Hugging Face's logo

AbstractPhil
/
geolip-hypersphere-experiments

TensorBoard
Model card Files
xet
 Metrics Community
1
AbstractPhil commited on
Commit
fd243cb
·
verified ·
1 Parent(s): 3bc2d4b

Create 1_1_constellation_adapted_kymatio_projected_output.txt

Browse files
Files changed (1) hide show
  1. spectral/experiment_1/1_1_constellation_adapted_kymatio_projected_output.txt +213 -0
spectral/experiment_1/1_1_constellation_adapted_kymatio_projected_output.txt ADDED
@@ -0,0 +1,213 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ============================================================
2
+ GeoLIP Scattering Constellation — Autopsy-Informed
3
+ Scattering: kymatio J=2, L=8, order 2 → (B, 243, 8, 8)
4
+ BN(243) → FLATTEN(15552) → proj(512) → S^511
5
+ Constellation: 64 anchors on S^511
6
+ Patchwork: 8×64 = 512d
7
+ Activation: SquaredReLU
8
+ Loss: CE + InfoNCE + attract + CV(0.22) + spread
9
+ Optimizer: SGD lr=0.05, momentum=0.9, wd=5e-4
10
+ Batch: 128, Epochs: 90
11
+ Device: cuda
12
+ ============================================================
13
+ Train: 50,000 Val: 10,000
14
+ Scattering output: torch.Size([2, 243, 8, 8]) (5D=True)
15
+ Total params: 17,094,640
16
+ BN: 486
17
+ Projection: 16,454,144
18
+ Constellation+PW+Clf: 640,010
19
+
20
+ ============================================================
21
+ TRAINING — 90 epochs
22
+ SGD lr=0.05, step decay 5x every 20 epochs
23
+ Anchor push: every 50 batches, lr=0.1
24
+ ============================================================
25
+ E 1/90: 100%|██████████| 390/390 [00:24<00:00, 16.18b/s, acc=11%, anch=10/64, cos=0.585, loss=2.9910, nce=0.92, ordered=1, push=7]
26
+ E 1: train=11.4% val=18.1% loss=2.9910 nce=0.92 cos=0.443 cv=0.2253(✓) anch=15/64 push=7 (24s) ★
27
+ E 2/90: 100%|██████████| 390/390 [00:23<00:00, 16.73b/s, acc=36%, anch=20/64, cos=0.580, loss=1.9575, nce=1.00, ordered=1, push=15]
28
+ E 2: train=35.7% val=48.1% loss=1.9575 nce=1.00 cos=0.578 cv=0.2524(✗) anch=31/64 push=15 (23s) ★
29
+ E 3/90: 100%|██████████| 390/390 [00:23<00:00, 16.59b/s, acc=53%, anch=31/64, cos=0.609, loss=1.6051, nce=1.00, ordered=1, push=23]
30
+ E 3: train=52.7% val=59.1% loss=1.6051 nce=1.00 cos=0.593 cv=0.2190(✓) anch=53/64 push=23 (24s) ★
31
+ E 4/90: 100%|██████████| 390/390 [00:23<00:00, 16.51b/s, acc=58%, anch=37/64, cos=0.623, loss=1.4624, nce=1.00, ordered=1, push=31]
32
+ E 4: train=58.1% val=62.9% loss=1.4624 nce=1.00 cos=0.604 cv=0.2471(✓) anch=58/64 push=31 (24s) ★
33
+ E 5/90: 100%|██████████| 390/390 [00:23<00:00, 16.61b/s, acc=61%, anch=44/64, cos=0.627, loss=1.3767, nce=1.00, ordered=1, push=39]
34
+ E 5: train=61.4% val=66.4% loss=1.3767 nce=1.00 cos=0.616 cv=0.2336(✓) anch=60/64 push=39 (23s) ★
35
+ E 6/90: 100%|██████████| 390/390 [00:24<00:00, 16.25b/s, acc=64%, anch=43/64, cos=0.636, loss=1.3137, nce=1.00, ordered=1, push=46]
36
+ E 6: train=63.7% val=68.7% loss=1.3137 nce=1.00 cos=0.623 cv=0.2199(✓) anch=61/64 push=46 (24s) ★
37
