Add EMP Claim2 multi-arch and ResNet18 artifacts with SHA-256
Browse files- results/CHECKSUMS.sha256 +4 -0
- results/emp_loss_bound.json +405 -0
- results/emp_loss_bound_multiarch.json +1498 -0
- results/emp_resnet18_synth_loss.json +419 -0
- results/emp_weight_stats.json +1676 -0
results/CHECKSUMS.sha256
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
31ae03b88eed6c0927e97a4df5e7e70782748889bff3195ca12f043868695014 emp_loss_bound_multiarch.json
|
| 2 |
+
6c51cc1eb6040d9626487f05e35d6b78cf4f1c7805869c2e88a53607770f5ee9 emp_resnet18_synth_loss.json
|
| 3 |
+
a02d78588f0fe9bb0af87a18ecf8e67776fb8227196392d9b768bc945087c920 emp_loss_bound.json
|
| 4 |
+
7f7e814f1d5ad5b4f8b3922b736adaedbad8f335ad947c64994de4ca2a7334c0 emp_weight_stats.json
|
results/emp_loss_bound.json
ADDED
|
@@ -0,0 +1,405 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"claim": 2,
|
| 3 |
+
"description": "Theorem 4.2: loss change after EMP pruning is bounded",
|
| 4 |
+
"model": "4-layer MLP (784->2048->1024->512->10, Xavier init)",
|
| 5 |
+
"n_samples": 500,
|
| 6 |
+
"seeds": [
|
| 7 |
+
0,
|
| 8 |
+
1,
|
| 9 |
+
2
|
| 10 |
+
],
|
| 11 |
+
"betas": [
|
| 12 |
+
0.25,
|
| 13 |
+
0.5,
|
| 14 |
+
0.75,
|
| 15 |
+
1.0,
|
| 16 |
+
1.25,
|
| 17 |
+
1.5,
|
| 18 |
+
2.0
|
| 19 |
+
],
|
| 20 |
+
"n_total_configs": 21,
|
| 21 |
+
"n_loss_bound_holds": 21,
|
| 22 |
+
"all_bounds_hold": true,
|
| 23 |
+
"bound_formula": "|\u0394L_MSE| \u2264 (C_lip * ||\u0394W||_F * sigma_x)^2 / n_samples",
|
| 24 |
+
"rows": [
|
| 25 |
+
{
|
| 26 |
+
"seed": 0,
|
| 27 |
+
"beta": 0.25,
|
| 28 |
+
"n_eff": 2649414.0,
|
| 29 |
+
"nu": 662353,
|
| 30 |
+
"s_eff": 0.37194,
|
| 31 |
+
"mass_removed": 0.62806,
|
| 32 |
+
"mass_bound_term": 0.842545,
|
| 33 |
+
"mass_bound_holds": true,
|
| 34 |
+
"delta_w_norm": 36.4881,
|
| 35 |
+
"w_norm": 56.5946,
|
| 36 |
+
"delta_w_frac": 0.644728,
|
| 37 |
+
"loss_change_mse": 0.163182,
|
| 38 |
+
"loss_bound": 297630.813,
|
| 39 |
+
"loss_bound_holds": true,
|
| 40 |
+
"C_lip_orig": 11.9171,
|
| 41 |
+
"sigma_x": 28.0544
|
| 42 |
+
},
|
| 43 |
+
{
|
| 44 |
+
"seed": 0,
|
| 45 |
+
"beta": 0.5,
|
| 46 |
+
"n_eff": 2649414.0,
|
| 47 |
+
"nu": 1324707,
|
| 48 |
+
"s_eff": 0.605295,
|
| 49 |
+
"mass_removed": 0.394705,
|
| 50 |
+
"mass_bound_term": 0.686498,
|
| 51 |
+
"mass_bound_holds": true,
|
| 52 |
+
"delta_w_norm": 25.0368,
|
| 53 |
+
"w_norm": 56.5946,
|
| 54 |
+
"delta_w_frac": 0.442388,
|
| 55 |
+
"loss_change_mse": 0.080525,
|
| 56 |
+
"loss_bound": 140130.2778,
|
| 57 |
+
"loss_bound_holds": true,
|
| 58 |
+
"C_lip_orig": 11.9171,
|
| 59 |
+
"sigma_x": 28.0544
|
| 60 |
+
},
|
| 61 |
+
{
|
| 62 |
+
"seed": 0,
|
| 63 |
+
"beta": 0.75,
|
| 64 |
+
"n_eff": 2649414.0,
|
| 65 |
+
"nu": 1987060,
|
| 66 |
+
"s_eff": 0.772446,
|
| 67 |
+
"mass_removed": 0.227554,
|
| 68 |
+
"mass_bound_term": 0.530215,
|
| 69 |
+
"mass_bound_holds": true,
|
| 70 |
+
"delta_w_norm": 16.2952,
|
| 71 |
+
"w_norm": 56.5946,
|
| 72 |
+
"delta_w_frac": 0.287929,
|
| 73 |
+
"loss_change_mse": 0.03736,
|
| 74 |
+
"loss_bound": 59360.2562,
|
| 75 |
+
"loss_bound_holds": true,
|
| 76 |
+
"C_lip_orig": 11.9171,
|
| 77 |
+
"sigma_x": 28.0544
|
| 78 |
+
},
|
| 79 |
+
{
|
| 80 |
+
"seed": 0,
|
| 81 |
+
"beta": 1.0,
|
| 82 |
+
"n_eff": 2649414.0,
|
| 83 |
+
"nu": 2649414,
|
| 84 |
+
"s_eff": 0.889327,
|
| 85 |
+
"mass_removed": 0.110673,
|
| 86 |
+
"mass_bound_term": 0.373845,
|
| 87 |
+
"mass_bound_holds": true,
|
| 88 |
+
"delta_w_norm": 9.3942,
|
| 89 |
+
"w_norm": 56.5946,
|
| 90 |
+
"delta_w_frac": 0.16599,
|
| 91 |
+
"loss_change_mse": 0.014338,
|
| 92 |
+
"loss_bound": 19728.3001,
|
| 93 |
+
"loss_bound_holds": true,
|
| 94 |
+
"C_lip_orig": 11.9171,
|
| 95 |
+
"sigma_x": 28.0544
|
| 96 |
+
},
|
| 97 |
+
{
|
| 98 |
+
"seed": 0,
|
| 99 |
+
"beta": 1.25,
|
| 100 |
+
"n_eff": 2649414.0,
|
| 101 |
+
"nu": 3311767,
|
| 102 |
+
"s_eff": 0.963046,
|
| 103 |
+
"mass_removed": 0.036954,
|
| 104 |
+
"mass_bound_term": 0.217426,
|
| 105 |
+
"mass_bound_holds": true,
|
| 106 |
+
"delta_w_norm": 4.102,
|
| 107 |
+
"w_norm": 56.5946,
|
| 108 |
+
"delta_w_frac": 0.072481,
|
| 109 |
+
"loss_change_mse": 0.002979,
|
| 110 |
+
"loss_bound": 3761.5689,
|
| 111 |
+
"loss_bound_holds": true,
|
| 112 |
+
"C_lip_orig": 11.9171,
|
| 113 |
+
"sigma_x": 28.0544
|
| 114 |
+
},
|
| 115 |
+
{
|
| 116 |
+
"seed": 0,
|
| 117 |
+
"beta": 1.5,
|
| 118 |
+
"n_eff": 2649414.0,
|
| 119 |
+
"nu": 3974121,
|
| 120 |
+
"s_eff": 0.997109,
|
| 121 |
+
"mass_removed": 0.002891,
|
| 122 |
+
"mass_bound_term": 0.060971,
|
| 123 |
+
"mass_bound_holds": true,
|
| 124 |
+
"delta_w_norm": 0.6053,
|
| 125 |
+
"w_norm": 56.5946,
|
| 126 |
+
"delta_w_frac": 0.010695,
|
| 127 |
+
"loss_change_mse": 7.5e-05,
|
| 128 |
+
"loss_bound": 81.9039,
|
| 129 |
+
"loss_bound_holds": true,
|
| 130 |
+
"C_lip_orig": 11.9171,
|
| 131 |
+
"sigma_x": 28.0544
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"seed": 0,
|
| 135 |
+
"beta": 2.0,
|
| 136 |
+
"n_eff": 2649414.0,
|
| 137 |
+
"nu": 4232191,
|
| 138 |
+
"s_eff": 1.0,
|
| 139 |
+
"mass_removed": 0.0,
|
| 140 |
+
"mass_bound_term": 0.0,
|
| 141 |
+
"mass_bound_holds": true,
|
| 142 |
+
"delta_w_norm": 0.0,
|
| 143 |
+
"w_norm": 56.5946,
|
| 144 |
+
"delta_w_frac": 0.0,
|
| 145 |
+
"loss_change_mse": 0.0,
|
| 146 |
+
"loss_bound": 0.0,
|
| 147 |
+
"loss_bound_holds": true,
|
| 148 |
+
"C_lip_orig": 11.9171,
|
| 149 |
+
"sigma_x": 28.0544
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"seed": 1,
|
| 153 |
+
"beta": 0.25,
|
| 154 |
+
"n_eff": 2649414.0,
|
| 155 |
+
"nu": 662353,
|
| 156 |
+
"s_eff": 0.37194,
|
| 157 |
+
"mass_removed": 0.62806,
|
| 158 |
+
"mass_bound_term": 0.842545,
|
| 159 |
+
"mass_bound_holds": true,
|
| 160 |
+
"delta_w_norm": 36.4881,
|
| 161 |
+
"w_norm": 56.5946,
|
| 162 |
+
"delta_w_frac": 0.644728,
|
| 163 |
+
"loss_change_mse": 0.16558,
|
| 164 |
+
"loss_bound": 296339.3515,
|
| 165 |
+
"loss_bound_holds": true,
|
| 166 |
+
"C_lip_orig": 11.9171,
|
| 167 |
+
"sigma_x": 27.9935
|
| 168 |
+
},
|
| 169 |
+
{
|
| 170 |
+
"seed": 1,
|
| 171 |
+
"beta": 0.5,
|
| 172 |
+
"n_eff": 2649414.0,
|
| 173 |
+
"nu": 1324707,
|
| 174 |
+
"s_eff": 0.605295,
|
| 175 |
+
"mass_removed": 0.394705,
|
| 176 |
+
"mass_bound_term": 0.686498,
|
| 177 |
+
"mass_bound_holds": true,
|
| 178 |
+
"delta_w_norm": 25.0368,
|
| 179 |
+
"w_norm": 56.5946,
|
| 180 |
+
"delta_w_frac": 0.442388,
|
| 181 |
+
"loss_change_mse": 0.081001,
|
| 182 |
+
"loss_bound": 139522.233,
|
| 183 |
+
"loss_bound_holds": true,
|
| 184 |
+
"C_lip_orig": 11.9171,
|
| 185 |
+
"sigma_x": 27.9935
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"seed": 1,
|
| 189 |
+
"beta": 0.75,
|
| 190 |
+
"n_eff": 2649414.0,
|
| 191 |
+
"nu": 1987060,
|
| 192 |
+
"s_eff": 0.772446,
|
| 193 |
+
"mass_removed": 0.227554,
|
| 194 |
+
"mass_bound_term": 0.530215,
|
| 195 |
+
"mass_bound_holds": true,
|
| 196 |
+
"delta_w_norm": 16.2952,
|
| 197 |
+
"w_norm": 56.5946,
|
| 198 |
+
"delta_w_frac": 0.287929,
|
| 199 |
+
"loss_change_mse": 0.038302,
|
| 200 |
+
"loss_bound": 59102.6838,
|
| 201 |
+
"loss_bound_holds": true,
|
| 202 |
+
"C_lip_orig": 11.9171,
|
| 203 |
+
"sigma_x": 27.9935
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"seed": 1,
|
| 207 |
+
"beta": 1.0,
|
| 208 |
+
"n_eff": 2649414.0,
|
| 209 |
+
"nu": 2649414,
|
| 210 |
+
"s_eff": 0.889327,
|
| 211 |
+
"mass_removed": 0.110673,
|
| 212 |
+
"mass_bound_term": 0.373845,
|
| 213 |
+
"mass_bound_holds": true,
|
| 214 |
+
"delta_w_norm": 9.3942,
|
| 215 |
+
"w_norm": 56.5946,
|
| 216 |
+
"delta_w_frac": 0.16599,
|
| 217 |
+
"loss_change_mse": 0.013751,
|
| 218 |
+
"loss_bound": 19642.6962,
|
| 219 |
+
"loss_bound_holds": true,
|
| 220 |
+
"C_lip_orig": 11.9171,
|
| 221 |
+
"sigma_x": 27.9935
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"seed": 1,
|
| 225 |
+
"beta": 1.25,
|
| 226 |
+
"n_eff": 2649414.0,
|
| 227 |
+
"nu": 3311767,
|
| 228 |
+
"s_eff": 0.963046,
|
| 229 |
+
"mass_removed": 0.036954,
|
| 230 |
+
"mass_bound_term": 0.217426,
|
| 231 |
+
"mass_bound_holds": true,
|
| 232 |
+
"delta_w_norm": 4.102,
|
| 233 |
+
"w_norm": 56.5946,
|
| 234 |
+
"delta_w_frac": 0.072481,
|
| 235 |
+
"loss_change_mse": 0.002912,
|
| 236 |
+
"loss_bound": 3745.2469,
|
| 237 |
+
"loss_bound_holds": true,
|
| 238 |
+
"C_lip_orig": 11.9171,
|
| 239 |
+
"sigma_x": 27.9935
|
| 240 |
+
},
|
| 241 |
+
{
|
| 242 |
+
"seed": 1,
|
| 243 |
+
"beta": 1.5,
|
| 244 |
+
"n_eff": 2649414.0,
|
| 245 |
+
"nu": 3974121,
|
| 246 |
+
"s_eff": 0.997109,
|
| 247 |
+
"mass_removed": 0.002891,
|
| 248 |
+
"mass_bound_term": 0.060971,
|
| 249 |
+
"mass_bound_holds": true,
|
| 250 |
+
"delta_w_norm": 0.6053,
|
| 251 |
+
"w_norm": 56.5946,
|
| 252 |
+
"delta_w_frac": 0.010695,
|
| 253 |
+
"loss_change_mse": 7.2e-05,
|
| 254 |
+
"loss_bound": 81.5485,
|
| 255 |
+
"loss_bound_holds": true,
|
| 256 |
+
"C_lip_orig": 11.9171,
|
| 257 |
+
"sigma_x": 27.9935
|
| 258 |
+
},
|
| 259 |
+
{
|
| 260 |
+
"seed": 1,
|
| 261 |
+
"beta": 2.0,
|
| 262 |
+
"n_eff": 2649414.0,
|
| 263 |
+
"nu": 4232191,
|
| 264 |
+
"s_eff": 1.0,
|
| 265 |
+
"mass_removed": 0.0,
|
| 266 |
+
"mass_bound_term": 0.0,
|
| 267 |
+
"mass_bound_holds": true,
|
| 268 |
+
"delta_w_norm": 0.0,
|
| 269 |
+
"w_norm": 56.5946,
|
| 270 |
+
"delta_w_frac": 0.0,
|
| 271 |
+
"loss_change_mse": 0.0,
|
| 272 |
+
"loss_bound": 0.0,
|
| 273 |
+
"loss_bound_holds": true,
|
| 274 |
+
"C_lip_orig": 11.9171,
|
| 275 |
+
"sigma_x": 27.9935
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"seed": 2,
|
| 279 |
+
"beta": 0.25,
|
| 280 |
+
"n_eff": 2649414.0,
|
| 281 |
+
"nu": 662353,
|
| 282 |
+
"s_eff": 0.37194,
|
| 283 |
+
"mass_removed": 0.62806,
|
| 284 |
+
"mass_bound_term": 0.842545,
|
| 285 |
+
"mass_bound_holds": true,
|
| 286 |
+
"delta_w_norm": 36.4881,
|
| 287 |
+
"w_norm": 56.5946,
|
| 288 |
+
"delta_w_frac": 0.644728,
|
| 289 |
+
"loss_change_mse": 0.164294,
|
| 290 |
+
"loss_bound": 295961.5327,
|
| 291 |
+
"loss_bound_holds": true,
|
| 292 |
+
"C_lip_orig": 11.9171,
|
| 293 |
+
"sigma_x": 27.9756
|
| 294 |
+
},
|
| 295 |
+
{
|
| 296 |
+
"seed": 2,
|
| 297 |
+
"beta": 0.5,
|
| 298 |
+
"n_eff": 2649414.0,
|
| 299 |
+
"nu": 1324707,
|
| 300 |
+
"s_eff": 0.605295,
|
| 301 |
+
"mass_removed": 0.394705,
|
| 302 |
+
"mass_bound_term": 0.686498,
|
| 303 |
+
"mass_bound_holds": true,
|
| 304 |
+
"delta_w_norm": 25.0368,
|
| 305 |
+
"w_norm": 56.5946,
|
| 306 |
+
"delta_w_frac": 0.442388,
|
| 307 |
+
"loss_change_mse": 0.082226,
|
| 308 |
+
"loss_bound": 139344.3487,
|
| 309 |
+
"loss_bound_holds": true,
|
| 310 |
+
"C_lip_orig": 11.9171,
|
| 311 |
+
"sigma_x": 27.9756
|
| 312 |
+
},
|
| 313 |
+
{
|
| 314 |
+
"seed": 2,
|
| 315 |
+
"beta": 0.75,
|
| 316 |
+
"n_eff": 2649414.0,
|
| 317 |
+
"nu": 1987060,
|
| 318 |
+
"s_eff": 0.772446,
|
| 319 |
+
"mass_removed": 0.227554,
|
| 320 |
+
"mass_bound_term": 0.530215,
|
| 321 |
+
"mass_bound_holds": true,
|
| 322 |
+
"delta_w_norm": 16.2952,
|
| 323 |
+
"w_norm": 56.5946,
|
| 324 |
+
"delta_w_frac": 0.287929,
|
| 325 |
+
"loss_change_mse": 0.038528,
|
| 326 |
+
"loss_bound": 59027.3306,
|
| 327 |
+
"loss_bound_holds": true,
|
| 328 |
+
"C_lip_orig": 11.9171,
|
| 329 |
+
"sigma_x": 27.9756
|
| 330 |
+
},
|
| 331 |
+
{
|
| 332 |
+
"seed": 2,
|
| 333 |
+
"beta": 1.0,
|
| 334 |
+
"n_eff": 2649414.0,
|
| 335 |
+
"nu": 2649414,
|
| 336 |
+
"s_eff": 0.889327,
|
| 337 |
+
"mass_removed": 0.110673,
|
| 338 |
+
"mass_bound_term": 0.373845,
|
| 339 |
+
"mass_bound_holds": true,
|
| 340 |
+
"delta_w_norm": 9.3942,
|
| 341 |
+
"w_norm": 56.5946,
|
| 342 |
+
"delta_w_frac": 0.16599,
|
| 343 |
+
"loss_change_mse": 0.014309,
|
| 344 |
+
"loss_bound": 19617.6527,
|
| 345 |
+
"loss_bound_holds": true,
|
| 346 |
+
"C_lip_orig": 11.9171,
|
| 347 |
+
"sigma_x": 27.9756
|
| 348 |
+
},
|
| 349 |
+
{
|
| 350 |
+
"seed": 2,
|
| 351 |
+
"beta": 1.25,
|
| 352 |
+
"n_eff": 2649414.0,
|
| 353 |
+
"nu": 3311767,
|
| 354 |
+
"s_eff": 0.963046,
|
| 355 |
+
"mass_removed": 0.036954,
|
| 356 |
+
"mass_bound_term": 0.217426,
|
| 357 |
+
"mass_bound_holds": true,
|
| 358 |
+
"delta_w_norm": 4.102,
|
| 359 |
+
"w_norm": 56.5946,
|
| 360 |
+
"delta_w_frac": 0.072481,
|
| 361 |
+
"loss_change_mse": 0.002925,
|
| 362 |
+
"loss_bound": 3740.4719,
|
| 363 |
+
"loss_bound_holds": true,
|
| 364 |
+
"C_lip_orig": 11.9171,
|
| 365 |
+
"sigma_x": 27.9756
|
| 366 |
+
},
|
| 367 |
+
{
|
| 368 |
+
"seed": 2,
|
| 369 |
+
"beta": 1.5,
|
| 370 |
+
"n_eff": 2649414.0,
|
| 371 |
+
"nu": 3974121,
|
| 372 |
+
"s_eff": 0.997109,
|
| 373 |
+
"mass_removed": 0.002891,
|
| 374 |
+
"mass_bound_term": 0.060971,
|
| 375 |
+
"mass_bound_holds": true,
|
| 376 |
+
"delta_w_norm": 0.6053,
|
| 377 |
+
"w_norm": 56.5946,
|
| 378 |
+
"delta_w_frac": 0.010695,
|
| 379 |
+
"loss_change_mse": 7.6e-05,
|
| 380 |
+
"loss_bound": 81.4445,
|
| 381 |
+
"loss_bound_holds": true,
|
| 382 |
+
"C_lip_orig": 11.9171,
|
| 383 |
+
"sigma_x": 27.9756
|
| 384 |
+
},
|
| 385 |
+
{
|
| 386 |
+
"seed": 2,
|
| 387 |
+
"beta": 2.0,
|
| 388 |
+
"n_eff": 2649414.0,
|
| 389 |
+
"nu": 4232191,
|
| 390 |
+
"s_eff": 1.0,
|
| 391 |
+
"mass_removed": 0.0,
|
| 392 |
+
"mass_bound_term": 0.0,
|
| 393 |
+
"mass_bound_holds": true,
|
| 394 |
+
"delta_w_norm": 0.0,
|
| 395 |
+
"w_norm": 56.5946,
