Ablation B-MID-B3_structured
Browse files
B/MID/B3_structured/seed0/final_report.json
ADDED
|
@@ -0,0 +1,198 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config": {
|
| 3 |
+
"group": "B",
|
| 4 |
+
"variant": "B3_structured",
|
| 5 |
+
"band": "MID",
|
| 6 |
+
"seed": 0,
|
| 7 |
+
"phase": 1,
|
| 8 |
+
"overrides": {
|
| 9 |
+
"noise_types": [
|
| 10 |
+
3,
|
| 11 |
+
4,
|
| 12 |
+
5,
|
| 13 |
+
11,
|
| 14 |
+
13
|
| 15 |
+
]
|
| 16 |
+
},
|
| 17 |
+
"description": "B-MID-B3_structured"
|
| 18 |
+
},
|
| 19 |
+
"run_config": {
|
| 20 |
+
"matrix_v": 64,
|
| 21 |
+
"D": 8,
|
| 22 |
+
"patch_size": 16,
|
| 23 |
+
"hidden": 64,
|
| 24 |
+
"depth": 1,
|
| 25 |
+
"n_cross_layers": 1,
|
| 26 |
+
"n_heads": 4,
|
| 27 |
+
"max_alpha": 0.2,
|
| 28 |
+
"alpha_init": -2.0,
|
| 29 |
+
"img_size": 64,
|
| 30 |
+
"batch_size": 128,
|
| 31 |
+
"lr": 0.0001,
|
| 32 |
+
"epochs": 1,
|
| 33 |
+
"weight_decay": 0.0,
|
| 34 |
+
"use_cv_ema": true,
|
| 35 |
+
"cv_ema_alpha": 0.01,
|
| 36 |
+
"cv_alignment_epochs": 0,
|
| 37 |
+
"cv_measure_every": 50,
|
| 38 |
+
"cv_sigma_scale": 0.3,
|
| 39 |
+
"boost": 0.5,
|
| 40 |
+
"cross_attn_clip": 0.5,
|
| 41 |
+
"allowed_types": [
|
| 42 |
+
3,
|
| 43 |
+
4,
|
| 44 |
+
5,
|
| 45 |
+
11,
|
| 46 |
+
13
|
| 47 |
+
],
|
| 48 |
+
"train_size": 1000000,
|
| 49 |
+
"val_size": 10000,
|
| 50 |
+
"num_workers": 2,
|
| 51 |
+
"report_every": 100,
|
| 52 |
+
"major_report_every": 10,
|
| 53 |
+
"save_every": 5,
|
| 54 |
+
"seed": 0,
|
| 55 |
+
"hf_repo": "AbstractPhil/geolip-svae-batteries",
|
| 56 |
+
"upload": false
|
| 57 |
+
},
|
| 58 |
+
"cv_ema_final": 0.34668383538245,
|
| 59 |
+
"cv_last": 0.3552471149291473,
|
| 60 |
+
"predicted_band": "MID",
|
| 61 |
+
"expected_band": "MID",
|
| 62 |
+
"band_match": true,
|
| 63 |
+
"test_mse": 0.7076205015182495,
|
| 64 |
+
"recon_ema": 0.7156597306622593,
|
| 65 |
+
"S0": 3.4786195755004883,
|
| 66 |
+
"SD": 2.1352601051330566,
|
| 67 |
+
"ratio": 1.629131528682032,
|
| 68 |
+
"erank": 7.9002275466918945,
|
| 69 |
+
"observed_sphere_cv": 0.3502177286616089,
|
| 70 |
+
"uniform_sphere_cv_prediction": 0.35679185069620456,
|
| 71 |
+
"band_deviation": -0.006574122034595675,
|
| 72 |
+
"params_count": 183339,
|
| 73 |
+
"wallclock_seconds": 35.102540493011475,
|
| 74 |
+
"batches_completed": 1000,
|
| 75 |
+
"batch_limit": 1000,
|
| 76 |
+
"cv_trajectory": [
|
| 77 |
+
{
|
| 78 |
+
"batch": 0,
|
| 79 |
+
"cv": 0.3444089512087921,
|
| 80 |
+
"cv_ema": 0.3444089512087921,
|
| 81 |
+
"recon": 1.6098403930664062
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"batch": 50,
|
| 85 |
+
"cv": 0.3479154294050981,
|
| 86 |
+
"cv_ema": 0.3444440159907552,
|
| 87 |
+
"recon": 1.848254680633545
|
| 88 |
+
},
|
| 89 |
+
{
|
| 90 |
+
"batch": 100,
|
| 91 |
+
"cv": 0.3581772123750436,
|
| 92 |
+
"cv_ema": 0.34458134795459805,
|
| 93 |
+
"recon": 1.306191325187683
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"batch": 150,
|
| 97 |
+
"cv": 0.3706876541684821,
|
| 98 |
+
"cv_ema": 0.3448424110167369,