+ E 7/90: 100%|██████████| 390/390 [00:24<00:00, 15.99b/s, acc=65%, anch=41/64, cos=0.638, loss=1.2652, nce=1.00, ordered=1, push=54]
38
+ E 7: train=65.3% val=66.8% loss=1.2652 nce=1.00 cos=0.627 cv=0.2298(✓) anch=60/64 push=54 (24s)
39
+ E 8/90: 100%|██████████| 390/390 [00:23<00:00, 16.66b/s, acc=67%, anch=44/64, cos=0.630, loss=1.2313, nce=1.00, ordered=1, push=62]
40
+ E 8: train=66.5% val=69.9% loss=1.2313 nce=1.00 cos=0.631 cv=0.2291(✓) anch=63/64 push=62 (23s) ★
41
+ E 9/90: 100%|██████████| 390/390 [00:23<00:00, 16.67b/s, acc=68%, anch=45/64, cos=0.650, loss=1.1982, nce=1.00, ordered=1, push=70]
42
+ E 9: train=67.6% val=70.8% loss=1.1982 nce=1.00 cos=0.632 cv=0.1920(✓) anch=62/64 push=70 (23s) ★
43
+ E 10/90: 100%|██████████| 390/390 [00:23<00:00, 16.68b/s, acc=68%, anch=42/64, cos=0.626, loss=1.1788, nce=1.00, ordered=1, push=78]
44
+ E 10: train=68.2% val=70.9% loss=1.1788 nce=1.00 cos=0.633 cv=0.2325(✓) anch=61/64 push=78 (23s) ★
45
+ E 11/90: 100%|██████████| 390/390 [00:23<00:00, 16.27b/s, acc=69%, anch=46/64, cos=0.629, loss=1.1529, nce=1.00, ordered=1, push=85]
46
+ E 11: train=69.2% val=72.6% loss=1.1529 nce=1.00 cos=0.634 cv=0.2451(✓) anch=61/64 push=85 (24s) ★
47
+ E 12/90: 100%|██████████| 390/390 [00:23<00:00, 16.53b/s, acc=70%, anch=51/64, cos=0.643, loss=1.1286, nce=1.00, ordered=1, push=93]
48
+ E 12: train=69.9% val=73.3% loss=1.1286 nce=1.00 cos=0.633 cv=0.2203(✓) anch=61/64 push=93 (24s) ★
49
+ E 13/90: 100%|██████████| 390/390 [00:24<00:00, 16.03b/s, acc=71%, anch=43/64, cos=0.611, loss=1.1052, nce=1.00, ordered=1, push=101]
50
+ E 13: train=70.8% val=73.0% loss=1.1052 nce=1.00 cos=0.631 cv=0.2340(✓) anch=59/64 push=101 (24s)
51
+ E 14/90: 100%|██████████| 390/390 [00:24<00:00, 16.19b/s, acc=71%, anch=46/64, cos=0.640, loss=1.0890, nce=1.00, ordered=1, push=109]
52
+ E 14: train=71.3% val=73.8% loss=1.0890 nce=1.00 cos=0.631 cv=0.2102(✓) anch=60/64 push=109 (24s) ★
53
+ E 15/90: 100%|██████████| 390/390 [00:23<00:00, 16.28b/s, acc=72%, anch=48/64, cos=0.626, loss=1.0676, nce=1.00, ordered=1, push=117]
54
+ E 15: train=72.0% val=72.0% loss=1.0676 nce=1.00 cos=0.631 cv=0.2125(✓) anch=55/64 push=117 (24s)
55
+ E 16/90: 100%|██████████| 390/390 [00:24<00:00, 16.03b/s, acc=73%, anch=48/64, cos=0.637, loss=1.0512, nce=1.00, ordered=1, push=124]
56
+ E 16: train=72.6% val=74.6% loss=1.0512 nce=1.00 cos=0.630 cv=0.2029(✓) anch=58/64 push=124 (24s) ★
57
+ E 17/90: 100%|██████████| 390/390 [00:24<00:00, 15.96b/s, acc=73%, anch=43/64, cos=0.634, loss=1.0435, nce=1.00, ordered=1, push=132]
58
+ E 17: train=72.8% val=75.4% loss=1.0435 nce=1.00 cos=0.629 cv=0.1959(✓) anch=59/64 push=132 (24s) ★
59
+ E 18/90: 100%|██████████| 390/390 [00:24<00:00, 15.85b/s, acc=73%, anch=44/64, cos=0.615, loss=1.0334, nce=1.00, ordered=1, push=140]
60
+ E 18: train=73.1% val=74.8% loss=1.0334 nce=1.00 cos=0.629 cv=0.2087(✓) anch=54/64 push=140 (25s)
61