|
| 396 |
+
"delta_w_frac": 0.0,
|
| 397 |
+
"loss_change_mse": 0.0,
|
| 398 |
+
"loss_bound": 0.0,
|
| 399 |
+
"loss_bound_holds": true,
|
| 400 |
+
"C_lip_orig": 11.9171,
|
| 401 |
+
"sigma_x": 27.9756
|
| 402 |
+
}
|
| 403 |
+
],
|
| 404 |
+
"elapsed_s": 30
|
| 405 |
+
}
|
results/emp_loss_bound_multiarch.json
ADDED
|
@@ -0,0 +1,1498 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"claim": 2,
|
| 3 |
+
"description": "Theorem 4.2 multi-arch: actual MSE loss change after EMP pruning \u2264 Lipschitz-style bound on MLP, CNN, and Transformer",
|
| 4 |
+
"disclosure": "scaled / comparable \u2014 synthetic teacher-label protocol; not full ImageNet/LLM/KAN suite",
|
| 5 |
+
"bound_formula": "|\u0394L_MSE| \u2264 (C_lip * ||\u0394W||_F * sigma_x)^2 / n_samples",
|
| 6 |
+
"environment": {
|
| 7 |
+
"os": "Windows 11",
|
| 8 |
+
"python": "3.13.3",
|
| 9 |
+
"torch": "2.6.0+cu124",
|
| 10 |
+
"numpy": "2.4.4",
|
| 11 |
+
"device": "cuda",
|
| 12 |
+
"gpu": "NVIDIA GeForce RTX 4060 Laptop GPU",
|
| 13 |
+
"timestamp_utc": "2026-07-19T07:14:28.527117+00:00"
|
| 14 |
+
},
|
| 15 |
+
"n_samples": 256,
|
| 16 |
+
"seeds": [
|
| 17 |
+
0,
|
| 18 |
+
1,
|
| 19 |
+
2
|
| 20 |
+
],
|
| 21 |
+
"betas": [
|
| 22 |
+
0.25,
|
| 23 |
+
0.5,
|
| 24 |
+
0.75,
|
| 25 |
+
1.0,
|
| 26 |
+
1.25,
|
| 27 |
+
1.5,
|
| 28 |
+
2.0
|
| 29 |
+
],
|
| 30 |
+
"n_total_configs": 63,
|
| 31 |
+
"n_loss_bound_holds": 63,
|
| 32 |
+
"all_bounds_hold": true,
|
| 33 |
+
"by_arch": {
|
| 34 |
+
"mlp": {
|
| 35 |
+
"arch": "mlp",
|
| 36 |
+
"model_desc": "4-layer MLP 784\u21921024\u2192512\u2192256\u219210",
|
| 37 |
+
"N_weights": 1460736,
|
| 38 |
+
"n_configs": 21,
|
| 39 |
+
"n_loss_bound_holds": 21,
|
| 40 |
+
"all_bounds_hold": true,
|
| 41 |
+
"seed0_table": [
|
| 42 |
+
{
|
| 43 |
+
"arch": "mlp",
|
| 44 |
+
"seed": 0,
|
| 45 |
+
"beta": 0.25,
|
| 46 |
+
"N": 1460736,
|
| 47 |
+
"n_eff": 909856.56,
|
| 48 |
+
"nu": 227464,
|
| 49 |
+
"retain_frac": 0.155719,
|
| 50 |
+
"s_eff": 0.371833,
|
| 51 |
+
"delta_w_norm": 28.260038,
|
| 52 |
+
"delta_w_frac": 0.643184,
|
| 53 |
+
"loss_change_mse": 0.21983676,
|
| 54 |
+
"loss_bound": 367489.3872,
|
| 55 |
+
"loss_bound_holds": true,
|
| 56 |
+
"C_lip_orig": 12.222287,
|
| 57 |
+
"sigma_x": 28.0813
|
| 58 |
+
},
|
| 59 |
+
{
|
| 60 |
+
"arch": "mlp",
|
| 61 |
+
"seed": 0,
|
| 62 |
+
"beta": 0.5,
|
| 63 |
+
"N": 1460736,
|
| 64 |
+
"n_eff": 909856.56,
|
| 65 |
+
"nu": 454928,
|
| 66 |
+
"retain_frac": 0.311438,
|
| 67 |
+
"s_eff": 0.604282,
|
| 68 |
+
"delta_w_norm": 19.42478,
|
| 69 |
+
"delta_w_frac": 0.442098,
|
| 70 |
+
"loss_change_mse": 0.12417114,
|
| 71 |
+
"loss_bound": 173624.7046,
|
| 72 |
+
"loss_bound_holds": true,
|
| 73 |
+
"C_lip_orig": 12.222287,
|
| 74 |
+
"sigma_x": 28.0813
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"arch": "mlp",
|
| 78 |
+
"seed": 0,
|
| 79 |
+
"beta": 0.75,
|
| 80 |
+
"N": 1460736,
|
| 81 |
+
"n_eff": 909856.56,
|
| 82 |
+
"nu": 682392,
|
| 83 |
+
"retain_frac": 0.467156,
|
| 84 |
+
"s_eff": 0.77107,
|
| 85 |
+
"delta_w_norm": 12.669997,
|
| 86 |
+
"delta_w_frac": 0.288363,
|
| 87 |
+
"loss_change_mse": 0.06934832,
|
| 88 |
+
"loss_bound": 73867.3206,
|
| 89 |
+
"loss_bound_holds": true,
|
| 90 |
+
"C_lip_orig": 12.222287,
|
| 91 |
+
"sigma_x": 28.0813
|
| 92 |
+
},
|
| 93 |
+
{
|
| 94 |
+
"arch": "mlp",
|
| 95 |
+
"seed": 0,
|
| 96 |
+
"beta": 1.0,
|
| 97 |
+
"N": 1460736,
|
| 98 |
+
"n_eff": 909856.56,
|
| 99 |
+
"nu": 909856,
|
| 100 |
+
"retain_frac": 0.622875,
|
| 101 |
+
"s_eff": 0.887834,
|
| 102 |
+
"delta_w_norm": 7.341904,
|
| 103 |
+
"delta_w_frac": 0.167098,
|
| 104 |
+
"loss_change_mse": 0.02752387,
|
| 105 |
+
"loss_bound": 24803.7148,
|
| 106 |
+
"loss_bound_holds": true,
|
| 107 |
+
"C_lip_orig": 12.222287,
|
| 108 |
+
"sigma_x": 28.0813
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"arch": "mlp",
|
| 112 |
+
"seed": 0,
|
| 113 |
+
"beta": 1.25,
|
| 114 |
+
"N": 1460736,
|
| 115 |
+
"n_eff": 909856.56,
|
| 116 |
+
"nu": 1137320,
|
| 117 |
+
"retain_frac": 0.778594,
|
| 118 |
+
"s_eff": 0.961812,
|
| 119 |
+
"delta_w_norm": 3.254779,
|
| 120 |
+
"delta_w_frac": 0.074077,
|
| 121 |
+
"loss_change_mse": 0.00497048,
|
| 122 |
+
"loss_bound": 4874.637,
|
| 123 |
+
"loss_bound_holds": true,
|
| 124 |
+
"C_lip_orig": 12.222287,
|
| 125 |
+
"sigma_x": 28.0813
|
| 126 |
+
},
|
| 127 |
+
{
|
| 128 |
+
"arch": "mlp",
|
| 129 |
+
"seed": 0,
|
| 130 |
+
"beta": 1.5,
|
| 131 |
+
"N": 1460736,
|
| 132 |
+
"n_eff": 909856.56,
|
| 133 |
+
"nu": 1364784,
|
| 134 |
+
"retain_frac": 0.934313,
|
| 135 |
+
"s_eff": 0.996662,
|
| 136 |
+
"delta_w_norm": 0.521827,
|
| 137 |
+
"delta_w_frac": 0.011877,
|
| 138 |
+
"loss_change_mse": 0.00014248,
|
| 139 |
+
"loss_bound": 125.3006,
|
| 140 |
+
"loss_bound_holds": true,
|
| 141 |
+
"C_lip_orig": 12.222287,
|
| 142 |
+
"sigma_x": 28.0813
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"arch": "mlp",
|
| 146 |
+
"seed": 0,
|
| 147 |
+
"beta": 2.0,
|
| 148 |
+
"N": 1460736,
|
| 149 |
+
"n_eff": 909856.56,
|
| 150 |
+
"nu": 1460735,
|
| 151 |
+
"retain_frac": 0.999999,
|
| 152 |
+
"s_eff": 1.0,
|
| 153 |
+
"delta_w_norm": 0.0,
|
| 154 |
+
"delta_w_frac": 0.0,
|
| 155 |
+
"loss_change_mse": 0.0,
|
| 156 |
+
"loss_bound": 0.0,
|
| 157 |
+
"loss_bound_holds": true,
|
| 158 |
+
"C_lip_orig": 12.222287,
|
| 159 |
+
"sigma_x": 28.0813
|
| 160 |
+
}
|
| 161 |
+
]
|
| 162 |
+
},
|
| 163 |
+
"cnn": {
|
| 164 |
+
"arch": "cnn",
|
| 165 |
+
"model_desc": "SmallCNN Conv64-128-256 + FC128\u219210 (3\u00d732\u00d732)",
|
| 166 |
+
"N_weights": 404416,
|
| 167 |
+
"n_configs": 21,
|
| 168 |
+
"n_loss_bound_holds": 21,
|
| 169 |
+
"all_bounds_hold": true,
|
| 170 |
+
"seed0_table": [
|
| 171 |
+
{
|
| 172 |
+
"arch": "cnn",
|
| 173 |
+
"seed": 0,
|
| 174 |
+
"beta": 0.25,
|
| 175 |
+
"N": 404416,
|
| 176 |
+
"n_eff": 210848.23,
|
| 177 |
+
"nu": 52712,
|
| 178 |
+
"retain_frac": 0.130341,
|
| 179 |
+
"s_eff": 0.371454,
|
| 180 |
+
"delta_w_norm": 12.438971,
|
| 181 |
+
"delta_w_frac": 0.586374,
|
| 182 |
+
"loss_change_mse": 0.00040632,
|
| 183 |
+
"loss_bound": 22736.0216,
|
| 184 |
+
"loss_bound_holds": true,
|
| 185 |
+
"C_lip_orig": 3.49684,
|
| 186 |
+
"sigma_x": 55.4647
|
| 187 |
+
},
|
| 188 |
+
{
|
| 189 |
+
"arch": "cnn",
|
| 190 |
+
"seed": 0,
|
| 191 |
+
"beta": 0.5,
|
| 192 |
+
"N": 404416,
|
| 193 |
+
"n_eff": 210848.23,
|
| 194 |
+
"nu": 105424,
|
| 195 |
+
"retain_frac": 0.260682,
|
| 196 |
+
"s_eff": 0.572948,
|
| 197 |
+
"delta_w_norm": 8.997647,
|
| 198 |
+
"delta_w_frac": 0.42415,
|
| 199 |
+
"loss_change_mse": 0.00067222,
|
| 200 |
+
"loss_bound": 11896.0703,
|
| 201 |
+
"loss_bound_holds": true,
|
| 202 |
+
"C_lip_orig": 3.49684,
|
| 203 |
+
"sigma_x": 55.4647
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"arch": "cnn",
|
| 207 |
+
"seed": 0,
|
| 208 |
+
"beta": 0.75,
|
| 209 |
+
"N": 404416,
|
| 210 |
+
"n_eff": 210848.23,
|
| 211 |
+
"nu": 158136,
|
| 212 |
+
"retain_frac": 0.391023,
|
| 213 |
+
"s_eff": 0.721444,
|
| 214 |
+
"delta_w_norm": 6.402608,
|
| 215 |
+
"delta_w_frac": 0.301819,
|
| 216 |
+
"loss_change_mse": 0.0005174,
|
| 217 |
+
"loss_bound": 6023.6455,
|
| 218 |
+
"loss_bound_holds": true,
|
| 219 |
+
"C_lip_orig": 3.49684,
|
| 220 |
+
"sigma_x": 55.4647
|
| 221 |
+
},
|
| 222 |
+
{
|
| 223 |
+
"arch": "cnn",
|
| 224 |
+
"seed": 0,
|
| 225 |
+
"beta": 1.0,
|
| 226 |
+
"N": 404416,
|
| 227 |
+
"n_eff": 210848.23,
|
| 228 |
+
"nu": 210848,
|
| 229 |
+
"retain_frac": 0.521364,
|
| 230 |
+
"s_eff": 0.832488,
|
| 231 |
+
"delta_w_norm": 4.316469,
|
| 232 |
+
"delta_w_frac": 0.203479,
|
| 233 |
+
"loss_change_mse": 0.0001716,
|
| 234 |
+
"loss_bound": 2737.8067,
|
| 235 |
+
"loss_bound_holds": true,
|
| 236 |
+
"C_lip_orig": 3.49684,
|
| 237 |
+
"sigma_x": 55.4647
|
| 238 |
+
},
|
| 239 |
+
{
|
| 240 |
+
"arch": "cnn",
|
| 241 |
+
"seed": 0,
|
| 242 |
+
"beta": 1.25,
|
| 243 |
+
"N": 404416,
|
| 244 |
+
"n_eff": 210848.23,
|
| 245 |
+
"nu": 263560,
|
| 246 |
+
"retain_frac": 0.651705,
|
| 247 |
+
"s_eff": 0.912934,
|
| 248 |
+
"delta_w_norm": 2.619597,
|
| 249 |
+
"delta_w_frac": 0.123488,
|
| 250 |
+
"loss_change_mse": 0.00011953,
|
| 251 |
+
"loss_bound": 1008.3579,
|
| 252 |
+
"loss_bound_holds": true,
|
| 253 |
+
"C_lip_orig": 3.49684,
|
| 254 |
+
"sigma_x": 55.4647
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"arch": "cnn",
|
| 258 |
+
"seed": 0,
|
| 259 |
+
"beta": 1.5,
|
| 260 |
+
"N": 404416,
|
| 261 |
+
"n_eff": 210848.23,
|
| 262 |
+
"nu": 316272,
|
| 263 |
+
"retain_frac": 0.782046,
|
| 264 |
+
"s_eff": 0.9663,
|
| 265 |
+
"delta_w_norm": 1.278642,
|
| 266 |
+
"delta_w_frac": 0.060275,
|
| 267 |
+
"loss_change_mse": 9.32e-06,
|
| 268 |
+
"loss_bound": 240.2389,
|
| 269 |
+
"loss_bound_holds": true,
|
| 270 |
+
"C_lip_orig": 3.49684,
|
| 271 |
+
"sigma_x": 55.4647
|
| 272 |
+
},
|
| 273 |
+
{
|
| 274 |
+
"arch": "cnn",
|
| 275 |
+
"seed": 0,
|
| 276 |
+
"beta": 2.0,
|
| 277 |
+
"N": 404416,
|
| 278 |
+
"n_eff": 210848.23,
|
| 279 |
+
"nu": 404415,
|
| 280 |
+
"retain_frac": 0.999998,
|
| 281 |
+
"s_eff": 1.0,
|
| 282 |
+
"delta_w_norm": 0.0,
|
| 283 |
+
"delta_w_frac": 0.0,
|
| 284 |
+
"loss_change_mse": 0.0,
|
| 285 |
+
"loss_bound": 0.0,
|
| 286 |
+
"loss_bound_holds": true,
|
| 287 |
+
"C_lip_orig": 3.49684,
|
| 288 |
+
"sigma_x": 55.4647
|
| 289 |
+
}
|
| 290 |
+
]
|
| 291 |
+
},
|
| 292 |
+
"transformer": {
|
| 293 |
+
"arch": "transformer",
|
| 294 |
+
"model_desc": "TinyTransformer seq=16 d=64 nhead=4 nlayers=2 \u219210",
|
| 295 |
+
"N_weights": 78464,
|
| 296 |
+
"n_configs": 21,
|
| 297 |
+
"n_loss_bound_holds": 21,
|
| 298 |
+
"all_bounds_hold": true,
|
| 299 |
+
"seed0_table": [
|
| 300 |
+
{
|
| 301 |
+
"arch": "transformer",
|
| 302 |
+
"seed": 0,
|
| 303 |
+
"beta": 0.25,
|
| 304 |
+
"N": 78464,
|
| 305 |
+
"n_eff": 51293.48,
|
| 306 |
+
"nu": 12823,
|
| 307 |
+
"retain_frac": 0.163425,
|
| 308 |
+
"s_eff": 0.363358,
|
| 309 |
+
"delta_w_norm": 19.467049,
|
| 310 |
+
"delta_w_frac": 0.6602,
|
| 311 |
+
"loss_change_mse": 0.36617145,
|
| 312 |
+
"loss_bound": 8488108374.8184,
|
| 313 |
+
"loss_bound_holds": true,
|
| 314 |
+
"C_lip_orig": 2358.180067,
|
| 315 |
+
"sigma_x": 32.1106
|
| 316 |
+
},
|
| 317 |
+
{
|
| 318 |
+
"arch": "transformer",
|
| 319 |
+
"seed": 0,
|
| 320 |
+
"beta": 0.5,
|
| 321 |
+
"N": 78464,
|
| 322 |
+
"n_eff": 51293.48,
|
| 323 |
+
"nu": 25646,
|
| 324 |
+
"retain_frac": 0.326851,
|
| 325 |
+
"s_eff": 0.599243,
|
| 326 |
+
"delta_w_norm": 13.571939,
|
| 327 |
+
"delta_w_frac": 0.460275,
|
| 328 |
+
"loss_change_mse": 0.17683111,
|
| 329 |
+
"loss_bound": 4125670271.6632,
|
| 330 |
+
"loss_bound_holds": true,
|
| 331 |
+
"C_lip_orig": 2358.180067,
|
| 332 |
+
"sigma_x": 32.1106
|
| 333 |
+
},
|
| 334 |
+
{
|
| 335 |
+
"arch": "transformer",
|
| 336 |
+
"seed": 0,
|
| 337 |
+
"beta": 0.75,
|
| 338 |
+
"N": 78464,
|
| 339 |
+
"n_eff": 51293.48,
|
| 340 |
+
"nu": 38470,
|
| 341 |
+
"retain_frac": 0.490289,
|
| 342 |
+
"s_eff": 0.774719,
|
| 343 |
+
"delta_w_norm": 8.728827,
|
| 344 |
+
"delta_w_frac": 0.296027,
|
| 345 |
+
"loss_change_mse": 0.11401466,
|
| 346 |
+
"loss_bound": 1706563341.9569,
|
| 347 |
+
"loss_bound_holds": true,
|
| 348 |
+
"C_lip_orig": 2358.180067,
|
| 349 |
+
"sigma_x": 32.1106
|
| 350 |
+
},
|
| 351 |
+
{
|
| 352 |
+
"arch": "transformer",
|
| 353 |
+
"seed": 0,
|
| 354 |
+
"beta": 1.0,
|
| 355 |
+
"N": 78464,
|
| 356 |
+
"n_eff": 51293.48,
|
| 357 |
+
"nu": 51293,
|
| 358 |
+
"retain_frac": 0.653714,
|
| 359 |
+
"s_eff": 0.89739,
|
| 360 |
+
"delta_w_norm": 4.811022,
|
| 361 |
+
"delta_w_frac": 0.16316,
|
| 362 |
+
"loss_change_mse": 0.03556463,
|
| 363 |
+
"loss_bound": 518424388.7459,
|
| 364 |
+
"loss_bound_holds": true,
|
| 365 |
+
"C_lip_orig": 2358.180067,
|
| 366 |
+
"sigma_x": 32.1106
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"arch": "transformer",
|
| 370 |
+
"seed": 0,
|
| 371 |
+
"beta": 1.25,
|
| 372 |
+
"N": 78464,
|
| 373 |
+
"n_eff": 51293.48,
|
| 374 |
+
"nu": 64116,
|
| 375 |
+
"retain_frac": 0.817139,
|
| 376 |
+
"s_eff": 0.97161,
|
| 377 |
+
"delta_w_norm": 1.825973,
|
| 378 |
+
"delta_w_frac": 0.061926,
|
| 379 |
+
"loss_change_mse": 0.00704481,
|
| 380 |
+
"loss_bound": 74679127.7396,
|
| 381 |
+
"loss_bound_holds": true,
|
| 382 |
+
"C_lip_orig": 2358.180067,
|
| 383 |
+
"sigma_x": 32.1106
|
| 384 |
+
},
|
| 385 |
+
{
|
| 386 |
+
"arch": "transformer",
|
| 387 |
+
"seed": 0,
|
| 388 |
+
"beta": 1.5,
|
| 389 |
+
"N": 78464,
|
| 390 |
+
"n_eff": 51293.48,
|
| 391 |
+
"nu": 76940,
|
| 392 |
+
"retain_frac": 0.980577,
|
| 393 |
+
"s_eff": 0.999679,
|
| 394 |
+
"delta_w_norm": 0.063212,
|
| 395 |
+
"delta_w_frac": 0.002144,
|
| 396 |
+
"loss_change_mse": 6.77e-06,
|
| 397 |
+
"loss_bound": 89496.65,
|
| 398 |
+
"loss_bound_holds": true,
|
| 399 |
+
"C_lip_orig": 2358.180067,
|
| 400 |
+
"sigma_x": 32.1106
|
| 401 |
+
},
|
| 402 |
+
{
|
| 403 |
+
"arch": "transformer",
|
| 404 |
+
"seed": 0,
|
| 405 |
+
"beta": 2.0,
|
| 406 |
+
"N": 78464,
|
| 407 |
+
"n_eff": 51293.48,
|
| 408 |
+
"nu": 78463,
|
| 409 |
+
"retain_frac": 0.999987,
|
| 410 |
+
"s_eff": 1.0,
|
| 411 |
+
"delta_w_norm": 0.0,
|
| 412 |
+
"delta_w_frac": 0.0,
|
| 413 |
+
"loss_change_mse": 0.0,
|
| 414 |
+
"loss_bound": 0.0,
