-1-18-14:00
    • Start 15bx192c training
    • Soft weights = 4.0
    • learning rate = 2.5e-3

  • 2025-2-1-7:00
    • Played 2100k games.
    • learning rate = 1.25e-3 (from 2.5e-3)

  • 2025-2-1-21:42
    • Played 2030k games.
    • Accumulate around
      2.70146×109       20bx256c eval queries.
    • The strengh is as same as Leela Zero with ELFv0 weights. Elo different is +9.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w elfv0.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 814 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          182             236          267  (51.35%)
​​​​​​​​Leela Zero 0.17        171             225          260  (48.65%)

  • 2024-2-10-14:15
    • Played 2330k games
    • Full Visits = 400 (from 150)
    • Fast Visits = 125 (from 50)
    • Gumbel Visits = 55 (from 32)
    • Resign Visits = 75 (from 55)

  • 2024-2-17-11:20
    • Played 2480k games.
    • Accumulate around
      4.9194×109       20bx256c eval queries.
    • Strength should be between ELFv0 and ELFv1.
    • Halt 15bx192c training.

  • 2024-2-25-1:36
    • Start 20bx256c training.
    • learning rate = 6.25e-4
    • The 20bx256c pretrain model is slightly stronger than ELFv2. Win-rate is 53.85% (played 882 games).

  • 2024-3-2 2 15:45
    • Played 2570k games.
    • Accumulate around
      6.4240×109       20bx256c eval queries.
    • The strengh is better than Leela Zero with ELFv2 weights. Elo different is +48.
      • Sayuri: -w current_weights -t 1 -b 1 -p 800 --cpuct-init 0.5 --use-optimistic-policy --random-moves-factor 0.1 --random-moves-temp 0.8
      • Leela Zero: -w elfv2.gz --noponder -v 800 -g -t 1 -b 1 --timemanage off --randomcnt 30 --randomtemp 0.8
      • Game result (played 709 games with Leela Zero):
      ​​​​​​​​Name             black won       white won       total (win-rate)
​​​​​​​​
​​​​​​​​Sayrui v0.8.1          173             230          403  (56.84%)
​​​​​​​​Leela Zero 0.17        124             182          306  (43.16%)
Last changed by 
 
CGLemon
0
270
Published on HackMD