+ E 19/90: 100%|██████████| 390/390 [00:23<00:00, 16.39b/s, acc=73%, anch=47/64, cos=0.623, loss=1.0244, nce=1.00, ordered=1, push=148]
62
+ E 19: train=73.3% val=75.5% loss=1.0244 nce=1.00 cos=0.628 cv=0.2193(✓) anch=53/64 push=148 (24s) ★
63
+ E 20/90: 100%|██████████| 390/390 [00:23<00:00, 16.58b/s, acc=74%, anch=43/64, cos=0.621, loss=1.0147, nce=1.00, ordered=1, push=156]
64
+ E 20: train=73.8% val=74.5% loss=1.0147 nce=1.00 cos=0.629 cv=0.1961(✓) anch=54/64 push=156 (24s)
65
+ E 21/90: 100%|██████████| 390/390 [00:23<00:00, 16.64b/s, acc=78%, anch=46/64, cos=0.643, loss=0.8636, nce=1.00, ordered=1, push=163]
66
+ E 21: train=78.5% val=78.4% loss=0.8636 nce=1.00 cos=0.634 cv=0.1963(✓) anch=52/64 push=163 (23s) ★
67
+ E 22/90: 100%|██████████| 390/390 [00:23<00:00, 16.88b/s, acc=80%, anch=43/64, cos=0.640, loss=0.8297, nce=1.00, ordered=1, push=171]
68
+ E 22: train=79.6% val=78.8% loss=0.8297 nce=1.00 cos=0.629 cv=0.1907(✓) anch=53/64 push=171 (23s) ★
69
+ E 23/90: 100%|██████████| 390/390 [00:23<00:00, 16.40b/s, acc=80%, anch=45/64, cos=0.629, loss=0.8118, nce=1.00, ordered=1, push=179]
70
+ E 23: train=80.5% val=79.2% loss=0.8118 nce=1.00 cos=0.625 cv=0.2046(✓) anch=55/64 push=179 (24s) ★
71
+ E 24/90: 100%|██████████| 390/390 [00:23<00:00, 16.81b/s, acc=81%, anch=44/64, cos=0.623, loss=0.7979, nce=1.00, ordered=1, push=187]
72
+ E 24: train=81.1% val=79.4% loss=0.7979 nce=1.00 cos=0.623 cv=0.2197(✓) anch=59/64 push=187 (23s) ★
73
+ E 25/90: 100%|██████████| 390/390 [00:23<00:00, 16.72b/s, acc=81%, anch=44/64, cos=0.606, loss=0.7901, nce=1.00, ordered=1, push=195]
74
+ E 25: train=81.4% val=79.4% loss=0.7901 nce=1.00 cos=0.622 cv=0.2064(✓) anch=59/64 push=195 (23s) ★
75
+ E 26/90: 100%|██████████| 390/390 [00:23<00:00, 16.60b/s, acc=82%, anch=39/64, cos=0.617, loss=0.7819, nce=1.00, ordered=1, push=202]
76
+ E 26: train=81.6% val=78.9% loss=0.7819 nce=1.00 cos=0.623 cv=0.1903(✓) anch=57/64 push=202 (23s)
77
+ E 27/90: 100%|██████████| 390/390 [00:24<00:00, 15.94b/s, acc=82%, anch=40/64, cos=0.623, loss=0.7722, nce=1.00, ordered=1, push=210]
78
+ E 27: train=81.9% val=79.9% loss=0.7722 nce=1.00 cos=0.622 cv=0.2022(✓) anch=58/64 push=210 (24s) ★
79
+ E 28/90: 100%|██████████| 390/390 [00:23<00:00, 16.79b/s, acc=82%, anch=46/64, cos=0.617, loss=0.7687, nce=1.00, ordered=1, push=218]
80
+ E 28: train=82.0% val=79.2% loss=0.7687 nce=1.00 cos=0.621 cv=0.1964(✓) anch=59/64 push=218 (23s)
81
+ E 29/90: 100%|██████████| 390/390 [00:23<00:00, 16.65b/s, acc=82%, anch=43/64, cos=0.621, loss=0.7600, nce=1.00, ordered=1, push=226]
82
+ E 29: train=82.4% val=79.9% loss=0.7600 nce=1.00 cos=0.621 cv=0.2047(✓) anch=60/64 push=226 (23s)
83
+ E 30/90: 100%|██████████| 390/390 [00:23<00:00, 16.75b/s, acc=83%, anch=42/64, cos=0.614, loss=0.7508, nce=1.00, ordered=1, push=234]
84
+ E 30: train=82.8% val=79.8% loss=0.7508 nce=1.00 cos=0.620 cv=0.1812(✓) anch=63/64 push=234 (23s)
85
+ E 31/90: 100%|██████████| 390/390 [00:24<00:00, 15.81b/s, acc=83%, anch=39/64, cos=0.608, loss=0.7491, nce=1.00, ordered=1, push=241]