|
| 415 |
+
"loss_bound_holds": true,
|
| 416 |
+
"C_lip_orig": 2358.180067,
|
| 417 |
+
"sigma_x": 32.1106
|
| 418 |
+
}
|
| 419 |
+
]
|
| 420 |
+
}
|
| 421 |
+
},
|
| 422 |
+
"rows": [
|
| 423 |
+
{
|
| 424 |
+
"arch": "mlp",
|
| 425 |
+
"seed": 0,
|
| 426 |
+
"beta": 0.25,
|
| 427 |
+
"N": 1460736,
|
| 428 |
+
"n_eff": 909856.56,
|
| 429 |
+
"nu": 227464,
|
| 430 |
+
"retain_frac": 0.155719,
|
| 431 |
+
"s_eff": 0.371833,
|
| 432 |
+
"delta_w_norm": 28.260038,
|
| 433 |
+
"delta_w_frac": 0.643184,
|
| 434 |
+
"loss_change_mse": 0.21983676,
|
| 435 |
+
"loss_bound": 367489.3872,
|
| 436 |
+
"loss_bound_holds": true,
|
| 437 |
+
"C_lip_orig": 12.222287,
|
| 438 |
+
"sigma_x": 28.0813
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"arch": "mlp",
|
| 442 |
+
"seed": 0,
|
| 443 |
+
"beta": 0.5,
|
| 444 |
+
"N": 1460736,
|
| 445 |
+
"n_eff": 909856.56,
|
| 446 |
+
"nu": 454928,
|
| 447 |
+
"retain_frac": 0.311438,
|
| 448 |
+
"s_eff": 0.604282,
|
| 449 |
+
"delta_w_norm": 19.42478,
|
| 450 |
+
"delta_w_frac": 0.442098,
|
| 451 |
+
"loss_change_mse": 0.12417114,
|
| 452 |
+
"loss_bound": 173624.7046,
|
| 453 |
+
"loss_bound_holds": true,
|
| 454 |
+
"C_lip_orig": 12.222287,
|
| 455 |
+
"sigma_x": 28.0813
|
| 456 |
+
},
|
| 457 |
+
{
|
| 458 |
+
"arch": "mlp",
|
| 459 |
+
"seed": 0,
|
| 460 |
+
"beta": 0.75,
|
| 461 |
+
"N": 1460736,
|
| 462 |
+
"n_eff": 909856.56,
|
| 463 |
+
"nu": 682392,
|
| 464 |
+
"retain_frac": 0.467156,
|
| 465 |
+
"s_eff": 0.77107,
|
| 466 |
+
"delta_w_norm": 12.669997,
|
| 467 |
+
"delta_w_frac": 0.288363,
|
| 468 |
+
"loss_change_mse": 0.06934832,
|
| 469 |
+
"loss_bound": 73867.3206,
|
| 470 |
+
"loss_bound_holds": true,
|
| 471 |
+
"C_lip_orig": 12.222287,
|
| 472 |
+
"sigma_x": 28.0813
|
| 473 |
+
},
|
| 474 |
+
{
|
| 475 |
+
"arch": "mlp",
|
| 476 |
+
"seed": 0,
|
| 477 |
+
"beta": 1.0,
|
| 478 |
+
"N": 1460736,
|
| 479 |
+
"n_eff": 909856.56,
|
| 480 |
+
"nu": 909856,
|
| 481 |
+
"retain_frac": 0.622875,
|
| 482 |
+
"s_eff": 0.887834,
|
| 483 |
+
"delta_w_norm": 7.341904,
|
| 484 |
+
"delta_w_frac": 0.167098,
|
| 485 |
+
"loss_change_mse": 0.02752387,
|
| 486 |
+
"loss_bound": 24803.7148,
|
| 487 |
+
"loss_bound_holds": true,
|
| 488 |
+
"C_lip_orig": 12.222287,
|
| 489 |
+
"sigma_x": 28.0813
|
| 490 |
+
},
|
| 491 |
+
{
|
| 492 |
+
"arch": "mlp",
|
| 493 |
+
"seed": 0,
|
| 494 |
+
"beta": 1.25,
|
| 495 |
+
"N": 1460736,
|
| 496 |
+
"n_eff": 909856.56,
|
| 497 |
+
"nu": 1137320,
|
| 498 |
+
"retain_frac": 0.778594,
|
| 499 |
+
"s_eff": 0.961812,
|
| 500 |
+
"delta_w_norm": 3.254779,
|
| 501 |
+
"delta_w_frac": 0.074077,
|
| 502 |
+
"loss_change_mse": 0.00497048,
|
| 503 |
+
"loss_bound": 4874.637,
|
| 504 |
+
"loss_bound_holds": true,
|
| 505 |
+
"C_lip_orig": 12.222287,
|
| 506 |
+
"sigma_x": 28.0813
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"arch": "mlp",
|
| 510 |
+
"seed": 0,
|
| 511 |
+
"beta": 1.5,
|
| 512 |
+
"N": 1460736,
|
| 513 |
+
"n_eff": 909856.56,
|
| 514 |
+
"nu": 1364784,
|
| 515 |
+
"retain_frac": 0.934313,
|
| 516 |
+
"s_eff": 0.996662,
|
| 517 |
+
"delta_w_norm": 0.521827,
|
| 518 |
+
"delta_w_frac": 0.011877,
|
| 519 |
+
"loss_change_mse": 0.00014248,
|
| 520 |
+
"loss_bound": 125.3006,
|
| 521 |
+
"loss_bound_holds": true,
|
| 522 |
+
"C_lip_orig": 12.222287,
|
| 523 |
+
"sigma_x": 28.0813
|
| 524 |
+
},
|
| 525 |
+
{
|
| 526 |
+
"arch": "mlp",
|
| 527 |
+
"seed": 0,
|
| 528 |
+
"beta": 2.0,
|
| 529 |
+
"N": 1460736,
|
| 530 |
+
"n_eff": 909856.56,
|
| 531 |
+
"nu": 1460735,
|
| 532 |
+
"retain_frac": 0.999999,
|
| 533 |
+
"s_eff": 1.0,
|
| 534 |
+
"delta_w_norm": 0.0,
|
| 535 |
+
"delta_w_frac": 0.0,
|
| 536 |
+
"loss_change_mse": 0.0,
|
| 537 |
+
"loss_bound": 0.0,
|
| 538 |
+
"loss_bound_holds": true,
|
| 539 |
+
"C_lip_orig": 12.222287,
|
| 540 |
+
"sigma_x": 28.0813
|
| 541 |
+
},
|
| 542 |
+
{
|
| 543 |
+
"arch": "mlp",
|
| 544 |
+
"seed": 1,
|
| 545 |
+
"beta": 0.25,
|
| 546 |
+
"N": 1460736,
|
| 547 |
+
"n_eff": 909856.56,
|
| 548 |
+
"nu": 227464,
|
| 549 |
+
"retain_frac": 0.155719,
|
| 550 |
+
"s_eff": 0.371833,
|
| 551 |
+
"delta_w_norm": 28.260038,
|
| 552 |
+
"delta_w_frac": 0.643184,
|
| 553 |
+
"loss_change_mse": 0.24220054,
|
| 554 |
+
"loss_bound": 366783.9857,
|
| 555 |
+
"loss_bound_holds": true,
|
| 556 |
+
"C_lip_orig": 12.222287,
|
| 557 |
+
"sigma_x": 28.0543
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"arch": "mlp",
|
| 561 |
+
"seed": 1,
|
| 562 |
+
"beta": 0.5,
|
| 563 |
+
"N": 1460736,
|
| 564 |
+
"n_eff": 909856.56,
|
| 565 |
+
"nu": 454928,
|
| 566 |
+
"retain_frac": 0.311438,
|
| 567 |
+
"s_eff": 0.604282,
|
| 568 |
+
"delta_w_norm": 19.42478,
|
| 569 |
+
"delta_w_frac": 0.442098,
|
| 570 |
+
"loss_change_mse": 0.13517503,
|
| 571 |
+
"loss_bound": 173291.4293,
|
| 572 |
+
"loss_bound_holds": true,
|
| 573 |
+
"C_lip_orig": 12.222287,
|
| 574 |
+
"sigma_x": 28.0543
|
| 575 |
+
},
|
| 576 |
+
{
|
| 577 |
+
"arch": "mlp",
|
| 578 |
+
"seed": 1,
|
| 579 |
+
"beta": 0.75,
|
| 580 |
+
"N": 1460736,
|
| 581 |
+
"n_eff": 909856.56,
|
| 582 |
+
"nu": 682392,
|
| 583 |
+
"retain_frac": 0.467156,
|
| 584 |
+
"s_eff": 0.77107,
|
| 585 |
+
"delta_w_norm": 12.669997,
|
| 586 |
+
"delta_w_frac": 0.288363,
|
| 587 |
+
"loss_change_mse": 0.07131672,
|
| 588 |
+
"loss_bound": 73725.5311,
|
| 589 |
+
"loss_bound_holds": true,
|
| 590 |
+
"C_lip_orig": 12.222287,
|
| 591 |
+
"sigma_x": 28.0543
|
| 592 |
+
},
|
| 593 |
+
{
|
| 594 |
+
"arch": "mlp",
|
| 595 |
+
"seed": 1,
|
| 596 |
+
"beta": 1.0,
|
| 597 |
+
"N": 1460736,
|
| 598 |
+
"n_eff": 909856.56,
|
| 599 |
+
"nu": 909856,
|
| 600 |
+
"retain_frac": 0.622875,
|
| 601 |
+
"s_eff": 0.887834,
|
| 602 |
+
"delta_w_norm": 7.341904,
|
| 603 |
+
"delta_w_frac": 0.167098,
|
| 604 |
+
"loss_change_mse": 0.0275071,
|
| 605 |
+
"loss_bound": 24756.1037,
|
| 606 |
+
"loss_bound_holds": true,
|
| 607 |
+
"C_lip_orig": 12.222287,
|
| 608 |
+
"sigma_x": 28.0543
|
| 609 |
+
},
|
| 610 |
+
{
|
| 611 |
+
"arch": "mlp",
|
| 612 |
+
"seed": 1,
|
| 613 |
+
"beta": 1.25,
|
| 614 |
+
"N": 1460736,
|
| 615 |
+
"n_eff": 909856.56,
|
| 616 |
+
"nu": 1137320,
|
| 617 |
+
"retain_frac": 0.778594,
|
| 618 |
+
"s_eff": 0.961812,
|
| 619 |
+
"delta_w_norm": 3.254779,
|
| 620 |
+
"delta_w_frac": 0.074077,
|
| 621 |
+
"loss_change_mse": 0.00512693,
|
| 622 |
+
"loss_bound": 4865.2801,
|
| 623 |
+
"loss_bound_holds": true,
|
| 624 |
+
"C_lip_orig": 12.222287,
|
| 625 |
+
"sigma_x": 28.0543
|
| 626 |
+
},
|
| 627 |
+
{
|
| 628 |
+
"arch": "mlp",
|
| 629 |
+
"seed": 1,
|
| 630 |
+
"beta": 1.5,
|
| 631 |
+
"N": 1460736,
|
| 632 |
+
"n_eff": 909856.56,
|
| 633 |
+
"nu": 1364784,
|
| 634 |
+
"retain_frac": 0.934313,
|
| 635 |
+
"s_eff": 0.996662,
|
| 636 |
+
"delta_w_norm": 0.521827,
|
| 637 |
+
"delta_w_frac": 0.011877,
|
| 638 |
+
"loss_change_mse": 0.00013892,
|
| 639 |
+
"loss_bound": 125.0601,
|
| 640 |
+
"loss_bound_holds": true,
|
| 641 |
+
"C_lip_orig": 12.222287,
|
| 642 |
+
"sigma_x": 28.0543
|
| 643 |
+
},
|
| 644 |
+
{
|
| 645 |
+
"arch": "mlp",
|
| 646 |
+
"seed": 1,
|
| 647 |
+
"beta": 2.0,
|
| 648 |
+
"N": 1460736,
|
| 649 |
+
"n_eff": 909856.56,
|
| 650 |
+
"nu": 1460735,
|
| 651 |
+
"retain_frac": 0.999999,
|
| 652 |
+
"s_eff": 1.0,
|
| 653 |
+
"delta_w_norm": 0.0,
|
| 654 |
+
"delta_w_frac": 0.0,
|
| 655 |
+
"loss_change_mse": 0.0,
|
| 656 |
+
"loss_bound": 0.0,
|
| 657 |
+
"loss_bound_holds": true,
|
| 658 |
+
"C_lip_orig": 12.222287,
|
| 659 |
+
"sigma_x": 28.0543
|
| 660 |
+
},
|
| 661 |
+
{
|
| 662 |
+
"arch": "mlp",
|
| 663 |
+
"seed": 2,
|
| 664 |
+
"beta": 0.25,
|
| 665 |
+
"N": 1460736,
|
| 666 |
+
"n_eff": 909856.56,
|
| 667 |
+
"nu": 227464,
|
| 668 |
+
"retain_frac": 0.155719,
|
| 669 |
+
"s_eff": 0.371833,
|
| 670 |
+
"delta_w_norm": 28.260038,
|
| 671 |
+
"delta_w_frac": 0.643184,
|
| 672 |
+
"loss_change_mse": 0.23691006,
|
| 673 |
+
"loss_bound": 365320.7607,
|
| 674 |
+
"loss_bound_holds": true,
|
| 675 |
+
"C_lip_orig": 12.222287,
|
| 676 |
+
"sigma_x": 27.9983
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"arch": "mlp",
|
| 680 |
+
"seed": 2,
|
| 681 |
+
"beta": 0.5,
|
| 682 |
+
"N": 1460736,
|
| 683 |
+
"n_eff": 909856.56,
|
| 684 |
+
"nu": 454928,
|
| 685 |
+
"retain_frac": 0.311438,
|
| 686 |
+
"s_eff": 0.604282,
|
| 687 |
+
"delta_w_norm": 19.42478,
|
| 688 |
+
"delta_w_frac": 0.442098,
|
| 689 |
+
"loss_change_mse": 0.1266737,
|
| 690 |
+
"loss_bound": 172600.1114,
|
| 691 |
+
"loss_bound_holds": true,
|
| 692 |
+
"C_lip_orig": 12.222287,
|
| 693 |
+
"sigma_x": 27.9983
|
| 694 |
+
},
|
| 695 |
+
{
|
| 696 |
+
"arch": "mlp",
|
| 697 |
+
"seed": 2,
|
| 698 |
+
"beta": 0.75,
|
| 699 |
+
"N": 1460736,
|
| 700 |
+
"n_eff": 909856.56,
|
| 701 |
+
"nu": 682392,
|
| 702 |
+
"retain_frac": 0.467156,
|
| 703 |
+
"s_eff": 0.77107,
|
| 704 |
+
"delta_w_norm": 12.669997,
|
| 705 |
+
"delta_w_frac": 0.288363,
|
| 706 |
+
"loss_change_mse": 0.07159347,
|
| 707 |
+
"loss_bound": 73431.4151,
|
| 708 |
+
"loss_bound_holds": true,
|
| 709 |
+
"C_lip_orig": 12.222287,
|
| 710 |
+
"sigma_x": 27.9983
|
| 711 |
+
},
|
| 712 |
+
{
|
| 713 |
+
"arch": "mlp",
|
| 714 |
+
"seed": 2,
|
| 715 |
+
"beta": 1.0,
|
| 716 |
+
"N": 1460736,
|
| 717 |
+
"n_eff": 909856.56,
|
| 718 |
+
"nu": 909856,
|
| 719 |
+
"retain_frac": 0.622875,
|
| 720 |
+
"s_eff": 0.887834,
|
| 721 |
+
"delta_w_norm": 7.341904,
|
| 722 |
+
"delta_w_frac": 0.167098,
|
| 723 |
+
"loss_change_mse": 0.02667378,
|
| 724 |
+
"loss_bound": 24657.3433,
|
| 725 |
+
"loss_bound_holds": true,
|
| 726 |
+
"C_lip_orig": 12.222287,
|
| 727 |
+
"sigma_x": 27.9983
|
| 728 |
+
},
|
| 729 |
+
{
|
| 730 |
+
"arch": "mlp",
|
| 731 |
+
"seed": 2,
|
| 732 |
+
"beta": 1.25,
|
| 733 |
+
"N": 1460736,
|
| 734 |
+
"n_eff": 909856.56,
|
| 735 |
+
"nu": 1137320,
|
| 736 |
+
"retain_frac": 0.778594,
|
| 737 |
+
"s_eff": 0.961812,
|
| 738 |
+
"delta_w_norm": 3.254779,
|
| 739 |
+
"delta_w_frac": 0.074077,
|
| 740 |
+
"loss_change_mse": 0.00511043,
|
| 741 |
+
"loss_bound": 4845.8708,
|
| 742 |
+
"loss_bound_holds": true,
|
| 743 |
+
"C_lip_orig": 12.222287,
|
| 744 |
+
"sigma_x": 27.9983
|
| 745 |
+
},
|
| 746 |
+
{
|
| 747 |
+
"arch": "mlp",
|
| 748 |
+
"seed": 2,
|
| 749 |
+
"beta": 1.5,
|
| 750 |
+
"N": 1460736,
|
| 751 |
+
"n_eff": 909856.56,
|
| 752 |
+
"nu": 1364784,
|
| 753 |
+
"retain_frac": 0.934313,
|
| 754 |
+
"s_eff": 0.996662,
|
| 755 |
+
"delta_w_norm": 0.521827,
|
| 756 |
+
"delta_w_frac": 0.011877,
|
| 757 |
+
"loss_change_mse": 0.00014355,
|
| 758 |
+
"loss_bound": 124.5612,
|
| 759 |
+
"loss_bound_holds": true,
|
| 760 |
+
"C_lip_orig": 12.222287,
|
| 761 |
+
"sigma_x": 27.9983
|
| 762 |
+
},
|
| 763 |
+
{
|
| 764 |
+
"arch": "mlp",
|
| 765 |
+
"seed": 2,
|
| 766 |
+
"beta": 2.0,
|
| 767 |
+
"N": 1460736,
|
| 768 |
+
"n_eff": 909856.56,
|
| 769 |
+
"nu": 1460735,
|
| 770 |
+
"retain_frac": 0.999999,
|
| 771 |
+
"s_eff": 1.0,
|
| 772 |
+
"delta_w_norm": 0.0,
|
| 773 |
+
"delta_w_frac": 0.0,
|
| 774 |
+
"loss_change_mse": 0.0,
|
| 775 |
+
"loss_bound": 0.0,
|
| 776 |
+
"loss_bound_holds": true,
|
| 777 |
+
"C_lip_orig": 12.222287,
|
| 778 |
+
"sigma_x": 27.9983
|
| 779 |
+
},
|
| 780 |
+
{
|
| 781 |
+
"arch": "cnn",
|
| 782 |
+
"seed": 0,
|
| 783 |
+
"beta": 0.25,
|
| 784 |
+
"N": 404416,
|
| 785 |
+
"n_eff": 210848.23,
|
| 786 |
+
"nu": 52712,
|
| 787 |
+
"retain_frac": 0.130341,
|
| 788 |
+
"s_eff": 0.371454,
|
| 789 |
+
"delta_w_norm": 12.438971,
|
| 790 |
+
"delta_w_frac": 0.586374,
|
| 791 |
+
"loss_change_mse": 0.00040632,
|
| 792 |
+
"loss_bound": 22736.0216,
|
| 793 |
+
"loss_bound_holds": true,
|
| 794 |
+
"C_lip_orig": 3.49684,
|
| 795 |
+
"sigma_x": 55.4647
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"arch": "cnn",
|
| 799 |
+
"seed": 0,
|
| 800 |
+
"beta": 0.5,
|
| 801 |
+
"N": 404416,
|
| 802 |
+
"n_eff": 210848.23,
|
| 803 |
+
"nu": 105424,
|
| 804 |
+
"retain_frac": 0.260682,
|
| 805 |
+
"s_eff": 0.572948,
|
| 806 |
+
"delta_w_norm": 8.997647,
|
| 807 |
+
"delta_w_frac": 0.42415,
|
| 808 |
+
"loss_change_mse": 0.00067222,
|
| 809 |
+
"loss_bound": 11896.0703,
|
| 810 |
+
"loss_bound_holds": true,
|
| 811 |
+
"C_lip_orig": 3.49684,
|
| 812 |
+
"sigma_x": 55.4647
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"arch": "cnn",
|
| 816 |
+
"seed": 0,
|
| 817 |
+
"beta": 0.75,
|
| 818 |
+
"N": 404416,
|
| 819 |
+
"n_eff": 210848.23,
|
| 820 |
+
"nu": 158136,
|
| 821 |
+
"retain_frac": 0.391023,
|
| 822 |
+
"s_eff": 0.721444,
|
| 823 |
+
"delta_w_norm": 6.402608,
|
| 824 |
+
"delta_w_frac": 0.301819,
|
| 825 |
+
"loss_change_mse": 0.0005174,
|
| 826 |
+
"loss_bound": 6023.6455,
|
| 827 |
+
"loss_bound_holds": true,
|
| 828 |
+
"C_lip_orig": 3.49684,
|
| 829 |
+
"sigma_x": 55.4647
|
| 830 |
+
},
|
| 831 |
+
{
|
| 832 |
+
"arch": "cnn",
|
| 833 |
+
"seed": 0,
|
| 834 |
+
"beta": 1.0,
|
| 835 |
+
"N": 404416,
|
| 836 |
+
"n_eff": 210848.23,
|
| 837 |
+
"nu": 210848,
|
| 838 |
+
"retain_frac": 0.521364,
|
| 839 |
+
"s_eff": 0.832488,
|
| 840 |
+
"delta_w_norm": 4.316469,
|
| 841 |
+
"delta_w_frac": 0.203479,
|
| 842 |
+
"loss_change_mse": 0.0001716,
|
| 843 |
+
"loss_bound": 2737.8067,
|
| 844 |
+
"loss_bound_holds": true,
|
| 845 |
+
"C_lip_orig": 3.49684,
|
| 846 |
+
"sigma_x": 55.4647
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"arch": "cnn",
