Finished
KovaxLeela-dag-previewdebl1M-k2hyb-100kdiff10.0+0.10LLR: 2.94 (-2.94, 2.94) [0.00, 3.00]
Games: 9740 W: 2267 L: 2088 D: 5385
Ptnml(0-2): 0, 971, 2753, 1142, 4
k2hyb recheck, as this net played vs 100k instead of 1M
KovaxLeela-dag-previewpolicy-blenddiff10.0+0.10LLR: 2.96 (-2.94, 2.94) [0.00, 3.00]
Games: 8412 W: 1970 L: 1798 D: 4644
Ptnml(0-2): 1, 775, 2483, 945, 2
Optimistic blend vs Master post softmax.
KovaxLeela-dag-previewpolicy-blenddiff10.0+0.10Tuning 6 Parameters
3612/12500 Iterations
57794/200000 Games Played
Tune policy blending on T1 sharp net.
KovaxLeela-dag-previewmasterdiffN=1000Elo: 11.91 +- 1.76 (95%) [N=40000]
Games: 40014 W: 10593 L: 9222 D: 20199
Ptnml(0-2): 181, 3889, 10566, 5120, 251
check how much PST and FPU is shift alone gains with the sharp target corpus.
KovaxLeela-dag-previewpolicy-blenddiffN=1000Elo: 57.21 +- 2.44 (95%) [N=40000]
Games: 21086 W: 7049 L: 3608 D: 10429
Ptnml(0-2): 30, 1176, 4936, 4125, 276
small node count initial check on sharp policy target + policy head blending tune resutls
KovaxLeela-dag-previewpolicy-blenddiffN=1000Tuning 6 Parameters
3000/3000 Iterations
192016/192000 Games Played
Policy blend tune on fix nodes
KovaxLeela-dag-previewmasterdiff10.0+0.10Tuning 2 Parameters
1000/1000 Iterations
16000/16000 Games Played
cleanvisit2 corpus less sharp than previous
KovaxLeela-dag-previewt1-cleanvisit2-polhead-25kdiff10.0+0.10LLR: 2.96 (-2.94, 2.94) [0.00, 3.00]
Games: 9518 W: 2167 L: 1986 D: 5365
Ptnml(0-2): 2, 963, 2646, 1148, 0
KovaxLeela-dag-previewdebl1M-cleanvisit-100k-muondiff40.0+0.40LLR: 0.31 (-2.94, 2.94) [0.00, 3.00]
Games: 2262 W: 503 L: 482 D: 1277
Ptnml(0-2): 0, 233, 644, 254, 0
higher pst
KovaxLeela-dag-previewdebl1M-cleanvisit-100k-muondiff40.0+0.40LLR: 0.30 (-2.94, 2.94) [0.00, 3.00]
Games: 2214 W: 503 L: 482 D: 1229
Ptnml(0-2): 0, 235, 616, 256, 0
scaling test
KovaxLeela-dag-previewdebl1M-cleanvisit-100k-muondiff10.0+0.10LLR: 2.96 (-2.94, 2.94) [0.00, 3.00]
Games: 5960 W: 1424 L: 1248 D: 3288
Ptnml(0-2): 1, 586, 1633, 756, 4
Muon vs Adam ELO comparison vs base
KovaxLeela-dag-previewmasterdiff10.0+0.10Tuning 2 Parameters
500/500 Iterations
8000/8000 Games Played
Muon PST,fpu further tune
KovaxLeela-dag-previewdebl1M-cleanvisit-100kdiff40.0+0.40LLR: 2.95 (-2.94, 2.94) [0.00, 3.00]
Games: 7314 W: 1671 L: 1505 D: 4138
Ptnml(0-2): 0, 642, 2207, 808, 0
KovaxLeela-dag-previewdebl1M-cleanvisit-100kdiff10.0+0.10LLR: 2.96 (-2.94, 2.94) [0.00, 3.00]
Games: 9206 W: 2153 L: 1975 D: 5078
Ptnml(0-2): 0, 903, 2626, 1067, 7
KovaxLeela-dag-previewmasterdiff10.0+0.10Tuning 2 Parameters
1000/1000 Iterations
16000/16000 Games Played
KovaxLeela-dag-previewmasterdiff10.0+0.10Tuning 2 Parameters
1000/1000 Iterations
16000/16000 Games Played
KovaxLeela-dag-previewt1-cleanvisit1.3Mgames-100kdiff10.0+0.10LLR: 2.95 (-2.94, 2.94) [0.00, 3.00]
Games: 20772 W: 4688 L: 4485 D: 11599
Ptnml(0-2): 2, 2184, 5822, 2365, 13
KovaxLeela-dag-previewmasterdiff10.0+0.10Tuning 2 Parameters
784/1000 Iterations
12550/16000 Games Played
KovaxLeela-dag-previewdebl1M-k2hyb-100kdiff40.0+0.40LLR: -0.08 (-2.94, 2.94) [0.00, 3.00]
Games: 128 W: 26 L: 30 D: 72
Ptnml(0-2): 0, 14, 40, 10, 0
KovaxLeela-dag-previewdebl1M-k2hyb-100kdiff10.0+0.10LLR: 1.81 (-2.94, 2.94) [0.00, 3.00]
Games: 1896 W: 452 L: 351 D: 1093
Ptnml(0-2): 0, 154, 539, 255, 0
KovaxLeela-dag-previewdebl1M-k2hyb-100kdiff10.0+0.10LLR: 0.47 (-2.94, 2.94) [0.00, 3.00]
Games: 720 W: 191 L: 165 D: 364
Ptnml(0-2): 0, 60, 214, 86, 0
KovaxLeela-dag-previewdebl1M-unfrozen-k2hyb-25kdiff10.0+0.10LLR: 2.95 (-2.94, 2.94) [0.00, 3.00]
Games: 3982 W: 959 L: 787 D: 2236
Ptnml(0-2): 1, 367, 1083, 539, 1
KovaxLeela-dag-previewdebl1M-hpol-k2hyb-25kdiff40.0+0.40LLR: 2.94 (-2.94, 2.94) [0.00, 3.00]
Games: 1000 W: 338 L: 100 D: 562
Ptnml(0-2): 0, 27, 208, 265, 0
KovaxLeela-dag-previewT1-256x-10-rl-base-2.5M-policy25k-t91diff40.0+0.40LLR: 0.28 (-2.94, 2.94) [0.00, 3.00]
Games: 1292 W: 261 L: 245 D: 786
Ptnml(0-2): 0, 108, 414, 124, 0
KovaxLeela-dag-previewdebl1M-hpol-grill9-25kdiff40.0+0.40LLR: 2.26 (-2.94, 2.94) [0.00, 3.00]
Games: 1664 W: 441 L: 310 D: 913
Ptnml(0-2): 1, 126, 446, 259, 0