86
+ E 31: train=82.7% val=80.0% loss=0.7491 nce=1.00 cos=0.620 cv=0.1851(✓) anch=63/64 push=241 (25s) ★
87
+ E 32/90: 100%|██████████| 390/390 [00:24<00:00, 15.91b/s, acc=83%, anch=41/64, cos=0.611, loss=0.7433, nce=1.00, ordered=1, push=249]
88
+ E 32: train=82.8% val=79.8% loss=0.7433 nce=1.00 cos=0.620 cv=0.2039(✓) anch=64/64 push=249 (25s)
89
+ E 33/90: 100%|██████████| 390/390 [00:23<00:00, 16.28b/s, acc=83%, anch=37/64, cos=0.619, loss=0.7354, nce=1.00, ordered=1, push=257]
90
+ E 33: train=83.2% val=79.7% loss=0.7354 nce=1.00 cos=0.619 cv=0.1954(✓) anch=61/64 push=257 (24s)
91
+ E 34/90: 100%|██████████| 390/390 [00:24<00:00, 16.22b/s, acc=84%, anch=42/64, cos=0.616, loss=0.7313, nce=1.00, ordered=1, push=265]
92
+ E 34: train=83.6% val=80.0% loss=0.7313 nce=1.00 cos=0.619 cv=0.2142(✓) anch=63/64 push=265 (24s)
93
+ E 35/90: 100%|██████████| 390/390 [00:23<00:00, 16.63b/s, acc=84%, anch=37/64, cos=0.628, loss=0.7243, nce=1.00, ordered=1, push=273]
94
+ E 35: train=83.7% val=79.8% loss=0.7243 nce=1.00 cos=0.619 cv=0.1853(✓) anch=62/64 push=273 (23s)
95
+ E 36/90: 100%|██████████| 390/390 [00:23<00:00, 16.55b/s, acc=84%, anch=38/64, cos=0.624, loss=0.7218, nce=1.00, ordered=1, push=280]
96
+ E 36: train=83.7% val=80.5% loss=0.7218 nce=1.00 cos=0.618 cv=0.2050(✓) anch=63/64 push=280 (24s) ★
97
+ E 37/90: 100%|██████████| 390/390 [00:24<00:00, 16.15b/s, acc=84%, anch=40/64, cos=0.623, loss=0.7178, nce=1.00, ordered=1, push=288]
98
+ E 37: train=84.0% val=80.1% loss=0.7178 nce=1.00 cos=0.618 cv=0.1927(✓) anch=63/64 push=288 (24s)
99
+ E 38/90: 100%|██████████| 390/390 [00:24<00:00, 16.01b/s, acc=84%, anch=35/64, cos=0.618, loss=0.7151, nce=1.00, ordered=1, push=296]
100
+ E 38: train=83.9% val=80.2% loss=0.7151 nce=1.00 cos=0.619 cv=0.1881(✓) anch=63/64 push=296 (24s)
101
+ E 39/90: 100%|██████████| 390/390 [00:24<00:00, 16.23b/s, acc=84%, anch=37/64, cos=0.616, loss=0.7109, nce=1.00, ordered=1, push=304]
102
+ E 39: train=84.2% val=80.1% loss=0.7109 nce=1.00 cos=0.618 cv=0.2067(✓) anch=62/64 push=304 (24s)
103
+ E 40/90: 100%|██████████| 390/390 [00:23<00:00, 16.54b/s, acc=84%, anch=43/64, cos=0.613, loss=0.7073, nce=1.00, ordered=1, push=312]
104
+ E 40: train=84.2% val=80.2% loss=0.7073 nce=1.00 cos=0.617 cv=0.1900(✓) anch=63/64 push=312 (24s)
105
+ E 41/90: 100%|██████████| 390/390 [00:23<00:00, 16.41b/s, acc=86%, anch=33/64, cos=0.612, loss=0.6508, nce=1.00, ordered=1, push=319]
106
+ E 41: train=86.1% val=80.9% loss=0.6508 nce=1.00 cos=0.620 cv=0.1901(✓) anch=64/64 push=319 (24s) ★
107
+ E 42/90: 100%|██████████| 390/390 [00:23<00:00, 16.36b/s, acc=87%, anch=31/64, cos=0.642, loss=0.6345, nce=1.00, ordered=1, push=327]
108
+ E 42: train=86.8% val=81.1% loss=0.6345 nce=1.00 cos=0.622 cv=0.2207(✓) anch=63/64 push=327 (24s) ★
109
+ E 43/90: 100%|██████████| 390/390 [00:24<00:00, 16.10b/s, acc=87%, anch=35/64, cos=0.625, loss=0.6283, nce=1.00, ordered=1, push=335]
110