|
| 850 |
+
"seed": 0,
|
| 851 |
+
"beta": 1.25,
|
| 852 |
+
"N": 404416,
|
| 853 |
+
"n_eff": 210848.23,
|
| 854 |
+
"nu": 263560,
|
| 855 |
+
"retain_frac": 0.651705,
|
| 856 |
+
"s_eff": 0.912934,
|
| 857 |
+
"delta_w_norm": 2.619597,
|
| 858 |
+
"delta_w_frac": 0.123488,
|
| 859 |
+
"loss_change_mse": 0.00011953,
|
| 860 |
+
"loss_bound": 1008.3579,
|
| 861 |
+
"loss_bound_holds": true,
|
| 862 |
+
"C_lip_orig": 3.49684,
|
| 863 |
+
"sigma_x": 55.4647
|
| 864 |
+
},
|
| 865 |
+
{
|
| 866 |
+
"arch": "cnn",
|
| 867 |
+
"seed": 0,
|
| 868 |
+
"beta": 1.5,
|
| 869 |
+
"N": 404416,
|
| 870 |
+
"n_eff": 210848.23,
|
| 871 |
+
"nu": 316272,
|
| 872 |
+
"retain_frac": 0.782046,
|
| 873 |
+
"s_eff": 0.9663,
|
| 874 |
+
"delta_w_norm": 1.278642,
|
| 875 |
+
"delta_w_frac": 0.060275,
|
| 876 |
+
"loss_change_mse": 9.32e-06,
|
| 877 |
+
"loss_bound": 240.2389,
|
| 878 |
+
"loss_bound_holds": true,
|
| 879 |
+
"C_lip_orig": 3.49684,
|
| 880 |
+
"sigma_x": 55.4647
|
| 881 |
+
},
|
| 882 |
+
{
|
| 883 |
+
"arch": "cnn",
|
| 884 |
+
"seed": 0,
|
| 885 |
+
"beta": 2.0,
|
| 886 |
+
"N": 404416,
|
| 887 |
+
"n_eff": 210848.23,
|
| 888 |
+
"nu": 404415,
|
| 889 |
+
"retain_frac": 0.999998,
|
| 890 |
+
"s_eff": 1.0,
|
| 891 |
+
"delta_w_norm": 0.0,
|
| 892 |
+
"delta_w_frac": 0.0,
|
| 893 |
+
"loss_change_mse": 0.0,
|
| 894 |
+
"loss_bound": 0.0,
|
| 895 |
+
"loss_bound_holds": true,
|
| 896 |
+
"C_lip_orig": 3.49684,
|
| 897 |
+
"sigma_x": 55.4647
|
| 898 |
+
},
|
| 899 |
+
{
|
| 900 |
+
"arch": "cnn",
|
| 901 |
+
"seed": 1,
|
| 902 |
+
"beta": 0.25,
|
| 903 |
+
"N": 404416,
|
| 904 |
+
"n_eff": 210848.23,
|
| 905 |
+
"nu": 52712,
|
| 906 |
+
"retain_frac": 0.130341,
|
| 907 |
+
"s_eff": 0.371454,
|
| 908 |
+
"delta_w_norm": 12.438971,
|
| 909 |
+
"delta_w_frac": 0.586374,
|
| 910 |
+
"loss_change_mse": 0.00040684,
|
| 911 |
+
"loss_bound": 22688.2534,
|
| 912 |
+
"loss_bound_holds": true,
|
| 913 |
+
"C_lip_orig": 3.49684,
|
| 914 |
+
"sigma_x": 55.4064
|
| 915 |
+
},
|
| 916 |
+
{
|
| 917 |
+
"arch": "cnn",
|
| 918 |
+
"seed": 1,
|
| 919 |
+
"beta": 0.5,
|
| 920 |
+
"N": 404416,
|
| 921 |
+
"n_eff": 210848.23,
|
| 922 |
+
"nu": 105424,
|
| 923 |
+
"retain_frac": 0.260682,
|
| 924 |
+
"s_eff": 0.572948,
|
| 925 |
+
"delta_w_norm": 8.997647,
|
| 926 |
+
"delta_w_frac": 0.42415,
|
| 927 |
+
"loss_change_mse": 0.00067211,
|
| 928 |
+
"loss_bound": 11871.0767,
|
| 929 |
+
"loss_bound_holds": true,
|
| 930 |
+
"C_lip_orig": 3.49684,
|
| 931 |
+
"sigma_x": 55.4064
|
| 932 |
+
},
|
| 933 |
+
{
|
| 934 |
+
"arch": "cnn",
|
| 935 |
+
"seed": 1,
|
| 936 |
+
"beta": 0.75,
|
| 937 |
+
"N": 404416,
|
| 938 |
+
"n_eff": 210848.23,
|
| 939 |
+
"nu": 158136,
|
| 940 |
+
"retain_frac": 0.391023,
|
| 941 |
+
"s_eff": 0.721444,
|
| 942 |
+
"delta_w_norm": 6.402608,
|
| 943 |
+
"delta_w_frac": 0.301819,
|
| 944 |
+
"loss_change_mse": 0.00051484,
|
| 945 |
+
"loss_bound": 6010.9899,
|
| 946 |
+
"loss_bound_holds": true,
|
| 947 |
+
"C_lip_orig": 3.49684,
|
| 948 |
+
"sigma_x": 55.4064
|
| 949 |
+
},
|
| 950 |
+
{
|
| 951 |
+
"arch": "cnn",
|
| 952 |
+
"seed": 1,
|
| 953 |
+
"beta": 1.0,
|
| 954 |
+
"N": 404416,
|
| 955 |
+
"n_eff": 210848.23,
|
| 956 |
+
"nu": 210848,
|
| 957 |
+
"retain_frac": 0.521364,
|
| 958 |
+
"s_eff": 0.832488,
|
| 959 |
+
"delta_w_norm": 4.316469,
|
| 960 |
+
"delta_w_frac": 0.203479,
|
| 961 |
+
"loss_change_mse": 0.00016967,
|
| 962 |
+
"loss_bound": 2732.0546,
|
| 963 |
+
"loss_bound_holds": true,
|
| 964 |
+
"C_lip_orig": 3.49684,
|
| 965 |
+
"sigma_x": 55.4064
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"arch": "cnn",
|
| 969 |
+
"seed": 1,
|
| 970 |
+
"beta": 1.25,
|
| 971 |
+
"N": 404416,
|
| 972 |
+
"n_eff": 210848.23,
|
| 973 |
+
"nu": 263560,
|
| 974 |
+
"retain_frac": 0.651705,
|
| 975 |
+
"s_eff": 0.912934,
|
| 976 |
+
"delta_w_norm": 2.619597,
|
| 977 |
+
"delta_w_frac": 0.123488,
|
| 978 |
+
"loss_change_mse": 0.00011876,
|
| 979 |
+
"loss_bound": 1006.2393,
|
| 980 |
+
"loss_bound_holds": true,
|
| 981 |
+
"C_lip_orig": 3.49684,
|
| 982 |
+
"sigma_x": 55.4064
|
| 983 |
+
},
|
| 984 |
+
{
|
| 985 |
+
"arch": "cnn",
|
| 986 |
+
"seed": 1,
|
| 987 |
+
"beta": 1.5,
|
| 988 |
+
"N": 404416,
|
| 989 |
+
"n_eff": 210848.23,
|
| 990 |
+
"nu": 316272,
|
| 991 |
+
"retain_frac": 0.782046,
|
| 992 |
+
"s_eff": 0.9663,
|
| 993 |
+
"delta_w_norm": 1.278642,
|
| 994 |
+
"delta_w_frac": 0.060275,
|
| 995 |
+
"loss_change_mse": 9.43e-06,
|
| 996 |
+
"loss_bound": 239.7342,
|
| 997 |
+
"loss_bound_holds": true,
|
| 998 |
+
"C_lip_orig": 3.49684,
|
| 999 |
+
"sigma_x": 55.4064
|
| 1000 |
+
},
|
| 1001 |
+
{
|
| 1002 |
+
"arch": "cnn",
|
| 1003 |
+
"seed": 1,
|
| 1004 |
+
"beta": 2.0,
|
| 1005 |
+
"N": 404416,
|
| 1006 |
+
"n_eff": 210848.23,
|
| 1007 |
+
"nu": 404415,
|
| 1008 |
+
"retain_frac": 0.999998,
|
| 1009 |
+
"s_eff": 1.0,
|
| 1010 |
+
"delta_w_norm": 0.0,
|
| 1011 |
+
"delta_w_frac": 0.0,
|
| 1012 |
+
"loss_change_mse": 0.0,
|
| 1013 |
+
"loss_bound": 0.0,
|
| 1014 |
+
"loss_bound_holds": true,
|
| 1015 |
+
"C_lip_orig": 3.49684,
|
| 1016 |
+
"sigma_x": 55.4064
|
| 1017 |
+
},
|
| 1018 |
+
{
|
| 1019 |
+
"arch": "cnn",
|
| 1020 |
+
"seed": 2,
|
| 1021 |
+
"beta": 0.25,
|
| 1022 |
+
"N": 404416,
|
| 1023 |
+
"n_eff": 210848.23,
|
| 1024 |
+
"nu": 52712,
|
| 1025 |
+
"retain_frac": 0.130341,
|
| 1026 |
+
"s_eff": 0.371454,
|
| 1027 |
+
"delta_w_norm": 12.438971,
|
| 1028 |
+
"delta_w_frac": 0.586374,
|
| 1029 |
+
"loss_change_mse": 0.00041493,
|
| 1030 |
+
"loss_bound": 22696.3206,
|
| 1031 |
+
"loss_bound_holds": true,
|
| 1032 |
+
"C_lip_orig": 3.49684,
|
| 1033 |
+
"sigma_x": 55.4163
|
| 1034 |
+
},
|
| 1035 |
+
{
|
| 1036 |
+
"arch": "cnn",
|
| 1037 |
+
"seed": 2,
|
| 1038 |
+
"beta": 0.5,
|
| 1039 |
+
"N": 404416,
|
| 1040 |
+
"n_eff": 210848.23,
|
| 1041 |
+
"nu": 105424,
|
| 1042 |
+
"retain_frac": 0.260682,
|
| 1043 |
+
"s_eff": 0.572948,
|
| 1044 |
+
"delta_w_norm": 8.997647,
|
| 1045 |
+
"delta_w_frac": 0.42415,
|
| 1046 |
+
"loss_change_mse": 0.00068492,
|
| 1047 |
+
"loss_bound": 11875.2977,
|
| 1048 |
+
"loss_bound_holds": true,
|
| 1049 |
+
"C_lip_orig": 3.49684,
|
| 1050 |
+
"sigma_x": 55.4163
|
| 1051 |
+
},
|
| 1052 |
+
{
|
| 1053 |
+
"arch": "cnn",
|
| 1054 |
+
"seed": 2,
|
| 1055 |
+
"beta": 0.75,
|
| 1056 |
+
"N": 404416,
|
| 1057 |
+
"n_eff": 210848.23,
|
| 1058 |
+
"nu": 158136,
|
| 1059 |
+
"retain_frac": 0.391023,
|
| 1060 |
+
"s_eff": 0.721444,
|
| 1061 |
+
"delta_w_norm": 6.402608,
|
| 1062 |
+
"delta_w_frac": 0.301819,
|
| 1063 |
+
"loss_change_mse": 0.000522,
|
| 1064 |
+
"loss_bound": 6013.1272,
|
| 1065 |
+
"loss_bound_holds": true,
|
| 1066 |
+
"C_lip_orig": 3.49684,
|
| 1067 |
+
"sigma_x": 55.4163
|
| 1068 |
+
},
|
| 1069 |
+
{
|
| 1070 |
+
"arch": "cnn",
|
| 1071 |
+
"seed": 2,
|
| 1072 |
+
"beta": 1.0,
|
| 1073 |
+
"N": 404416,
|
| 1074 |
+
"n_eff": 210848.23,
|
| 1075 |
+
"nu": 210848,
|
| 1076 |
+
"retain_frac": 0.521364,
|
| 1077 |
+
"s_eff": 0.832488,
|
| 1078 |
+
"delta_w_norm": 4.316469,
|
| 1079 |
+
"delta_w_frac": 0.203479,
|
| 1080 |
+
"loss_change_mse": 0.00016988,
|
| 1081 |
+
"loss_bound": 2733.0261,
|
| 1082 |
+
"loss_bound_holds": true,
|
| 1083 |
+
"C_lip_orig": 3.49684,
|
| 1084 |
+
"sigma_x": 55.4163
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"arch": "cnn",
|
| 1088 |
+
"seed": 2,
|
| 1089 |
+
"beta": 1.25,
|
| 1090 |
+
"N": 404416,
|
| 1091 |
+
"n_eff": 210848.23,
|
| 1092 |
+
"nu": 263560,
|
| 1093 |
+
"retain_frac": 0.651705,
|
| 1094 |
+
"s_eff": 0.912934,
|
| 1095 |
+
"delta_w_norm": 2.619597,
|
| 1096 |
+
"delta_w_frac": 0.123488,
|
| 1097 |
+
"loss_change_mse": 0.00011645,
|
| 1098 |
+
"loss_bound": 1006.5971,
|
| 1099 |
+
"loss_bound_holds": true,
|
| 1100 |
+
"C_lip_orig": 3.49684,
|
| 1101 |
+
"sigma_x": 55.4163
|
| 1102 |
+
},
|
| 1103 |
+
{
|
| 1104 |
+
"arch": "cnn",
|
| 1105 |
+
"seed": 2,
|
| 1106 |
+
"beta": 1.5,
|
| 1107 |
+
"N": 404416,
|
| 1108 |
+
"n_eff": 210848.23,
|
| 1109 |
+
"nu": 316272,
|
| 1110 |
+
"retain_frac": 0.782046,
|
| 1111 |
+
"s_eff": 0.9663,
|
| 1112 |
+
"delta_w_norm": 1.278642,
|
| 1113 |
+
"delta_w_frac": 0.060275,
|
| 1114 |
+
"loss_change_mse": 9.66e-06,
|
| 1115 |
+
"loss_bound": 239.8194,
|
| 1116 |
+
"loss_bound_holds": true,
|
| 1117 |
+
"C_lip_orig": 3.49684,
|
| 1118 |
+
"sigma_x": 55.4163
|
| 1119 |
+
},
|
| 1120 |
+
{
|
| 1121 |
+
"arch": "cnn",
|
| 1122 |
+
"seed": 2,
|
| 1123 |
+
"beta": 2.0,
|
| 1124 |
+
"N": 404416,
|
| 1125 |
+
"n_eff": 210848.23,
|
| 1126 |
+
"nu": 404415,
|
| 1127 |
+
"retain_frac": 0.999998,
|
| 1128 |
+
"s_eff": 1.0,
|
| 1129 |
+
"delta_w_norm": 0.0,
|
| 1130 |
+
"delta_w_frac": 0.0,
|
| 1131 |
+
"loss_change_mse": 0.0,
|
| 1132 |
+
"loss_bound": 0.0,
|
| 1133 |
+
"loss_bound_holds": true,
|
| 1134 |
+
"C_lip_orig": 3.49684,
|
| 1135 |
+
"sigma_x": 55.4163
|
| 1136 |
+
},
|
| 1137 |
+
{
|
| 1138 |
+
"arch": "transformer",
|
| 1139 |
+
"seed": 0,
|
| 1140 |
+
"beta": 0.25,
|
| 1141 |
+
"N": 78464,
|
| 1142 |
+
"n_eff": 51293.48,
|
| 1143 |
+
"nu": 12823,
|
| 1144 |
+
"retain_frac": 0.163425,
|
| 1145 |
+
"s_eff": 0.363358,
|
| 1146 |
+
"delta_w_norm": 19.467049,
|
| 1147 |
+
"delta_w_frac": 0.6602,
|
| 1148 |
+
"loss_change_mse": 0.36617145,
|
| 1149 |
+
"loss_bound": 8488108374.8184,
|
| 1150 |
+
"loss_bound_holds": true,
|
| 1151 |
+
"C_lip_orig": 2358.180067,
|
| 1152 |
+
"sigma_x": 32.1106
|
| 1153 |
+
},
|
| 1154 |
+
{
|
| 1155 |
+
"arch": "transformer",
|
| 1156 |
+
"seed": 0,
|
| 1157 |
+
"beta": 0.5,
|
| 1158 |
+
"N": 78464,
|
| 1159 |
+
"n_eff": 51293.48,
|
| 1160 |
+
"nu": 25646,
|
| 1161 |
+
"retain_frac": 0.326851,
|
| 1162 |
+
"s_eff": 0.599243,
|
| 1163 |
+
"delta_w_norm": 13.571939,
|
| 1164 |
+
"delta_w_frac": 0.460275,
|
| 1165 |
+
"loss_change_mse": 0.17683111,
|
| 1166 |
+
"loss_bound": 4125670271.6632,
|
| 1167 |
+
"loss_bound_holds": true,
|
| 1168 |
+
"C_lip_orig": 2358.180067,
|
| 1169 |
+
"sigma_x": 32.1106
|
| 1170 |
+
},
|
| 1171 |
+
{
|
| 1172 |
+
"arch": "transformer",
|
| 1173 |
+
"seed": 0,
|
| 1174 |
+
"beta": 0.75,
|
| 1175 |
+
"N": 78464,
|
| 1176 |
+
"n_eff": 51293.48,
|
| 1177 |
+
"nu": 38470,
|
| 1178 |
+
"retain_frac": 0.490289,
|
| 1179 |
+
"s_eff": 0.774719,
|
| 1180 |
+
"delta_w_norm": 8.728827,
|
| 1181 |
+
"delta_w_frac": 0.296027,
|
| 1182 |
+
"loss_change_mse": 0.11401466,
|
| 1183 |
+
"loss_bound": 1706563341.9569,
|
| 1184 |
+
"loss_bound_holds": true,
|
| 1185 |
+
"C_lip_orig": 2358.180067,
|
| 1186 |
+
"sigma_x": 32.1106
|
| 1187 |
+
},
|
| 1188 |
+
{
|
| 1189 |
+
"arch": "transformer",
|
| 1190 |
+
"seed": 0,
|
| 1191 |
+
"beta": 1.0,
|
| 1192 |
+
"N": 78464,
|
| 1193 |
+
"n_eff": 51293.48,
|
| 1194 |
+
"nu": 51293,
|
| 1195 |
+
"retain_frac": 0.653714,
|
| 1196 |
+
"s_eff": 0.89739,
|
| 1197 |
+
"delta_w_norm": 4.811022,
|
| 1198 |
+
"delta_w_frac": 0.16316,
|
| 1199 |
+
"loss_change_mse": 0.03556463,
|
| 1200 |
+
"loss_bound": 518424388.7459,
|
| 1201 |
+
"loss_bound_holds": true,
|
| 1202 |
+
"C_lip_orig": 2358.180067,
|
| 1203 |
+
"sigma_x": 32.1106
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"arch": "transformer",
|
| 1207 |
+
"seed": 0,
|
| 1208 |
+
"beta": 1.25,
|
| 1209 |
+
"N": 78464,
|
| 1210 |
+
"n_eff": 51293.48,
|
| 1211 |
+
"nu": 64116,
|
| 1212 |
+
"retain_frac": 0.817139,
|
| 1213 |
+
"s_eff": 0.97161,
|
| 1214 |
+
"delta_w_norm": 1.825973,
|
| 1215 |
+
"delta_w_frac": 0.061926,
|
| 1216 |
+
"loss_change_mse": 0.00704481,
|
| 1217 |
+
"loss_bound": 74679127.7396,
|
| 1218 |
+
"loss_bound_holds": true,
|
| 1219 |
+
"C_lip_orig": 2358.180067,
|
| 1220 |
+
"sigma_x": 32.1106
|
| 1221 |
+
},
|
| 1222 |
+
{
|
| 1223 |
+
"arch": "transformer",
|
| 1224 |
+
"seed": 0,
|
| 1225 |
+
"beta": 1.5,
|
| 1226 |
+
"N": 78464,
|
| 1227 |
+
"n_eff": 51293.48,
|
| 1228 |
+
"nu": 76940,
|
| 1229 |
+
"retain_frac": 0.980577,
|
| 1230 |
+
"s_eff": 0.999679,
|
| 1231 |
+
"delta_w_norm": 0.063212,
|
| 1232 |
+
"delta_w_frac": 0.002144,
|
| 1233 |
+
"loss_change_mse": 6.77e-06,
|
| 1234 |
+
"loss_bound": 89496.65,
|
| 1235 |
+
"loss_bound_holds": true,
|
| 1236 |
+
"C_lip_orig": 2358.180067,
|
| 1237 |
+
"sigma_x": 32.1106
|
| 1238 |
+
},
|
| 1239 |
+
{
|
| 1240 |
+
"arch": "transformer",
|
| 1241 |
+
"seed": 0,
|
| 1242 |
+
"beta": 2.0,
|
| 1243 |
+
"N": 78464,
|
| 1244 |
+
"n_eff": 51293.48,
|
| 1245 |
+
"nu": 78463,
|
| 1246 |
+
"retain_frac": 0.999987,
|
| 1247 |
+
"s_eff": 1.0,
|
| 1248 |
+
"delta_w_norm": 0.0,
|
| 1249 |
+
"delta_w_frac": 0.0,
|
| 1250 |
+
"loss_change_mse": 0.0,
|
| 1251 |
+
"loss_bound": 0.0,
|
| 1252 |
+
"loss_bound_holds": true,
|
| 1253 |
+
"C_lip_orig": 2358.180067,
|
| 1254 |
+
"sigma_x": 32.1106
|
| 1255 |
+
},
|
| 1256 |
+
{
|
| 1257 |
+
"arch": "transformer",
|
| 1258 |
+
"seed": 1,
|
| 1259 |
+
"beta": 0.25,
|
| 1260 |
+
"N": 78464,
|
| 1261 |
+
"n_eff": 51293.48,
|
| 1262 |
+
"nu": 12823,
|
| 1263 |
+
"retain_frac": 0.163425,
|
| 1264 |
+
"s_eff": 0.363358,
|
| 1265 |
+
"delta_w_norm": 19.467049,
|
| 1266 |
+
"delta_w_frac": 0.6602,
|
| 1267 |
+
"loss_change_mse": 0.35847807,
|
| 1268 |
+
"loss_bound": 8449857467.2591,
|
| 1269 |
+
"loss_bound_holds": true,
|
| 1270 |
+
"C_lip_orig": 2358.180067,
|
| 1271 |
+
"sigma_x": 32.0382
|
| 1272 |
+
},
|
| 1273 |
+
{
|
| 1274 |
+