|
| 99 |
+
"recon": 1.1080111265182495
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"batch": 200,
|
| 103 |
+
"cv": 0.3578740946417933,
|
| 104 |
+
"cv_ema": 0.34497272785298744,
|
| 105 |
+
"recon": 0.9397088885307312
|
| 106 |
+
},
|
| 107 |
+
{
|
| 108 |
+
"batch": 250,
|
| 109 |
+
"cv": 0.3779429916415793,
|
| 110 |
+
"cv_ema": 0.34530243049087334,
|
| 111 |
+
"recon": 1.019446849822998
|
| 112 |
+
},
|
| 113 |
+
{
|
| 114 |
+
"batch": 300,
|
| 115 |
+
"cv": 0.3395249685330748,
|
| 116 |
+
"cv_ema": 0.3452446558712953,
|
| 117 |
+
"recon": 0.5680596232414246
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"batch": 350,
|
| 121 |
+
"cv": 0.3509648631414789,
|
| 122 |
+
"cv_ema": 0.34530185794399715,
|
| 123 |
+
"recon": 0.6875750422477722
|
| 124 |
+
},
|
| 125 |
+
{
|
| 126 |
+
"batch": 400,
|
| 127 |
+
"cv": 0.3697870560722546,
|
| 128 |
+
"cv_ema": 0.34554670992527975,
|
| 129 |
+
"recon": 0.5586971044540405
|
| 130 |
+
},
|
| 131 |
+
{
|
| 132 |
+
"batch": 450,
|
| 133 |
+
"cv": 0.36749439147604024,
|
| 134 |
+
"cv_ema": 0.34576618674078735,
|
| 135 |
+
"recon": 0.565224826335907
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"batch": 500,
|
| 139 |
+
"cv": 0.33349194259729187,
|
| 140 |
+
"cv_ema": 0.3456434442993524,
|
| 141 |
+
"recon": 0.6544479727745056
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"batch": 550,
|
| 145 |
+
"cv": 0.3751535063499351,
|
| 146 |
+
"cv_ema": 0.3459385449198582,
|
| 147 |
+
"recon": 0.6961528062820435
|
| 148 |
+
},
|
| 149 |
+
{
|
| 150 |
+
"batch": 600,
|
| 151 |
+
"cv": 0.3724817396326667,
|
| 152 |
+
"cv_ema": 0.3462039768669863,
|
| 153 |
+
"recon": 0.6672663688659668
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"batch": 650,
|
| 157 |
+
"cv": 0.3793632653309241,
|
| 158 |
+
"cv_ema": 0.34653556975162564,
|
| 159 |
+
"recon": 0.6253349781036377
|
| 160 |
+
},
|
| 161 |
+
{
|
| 162 |
+
"batch": 700,
|
| 163 |
+
"cv": 0.3479098005762332,
|
| 164 |
+
"cv_ema": 0.3465493120598717,
|
| 165 |
+
"recon": 0.6695920825004578
|
| 166 |
+
},
|
| 167 |
+
{
|
| 168 |
+
"batch": 750,
|
| 169 |
+
"cv": 0.322112884964632,
|
| 170 |
+
"cv_ema": 0.3463049477889193,
|
| 171 |
+
"recon": 0.6097941398620605
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"batch": 800,
|
| 175 |
+
"cv": 0.3550754120588618,
|
| 176 |
+
"cv_ema": 0.34639265243161876,
|
| 177 |
+
"recon": 0.7665531635284424
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"batch": 850,
|
| 181 |
+
"cv": 0.3429315539760714,
|
| 182 |
+
"cv_ema": 0.3463580414470633,
|
| 183 |
+
"recon": 0.6992430090904236
|
| 184 |
+
},
|
| 185 |
+
{
|
| 186 |
+
"batch": 900,
|
| 187 |
+
"cv": 0.3702876576658421,
|
| 188 |
+
"cv_ema": 0.34659733760925104,
|
| 189 |
+
"recon": 0.4272443652153015
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"batch": 950,
|
| 193 |
+
"cv": 0.3552471149291473,
|
| 194 |
+
"cv_ema": 0.34668383538245,
|
| 195 |
+
"recon": 0.7216233015060425
|
| 196 |
+
}
|
| 197 |
+
]
|
| 198 |
+
}
|