+ E 43: train=86.9% val=81.2% loss=0.6283 nce=1.00 cos=0.623 cv=0.2033(✓) anch=63/64 push=335 (24s) ★
111
+ E 44/90: 100%|██████████| 390/390 [00:24<00:00, 15.94b/s, acc=87%, anch=25/64, cos=0.630, loss=0.6269, nce=1.00, ordered=1, push=343]
112
+ E 44: train=86.9% val=81.5% loss=0.6269 nce=1.00 cos=0.624 cv=0.1830(✓) anch=63/64 push=343 (24s) ★
113
+ E 45/90: 100%|██████████| 390/390 [00:24<00:00, 16.09b/s, acc=87%, anch=21/64, cos=0.637, loss=0.6188, nce=1.00, ordered=1, push=351]
114
+ E 45: train=87.4% val=81.3% loss=0.6188 nce=1.00 cos=0.625 cv=0.1910(✓) anch=61/64 push=351 (24s)
115
+ E 46/90: 100%|██████████| 390/390 [00:24<00:00, 16.18b/s, acc=87%, anch=29/64, cos=0.606, loss=0.6182, nce=1.00, ordered=1, push=358]
116
+ E 46: train=87.3% val=81.8% loss=0.6182 nce=1.00 cos=0.624 cv=0.1909(✓) anch=61/64 push=358 (24s) ★
117
+ E 47/90: 100%|██████████| 390/390 [00:24<00:00, 16.07b/s, acc=88%, anch=28/64, cos=0.625, loss=0.6115, nce=1.00, ordered=1, push=366]
118
+ E 47: train=87.7% val=81.6% loss=0.6115 nce=1.00 cos=0.625 cv=0.1990(✓) anch=62/64 push=366 (24s)
119
+ E 48/90: 100%|██████████| 390/390 [00:23<00:00, 16.37b/s, acc=88%, anch=28/64, cos=0.611, loss=0.6109, nce=1.00, ordered=1, push=374]
120
+ E 48: train=87.6% val=81.4% loss=0.6109 nce=1.00 cos=0.625 cv=0.1917(✓) anch=60/64 push=374 (24s)
121
+ E 49/90: 100%|██████████| 390/390 [00:24<00:00, 16.09b/s, acc=88%, anch=29/64, cos=0.633, loss=0.6050, nce=1.00, ordered=1, push=382]
122
+ E 49: train=87.8% val=81.6% loss=0.6050 nce=1.00 cos=0.625 cv=0.1901(✓) anch=61/64 push=382 (24s)
123
+ E 50/90: 100%|██████████| 390/390 [00:24<00:00, 16.12b/s, acc=88%, anch=24/64, cos=0.628, loss=0.6046, nce=1.00, ordered=1, push=390]
124
+ E 50: train=87.8% val=81.4% loss=0.6046 nce=1.00 cos=0.625 cv=0.1782(✗) anch=60/64 push=390 (24s)
125
+ E 51/90: 100%|██████████| 390/390 [00:23<00:00, 16.77b/s, acc=88%, anch=25/64, cos=0.618, loss=0.6004, nce=1.00, ordered=1, push=397]
126
+ E 51: train=87.8% val=81.2% loss=0.6004 nce=1.00 cos=0.625 cv=0.1873(✓) anch=61/64 push=397 (23s)
127
+ E 52/90: 100%|██████████| 390/390 [00:23<00:00, 16.46b/s, acc=88%, anch=26/64, cos=0.618, loss=0.6003, nce=1.00, ordered=1, push=405]
128
+ E 52: train=88.0% val=81.5% loss=0.6003 nce=1.00 cos=0.625 cv=0.1935(✓) anch=63/64 push=405 (24s)
129
+ E 53/90: 100%|██████████| 390/390 [00:23<00:00, 16.54b/s, acc=88%, anch=29/64, cos=0.617, loss=0.5968, nce=1.00, ordered=1, push=413]
130
+ E 53: train=88.2% val=81.3% loss=0.5968 nce=1.00 cos=0.624 cv=0.1905(✓) anch=62/64 push=413 (24s)
131
+ E 54/90: 100%|██████████| 390/390 [00:23<00:00, 16.33b/s, acc=88%, anch=26/64, cos=0.617, loss=0.5901, nce=1.00, ordered=1, push=421]
132
+ E 54: train=88.3% val=81.8% loss=0.5901 nce=1.00 cos=0.625 cv=0.1917(✓) anch=62/64 push=421 (24s) ★
133
+ E 55/90: 100%|██████████| 390/390 [00:24<00:00, 16.10b/s, acc=88%, anch=31/64, cos=0.623, loss=0.5934, nce=1.00, ordered=1, push=429]
134
+ E 55: train=88.1% val=81.4% loss=0.5934 nce=1.00 cos=0.625 cv=0.1939(✓) anch=61/64 push=429 (24s)