"arch": "transformer",
|
| 1275 |
+
"seed": 1,
|
| 1276 |
+
"beta": 0.5,
|
| 1277 |
+
"N": 78464,
|
| 1278 |
+
"n_eff": 51293.48,
|
| 1279 |
+
"nu": 25646,
|
| 1280 |
+
"retain_frac": 0.326851,
|
| 1281 |
+
"s_eff": 0.599243,
|
| 1282 |
+
"delta_w_norm": 13.571939,
|
| 1283 |
+
"delta_w_frac": 0.460275,
|
| 1284 |
+
"loss_change_mse": 0.16661619,
|
| 1285 |
+
"loss_bound": 4107078304.5001,
|
| 1286 |
+
"loss_bound_holds": true,
|
| 1287 |
+
"C_lip_orig": 2358.180067,
|
| 1288 |
+
"sigma_x": 32.0382
|
| 1289 |
+
},
|
| 1290 |
+
{
|
| 1291 |
+
"arch": "transformer",
|
| 1292 |
+
"seed": 1,
|
| 1293 |
+
"beta": 0.75,
|
| 1294 |
+
"N": 78464,
|
| 1295 |
+
"n_eff": 51293.48,
|
| 1296 |
+
"nu": 38470,
|
| 1297 |
+
"retain_frac": 0.490289,
|
| 1298 |
+
"s_eff": 0.774719,
|
| 1299 |
+
"delta_w_norm": 8.728827,
|
| 1300 |
+
"delta_w_frac": 0.296027,
|
| 1301 |
+
"loss_change_mse": 0.10661874,
|
| 1302 |
+
"loss_bound": 1698872865.6158,
|
| 1303 |
+
"loss_bound_holds": true,
|
| 1304 |
+
"C_lip_orig": 2358.180067,
|
| 1305 |
+
"sigma_x": 32.0382
|
| 1306 |
+
},
|
| 1307 |
+
{
|
| 1308 |
+
"arch": "transformer",
|
| 1309 |
+
"seed": 1,
|
| 1310 |
+
"beta": 1.0,
|
| 1311 |
+
"N": 78464,
|
| 1312 |
+
"n_eff": 51293.48,
|
| 1313 |
+
"nu": 51293,
|
| 1314 |
+
"retain_frac": 0.653714,
|
| 1315 |
+
"s_eff": 0.89739,
|
| 1316 |
+
"delta_w_norm": 4.811022,
|
| 1317 |
+
"delta_w_frac": 0.16316,
|
| 1318 |
+
"loss_change_mse": 0.03537756,
|
| 1319 |
+
"loss_bound": 516088155.2184,
|
| 1320 |
+
"loss_bound_holds": true,
|
| 1321 |
+
"C_lip_orig": 2358.180067,
|
| 1322 |
+
"sigma_x": 32.0382
|
| 1323 |
+
},
|
| 1324 |
+
{
|
| 1325 |
+
"arch": "transformer",
|
| 1326 |
+
"seed": 1,
|
| 1327 |
+
"beta": 1.25,
|
| 1328 |
+
"N": 78464,
|
| 1329 |
+
"n_eff": 51293.48,
|
| 1330 |
+
"nu": 64116,
|
| 1331 |
+
"retain_frac": 0.817139,
|
| 1332 |
+
"s_eff": 0.97161,
|
| 1333 |
+
"delta_w_norm": 1.825973,
|
| 1334 |
+
"delta_w_frac": 0.061926,
|
| 1335 |
+
"loss_change_mse": 0.00703152,
|
| 1336 |
+
"loss_bound": 74342592.8739,
|
| 1337 |
+
"loss_bound_holds": true,
|
| 1338 |
+
"C_lip_orig": 2358.180067,
|
| 1339 |
+
"sigma_x": 32.0382
|
| 1340 |
+
},
|
| 1341 |
+
{
|
| 1342 |
+
"arch": "transformer",
|
| 1343 |
+
"seed": 1,
|
| 1344 |
+
"beta": 1.5,
|
| 1345 |
+
"N": 78464,
|
| 1346 |
+
"n_eff": 51293.48,
|
| 1347 |
+
"nu": 76940,
|
| 1348 |
+
"retain_frac": 0.980577,
|
| 1349 |
+
"s_eff": 0.999679,
|
| 1350 |
+
"delta_w_norm": 0.063212,
|
| 1351 |
+
"delta_w_frac": 0.002144,
|
| 1352 |
+
"loss_change_mse": 6.44e-06,
|
| 1353 |
+
"loss_bound": 89093.3413,
|
| 1354 |
+
"loss_bound_holds": true,
|
| 1355 |
+
"C_lip_orig": 2358.180067,
|
| 1356 |
+
"sigma_x": 32.0382
|
| 1357 |
+
},
|
| 1358 |
+
{
|
| 1359 |
+
"arch": "transformer",
|
| 1360 |
+
"seed": 1,
|
| 1361 |
+
"beta": 2.0,
|
| 1362 |
+
"N": 78464,
|
| 1363 |
+
"n_eff": 51293.48,
|
| 1364 |
+
"nu": 78463,
|
| 1365 |
+
"retain_frac": 0.999987,
|
| 1366 |
+
"s_eff": 1.0,
|
| 1367 |
+
"delta_w_norm": 0.0,
|
| 1368 |
+
"delta_w_frac": 0.0,
|
| 1369 |
+
"loss_change_mse": 0.0,
|
| 1370 |
+
"loss_bound": 0.0,
|
| 1371 |
+
"loss_bound_holds": true,
|
| 1372 |
+
"C_lip_orig": 2358.180067,
|
| 1373 |
+
"sigma_x": 32.0382
|
| 1374 |
+
},
|
| 1375 |
+
{
|
| 1376 |
+
"arch": "transformer",
|
| 1377 |
+
"seed": 2,
|
| 1378 |
+
"beta": 0.25,
|
| 1379 |
+
"N": 78464,
|
| 1380 |
+
"n_eff": 51293.48,
|
| 1381 |
+
"nu": 12823,
|
| 1382 |
+
"retain_frac": 0.163425,
|
| 1383 |
+
"s_eff": 0.363358,
|
| 1384 |
+
"delta_w_norm": 19.467049,
|
| 1385 |
+
"delta_w_frac": 0.6602,
|
| 1386 |
+
"loss_change_mse": 0.37094107,
|
| 1387 |
+
"loss_bound": 8438307345.4753,
|
| 1388 |
+
"loss_bound_holds": true,
|
| 1389 |
+
"C_lip_orig": 2358.180067,
|
| 1390 |
+
"sigma_x": 32.0163
|
| 1391 |
+
},
|
| 1392 |
+
{
|
| 1393 |
+
"arch": "transformer",
|
| 1394 |
+
"seed": 2,
|
| 1395 |
+
"beta": 0.5,
|
| 1396 |
+
"N": 78464,
|
| 1397 |
+
"n_eff": 51293.48,
|
| 1398 |
+
"nu": 25646,
|
| 1399 |
+
"retain_frac": 0.326851,
|
| 1400 |
+
"s_eff": 0.599243,
|
| 1401 |
+
"delta_w_norm": 13.571939,
|
| 1402 |
+
"delta_w_frac": 0.460275,
|
| 1403 |
+
"loss_change_mse": 0.17148171,
|
| 1404 |
+
"loss_bound": 4101464333.5218,
|
| 1405 |
+
"loss_bound_holds": true,
|
| 1406 |
+
"C_lip_orig": 2358.180067,
|
| 1407 |
+
"sigma_x": 32.0163
|
| 1408 |
+
},
|
| 1409 |
+
{
|
| 1410 |
+
"arch": "transformer",
|
| 1411 |
+
"seed": 2,
|
| 1412 |
+
"beta": 0.75,
|
| 1413 |
+
"N": 78464,
|
| 1414 |
+
"n_eff": 51293.48,
|
| 1415 |
+
"nu": 38470,
|
| 1416 |
+
"retain_frac": 0.490289,
|
| 1417 |
+
"s_eff": 0.774719,
|
| 1418 |
+
"delta_w_norm": 8.728827,
|
| 1419 |
+
"delta_w_frac": 0.296027,
|
| 1420 |
+
"loss_change_mse": 0.11324503,
|
| 1421 |
+
"loss_bound": 1696550673.9612,
|
| 1422 |
+
"loss_bound_holds": true,
|
| 1423 |
+
"C_lip_orig": 2358.180067,
|
| 1424 |
+
"sigma_x": 32.0163
|
| 1425 |
+
},
|
| 1426 |
+
{
|
| 1427 |
+
"arch": "transformer",
|
| 1428 |
+
"seed": 2,
|
| 1429 |
+
"beta": 1.0,
|
| 1430 |
+
"N": 78464,
|
| 1431 |
+
"n_eff": 51293.48,
|
| 1432 |
+
"nu": 51293,
|
| 1433 |
+
"retain_frac": 0.653714,
|
| 1434 |
+
"s_eff": 0.89739,
|
| 1435 |
+
"delta_w_norm": 4.811022,
|
| 1436 |
+
"delta_w_frac": 0.16316,
|
| 1437 |
+
"loss_change_mse": 0.03405446,
|
| 1438 |
+
"loss_bound": 515382713.6098,
|
| 1439 |
+
"loss_bound_holds": true,
|
| 1440 |
+
"C_lip_orig": 2358.180067,
|
| 1441 |
+
"sigma_x": 32.0163
|
| 1442 |
+
},
|
| 1443 |
+
{
|
| 1444 |
+
"arch": "transformer",
|
| 1445 |
+
"seed": 2,
|
| 1446 |
+
"beta": 1.25,
|
| 1447 |
+
"N": 78464,
|
| 1448 |
+
"n_eff": 51293.48,
|
| 1449 |
+
"nu": 64116,
|
| 1450 |
+
"retain_frac": 0.817139,
|
| 1451 |
+
"s_eff": 0.97161,
|
| 1452 |
+
"delta_w_norm": 1.825973,
|
| 1453 |
+
"delta_w_frac": 0.061926,
|
| 1454 |
+
"loss_change_mse": 0.00656284,
|
| 1455 |
+
"loss_bound": 74240973.8816,
|
| 1456 |
+
"loss_bound_holds": true,
|
| 1457 |
+
"C_lip_orig": 2358.180067,
|
| 1458 |
+
"sigma_x": 32.0163
|
| 1459 |
+
},
|
| 1460 |
+
{
|
| 1461 |
+
"arch": "transformer",
|
| 1462 |
+
"seed": 2,
|
| 1463 |
+
"beta": 1.5,
|
| 1464 |
+
"N": 78464,
|
| 1465 |
+
"n_eff": 51293.48,
|
| 1466 |
+
"nu": 76940,
|
| 1467 |
+
"retain_frac": 0.980577,
|
| 1468 |
+
"s_eff": 0.999679,
|
| 1469 |
+
"delta_w_norm": 0.063212,
|
| 1470 |
+
"delta_w_frac": 0.002144,
|
| 1471 |
+
"loss_change_mse": 6.58e-06,
|
| 1472 |
+
"loss_bound": 88971.5595,
|
| 1473 |
+
"loss_bound_holds": true,
|
| 1474 |
+
"C_lip_orig": 2358.180067,
|
| 1475 |
+
"sigma_x": 32.0163
|
| 1476 |
+
},
|
| 1477 |
+
{
|
| 1478 |
+
"arch": "transformer",
|
| 1479 |
+
"seed": 2,
|
| 1480 |
+
"beta": 2.0,
|
| 1481 |
+
"N": 78464,
|
| 1482 |
+
"n_eff": 51293.48,
|
| 1483 |
+
"nu": 78463,
|
| 1484 |
+
"retain_frac": 0.999987,
|
| 1485 |
+
"s_eff": 1.0,
|
| 1486 |
+
"delta_w_norm": 0.0,
|
| 1487 |
+
"delta_w_frac": 0.0,
|
| 1488 |
+
"loss_change_mse": 0.0,
|
| 1489 |
+
"loss_bound": 0.0,
|
| 1490 |
+
"loss_bound_holds": true,
|
| 1491 |
+
"C_lip_orig": 2358.180067,
|
| 1492 |
+
"sigma_x": 32.0163
|
| 1493 |
+
}
|
| 1494 |
+
],
|
| 1495 |
+
"elapsed_s": 3.62,
|
| 1496 |
+
"upstream_commit": "4fdbca2",
|
| 1497 |
+
"sha256": "ef7f2aa340d4ce2e80c2edd651bbba8a367355e4833f5592e378ebb93edb555f"
|
| 1498 |
+
}
|
results/emp_resnet18_synth_loss.json
ADDED
|
@@ -0,0 +1,419 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"claim": 2,
|
| 3 |
+
"description": "ResNet18 EMP prune: actual logit MSE / agreement vs dense teacher",
|
| 4 |
+
"disclosure": "scaled / comparable \u2014 ImageNet-pretrained ResNet18; synthetic normalized 224 inputs (no CIFAR download); metrics vs dense teacher logits",
|
| 5 |
+
"model": "torchvision resnet18 IMAGENET1K_V1",
|
| 6 |
+
"n_samples": 128,
|
| 7 |
+
"seeds": [
|
| 8 |
+
0,
|
| 9 |
+
1,
|
| 10 |
+
2
|
| 11 |
+
],
|
| 12 |
+
"betas": [
|
| 13 |
+
0.25,
|
| 14 |
+
0.5,
|
| 15 |
+
0.75,
|
| 16 |
+
1.0,
|
| 17 |
+
1.25,
|
| 18 |
+
1.5,
|
| 19 |
+
2.0
|
| 20 |
+
],
|
| 21 |
+
"environment": {
|
| 22 |
+
"os": "Windows 11",
|
| 23 |
+
"python": "3.13.3",
|
| 24 |
+
"torch": "2.6.0+cu124",
|
| 25 |
+
"torchvision": "0.21.0+cu124",
|
| 26 |
+
"numpy": "2.4.4",
|
| 27 |
+
"device": "cuda",
|
| 28 |
+
"gpu": "NVIDIA GeForce RTX 4060 Laptop GPU",
|
| 29 |
+
"timestamp_utc": "2026-07-19T07:25:55.667242+00:00"
|
| 30 |
+
},
|
| 31 |
+
"rows": [
|
| 32 |
+
{
|
| 33 |
+
"seed": 0,
|
| 34 |
+
"beta": 0.25,
|
| 35 |
+
"N": 11678912,
|
| 36 |
+
"n_eff": 4970630.0,
|
| 37 |
+
"nu": 1242657,
|
| 38 |
+
"s_eff": 0.359672,
|
| 39 |
+
"sparsity": 0.893598,
|
| 40 |
+
"retain_frac": 0.106402,
|
| 41 |
+
"logit_mse": 48.4118537902832,
|
| 42 |
+
"agree_with_dense": 0.0,
|
| 43 |
+
"ce_vs_dense_pseudo": 29.268400192260742
|
| 44 |
+
},
|
| 45 |
+
{
|
| 46 |
+
"seed": 0,
|
| 47 |
+
"beta": 0.5,
|
| 48 |
+
"N": 11678912,
|
| 49 |
+
"n_eff": 4970630.0,
|
| 50 |
+
"nu": 2485315,
|
| 51 |
+
"s_eff": 0.537804,
|
| 52 |
+
"sparsity": 0.787196,
|
| 53 |
+
"retain_frac": 0.212804,
|
| 54 |
+
"logit_mse": 10.004705429077148,
|
| 55 |
+
"agree_with_dense": 0.0,
|
| 56 |
+
"ce_vs_dense_pseudo": 13.888275146484375
|
| 57 |
+
},
|
| 58 |
+
{
|
| 59 |
+
"seed": 0,
|
| 60 |
+
"beta": 0.75,
|
| 61 |
+
"N": 11678912,
|
| 62 |
+
"n_eff": 4970630.0,
|
| 63 |
+
"nu": 3727972,
|
| 64 |
+
"s_eff": 0.670098,
|
| 65 |
+
"sparsity": 0.680795,
|
| 66 |
+
"retain_frac": 0.319205,
|
| 67 |
+
"logit_mse": 4.766587734222412,
|
| 68 |
+
"agree_with_dense": 0.0,
|
| 69 |
+
"ce_vs_dense_pseudo": 6.906761169433594
|
| 70 |
+
},
|
| 71 |
+
{
|
| 72 |
+
"seed": 0,
|
| 73 |
+
"beta": 1.0,
|
| 74 |
+
"N": 11678912,
|
| 75 |
+
"n_eff": 4970630.0,
|
| 76 |
+
"nu": 4970630,
|
| 77 |
+
"s_eff": 0.772863,
|
| 78 |
+
"sparsity": 0.574393,
|
| 79 |
+
"retain_frac": 0.425607,
|
| 80 |
+
"logit_mse": 1.5024768114089966,
|
| 81 |
+
"agree_with_dense": 0.984375,
|
| 82 |
+
"ce_vs_dense_pseudo": 1.5291050672531128
|
| 83 |
+
},
|
| 84 |
+
{
|
| 85 |
+
"seed": 0,
|
| 86 |
+
"beta": 1.25,
|
| 87 |
+
"N": 11678912,
|
| 88 |
+
"n_eff": 4970630.0,
|
| 89 |
+
"nu": 6213287,
|
| 90 |
+
"s_eff": 0.852904,
|
| 91 |
+
"sparsity": 0.467991,
|
| 92 |
+
"retain_frac": 0.532009,
|
| 93 |
+
"logit_mse": 0.5345480442047119,
|
| 94 |
+
"agree_with_dense": 1.0,
|
| 95 |
+
"ce_vs_dense_pseudo": 0.7689579725265503
|
| 96 |
+
},
|
| 97 |
+
{
|
| 98 |
+
"seed": 0,
|
| 99 |
+
"beta": 1.5,
|
| 100 |
+
"N": 11678912,
|
| 101 |
+
"n_eff": 4970630.0,
|
| 102 |
+
"nu": 7455945,
|
| 103 |
+
"s_eff": 0.913787,
|
| 104 |
+
"sparsity": 0.361589,
|
| 105 |
+
"retain_frac": 0.638411,
|
| 106 |
+
"logit_mse": 0.32631343603134155,
|
| 107 |
+
"agree_with_dense": 1.0,
|
| 108 |
+
"ce_vs_dense_pseudo": 0.8859466910362244
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"seed": 0,
|
| 112 |
+
"beta": 2.0,
|
| 113 |
+
"N": 11678912,
|
| 114 |
+
"n_eff": 4970630.0,
|
| 115 |
+
"nu": 9941260,
|
| 116 |
+
"s_eff": 0.98575,
|
| 117 |
+
"sparsity": 0.148785,
|
| 118 |
+
"retain_frac": 0.851215,
|
| 119 |
+
"logit_mse": 0.013040440157055855,
|
| 120 |
+
"agree_with_dense": 1.0,
|
| 121 |
+
"ce_vs_dense_pseudo": 0.33175161480903625
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"seed": 1,
|
| 125 |
+
"beta": 0.25,
|
| 126 |
+
"N": 11678912,
|
| 127 |
+
"n_eff": 4970630.0,
|
| 128 |
+
"nu": 1242657,
|
| 129 |
+
"s_eff": 0.359672,
|
| 130 |
+
"sparsity": 0.893598,
|
| 131 |
+
"retain_frac": 0.106402,
|
| 132 |
+
"logit_mse": 48.27888107299805,
|
| 133 |
+
"agree_with_dense": 0.0,
|
| 134 |
+
"ce_vs_dense_pseudo": 29.21817970275879
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"seed": 1,
|
| 138 |
+
"beta": 0.5,
|
| 139 |
+
"N": 11678912,
|
| 140 |
+
"n_eff": 4970630.0,
|
| 141 |
+
"nu": 2485315,
|
| 142 |
+
"s_eff": 0.537804,
|
| 143 |
+
"sparsity": 0.787196,
|
| 144 |
+
"retain_frac": 0.212804,
|
| 145 |
+
"logit_mse": 9.95286750793457,
|
| 146 |
+
"agree_with_dense": 0.0,
|
| 147 |
+
"ce_vs_dense_pseudo": 13.783380508422852
|
| 148 |
+
},
|
| 149 |
+
{
|
| 150 |
+
"seed": 1,
|
| 151 |
+
"beta": 0.75,
|
| 152 |
+
"N": 11678912,
|
| 153 |
+
"n_eff": 4970630.0,
|
| 154 |
+
"nu": 3727972,
|
| 155 |
+
"s_eff": 0.670098,
|
| 156 |
+
"sparsity": 0.680795,
|
| 157 |
+
"retain_frac": 0.319205,
|
| 158 |
+
"logit_mse": 4.763496398925781,
|
| 159 |
+
"agree_with_dense": 0.0,
|
| 160 |
+
"ce_vs_dense_pseudo": 6.907163143157959
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"seed": 1,
|
| 164 |
+
"beta": 1.0,
|
| 165 |
+
"N": 11678912,
|
| 166 |
+
"n_eff": 4970630.0,
|
| 167 |
+
"nu": 4970630,
|
| 168 |
+
"s_eff": 0.772863,
|
| 169 |
+
"sparsity": 0.574393,
|
| 170 |
+
"retain_frac": 0.425607,
|
| 171 |
+
"logit_mse": 1.4761651754379272,
|
| 172 |
+
"agree_with_dense": 0.96875,
|
| 173 |
+
"ce_vs_dense_pseudo": 1.5737754106521606
|
| 174 |
+
},
|
| 175 |
+
{
|
| 176 |
+
"seed": 1,
|
| 177 |
+
"beta": 1.25,
|
| 178 |
+
"N": 11678912,
|
| 179 |
+
"n_eff": 4970630.0,
|
| 180 |
+
"nu": 6213287,
|
| 181 |
+
"s_eff": 0.852904,
|
| 182 |
+
"sparsity": 0.467991,
|
| 183 |
+
"retain_frac": 0.532009,
|
| 184 |
+
"logit_mse": 0.5283772349357605,
|
| 185 |
+
"agree_with_dense": 1.0,
|
| 186 |
+
"ce_vs_dense_pseudo": 0.7740398645401001
|
| 187 |
+
},
|
| 188 |
+
{
|
| 189 |
+
"seed": 1,
|
| 190 |
+
"beta": 1.5,
|
| 191 |
+
"N": 11678912,
|
| 192 |
+
"n_eff": 4970630.0,
