AbstractPhil commited on
Commit
91c7a47
·
verified ·
1 Parent(s): 147ad82

Ablation L-LOW-L2_kaiming

Browse files
L/LOW/L2_kaiming/seed0/final_report.json ADDED
@@ -0,0 +1,203 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "config": {
3
+ "group": "L",
4
+ "variant": "L2_kaiming",
5
+ "band": "LOW",
6
+ "seed": 0,
7
+ "phase": 1,
8
+ "overrides": {
9
+ "init": "kaiming_normal"
10
+ },
11
+ "description": "L-LOW-L2_kaiming"
12
+ },
13
+ "run_config": {
14
+ "matrix_v": 64,
15
+ "D": 16,
16
+ "patch_size": 16,
17
+ "hidden": 64,
18
+ "depth": 1,
19
+ "n_cross_layers": 1,
20
+ "n_heads": 4,
21
+ "max_alpha": 0.2,
22
+ "alpha_init": -2.0,
23
+ "img_size": 64,
24
+ "batch_size": 128,
25
+ "lr": 0.0001,
26
+ "epochs": 1,
27
+ "weight_decay": 0.0,
28
+ "use_cv_ema": true,
29
+ "cv_ema_alpha": 0.01,
30
+ "cv_alignment_epochs": 0,
31
+ "cv_measure_every": 50,
32
+ "cv_sigma_scale": 0.3,
33
+ "boost": 0.5,
34
+ "cross_attn_clip": 0.5,
35
+ "allowed_types": [
36
+ 0,
37
+ 1,
38
+ 2,
39
+ 3,
40
+ 4,
41
+ 5,
42
+ 6,
43
+ 7,
44
+ 8,
45
+ 9,
46
+ 10,
47
+ 11,
48
+ 12,
49
+ 13,
50
+ 14,
51
+ 15
52
+ ],
53
+ "train_size": 1000000,
54
+ "val_size": 10000,
55
+ "num_workers": 2,
56
+ "report_every": 100,
57
+ "major_report_every": 10,
58
+ "save_every": 5,
59
+ "seed": 0,
60
+ "hf_repo": "AbstractPhil/geolip-svae-batteries",
61
+ "upload": false
62
+ },
63
+ "cv_ema_final": 0.20268048310081777,
64
+ "cv_last": 0.18835150554378466,
65
+ "predicted_band": "LOW",
66
+ "expected_band": "LOW",
67
+ "band_match": true,
68
+ "test_mse": 1.2421412467956543,
69
+ "recon_ema": 1.270170715314491,
70
+ "S0": 2.765903949737549,
71
+ "SD": 1.1466048955917358,
72
+ "ratio": 2.4122554650244856,
73
+ "erank": 15.508450508117676,
74
+ "observed_sphere_cv": 0.201073350909783,
75
+ "uniform_sphere_cv_prediction": 0.19902003258265127,
76
+ "band_deviation": 0.002053318327131737,
77
+ "params_count": 250211,
78
+ "wallclock_seconds": 14.140987873077393,
79
+ "batches_completed": 1000,
80
+ "batch_limit": 1000,
81
+ "cv_trajectory": [
82
+ {
83
+ "batch": 0,
84
+ "cv": 0.2029791691088925,
85
+ "cv_ema": 0.2029791691088925,
86
+ "recon": 4.668613433837891
87
+ },
88
+ {
89
+ "batch": 50,
90
+ "cv": 0.21648394724882528,
91
+ "cv_ema": 0.2031142168902918,
92
+ "recon": 1.945159673690796
93
+ },
94
+ {
95
+ "batch": 100,
96
+ "cv": 0.1893611434481257,
97
+ "cv_ema": 0.2029766861558701,
98
+ "recon": 1.7791846990585327
99
+ },
100
+ {
101
+ "batch": 150,
102
+ "cv": 0.19954824400231091,
103
+ "cv_ema": 0.20294240173433453,
104
+ "recon": 1.8498338460922241
105
+ },
106
+ {
107
+ "batch": 200,
108
+ "cv": 0.20560903661816743,
109
+ "cv_ema": 0.20296906808317286,
110
+ "recon": 1.5357303619384766
111
+ },
112
+ {
113
+ "batch": 250,
114
+ "cv": 0.18203405358260166,
115
+ "cv_ema": 0.20275971793816716,
116
+ "recon": 1.6213998794555664
117
+ },
118
+ {
119
+ "batch": 300,
120
+ "cv": 0.21207436985849626,
121
+ "cv_ema": 0.20285286445737047,
122
+ "recon": 1.3911848068237305
123
+ },
124
+ {
125
+ "batch": 350,
126
+ "cv": 0.20268154998098964,
127
+ "cv_ema": 0.20285115131260667,
128
+ "recon": 1.5657209157943726
129
+ },
130
+ {
131
+ "batch": 400,
132
+ "cv": 0.21679204020223014,
133
+ "cv_ema": 0.20299056020150288,
134
+ "recon": 1.4704399108886719
135
+ },
136
+ {
137
+ "batch": 450,
138
+ "cv": 0.19471272207631984,
139
+ "cv_ema": 0.20290778182025107,
140
+ "recon": 1.3864983320236206
141
+ },
142
+ {
143
+ "batch": 500,
144
+ "cv": 0.20892353833497654,
145
+ "cv_ema": 0.2029679393853983,
146
+ "recon": 1.5340595245361328
147
+ },
148
+ {
149
+ "batch": 550,
150
+ "cv": 0.20015249370345908,
151
+ "cv_ema": 0.20293978492857892,
152
+ "recon": 1.3921970129013062
153
+ },
154
+ {
155
+ "batch": 600,
156
+ "cv": 0.18468106461683198,
157
+ "cv_ema": 0.20275719772546144,
158
+ "recon": 1.3303794860839844
159
+ },
160
+ {
161
+ "batch": 650,
162
+ "cv": 0.20355613178878587,
163
+ "cv_ema": 0.20276518706609467,
164
+ "recon": 1.257645606994629
165
+ },
166
+ {
167
+ "batch": 700,
168
+ "cv": 0.208737057486876,
169
+ "cv_ema": 0.2028249057703025,
170
+ "recon": 1.3407225608825684
171
+ },
172
+ {
173
+ "batch": 750,
174
+ "cv": 0.2158772252760689,
175
+ "cv_ema": 0.20295542896536015,
176
+ "recon": 1.1453542709350586
177
+ },
178
+ {
179
+ "batch": 800,
180
+ "cv": 0.2100931847930035,
181
+ "cv_ema": 0.20302680652363658,
182
+ "recon": 1.2153438329696655
183
+ },
184
+ {
185
+ "batch": 850,
186
+ "cv": 0.18833138964317195,
187
+ "cv_ema": 0.20287985235483194,
188
+ "recon": 1.3827879428863525
189
+ },
190
+ {
191
+ "batch": 900,
192
+ "cv": 0.19741664165748732,
193
+ "cv_ema": 0.2028252202478585,
194
+ "recon": 1.2571738958358765
195
+ },
196
+ {
197
+ "batch": 950,
198
+ "cv": 0.18835150554378466,
199
+ "cv_ema": 0.20268048310081777,
200
+ "recon": 1.2900829315185547
201
+ }
202
+ ]
203
+ }