135
+ E 56/90: 100%|██████████| 390/390 [00:24<00:00, 16.22b/s, acc=88%, anch=23/64, cos=0.630, loss=0.5866, nce=1.00, ordered=1, push=436]
136
+ E 56: train=88.5% val=81.5% loss=0.5866 nce=1.00 cos=0.626 cv=0.1983(✓) anch=62/64 push=436 (24s)
137
+ E 57/90: 100%|██████████| 390/390 [00:23<00:00, 16.30b/s, acc=88%, anch=27/64, cos=0.632, loss=0.5821, nce=1.00, ordered=1, push=444]
138
+ E 57: train=88.5% val=81.6% loss=0.5821 nce=1.00 cos=0.625 cv=0.1756(✗) anch=62/64 push=444 (24s)
139
+ E 58/90: 100%|██████████| 390/390 [00:24<00:00, 16.05b/s, acc=88%, anch=29/64, cos=0.620, loss=0.5856, nce=1.00, ordered=1, push=452]
140
+ E 58: train=88.4% val=81.5% loss=0.5856 nce=1.00 cos=0.626 cv=0.1938(✓) anch=62/64 push=452 (24s)
141
+ E 59/90: 100%|██████████| 390/390 [00:24<00:00, 16.07b/s, acc=89%, anch=24/64, cos=0.624, loss=0.5828, nce=1.00, ordered=1, push=460]
142
+ E 59: train=88.5% val=81.6% loss=0.5828 nce=1.00 cos=0.625 cv=0.1837(✓) anch=60/64 push=460 (24s)
143
+ E 60/90: 100%|██████████| 390/390 [00:23<00:00, 16.67b/s, acc=89%, anch=32/64, cos=0.613, loss=0.5814, nce=1.00, ordered=1, push=468]
144
+ E 60: train=88.5% val=81.5% loss=0.5814 nce=1.00 cos=0.624 cv=0.2070(✓) anch=63/64 push=468 (23s)
145
+ E 61/90: 100%|██████████| 390/390 [00:23<00:00, 16.26b/s, acc=89%, anch=26/64, cos=0.631, loss=0.5664, nce=1.00, ordered=1, push=475]
146
+ E 61: train=89.1% val=81.6% loss=0.5664 nce=1.00 cos=0.625 cv=0.1843(✓) anch=60/64 push=475 (24s)
147
+ E 62/90: 100%|██████████| 390/390 [00:23<00:00, 16.85b/s, acc=89%, anch=28/64, cos=0.635, loss=0.5647, nce=1.00, ordered=1, push=483]
148
+ E 62: train=89.2% val=81.8% loss=0.5647 nce=1.00 cos=0.625 cv=0.1873(✓) anch=59/64 push=483 (23s)
149
+ E 63/90: 100%|██████████| 390/390 [00:23<00:00, 16.72b/s, acc=89%, anch=28/64, cos=0.629, loss=0.5649, nce=1.00, ordered=1, push=491]
150
+ E 63: train=89.2% val=81.8% loss=0.5649 nce=1.00 cos=0.625 cv=0.1950(✓) anch=62/64 push=491 (23s)
151
+ E 64/90: 100%|██████████| 390/390 [00:24<00:00, 16.20b/s, acc=89%, anch=27/64, cos=0.619, loss=0.5579, nce=1.00, ordered=1, push=499]
152
+ E 64: train=89.3% val=81.7% loss=0.5579 nce=1.00 cos=0.626 cv=0.2015(✓) anch=58/64 push=499 (24s)
153
+ E 65/90: 100%|██████████| 390/390 [00:23<00:00, 16.64b/s, acc=89%, anch=32/64, cos=0.620, loss=0.5610, nce=1.00, ordered=1, push=507]
154
+ E 65: train=89.3% val=81.7% loss=0.5610 nce=1.00 cos=0.626 cv=0.1996(✓) anch=60/64 push=507 (23s)
155
+ E 66/90: 100%|██████████| 390/390 [00:23<00:00, 16.57b/s, acc=89%, anch=26/64, cos=0.619, loss=0.5583, nce=1.00, ordered=1, push=514]
156
+ E 66: train=89.3% val=81.7% loss=0.5583 nce=1.00 cos=0.626 cv=0.1914(✓) anch=62/64 push=514 (24s)
157
+ E 67/90: 100%|██████████| 390/390 [00:23<00:00, 16.66b/s, acc=89%, anch=28/64, cos=0.622, loss=0.5559, nce=1.00, ordered=1, push=522]
158
+ E 67: train=89.4% val=81.7% loss=0.5559 nce=1.00 cos=0.626 cv=0.2022(✓) anch=63/64 push=522 (23s)