|
| 193 |
+
"nu": 7455945,
|
| 194 |
+
"s_eff": 0.913787,
|
| 195 |
+
"sparsity": 0.361589,
|
| 196 |
+
"retain_frac": 0.638411,
|
| 197 |
+
"logit_mse": 0.3183683753013611,
|
| 198 |
+
"agree_with_dense": 1.0,
|
| 199 |
+
"ce_vs_dense_pseudo": 0.8864942193031311
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"seed": 1,
|
| 203 |
+
"beta": 2.0,
|
| 204 |
+
"N": 11678912,
|
| 205 |
+
"n_eff": 4970630.0,
|
| 206 |
+
"nu": 9941260,
|
| 207 |
+
"s_eff": 0.98575,
|
| 208 |
+
"sparsity": 0.148785,
|
| 209 |
+
"retain_frac": 0.851215,
|
| 210 |
+
"logit_mse": 0.012938400730490685,
|
| 211 |
+
"agree_with_dense": 1.0,
|
| 212 |
+
"ce_vs_dense_pseudo": 0.33158478140830994
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"seed": 2,
|
| 216 |
+
"beta": 0.25,
|
| 217 |
+
"N": 11678912,
|
| 218 |
+
"n_eff": 4970630.0,
|
| 219 |
+
"nu": 1242657,
|
| 220 |
+
"s_eff": 0.359672,
|
| 221 |
+
"sparsity": 0.893598,
|
| 222 |
+
"retain_frac": 0.106402,
|
| 223 |
+
"logit_mse": 48.321895599365234,
|
| 224 |
+
"agree_with_dense": 0.0,
|
| 225 |
+
"ce_vs_dense_pseudo": 29.21956443786621
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"seed": 2,
|
| 229 |
+
"beta": 0.5,
|
| 230 |
+
"N": 11678912,
|
| 231 |
+
"n_eff": 4970630.0,
|
| 232 |
+
"nu": 2485315,
|
| 233 |
+
"s_eff": 0.537804,
|
| 234 |
+
"sparsity": 0.787196,
|
| 235 |
+
"retain_frac": 0.212804,
|
| 236 |
+
"logit_mse": 10.018916130065918,
|
| 237 |
+
"agree_with_dense": 0.0,
|
| 238 |
+
"ce_vs_dense_pseudo": 13.871973037719727
|
| 239 |
+
},
|
| 240 |
+
{
|
| 241 |
+
"seed": 2,
|
| 242 |
+
"beta": 0.75,
|
| 243 |
+
"N": 11678912,
|
| 244 |
+
"n_eff": 4970630.0,
|
| 245 |
+
"nu": 3727972,
|
| 246 |
+
"s_eff": 0.670098,
|
| 247 |
+
"sparsity": 0.680795,
|
| 248 |
+
"retain_frac": 0.319205,
|
| 249 |
+
"logit_mse": 4.757717132568359,
|
| 250 |
+
"agree_with_dense": 0.0,
|
| 251 |
+
"ce_vs_dense_pseudo": 6.8995208740234375
|
| 252 |
+
},
|
| 253 |
+
{
|
| 254 |
+
"seed": 2,
|
| 255 |
+
"beta": 1.0,
|
| 256 |
+
"N": 11678912,
|
| 257 |
+
"n_eff": 4970630.0,
|
| 258 |
+
"nu": 4970630,
|
| 259 |
+
"s_eff": 0.772863,
|
| 260 |
+
"sparsity": 0.574393,
|
| 261 |
+
"retain_frac": 0.425607,
|
| 262 |
+
"logit_mse": 1.4677212238311768,
|
| 263 |
+
"agree_with_dense": 0.984375,
|
| 264 |
+
"ce_vs_dense_pseudo": 1.527180552482605
|
| 265 |
+
},
|
| 266 |
+
{
|
| 267 |
+
"seed": 2,
|
| 268 |
+
"beta": 1.25,
|
| 269 |
+
"N": 11678912,
|
| 270 |
+
"n_eff": 4970630.0,
|
| 271 |
+
"nu": 6213287,
|
| 272 |
+
"s_eff": 0.852904,
|
| 273 |
+
"sparsity": 0.467991,
|
| 274 |
+
"retain_frac": 0.532009,
|
| 275 |
+
"logit_mse": 0.5238240957260132,
|
| 276 |
+
"agree_with_dense": 1.0,
|
| 277 |
+
"ce_vs_dense_pseudo": 0.776103138923645
|
| 278 |
+
},
|
| 279 |
+
{
|
| 280 |
+
"seed": 2,
|
| 281 |
+
"beta": 1.5,
|
| 282 |
+
"N": 11678912,
|
| 283 |
+
"n_eff": 4970630.0,
|
| 284 |
+
"nu": 7455945,
|
| 285 |
+
"s_eff": 0.913787,
|
| 286 |
+
"sparsity": 0.361589,
|
| 287 |
+
"retain_frac": 0.638411,
|
| 288 |
+
"logit_mse": 0.31901127099990845,
|
| 289 |
+
"agree_with_dense": 1.0,
|
| 290 |
+
"ce_vs_dense_pseudo": 0.8787475824356079
|
| 291 |
+
},
|
| 292 |
+
{
|
| 293 |
+
"seed": 2,
|
| 294 |
+
"beta": 2.0,
|
| 295 |
+
"N": 11678912,
|
| 296 |
+
"n_eff": 4970630.0,
|
| 297 |
+
"nu": 9941260,
|
| 298 |
+
"s_eff": 0.98575,
|
| 299 |
+
"sparsity": 0.148785,
|
| 300 |
+
"retain_frac": 0.851215,
|
| 301 |
+
"logit_mse": 0.012943010777235031,
|
| 302 |
+
"agree_with_dense": 1.0,
|
| 303 |
+
"ce_vs_dense_pseudo": 0.32412609457969666
|
| 304 |
+
}
|
| 305 |
+
],
|
| 306 |
+
"seed0_table": [
|
| 307 |
+
{
|
| 308 |
+
"seed": 0,
|
| 309 |
+
"beta": 0.25,
|
| 310 |
+
"N": 11678912,
|
| 311 |
+
"n_eff": 4970630.0,
|
| 312 |
+
"nu": 1242657,
|
| 313 |
+
"s_eff": 0.359672,
|
| 314 |
+
"sparsity": 0.893598,
|
| 315 |
+
"retain_frac": 0.106402,
|
| 316 |
+
"logit_mse": 48.4118537902832,
|
| 317 |
+
"agree_with_dense": 0.0,
|
| 318 |
+
"ce_vs_dense_pseudo": 29.268400192260742
|
| 319 |
+
},
|
| 320 |
+
{
|
| 321 |
+
"seed": 0,
|
| 322 |
+
"beta": 0.5,
|
| 323 |
+
"N": 11678912,
|
| 324 |
+
"n_eff": 4970630.0,
|
| 325 |
+
"nu": 2485315,
|
| 326 |
+
"s_eff": 0.537804,
|
| 327 |
+
"sparsity": 0.787196,
|
| 328 |
+
"retain_frac": 0.212804,
|
| 329 |
+
"logit_mse": 10.004705429077148,
|
| 330 |
+
"agree_with_dense": 0.0,
|
| 331 |
+
"ce_vs_dense_pseudo": 13.888275146484375
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"seed": 0,
|
| 335 |
+
"beta": 0.75,
|
| 336 |
+
"N": 11678912,
|
| 337 |
+
"n_eff": 4970630.0,
|
| 338 |
+
"nu": 3727972,
|
| 339 |
+
"s_eff": 0.670098,
|
| 340 |
+
"sparsity": 0.680795,
|
| 341 |
+
"retain_frac": 0.319205,
|
| 342 |
+
"logit_mse": 4.766587734222412,
|
| 343 |
+
"agree_with_dense": 0.0,
|
| 344 |
+
"ce_vs_dense_pseudo": 6.906761169433594
|
| 345 |
+
},
|
| 346 |
+
{
|
| 347 |
+
"seed": 0,
|
| 348 |
+
"beta": 1.0,
|
| 349 |
+
"N": 11678912,
|
| 350 |
+
"n_eff": 4970630.0,
|
| 351 |
+
"nu": 4970630,
|
| 352 |
+
"s_eff": 0.772863,
|
| 353 |
+
"sparsity": 0.574393,
|
| 354 |
+
"retain_frac": 0.425607,
|
| 355 |
+
"logit_mse": 1.5024768114089966,
|
| 356 |
+
"agree_with_dense": 0.984375,
|
| 357 |
+
"ce_vs_dense_pseudo": 1.5291050672531128
|
| 358 |
+
},
|
| 359 |
+
{
|
| 360 |
+
"seed": 0,
|
| 361 |
+
"beta": 1.25,
|
| 362 |
+
"N": 11678912,
|
| 363 |
+
"n_eff": 4970630.0,
|
| 364 |
+
"nu": 6213287,
|
| 365 |
+
"s_eff": 0.852904,
|
| 366 |
+
"sparsity": 0.467991,
|
| 367 |
+
"retain_frac": 0.532009,
|
| 368 |
+
"logit_mse": 0.5345480442047119,
|
| 369 |
+
"agree_with_dense": 1.0,
|
| 370 |
+
"ce_vs_dense_pseudo": 0.7689579725265503
|
| 371 |
+
},
|
| 372 |
+
{
|
| 373 |
+
"seed": 0,
|
| 374 |
+
"beta": 1.5,
|
| 375 |
+
"N": 11678912,
|
| 376 |
+
"n_eff": 4970630.0,
|
| 377 |
+
"nu": 7455945,
|
| 378 |
+
"s_eff": 0.913787,
|
| 379 |
+
"sparsity": 0.361589,
|
| 380 |
+
"retain_frac": 0.638411,
|
| 381 |
+
"logit_mse": 0.32631343603134155,
|
| 382 |
+
"agree_with_dense": 1.0,
|
| 383 |
+
"ce_vs_dense_pseudo": 0.8859466910362244
|
| 384 |
+
},
|
| 385 |
+
{
|
| 386 |
+
"seed": 0,
|
| 387 |
+
"beta": 2.0,
|
| 388 |
+
"N": 11678912,
|
| 389 |
+
"n_eff": 4970630.0,
|
| 390 |
+
"nu": 9941260,
|
| 391 |
+
"s_eff": 0.98575,
|
| 392 |
+
"sparsity": 0.148785,
|
| 393 |
+
"retain_frac": 0.851215,
|
| 394 |
+
"logit_mse": 0.013040440157055855,
|
| 395 |
+
"agree_with_dense": 1.0,
|
| 396 |
+
"ce_vs_dense_pseudo": 0.33175161480903625
|
| 397 |
+
}
|
| 398 |
+
],
|
| 399 |
+
"highlight": {
|
| 400 |
+
"beta_0.25": {
|
| 401 |
+
"logit_mse": 48.4118537902832,
|
| 402 |
+
"agree_with_dense": 0.0,
|
| 403 |
+
"retain_frac": 0.106402
|
| 404 |
+
},
|
| 405 |
+
"beta_1.0": {
|
| 406 |
+
"logit_mse": 1.5024768114089966,
|
| 407 |
+
"agree_with_dense": 0.984375,
|
| 408 |
+
"retain_frac": 0.425607
|
| 409 |
+
},
|
| 410 |
+
"beta_1.5": {
|
| 411 |
+
"logit_mse": 0.32631343603134155,
|
| 412 |
+
"agree_with_dense": 1.0,
|
| 413 |
+
"retain_frac": 0.638411
|
| 414 |
+
}
|
| 415 |
+
},
|
| 416 |
+
"elapsed_s": 10.72,
|
| 417 |
+
"upstream_commit": "4fdbca2",
|
| 418 |
+
"sha256": "1d8e8021e74d47e4a7e168b24e6b292f0e925a59388f45e6754448205320a403"
|
| 419 |
+
}
|
results/emp_weight_stats.json
ADDED
|
@@ -0,0 +1,1676 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"experiment": "EMP weight statistics (multi-arch: MLP + CNN + Transformer)",
|
| 3 |
+
"paper": "Effective Model Pruning (c2CdXdEfqk)",
|
| 4 |
+
"upstream_commit": "4fdbca2",
|
| 5 |
+
"architecture_families": [
|
| 6 |
+
"MLP",
|
| 7 |
+
"CNN",
|
| 8 |
+
"Transformer"
|
| 9 |
+
],
|
| 10 |
+
"environment": {
|
| 11 |
+
"os": "Windows 11",
|
| 12 |
+
"python": "3.13.3",
|
| 13 |
+
"torch": "2.6.0+cu124",
|
| 14 |
+
"torchvision": "0.21.0+cu124",
|
| 15 |
+
"numpy": "2.4.4",
|
| 16 |
+
"device": "cpu"
|
| 17 |
+
},
|
| 18 |
+
"timestamp": "2026-07-18T22:51:34.777328+00:00",
|
| 19 |
+
"models": {
|
| 20 |
+
"MLP-4layer": {
|
| 21 |
+
"model": "MLP-4layer (784->2048->1024->512->10)",
|
| 22 |
+
"total_params": 4232192,
|
| 23 |
+
"mean_abs_weight": 0.021768247708678246,
|
| 24 |
+
"weight_std": 0.027512572705745697,
|
| 25 |
+
"layer_stats": [
|
| 26 |
+
{
|
| 27 |
+
"name": "0",
|
| 28 |
+
"type": "Linear",
|
| 29 |
+
"shape": [
|
| 30 |
+
2048,
|
| 31 |
+
784
|
| 32 |
+
],
|
| 33 |
+
"numel": 1605632,
|
| 34 |
+
"mean_abs": 0.021207734942436218,
|
| 35 |
+
"std": 0.02658386342227459
|
| 36 |
+
},
|
| 37 |
+
{
|
| 38 |
+
"name": "1",
|
| 39 |
+
"type": "Linear",
|
| 40 |
+
"shape": [
|
| 41 |
+
1024,
|
| 42 |
+
2048
|
| 43 |
+
],
|
| 44 |
+
"numel": 2097152,
|
| 45 |
+
"mean_abs": 0.02037639357149601,
|
| 46 |
+
"std": 0.02553321234881878
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"name": "2",
|
| 50 |
+
"type": "Linear",
|
| 51 |
+
"shape": [
|
| 52 |
+
512,
|
| 53 |
+
1024
|
| 54 |
+
],
|
| 55 |
+
"numel": 524288,
|
| 56 |
+
"mean_abs": 0.028783395886421204,
|
| 57 |
+
"std": 0.03607384115457535
|
| 58 |
+
},
|
| 59 |
+
{
|
| 60 |
+
"name": "3",
|
| 61 |
+
"type": "Linear",
|
| 62 |
+
"shape": [
|
| 63 |
+
10,
|
| 64 |
+
512
|
| 65 |
+
],
|
| 66 |
+
"numel": 5120,
|
| 67 |
+
"mean_abs": 0.04929830878973007,
|
| 68 |
+
"std": 0.06144307553768158
|
| 69 |
+
}
|
| 70 |
+
],
|
| 71 |
+
"beta_sweep": [
|
| 72 |
+
{
|
| 73 |
+
"beta": 0.25,
|
| 74 |
+
"N": 4232192,
|
| 75 |
+
"n_eff": 2649414.0,
|
| 76 |
+
"nu": 662353,
|
| 77 |
+
"sparsity": 0.8434964670789983,
|
| 78 |
+
"s_eff": 0.3719400465488434,
|
| 79 |
+
"actual_1_minus_seff": 0.6280599534511566,
|
| 80 |
+
"theoretical_bound_1_minus_seff": 0.8425445923201286,
|
| 81 |
+
"bound_holds": true,
|
| 82 |
+
"bound_margin": 0.21448463886897196,
|
| 83 |
+
"l2_preserved_fraction": 0.7644939287955947
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"beta": 0.5,
|
| 87 |
+
"N": 4232192,
|
| 88 |
+
"n_eff": 2649414.0,
|
| 89 |
+
"nu": 1324707,
|
| 90 |
+
"sparsity": 0.6869926978738204,
|
| 91 |
+
"s_eff": 0.6052946448326111,
|
| 92 |
+
"actual_1_minus_seff": 0.3947053551673889,
|
| 93 |
+
"theoretical_bound_1_minus_seff": 0.6864979680651238,
|
| 94 |
+
"bound_holds": true,
|
| 95 |
+
"bound_margin": 0.29179261289773484,
|
| 96 |
+
"l2_preserved_fraction": 0.8969132112729524
|
| 97 |
+
},
|
| 98 |
+
{
|
| 99 |
+
"beta": 0.75,
|
| 100 |
+
"N": 4232192,
|
| 101 |
+
"n_eff": 2649414.0,
|
| 102 |
+
"nu": 1987060,
|
| 103 |
+
"sparsity": 0.5304891649528187,
|
| 104 |
+
"s_eff": 0.7724460959434509,
|
| 105 |
+
"actual_1_minus_seff": 0.22755390405654907,
|
| 106 |
+
"theoretical_bound_1_minus_seff": 0.5302150648920348,
|
| 107 |
+
"bound_holds": true,
|
| 108 |
+
"bound_margin": 0.3026611608354858,
|
| 109 |
+
"l2_preserved_fraction": 0.9577440210041348
|
| 110 |
+
},
|
| 111 |
+
{
|
| 112 |
+
"beta": 1.0,
|
| 113 |
+
"N": 4232192,
|
| 114 |
+
"n_eff": 2649414.0,
|
| 115 |
+
"nu": 2649414,
|
| 116 |
+
"sparsity": 0.3739853957476409,
|
| 117 |
+
"s_eff": 0.8893267512321472,
|
| 118 |
+
"actual_1_minus_seff": 0.11067324876785278,
|
| 119 |
+
"theoretical_bound_1_minus_seff": 0.3738448858609692,
|
| 120 |
+
"bound_holds": true,
|
| 121 |
+
"bound_margin": 0.2631716370931164,
|
| 122 |
+
"l2_preserved_fraction": 0.9862176542147392
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"beta": 1.25,
|
| 126 |
+
"N": 4232192,
|
| 127 |
+
"n_eff": 2649414.0,
|
| 128 |
+
"nu": 3311767,
|
| 129 |
+
"sparsity": 0.21748186282663928,
|
| 130 |
+
"s_eff": 0.963045597076416,
|
| 131 |
+
"actual_1_minus_seff": 0.036954402923583984,
|
| 132 |
+
"theoretical_bound_1_minus_seff": 0.21742613085222498,
|
| 133 |
+
"bound_holds": true,
|
| 134 |
+
"bound_margin": 0.180471727928641,
|
| 135 |
+
"l2_preserved_fraction": 0.9974516606164923
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"beta": 1.5,
|
| 139 |
+
"N": 4232192,
|
| 140 |
+
"n_eff": 2649414.0,
|
| 141 |
+
"nu": 3974121,
|
| 142 |
+
"sparsity": 0.06097809362146145,
|
| 143 |
+
"s_eff": 0.9971094727516174,
|
| 144 |
+
"actual_1_minus_seff": 0.0028905272483825684,
|
| 145 |
+
"theoretical_bound_1_minus_seff": 0.06097054027550846,
|
| 146 |
+
"bound_holds": true,
|
| 147 |
+
"bound_margin": 0.05808001302712589,
|
| 148 |
+
"l2_preserved_fraction": 0.9999877998934479
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"beta": 2.0,
|
| 152 |
+
"N": 4232192,
|
| 153 |
+
"n_eff": 2649414.0,
|
| 154 |
+
"nu": 4232191,
|
| 155 |
+
"sparsity": 2.3628417611387675e-07,
|
| 156 |
+
"s_eff": 1.0,
|
| 157 |
+
"actual_1_minus_seff": 0.0,
|
| 158 |
+
"theoretical_bound_1_minus_seff": 0.0,
|
| 159 |
+
"bound_holds": true,
|
| 160 |
+
"bound_margin": 0.0,
|
| 161 |
+
"l2_preserved_fraction": 1.0
|
| 162 |
+
}
|
| 163 |
+
],
|
| 164 |
+
"uniform_comparison": [
|
| 165 |
+
{
|
| 166 |
+
"target_sparsity": 0.8434964670789983,
|
| 167 |
+
"actual_nu": 662353,
|
| 168 |
+
"s_eff": 0.3719400465488434,
|
| 169 |
+
"l2_preserved": 0.7644939287955947,
|
| 170 |
+
"beta": 0.25
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"target_sparsity": 0.6869926978738204,
|
| 174 |
+
"actual_nu": 1324707,
|
| 175 |
+
"s_eff": 0.6052946448326111,
|
| 176 |
+
"l2_preserved": 0.8969132112729524,
|
| 177 |
+
"beta": 0.5
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"target_sparsity": 0.5304891649528187,
|
| 181 |
+
"actual_nu": 1987060,
|
| 182 |
+
"s_eff": 0.7724460959434509,
|
| 183 |
+
