Ablation L-MID-L3_xavier
Browse files
L/MID/L3_xavier/seed0/final_report.json
ADDED
|
@@ -0,0 +1,203 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config": {
|
| 3 |
+
"group": "L",
|
| 4 |
+
"variant": "L3_xavier",
|
| 5 |
+
"band": "MID",
|
| 6 |
+
"seed": 0,
|
| 7 |
+
"phase": 1,
|
| 8 |
+
"overrides": {
|
| 9 |
+
"init": "xavier_uniform"
|
| 10 |
+
},
|
| 11 |
+
"description": "L-MID-L3_xavier"
|
| 12 |
+
},
|
| 13 |
+
"run_config": {
|
| 14 |
+
"matrix_v": 64,
|
| 15 |
+
"D": 8,
|
| 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": 0.33634778696775336,
|
| 64 |
+
"cv_last": 0.333630403265162,
|
| 65 |
+
"predicted_band": "MID",
|
| 66 |
+
"expected_band": "MID",
|
| 67 |
+
"band_match": true,
|
| 68 |
+
"test_mse": 1.018330454826355,
|
| 69 |
+
"recon_ema": 1.01714110050452,
|
| 70 |
+
"S0": 3.4833648204803467,
|
| 71 |
+
"SD": 2.0846481323242188,
|
| 72 |
+
"ratio": 1.670960556728134,
|
| 73 |
+
"erank": 7.889909744262695,
|
| 74 |
+
"observed_sphere_cv": 0.3627779736367799,
|
| 75 |
+
"uniform_sphere_cv_prediction": 0.35679185069620456,
|
| 76 |
+
"band_deviation": 0.0059861229405753225,
|
| 77 |
+
"params_count": 183339,
|
| 78 |
+
"wallclock_seconds": 34.224008560180664,
|
| 79 |
+
"batches_completed": 1000,
|
| 80 |
+
"batch_limit": 1000,
|
| 81 |
+
"cv_trajectory": [
|
| 82 |
+
{
|
| 83 |
+
"batch": 0,
|
| 84 |
+
"cv": 0.33171946166262445,
|
| 85 |
+
"cv_ema": 0.33171946166262445,
|
| 86 |
+
"recon": 1.6609481573104858
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"batch": 50,
|
| 90 |
+
"cv": 0.3452369626710614,
|
| 91 |
+
"cv_ema": 0.3318546366727088,
|
| 92 |
+
"recon": 1.6308587789535522
|
| 93 |
+
},
|
| 94 |
+
{
|
| 95 |
+
"batch": 100,
|
| 96 |
+
"cv": 0.35610982543929126,
|
| 97 |
+
"cv_ema": 0.3320971885603747,
|
| 98 |
+
"recon": 1.329087257385254
|
| 99 |
+
},
|
| 100 |
+
{
|
| 101 |
+
"batch": 150,
|
| 102 |
+
"cv": 0.36983823392864357,
|
| 103 |
+
"cv_ema": 0.33247459901405735,
|
| 104 |
+
"recon": 1.0412639379501343
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"batch": 200,
|
| 108 |
+
"cv": 0.35326147746116116,
|
| 109 |
+
"cv_ema": 0.33268246779852834,
|
| 110 |
+
"recon": 1.0904600620269775
|
| 111 |
+
},
|
| 112 |
+
{
|
| 113 |
+
"batch": 250,
|
| 114 |
+
"cv": 0.33724323995469085,
|
| 115 |
+
"cv_ema": 0.33272807552009,
|
| 116 |
+
"recon": 1.1561626195907593
|
| 117 |
+
},
|
| 118 |
+
{
|
| 119 |
+
"batch": 300,
|
| 120 |
+
"cv": 0.32154728055134324,
|
| 121 |
+
"cv_ema": 0.3326162675704025,
|
| 122 |
+
"recon": 1.1401513814926147
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"batch": 350,
|
| 126 |
+
"cv": 0.34278750207344727,
|
| 127 |
+
"cv_ema": 0.332717979915433,
|
| 128 |
+
"recon": 1.1519372463226318
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"batch": 400,
|
| 132 |
+
"cv": 0.3412123653421996,
|
| 133 |
+
"cv_ema": 0.33280292376970066,
|
| 134 |
+
"recon": 0.9262759685516357
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"batch": 450,
|
| 138 |
+
"cv": 0.3412980567674915,
|
| 139 |
+
"cv_ema": 0.3328878750996786,
|
| 140 |
+
"recon": 1.0899786949157715
|
| 141 |
+
},
|
| 142 |
+
{
|
| 143 |
+
"batch": 500,
|
| 144 |
+
"cv": 0.36789982240114866,
|
| 145 |
+
"cv_ema": 0.33323799457269326,
|
| 146 |
+
"recon": 0.9714137315750122
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"batch": 550,
|
| 150 |
+
"cv": 0.36070938122888513,
|
| 151 |
+
"cv_ema": 0.3335127084392552,
|
| 152 |
+
"recon": 1.048335075378418
|
| 153 |
+
},
|
| 154 |
+
{
|
| 155 |
+
"batch": 600,
|
| 156 |
+
"cv": 0.36453206729808224,
|
| 157 |
+
"cv_ema": 0.3338229020278435,
|
| 158 |
+
"recon": 1.1499202251434326
|
| 159 |
+
},
|
| 160 |
+
{
|
| 161 |
+
"batch": 650,
|
| 162 |
+
"cv": 0.363773820969309,
|
| 163 |
+
"cv_ema": 0.33412241121725816,
|
| 164 |
+
"recon": 0.9600512385368347
|
| 165 |
+
},
|
| 166 |
+
{
|
| 167 |
+
"batch": 700,
|
| 168 |
+
"cv": 0.3947727357450692,
|
| 169 |
+
"cv_ema": 0.3347289144625363,
|
| 170 |
+
"recon": 0.9712020754814148
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"batch": 750,
|
| 174 |
+
"cv": 0.4056490974041256,
|
| 175 |
+
"cv_ema": 0.3354381162919522,
|
| 176 |
+
"recon": 1.0953559875488281
|
| 177 |
+
},
|
| 178 |
+
{
|
| 179 |
+
"batch": 800,
|
| 180 |
+
"cv": 0.38253870233266657,
|
| 181 |
+
"cv_ema": 0.33590912215235935,
|
| 182 |
+
"recon": 1.1096006631851196
|
| 183 |
+
},
|
| 184 |
+
{
|
| 185 |
+
"batch": 850,
|
| 186 |
+
"cv": 0.3728233252326867,
|
| 187 |
+
"cv_ema": 0.3362782641831626,
|
| 188 |
+
"recon": 1.079160213470459
|
| 189 |
+
},
|
| 190 |
+
{
|
| 191 |
+
"batch": 900,
|
| 192 |
+
"cv": 0.3459753746650658,
|
| 193 |
+
"cv_ema": 0.3363752352879816,
|
| 194 |
+
"recon": 0.9140113592147827
|
| 195 |
+
},
|
| 196 |
+
{
|
| 197 |
+
"batch": 950,
|
| 198 |
+
"cv": 0.333630403265162,
|
| 199 |
+
"cv_ema": 0.33634778696775336,
|
| 200 |
+
"recon": 0.9870057106018066
|
| 201 |
+
}
|
| 202 |
+
]
|
| 203 |
+
}
|