159
+ E 68/90: 100%|██████████| 390/390 [00:24<00:00, 16.09b/s, acc=89%, anch=32/64, cos=0.610, loss=0.5560, nce=1.00, ordered=1, push=530]
160
+ E 68: train=89.5% val=81.7% loss=0.5560 nce=1.00 cos=0.626 cv=0.1947(✓) anch=61/64 push=530 (24s)
161
+ E 69/90: 100%|██████████| 390/390 [00:23<00:00, 16.85b/s, acc=89%, anch=33/64, cos=0.631, loss=0.5556, nce=1.00, ordered=1, push=538]
162
+ E 69: train=89.5% val=81.7% loss=0.5556 nce=1.00 cos=0.626 cv=0.1793(✗) anch=64/64 push=538 (23s)
163
+ E 70/90: 100%|██████████| 390/390 [00:23<00:00, 16.37b/s, acc=89%, anch=21/64, cos=0.623, loss=0.5577, nce=1.00, ordered=1, push=546]
164
+ E 70: train=89.5% val=81.7% loss=0.5577 nce=1.00 cos=0.627 cv=0.1937(✓) anch=64/64 push=546 (24s)
165
+ E 71/90: 100%|██████████| 390/390 [00:23<00:00, 16.40b/s, acc=89%, anch=28/64, cos=0.630, loss=0.5544, nce=1.00, ordered=1, push=553]
166
+ E 71: train=89.4% val=81.8% loss=0.5544 nce=1.00 cos=0.626 cv=0.1829(✓) anch=63/64 push=553 (24s) ★
167
+ E 72/90: 100%|██████████| 390/390 [00:23<00:00, 16.30b/s, acc=90%, anch=31/64, cos=0.634, loss=0.5526, nce=1.00, ordered=1, push=561]
168
+ E 72: train=89.6% val=81.8% loss=0.5526 nce=1.00 cos=0.627 cv=0.1855(✓) anch=61/64 push=561 (24s)
169
+ E 73/90: 100%|██████████| 390/390 [00:23<00:00, 16.59b/s, acc=90%, anch=25/64, cos=0.621, loss=0.5520, nce=1.00, ordered=1, push=569]
170
+ E 73: train=89.7% val=81.8% loss=0.5520 nce=1.00 cos=0.627 cv=0.1864(✓) anch=60/64 push=569 (24s)
171
+ E 74/90: 100%|██████████| 390/390 [00:23<00:00, 16.71b/s, acc=90%, anch=28/64, cos=0.630, loss=0.5493, nce=1.00, ordered=1, push=577]
172
+ E 74: train=89.6% val=81.8% loss=0.5493 nce=1.00 cos=0.627 cv=0.1882(✓) anch=63/64 push=577 (23s)
173
+ E 75/90: 100%|██████████| 390/390 [00:24<00:00, 16.04b/s, acc=90%, anch=28/64, cos=0.638, loss=0.5469, nce=1.00, ordered=1, push=585]
174
+ E 75: train=89.8% val=81.6% loss=0.5469 nce=1.00 cos=0.627 cv=0.1791(✗) anch=60/64 push=585 (24s)
175
+ E 76/90: 100%|██████████| 390/390 [00:23<00:00, 16.40b/s, acc=90%, anch=26/64, cos=0.626, loss=0.5497, nce=1.00, ordered=1, push=592]
176
+ E 76: train=89.7% val=81.8% loss=0.5497 nce=1.00 cos=0.627 cv=0.1965(✓) anch=63/64 push=592 (24s)
177
+ E 77/90: 100%|██████████| 390/390 [00:23<00:00, 16.60b/s, acc=90%, anch=27/64, cos=0.609, loss=0.5500, nce=1.00, ordered=1, push=600]
178
+ E 77: train=89.7% val=81.6% loss=0.5500 nce=1.00 cos=0.627 cv=0.1740(✗) anch=61/64 push=600 (23s)
179
+ E 78/90: 100%|██████████| 390/390 [00:23<00:00, 16.44b/s, acc=90%, anch=31/64, cos=0.628, loss=0.5518, nce=1.00, ordered=1, push=608]
180
+ E 78: train=89.6% val=81.9% loss=0.5518 nce=1.00 cos=0.627 cv=0.1878(✓) anch=61/64 push=608 (24s) ★
181
+ E 79/90: 100%|██████████| 390/390 [00:24<00:00, 16.16b/s, acc=90%, anch=28/64, cos=0.634, loss=0.5478, nce=1.00, ordered=1, push=616]
182
+ E 79: train=90.0% val=81.7% loss=0.5478 nce=1.00 cos=0.627 cv=0.1895(✓) anch=62/64 push=616 (24s)
183