"l2_preserved": 0.9577440210041348,
|
| 184 |
+
"beta": 0.75
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"target_sparsity": 0.3739853957476409,
|
| 188 |
+
"actual_nu": 2649414,
|
| 189 |
+
"s_eff": 0.8893267512321472,
|
| 190 |
+
"l2_preserved": 0.9862176542147392,
|
| 191 |
+
"beta": 1.0
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"target_sparsity": 0.21748186282663928,
|
| 195 |
+
"actual_nu": 3311767,
|
| 196 |
+
"s_eff": 0.963045597076416,
|
| 197 |
+
"l2_preserved": 0.9974516606164923,
|
| 198 |
+
"beta": 1.25
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"target_sparsity": 0.06097809362146145,
|
| 202 |
+
"actual_nu": 3974121,
|
| 203 |
+
"s_eff": 0.9971094727516174,
|
| 204 |
+
"l2_preserved": 0.9999877998934479,
|
| 205 |
+
"beta": 1.5
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"target_sparsity": 2.3628417611387675e-07,
|
| 209 |
+
"actual_nu": 4232191,
|
| 210 |
+
"s_eff": 1.0,
|
| 211 |
+
"l2_preserved": 1.0,
|
| 212 |
+
"beta": 2.0
|
| 213 |
+
}
|
| 214 |
+
]
|
| 215 |
+
},
|
| 216 |
+
"ResNet18": {
|
| 217 |
+
"model": "ResNet18",
|
| 218 |
+
"total_params": 11678912,
|
| 219 |
+
"mean_abs_weight": 0.016334563493728638,
|
| 220 |
+
"weight_std": 0.025007877498865128,
|
| 221 |
+
"layer_stats": [
|
| 222 |
+
{
|
| 223 |
+
"name": "conv1",
|
| 224 |
+
"type": "Conv2d",
|
| 225 |
+
"shape": [
|
| 226 |
+
64,
|
| 227 |
+
3,
|
| 228 |
+
7,
|
| 229 |
+
7
|
| 230 |
+
],
|
| 231 |
+
"numel": 9408,
|
| 232 |
+
"mean_abs": 0.0761537179350853,
|
| 233 |
+
"std": 0.1296982318162918
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"name": "layer1.0.conv1",
|
| 237 |
+
"type": "Conv2d",
|
| 238 |
+
"shape": [
|
| 239 |
+
64,
|
| 240 |
+
64,
|
| 241 |
+
3,
|
| 242 |
+
3
|
| 243 |
+
],
|
| 244 |
+
"numel": 36864,
|
| 245 |
+
"mean_abs": 0.031629934906959534,
|
| 246 |
+
"std": 0.0533977635204792
|
| 247 |
+
},
|
| 248 |
+
{
|
| 249 |
+
"name": "layer1.0.conv2",
|
| 250 |
+
"type": "Conv2d",
|
| 251 |
+
"shape": [
|
| 252 |
+
64,
|
| 253 |
+
64,
|
| 254 |
+
3,
|
| 255 |
+
3
|
| 256 |
+
],
|
| 257 |
+
"numel": 36864,
|
| 258 |
+
"mean_abs": 0.03200514614582062,
|
| 259 |
+
"std": 0.04520029574632645
|
| 260 |
+
},
|
| 261 |
+
{
|
| 262 |
+
"name": "layer1.1.conv1",
|
| 263 |
+
"type": "Conv2d",
|
| 264 |
+
"shape": [
|
| 265 |
+
64,
|
| 266 |
+
64,
|
| 267 |
+
3,
|
| 268 |
+
3
|
| 269 |
+
],
|
| 270 |
+
"numel": 36864,
|
| 271 |
+
"mean_abs": 0.03426660597324371,
|
| 272 |
+
"std": 0.05083491653203964
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
"name": "layer1.1.conv2",
|
| 276 |
+
"type": "Conv2d",
|
| 277 |
+
"shape": [
|
| 278 |
+
64,
|
| 279 |
+
64,
|
| 280 |
+
3,
|
| 281 |
+
3
|
| 282 |
+
],
|
| 283 |
+
"numel": 36864,
|
| 284 |
+
"mean_abs": 0.03150966390967369,
|
| 285 |
+
"std": 0.04397609457373619
|
| 286 |
+
},
|
| 287 |
+
{
|
| 288 |
+
"name": "layer2.0.conv1",
|
| 289 |
+
"type": "Conv2d",
|
| 290 |
+
"shape": [
|
| 291 |
+
128,
|
| 292 |
+
64,
|
| 293 |
+
3,
|
| 294 |
+
3
|
| 295 |
+
],
|
| 296 |
+
"numel": 73728,
|
| 297 |
+
"mean_abs": 0.02940523996949196,
|
| 298 |
+
"std": 0.041583266109228134
|
| 299 |
+
},
|
| 300 |
+
{
|
| 301 |
+
"name": "layer2.0.conv2",
|
| 302 |
+
"type": "Conv2d",
|
| 303 |
+
"shape": [
|
| 304 |
+
128,
|
| 305 |
+
128,
|
| 306 |
+
3,
|
| 307 |
+
3
|
| 308 |
+
],
|
| 309 |
+
"numel": 147456,
|
| 310 |
+
"mean_abs": 0.023948440328240395,
|
| 311 |
+
"std": 0.03401258587837219
|
| 312 |
+
},
|
| 313 |
+
{
|
| 314 |
+
"name": "layer2.0.downsample.0",
|
| 315 |
+
"type": "Conv2d",
|
| 316 |
+
"shape": [
|
| 317 |
+
128,
|
| 318 |
+
64,
|
| 319 |
+
1,
|
| 320 |
+
1
|
| 321 |
+
],
|
| 322 |
+
"numel": 8192,
|
| 323 |
+
"mean_abs": 0.044411800801754,
|
| 324 |
+
"std": 0.0706208124756813
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"name": "layer2.1.conv1",
|
| 328 |
+
"type": "Conv2d",
|
| 329 |
+
"shape": [
|
| 330 |
+
128,
|
| 331 |
+
128,
|
| 332 |
+
3,
|
| 333 |
+
3
|
| 334 |
+
],
|
| 335 |
+
"numel": 147456,
|
| 336 |
+
"mean_abs": 0.02426033839583397,
|
| 337 |
+
"std": 0.034170422703027725
|
| 338 |
+
},
|
| 339 |
+
{
|
| 340 |
+
"name": "layer2.1.conv2",
|
| 341 |
+
"type": "Conv2d",
|
| 342 |
+
"shape": [
|
| 343 |
+
128,
|
| 344 |
+
128,
|
| 345 |
+
3,
|
| 346 |
+
3
|
| 347 |
+
],
|
| 348 |
+
"numel": 147456,
|
| 349 |
+
"mean_abs": 0.022247975692152977,
|
| 350 |
+
"std": 0.03005100041627884
|
| 351 |
+
},
|
| 352 |
+
{
|
| 353 |
+
"name": "layer3.0.conv1",
|
| 354 |
+
"type": "Conv2d",
|
| 355 |
+
"shape": [
|
| 356 |
+
256,
|
| 357 |
+
128,
|
| 358 |
+
3,
|
| 359 |
+
3
|
| 360 |
+
],
|
| 361 |
+
"numel": 294912,
|
| 362 |
+
"mean_abs": 0.0209345743060112,
|
| 363 |
+
"std": 0.029047327116131783
|
| 364 |
+
},
|
| 365 |
+
{
|
| 366 |
+
"name": "layer3.0.conv2",
|
| 367 |
+
"type": "Conv2d",
|
| 368 |
+
"shape": [
|
| 369 |
+
256,
|
| 370 |
+
256,
|
| 371 |
+
3,
|
| 372 |
+
3
|
| 373 |
+
],
|
| 374 |
+
"numel": 589824,
|
| 375 |
+
"mean_abs": 0.018351763486862183,
|
| 376 |
+
"std": 0.02504604496061802
|
| 377 |
+
},
|
| 378 |
+
{
|
| 379 |
+
"name": "layer3.0.downsample.0",
|
| 380 |
+
"type": "Conv2d",
|
| 381 |
+
"shape": [
|
| 382 |
+
256,
|
| 383 |
+
128,
|
| 384 |
+
1,
|
| 385 |
+
1
|
| 386 |
+
],
|
| 387 |
+
"numel": 32768,
|
| 388 |
+
"mean_abs": 0.02377864345908165,
|
| 389 |
+
"std": 0.03294449672102928
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"name": "layer3.1.conv1",
|
| 393 |
+
"type": "Conv2d",
|
| 394 |
+
"shape": [
|
| 395 |
+
256,
|
| 396 |
+
256,
|
| 397 |
+
3,
|
| 398 |
+
3
|
| 399 |
+
],
|
| 400 |
+
"numel": 589824,
|
| 401 |
+
"mean_abs": 0.016645917668938637,
|
| 402 |
+
"std": 0.022360969334840775
|
| 403 |
+
},
|
| 404 |
+
{
|
| 405 |
+
"name": "layer3.1.conv2",
|
| 406 |
+
"type": "Conv2d",
|
| 407 |
+
"shape": [
|
| 408 |
+
256,
|
| 409 |
+
256,
|
| 410 |
+
3,
|
| 411 |
+
3
|
| 412 |
+
],
|
| 413 |
+
"numel": 589824,
|
| 414 |
+
"mean_abs": 0.015521775931119919,
|
| 415 |
+
"std": 0.020670222118496895
|
| 416 |
+
},
|
| 417 |
+
{
|
| 418 |
+
"name": "layer4.0.conv1",
|
| 419 |
+
"type": "Conv2d",
|
| 420 |
+
"shape": [
|
| 421 |
+
512,
|
| 422 |
+
256,
|
| 423 |
+
3,
|
| 424 |
+
3
|
| 425 |
+
],
|
| 426 |
+
"numel": 1179648,
|
| 427 |
+
"mean_abs": 0.01521181222051382,
|
| 428 |
+
"std": 0.01987350732088089
|
| 429 |
+
},
|
| 430 |
+
{
|
| 431 |
+
"name": "layer4.0.conv2",
|
| 432 |
+
"type": "Conv2d",
|
| 433 |
+
"shape": [
|
| 434 |
+
512,
|
| 435 |
+
512,
|
| 436 |
+
3,
|
| 437 |
+
3
|
| 438 |
+
],
|
| 439 |
+
"numel": 2359296,
|
| 440 |
+
"mean_abs": 0.013380504213273525,
|
| 441 |
+
"std": 0.017324835062026978
|
| 442 |
+
},
|
| 443 |
+
{
|
| 444 |
+
"name": "layer4.0.downsample.0",
|
| 445 |
+
"type": "Conv2d",
|
| 446 |
+
"shape": [
|
| 447 |
+
512,
|
| 448 |
+
256,
|
| 449 |
+
1,
|
| 450 |
+
1
|
| 451 |
+
],
|
| 452 |
+
"numel": 131072,
|
| 453 |
+
"mean_abs": 0.02387506142258644,
|
| 454 |
+
"std": 0.032782770693302155
|
| 455 |
+
},
|
| 456 |
+
{
|
| 457 |
+
"name": "layer4.1.conv1",
|
| 458 |
+
"type": "Conv2d",
|
| 459 |
+
"shape": [
|
| 460 |
+
512,
|
| 461 |
+
512,
|
| 462 |
+
3,
|
| 463 |
+
3
|
| 464 |
+
],
|
| 465 |
+
"numel": 2359296,
|
| 466 |
+
"mean_abs": 0.014033002778887749,
|
| 467 |
+
"std": 0.017829064279794693
|
| 468 |
+
},
|
| 469 |
+
{
|
| 470 |
+
"name": "layer4.1.conv2",
|
| 471 |
+
"type": "Conv2d",
|
| 472 |
+
"shape": [
|
| 473 |
+
512,
|
| 474 |
+
512,
|
| 475 |
+
3,
|
| 476 |
+
3
|
| 477 |
+
],
|
| 478 |
+
"numel": 2359296,
|
| 479 |
+
"mean_abs": 0.010081645101308823,
|
| 480 |
+
"std": 0.013201371766626835
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"name": "fc",
|
| 484 |
+
"type": "Linear",
|
| 485 |
+
"shape": [
|
| 486 |
+
1000,
|
| 487 |
+
512
|
| 488 |
+
],
|
| 489 |
+
"numel": 512000,
|
| 490 |
+
"mean_abs": 0.05092787370085716,
|
| 491 |
+
"std": 0.06945901364088058
|
| 492 |
+
}
|
| 493 |
+
],
|
| 494 |
+
"beta_sweep": [
|
| 495 |
+
{
|
| 496 |
+
"beta": 0.25,
|
| 497 |
+
"N": 11678912,
|
| 498 |
+
"n_eff": 4970629.5,
|
| 499 |
+
"nu": 1242657,
|
| 500 |
+
"sparsity": 0.8935982221631604,
|
| 501 |
+
"s_eff": 0.35967200994491577,
|
| 502 |
+
"actual_1_minus_seff": 0.6403279900550842,
|
| 503 |
+
"theoretical_bound_1_minus_seff": 0.8928404517719254,
|
| 504 |
+
"bound_holds": true,
|
| 505 |
+
"bound_margin": 0.25251246171684116,
|
| 506 |
+
"l2_preserved_fraction": 0.8410255595851589
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"beta": 0.5,
|
| 510 |
+
"N": 11678912,
|
| 511 |
+
"n_eff": 4970629.5,
|
| 512 |
+
"nu": 2485314,
|
| 513 |
+
"sparsity": 0.7871964443263209,
|
| 514 |
+
"s_eff": 0.5378041863441467,
|
| 515 |
+
"actual_1_minus_seff": 0.46219581365585327,
|
| 516 |
+
"theoretical_bound_1_minus_seff": 0.7867534134852929,
|
| 517 |
+
"bound_holds": true,
|
| 518 |
+
"bound_margin": 0.32455759982943966,
|
| 519 |
+
"l2_preserved_fraction": 0.9141560058730758
|
| 520 |
+
},
|
| 521 |
+
{
|
| 522 |
+
"beta": 0.75,
|
| 523 |
+
"N": 11678912,
|
| 524 |
+
"n_eff": 4970629.5,
|
| 525 |
+
"nu": 3727972,
|
| 526 |
+
"sparsity": 0.6807945808650668,
|
| 527 |
+
"s_eff": 0.670098066329956,
|
| 528 |
+
"actual_1_minus_seff": 0.32990193367004395,
|
| 529 |
+
"theoretical_bound_1_minus_seff": 0.6805036513739643,
|
| 530 |
+
"bound_holds": true,
|
| 531 |
+
"bound_margin": 0.35060171770392035,
|
| 532 |
+
"l2_preserved_fraction": 0.9519361174086984
|
| 533 |
+
},
|
| 534 |
+
{
|
| 535 |
+
"beta": 1.0,
|
| 536 |
+
"N": 11678912,
|
| 537 |
+
"n_eff": 4970629.5,
|
| 538 |
+
"nu": 4970629,
|
| 539 |
+
"sparsity": 0.5743928030282273,
|
| 540 |
+
"s_eff": 0.7728628516197205,
|
| 541 |
+
"actual_1_minus_seff": 0.22713714838027954,
|
| 542 |
+
"theoretical_bound_1_minus_seff": 0.5741975455156613,
|
| 543 |
+
"bound_holds": true,
|
| 544 |
+
"bound_margin": 0.3470603971353817,
|
| 545 |
+
"l2_preserved_fraction": 0.973983437655045
|
| 546 |
+
},
|
| 547 |
+
{
|
| 548 |
+
"beta": 1.25,
|
| 549 |
+
"N": 11678912,
|
| 550 |
+
"n_eff": 4970629.5,
|
| 551 |
+
"nu": 6213287,
|
| 552 |
+
"sparsity": 0.46799093956697335,
|
| 553 |
+
"s_eff": 0.8529043197631836,
|
| 554 |
+
"actual_1_minus_seff": 0.1470956802368164,
|
| 555 |
+
"theoretical_bound_1_minus_seff": 0.4678625010139507,
|
| 556 |
+
"bound_holds": true,
|
| 557 |
+
"bound_margin": 0.3207668207771343,
|
| 558 |
+
"l2_preserved_fraction": 0.9870897253559975
|
| 559 |
+
},
|
| 560 |
+
{
|
| 561 |
+
"beta": 1.5,
|
| 562 |
+
"N": 11678912,
|
| 563 |
+
"n_eff": 4970629.5,
|
| 564 |
+
"nu": 7455944,
|
| 565 |
+
"sparsity": 0.3615891617301338,
|
| 566 |
+
"s_eff": 0.9137875437736511,
|
| 567 |
+
"actual_1_minus_seff": 0.08621245622634888,
|
| 568 |
+
"theoretical_bound_1_minus_seff": 0.36150953264501795,
|
| 569 |
+
"bound_holds": true,
|
| 570 |
+
"bound_margin": 0.2752970764186691,
|
| 571 |
+
"l2_preserved_fraction": 0.9945489277407724
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"beta": 2.0,
|
| 575 |
+
"N": 11678912,
|
| 576 |
+
"n_eff": 4970629.5,
|
| 577 |
+
"nu": 9941259,
|
| 578 |
+
"sparsity": 0.1487855204320403,
|
| 579 |
+
"s_eff": 0.9857496619224548,
|
| 580 |
+
"actual_1_minus_seff": 0.014250338077545166,
|
| 581 |
+
"theoretical_bound_1_minus_seff": 0.1487673183871917,
|
| 582 |
+
"bound_holds": true,
|
| 583 |
+
"bound_margin": 0.13451698030964654,
|
| 584 |
+
"l2_preserved_fraction": 0.9999661861863522
|
| 585 |
+
}
|
| 586 |
+
],
|
| 587 |
+
"uniform_comparison": [
|
| 588 |
+
{
|
| 589 |
+
"target_sparsity": 0.8935982221631604,
|
| 590 |
+
"actual_nu": 1242656,
|
| 591 |
+
"s_eff": 0.35967186093330383,
|
| 592 |
+
"l2_preserved": 0.8410253811481476,
|
| 593 |
+
"beta": 0.25
|
| 594 |
+
},
|
| 595 |
+
{
|
| 596 |
+
"target_sparsity": 0.7871964443263209,
|
| 597 |
+
"actual_nu": 2485313,
|
| 598 |
+
"s_eff": 0.5378040671348572,
|
| 599 |
+
"l2_preserved": 0.9141560950915815,
|
| 600 |
+
"beta": 0.5
|
| 601 |
+
},
|
| 602 |
+
{
|
| 603 |
+
"target_sparsity": 0.6807945808650668,
|
| 604 |
+
"actual_nu": 3727972,
|
| 605 |
+
"s_eff": 0.670098066329956,
|
| 606 |
+
"l2_preserved": 0.9519361174086984,
|
| 607 |
+
"beta": 0.75
|
| 608 |
+
},
|
| 609 |
+
{
|
| 610 |
+
"target_sparsity": 0.5743928030282273,
|
| 611 |
+
"actual_nu": 4970629,
|
| 612 |
+
"s_eff": 0.7728628516197205,
|
| 613 |
+
"l2_preserved": 0.973983437655045,
|
| 614 |
+
"beta": 1.0
|
| 615 |
+
},
|
| 616 |
+
{
|
| 617 |
+
"target_sparsity": 0.46799093956697335,
|
| 618 |
+
"actual_nu": 6213287,
|
| 619 |
+
"s_eff": 0.8529043197631836,
|
| 620 |
+
"l2_preserved": 0.9870897253559975,
|
| 621 |
+
"beta": 1.25
|
| 622 |
+
},
|
| 623 |
+
{
|
| 624 |
+
"target_sparsity": 0.3615891617301338,
|
| 625 |
+
"actual_nu": 7455944,
|
| 626 |
+
"s_eff": 0.9137875437736511,
|
| 627 |
+
"l2_preserved": 0.9945489277407724,
|
| 628 |
+
"beta": 1.5
|
| 629 |
+
},
|
| 630 |
+
{
|
| 631 |
+
"target_sparsity": 0.1487855204320403,
|
| 632 |
+
"actual_nu": 9941259,
|
| 633 |
+
"s_eff": 0.9857496619224548,
|
| 634 |
+
"l2_preserved": 0.9999661861863522,
|
| 635 |
+
"beta": 2.0
|
| 636 |
+
}
|
| 637 |
+
]
|
| 638 |
+
},
|
| 639 |
+
"VGG16-BN": {
|
| 640 |
+
"model": "VGG16-BN",
|
| 641 |
+
"total_params": 138344128,
|
| 642 |
+
"mean_abs_weight": 0.005381909664720297,
|
| 643 |
+
"weight_std": 0.008436962962150574,
|
| 644 |
+
"layer_stats": [