+ E 80/90: 100%|██████████| 390/390 [00:23<00:00, 16.91b/s, acc=90%, anch=29/64, cos=0.621, loss=0.5501, nce=1.00, ordered=1, push=624]
184
+ E 80: train=89.7% val=81.8% loss=0.5501 nce=1.00 cos=0.627 cv=0.2013(✓) anch=63/64 push=624 (23s)
185
+ E 81/90: 100%|██████████| 390/390 [00:24<00:00, 16.11b/s, acc=90%, anch=28/64, cos=0.630, loss=0.5469, nce=1.00, ordered=1, push=631]
186
+ E 81: train=89.7% val=81.9% loss=0.5469 nce=1.00 cos=0.627 cv=0.1833(✓) anch=63/64 push=631 (24s)
187
+ E 82/90: 100%|██████████| 390/390 [00:24<00:00, 16.16b/s, acc=90%, anch=31/64, cos=0.620, loss=0.5448, nce=1.00, ordered=1, push=639]
188
+ E 82: train=89.8% val=81.8% loss=0.5448 nce=1.00 cos=0.627 cv=0.1781(✗) anch=62/64 push=639 (24s)
189
+ E 83/90: 100%|██████████| 390/390 [00:23<00:00, 16.77b/s, acc=90%, anch=22/64, cos=0.629, loss=0.5433, nce=1.00, ordered=1, push=647]
190
+ E 83: train=90.0% val=81.8% loss=0.5433 nce=1.00 cos=0.627 cv=0.1843(✓) anch=62/64 push=647 (23s)
191
+ E 84/90: 100%|██████████| 390/390 [00:23<00:00, 16.26b/s, acc=90%, anch=26/64, cos=0.625, loss=0.5440, nce=1.00, ordered=1, push=655]
192
+ E 84: train=89.8% val=81.7% loss=0.5440 nce=1.00 cos=0.627 cv=0.1974(✓) anch=63/64 push=655 (24s)
193
+ E 85/90: 100%|██████████| 390/390 [00:23<00:00, 16.25b/s, acc=90%, anch=24/64, cos=0.627, loss=0.5428, nce=1.00, ordered=1, push=663]
194
+ E 85: train=89.9% val=81.8% loss=0.5428 nce=1.00 cos=0.627 cv=0.1937(✓) anch=61/64 push=663 (24s)
195
+ E 86/90: 100%|██████████| 390/390 [00:24<00:00, 16.02b/s, acc=90%, anch=26/64, cos=0.636, loss=0.5450, nce=1.00, ordered=1, push=670]
196
+ E 86: train=89.8% val=81.7% loss=0.5450 nce=1.00 cos=0.627 cv=0.1844(✓) anch=62/64 push=670 (24s)
197
+ E 87/90: 100%|██████████| 390/390 [00:23<00:00, 16.47b/s, acc=90%, anch=27/64, cos=0.619, loss=0.5460, nce=1.00, ordered=1, push=678]
198
+ E 87: train=89.7% val=81.8% loss=0.5460 nce=1.00 cos=0.627 cv=0.1832(✓) anch=63/64 push=678 (24s)
199
+ E 88/90: 100%|██████████| 390/390 [00:24<00:00, 15.91b/s, acc=90%, anch=26/64, cos=0.615, loss=0.5447, nce=1.00, ordered=1, push=686]
200
+ E 88: train=89.9% val=81.5% loss=0.5447 nce=1.00 cos=0.627 cv=0.1934(✓) anch=59/64 push=686 (25s)
201
+ E 89/90: 100%|██████████| 390/390 [00:23<00:00, 16.44b/s, acc=90%, anch=27/64, cos=0.610, loss=0.5431, nce=1.00, ordered=1, push=694]
202
+ E 89: train=89.9% val=81.7% loss=0.5431 nce=1.00 cos=0.627 cv=0.1902(✓) anch=63/64 push=694 (24s)
203
+ E 90/90: 100%|██████████| 390/390 [00:24<00:00, 16.04b/s, acc=90%, anch=31/64, cos=0.614, loss=0.5434, nce=1.00, ordered=1, push=702]
204
+ E 90: train=89.9% val=81.8% loss=0.5434 nce=1.00 cos=0.627 cv=0.2057(✓) anch=60/64 push=702 (24s)
205
+
206
+ Best val accuracy: 81.9%
207
+ Total params: 17,094,640
208
+ Baseline (BN+linear): 70.8%
209
+ Target: >70.8% (constellation must add value over linear)
210
+
211
+ ============================================================
212
+ DONE
213
+ ============================================================