|
| 645 |
+
{
|
| 646 |
+
"name": "features.0",
|
| 647 |
+
"type": "Conv2d",
|
| 648 |
+
"shape": [
|
| 649 |
+
64,
|
| 650 |
+
3,
|
| 651 |
+
3,
|
| 652 |
+
3
|
| 653 |
+
],
|
| 654 |
+
"numel": 1728,
|
| 655 |
+
"mean_abs": 0.07537371665239334,
|
| 656 |
+
"std": 0.18069274723529816
|
| 657 |
+
},
|
| 658 |
+
{
|
| 659 |
+
"name": "features.3",
|
| 660 |
+
"type": "Conv2d",
|
| 661 |
+
"shape": [
|
| 662 |
+
64,
|
| 663 |
+
64,
|
| 664 |
+
3,
|
| 665 |
+
3
|
| 666 |
+
],
|
| 667 |
+
"numel": 36864,
|
| 668 |
+
"mean_abs": 0.0152515834197402,
|
| 669 |
+
"std": 0.03547147661447525
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"name": "features.7",
|
| 673 |
+
"type": "Conv2d",
|
| 674 |
+
"shape": [
|
| 675 |
+
128,
|
| 676 |
+
64,
|
| 677 |
+
3,
|
| 678 |
+
3
|
| 679 |
+
],
|
| 680 |
+
"numel": 73728,
|
| 681 |
+
"mean_abs": 0.022179871797561646,
|
| 682 |
+
"std": 0.03943381831049919
|
| 683 |
+
},
|
| 684 |
+
{
|
| 685 |
+
"name": "features.10",
|
| 686 |
+
"type": "Conv2d",
|
| 687 |
+
"shape": [
|
| 688 |
+
128,
|
| 689 |
+
128,
|
| 690 |
+
3,
|
| 691 |
+
3
|
| 692 |
+
],
|
| 693 |
+
"numel": 147456,
|
| 694 |
+
"mean_abs": 0.01824381574988365,
|
| 695 |
+
"std": 0.028957776725292206
|
| 696 |
+
},
|
| 697 |
+
{
|
| 698 |
+
"name": "features.14",
|
| 699 |
+
"type": "Conv2d",
|
| 700 |
+
"shape": [
|
| 701 |
+
256,
|
| 702 |
+
128,
|
| 703 |
+
3,
|
| 704 |
+
3
|
| 705 |
+
],
|
| 706 |
+
"numel": 294912,
|
| 707 |
+
"mean_abs": 0.017823530361056328,
|
| 708 |
+
"std": 0.025923190638422966
|
| 709 |
+
},
|
| 710 |
+
{
|
| 711 |
+
"name": "features.17",
|
| 712 |
+
"type": "Conv2d",
|
| 713 |
+
"shape": [
|
| 714 |
+
256,
|
| 715 |
+
256,
|
| 716 |
+
3,
|
| 717 |
+
3
|
| 718 |
+
],
|
| 719 |
+
"numel": 589824,
|
| 720 |
+
"mean_abs": 0.014825817197561264,
|
| 721 |
+
"std": 0.020382367074489594
|
| 722 |
+
},
|
| 723 |
+
{
|
| 724 |
+
"name": "features.20",
|
| 725 |
+
"type": "Conv2d",
|
| 726 |
+
"shape": [
|
| 727 |
+
256,
|
| 728 |
+
256,
|
| 729 |
+
3,
|
| 730 |
+
3
|
| 731 |
+
],
|
| 732 |
+
"numel": 589824,
|
| 733 |
+
"mean_abs": 0.01604265160858631,
|
| 734 |
+
"std": 0.021554531529545784
|
| 735 |
+
},
|
| 736 |
+
{
|
| 737 |
+
"name": "features.24",
|
| 738 |
+
"type": "Conv2d",
|
| 739 |
+
"shape": [
|
| 740 |
+
512,
|
| 741 |
+
256,
|
| 742 |
+
3,
|
| 743 |
+
3
|
| 744 |
+
],
|
| 745 |
+
"numel": 1179648,
|
| 746 |
+
"mean_abs": 0.014152848161756992,
|
| 747 |
+
"std": 0.019217344000935555
|
| 748 |
+
},
|
| 749 |
+
{
|
| 750 |
+
"name": "features.27",
|
| 751 |
+
"type": "Conv2d",
|
| 752 |
+
"shape": [
|
| 753 |
+
512,
|
| 754 |
+
512,
|
| 755 |
+
3,
|
| 756 |
+
3
|
| 757 |
+
],
|
| 758 |
+
"numel": 2359296,
|
| 759 |
+
"mean_abs": 0.011720959097146988,
|
| 760 |
+
"std": 0.015424084849655628
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"name": "features.30",
|
| 764 |
+
"type": "Conv2d",
|
| 765 |
+
"shape": [
|
| 766 |
+
512,
|
| 767 |
+
512,
|
| 768 |
+
3,
|
| 769 |
+
3
|
| 770 |
+
],
|
| 771 |
+
"numel": 2359296,
|
| 772 |
+
"mean_abs": 0.01195573527365923,
|
| 773 |
+
"std": 0.01562708057463169
|
| 774 |
+
},
|
| 775 |
+
{
|
| 776 |
+
"name": "features.34",
|
| 777 |
+
"type": "Conv2d",
|
| 778 |
+
"shape": [
|
| 779 |
+
512,
|
| 780 |
+
512,
|
| 781 |
+
3,
|
| 782 |
+
3
|
| 783 |
+
],
|
| 784 |
+
"numel": 2359296,
|
| 785 |
+
"mean_abs": 0.012774649076163769,
|
| 786 |
+
"std": 0.01662408374249935
|
| 787 |
+
},
|
| 788 |
+
{
|
| 789 |
+
"name": "features.37",
|
| 790 |
+
"type": "Conv2d",
|
| 791 |
+
"shape": [
|
| 792 |
+
512,
|
| 793 |
+
512,
|
| 794 |
+
3,
|
| 795 |
+
3
|
| 796 |
+
],
|
| 797 |
+
"numel": 2359296,
|
| 798 |
+
"mean_abs": 0.012733221054077148,
|
| 799 |
+
"std": 0.016091646626591682
|
| 800 |
+
},
|
| 801 |
+
{
|
| 802 |
+
"name": "features.40",
|
| 803 |
+
"type": "Conv2d",
|
| 804 |
+
"shape": [
|
| 805 |
+
512,
|
| 806 |
+
512,
|
| 807 |
+
3,
|
| 808 |
+
3
|
| 809 |
+
],
|
| 810 |
+
"numel": 2359296,
|
| 811 |
+
"mean_abs": 0.011355762369930744,
|
| 812 |
+
"std": 0.01471445057541132
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"name": "classifier.0",
|
| 816 |
+
"type": "Linear",
|
| 817 |
+
"shape": [
|
| 818 |
+
4096,
|
| 819 |
+
25088
|
| 820 |
+
],
|
| 821 |
+
"numel": 102760448,
|
| 822 |
+
"mean_abs": 0.0035946827847510576,
|
| 823 |
+
"std": 0.004716749768704176
|
| 824 |
+
},
|
| 825 |
+
{
|
| 826 |
+
"name": "classifier.3",
|
| 827 |
+
"type": "Linear",
|
| 828 |
+
"shape": [
|
| 829 |
+
4096,
|
| 830 |
+
4096
|
| 831 |
+
],
|
| 832 |
+
"numel": 16777216,
|
| 833 |
+
"mean_abs": 0.00743697676807642,
|
| 834 |
+
"std": 0.009650247171521187
|
| 835 |
+
},
|
| 836 |
+
{
|
| 837 |
+
"name": "classifier.6",
|
| 838 |
+
"type": "Linear",
|
| 839 |
+
"shape": [
|
| 840 |
+
1000,
|
| 841 |
+
4096
|
| 842 |
+
],
|
| 843 |
+
"numel": 4096000,
|
| 844 |
+
"mean_abs": 0.015230334363877773,
|
| 845 |
+
"std": 0.019549593329429626
|
| 846 |
+
}
|
| 847 |
+
],
|
| 848 |
+
"beta_sweep": [
|
| 849 |
+
{
|
| 850 |
+
"beta": 0.25,
|
| 851 |
+
"N": 138344128,
|
| 852 |
+
"n_eff": 56201752.0,
|
| 853 |
+
"nu": 14050438,
|
| 854 |
+
"sparsity": 0.8984384938983461,
|
| 855 |
+
"s_eff": 0.3791695237159729,
|
| 856 |
+
"actual_1_minus_seff": 0.6208304762840271,
|
| 857 |
+
"theoretical_bound_1_minus_seff": 0.8982113047520335,
|
| 858 |
+
"bound_holds": true,
|
| 859 |
+
"bound_margin": 0.2773808284680064,
|
| 860 |
+
"l2_preserved_fraction": 0.8672151974040192
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"beta": 0.5,
|
| 864 |
+
"N": 138344128,
|
| 865 |
+
"n_eff": 56201752.0,
|
| 866 |
+
"nu": 28100876,
|
| 867 |
+
"sparsity": 0.796876987796692,
|
| 868 |
+
"s_eff": 0.5589225888252258,
|
| 869 |
+
"actual_1_minus_seff": 0.44107741117477417,
|
| 870 |
+
"theoretical_bound_1_minus_seff": 0.7967427956838784,
|
| 871 |
+
"bound_holds": true,
|
| 872 |
+
"bound_margin": 0.35566538450910423,
|
| 873 |
+
"l2_preserved_fraction": 0.940617946174759
|
| 874 |
+
},
|
| 875 |
+
{
|
| 876 |
+
"beta": 0.75,
|
| 877 |
+
"N": 138344128,
|
| 878 |
+
"n_eff": 56201752.0,
|
| 879 |
+
"nu": 42151312,
|
| 880 |
+
"sparsity": 0.6953154961517413,
|
| 881 |
+
"s_eff": 0.6873372793197632,
|
| 882 |
+
"actual_1_minus_seff": 0.3126627206802368,
|
| 883 |
+
"theoretical_bound_1_minus_seff": 0.695226192892721,
|
| 884 |
+
"bound_holds": true,
|
| 885 |
+
"bound_margin": 0.3825634722124842,
|
| 886 |
+
"l2_preserved_fraction": 0.9868536606785648
|
| 887 |
+
},
|
| 888 |
+
{
|
| 889 |
+
"beta": 1.0,
|
| 890 |
+
"N": 138344128,
|
| 891 |
+
"n_eff": 56201752.0,
|
| 892 |
+
"nu": 56201752,
|
| 893 |
+
"sparsity": 0.593753975593384,
|
| 894 |
+
"s_eff": 0.7856668829917908,
|
| 895 |
+
"actual_1_minus_seff": 0.21433311700820923,
|
| 896 |
+
"theoretical_bound_1_minus_seff": 0.5936929467887193,
|
| 897 |
+
"bound_holds": true,
|
| 898 |
+
"bound_margin": 0.3793598297805101,
|
| 899 |
+
"l2_preserved_fraction": 0.9991829327293085
|
| 900 |
+
},
|
| 901 |
+
{
|
| 902 |
+
"beta": 1.25,
|
| 903 |
+
"N": 138344128,
|
| 904 |
+
"n_eff": 56201752.0,
|
| 905 |
+
"nu": 70252192,
|
| 906 |
+
"sparsity": 0.49219245503502684,
|
| 907 |
+
"s_eff": 0.861655056476593,
|
| 908 |
+
"actual_1_minus_seff": 0.13834494352340698,
|
| 909 |
+
"theoretical_bound_1_minus_seff": 0.4921512573812512,
|
| 910 |
+
"bound_holds": true,
|
| 911 |
+
"bound_margin": 0.3538063138578442,
|
| 912 |
+
"l2_preserved_fraction": 0.9998382076173994
|
| 913 |
+
},
|
| 914 |
+
{
|
| 915 |
+
"beta": 1.5,
|
| 916 |
+
"N": 138344128,
|
| 917 |
+
"n_eff": 56201752.0,
|
| 918 |
+
"nu": 84302624,
|
| 919 |
+
"sparsity": 0.3906309923034825,
|
| 920 |
+
"s_eff": 0.9191533923149109,
|
| 921 |
+
"actual_1_minus_seff": 0.08084660768508911,
|
| 922 |
+
"theoretical_bound_1_minus_seff": 0.39060440162040416,
|
| 923 |
+
"bound_holds": true,
|
| 924 |
+
"bound_margin": 0.30975779393531505,
|
| 925 |
+
"l2_preserved_fraction": 0.9999820580431746
|
| 926 |
+
},
|
| 927 |
+
{
|
| 928 |
+
"beta": 2.0,
|
| 929 |
+
"N": 138344128,
|
| 930 |
+
"n_eff": 56201752.0,
|
| 931 |
+
"nu": 112403504,
|
| 932 |
+
"sparsity": 0.1875079511867681,
|
| 933 |
+
"s_eff": 0.9866164326667786,
|
| 934 |
+
"actual_1_minus_seff": 0.013383567333221436,
|
| 935 |
+
"theoretical_bound_1_minus_seff": 0.18750029275937904,
|
| 936 |
+
"bound_holds": true,
|
| 937 |
+
"bound_margin": 0.1741167254261576,
|
| 938 |
+
"l2_preserved_fraction": 1.0000012590846894
|
| 939 |
+
}
|
| 940 |
+
],
|
| 941 |
+
"uniform_comparison": [
|
| 942 |
+
{
|
| 943 |
+
"target_sparsity": 0.8984384938983461,
|
| 944 |
+
"actual_nu": 14050437,
|
| 945 |
+
"s_eff": 0.3791695237159729,
|
| 946 |
+
"l2_preserved": 0.8672151187112261,
|
| 947 |
+
"beta": 0.25
|
| 948 |
+
},
|
| 949 |
+
{
|
| 950 |
+
"target_sparsity": 0.796876987796692,
|
| 951 |
+
"actual_nu": 28100876,
|
| 952 |
+
"s_eff": 0.5589225888252258,
|
| 953 |
+
"l2_preserved": 0.940617946174759,
|
| 954 |
+
"beta": 0.5
|
| 955 |
+
},
|
| 956 |
+
{
|
| 957 |
+
"target_sparsity": 0.6953154961517413,
|
| 958 |
+
"actual_nu": 42151312,
|
| 959 |
+
"s_eff": 0.6873372793197632,
|
| 960 |
+
"l2_preserved": 0.9868536606785648,
|
| 961 |
+
"beta": 0.75
|
| 962 |
+
},
|
| 963 |
+
{
|
| 964 |
+
"target_sparsity": 0.593753975593384,
|
| 965 |
+
"actual_nu": 56201752,
|
| 966 |
+
"s_eff": 0.7856668829917908,
|
| 967 |
+
"l2_preserved": 0.9991829327293085,
|
| 968 |
+
"beta": 1.0
|
| 969 |
+
},
|
| 970 |
+
{
|
| 971 |
+
"target_sparsity": 0.49219245503502684,
|
| 972 |
+
"actual_nu": 70252192,
|
| 973 |
+
"s_eff": 0.861655056476593,
|
| 974 |
+
"l2_preserved": 0.9998382076173994,
|
| 975 |
+
"beta": 1.25
|
| 976 |
+
},
|
| 977 |
+
{
|
| 978 |
+
"target_sparsity": 0.3906309923034825,
|
| 979 |
+
"actual_nu": 84302624,
|
| 980 |
+
"s_eff": 0.9191533923149109,
|
| 981 |
+
"l2_preserved": 0.9999820580431746,
|
| 982 |
+
"beta": 1.5
|
| 983 |
+
},
|
| 984 |
+
{
|
| 985 |
+
"target_sparsity": 0.1875079511867681,
|
| 986 |
+
"actual_nu": 112403504,
|
| 987 |
+
"s_eff": 0.9866164326667786,
|
| 988 |
+
"l2_preserved": 1.0000012590846894,
|
| 989 |
+
"beta": 2.0
|
| 990 |
+
}
|
| 991 |
+
]
|
| 992 |
+
},
|
| 993 |
+
"GPT-2-small": {
|
| 994 |
+
"model": "GPT-2 small (124M)",
|
| 995 |
+
"total_params": 84934656,
|
| 996 |
+
"mean_abs_weight": 0.09779595583677292,
|
| 997 |
+
"weight_std": 0.13065941631793976,
|
| 998 |
+
"layer_stats": [
|
| 999 |
+
{
|
| 1000 |
+
"name": "h.0.attn.c_attn",
|
| 1001 |
+
"type": "Conv1D",
|
| 1002 |
+
"shape": [
|
| 1003 |
+
768,
|
| 1004 |
+
2304
|
| 1005 |
+
],
|
| 1006 |
+
"numel": 1769472,
|
| 1007 |
+
"mean_abs": 0.13464289903640747,
|
| 1008 |
+
"std": 0.1996196210384369
|
| 1009 |
+
},
|
| 1010 |
+
{
|
| 1011 |
+
"name": "h.0.attn.c_proj",
|
| 1012 |
+
"type": "Conv1D",
|
| 1013 |
+
"shape": [
|
| 1014 |
+
768,
|
| 1015 |
+
768
|
| 1016 |
+
],
|
| 1017 |
+
"numel": 589824,
|
| 1018 |
+
"mean_abs": 0.07300904393196106,
|
| 1019 |
+
"std": 0.1474614143371582
|
| 1020 |
+
},
|
| 1021 |
+
{
|
| 1022 |
+
"name": "h.0.mlp.c_fc",
|
| 1023 |
+
"type": "Conv1D",
|
| 1024 |
+
"shape": [
|
| 1025 |
+
768,
|
| 1026 |
+
3072
|
| 1027 |
+
],
|
| 1028 |
+
"numel": 2359296,
|
| 1029 |
+
"mean_abs": 0.10516798496246338,
|
| 1030 |
+
"std": 0.14116929471492767
|
| 1031 |
+
},
|
| 1032 |
+
{
|
| 1033 |
+
"name": "h.0.mlp.c_proj",
|
| 1034 |
+
"type": "Conv1D",
|
| 1035 |
+
"shape": [
|
| 1036 |
+
3072,
|
| 1037 |
+
768
|
| 1038 |
+
],
|
| 1039 |
+
"numel": 2359296,
|
| 1040 |
+
"mean_abs": 0.06414201855659485,
|
| 1041 |
+
"std": 0.0879654809832573
|
| 1042 |
+
},
|
| 1043 |
+
{
|
| 1044 |
+
"name": "h.1.attn.c_attn",
|
| 1045 |
+
"type": "Conv1D",
|
| 1046 |
+
"shape": [
|
| 1047 |
+
768,
|
| 1048 |
+
2304
|
| 1049 |
+
],
|
| 1050 |
+
"numel": 1769472,
|
| 1051 |
+
"mean_abs": 0.10081971436738968,
|
| 1052 |
+
"std": 0.1400582492351532
|
| 1053 |
+
},
|
| 1054 |
+
{
|
| 1055 |
+
"name": "h.1.attn.c_proj",
|
| 1056 |
+
"type": "Conv1D",
|
| 1057 |
+
"shape": [
|
| 1058 |
+
768,
|
| 1059 |
+
768
|
| 1060 |
+
],
|
| 1061 |
+
"numel": 589824,
|
| 1062 |
+
"mean_abs": 0.06751096993684769,
|
| 1063 |
+
"std": 0.10191785544157028
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"name": "h.1.mlp.c_fc",
|
| 1067 |
+
"type": "Conv1D",
|
| 1068 |
+
"shape": [
|
| 1069 |
+
768,
|
| 1070 |
+
3072
|
| 1071 |
+
],
|
| 1072 |
+
"numel": 2359296,
|
| 1073 |
+
"mean_abs": 0.09996536374092102,
|
| 1074 |
+
"std": 0.1307205706834793
|
| 1075 |
+
},
|
| 1076 |
+
{
|
| 1077 |
+
"name": "h.1.mlp.c_proj",
|
| 1078 |
+
"type": "Conv1D",
|
| 1079 |
+
"shape": [
|
| 1080 |
+
3072,
|
| 1081 |
+
768
|
| 1082 |
+
],
|
| 1083 |
+
"numel": 2359296,
|
| 1084 |
+
"mean_abs": 0.06410805881023407,
|
| 1085 |
+
"std": 0.08719106763601303
|
| 1086 |
+
},
|
| 1087 |
+
{
|
| 1088 |
+
"name": "h.2.attn.c_attn",
|
| 1089 |
+
"type": "Conv1D",
|
| 1090 |
+
"shape": [
|
| 1091 |
+
768,
|
| 1092 |
+
2304
|
| 1093 |
+
],
|
| 1094 |
+
"numel": 1769472,
|
| 1095 |
+
"mean_abs": 0.1112792044878006,
|
| 1096 |
+
"std": 0.15266869962215424
|
| 1097 |
+
},
|
| 1098 |
+
{
|
| 1099 |
+
"name": "h.2.attn.c_proj",
|
| 1100 |
+
"type": "Conv1D",
|
| 1101 |
+
"shape": [
|
| 1102 |
+
768,
|
| 1103 |
+
768
|
| 1104 |
+
],
|
| 1105 |
+
"numel": 589824,
|
| 1106 |
+
"mean_abs": 0.06107465550303459,
|
| 1107 |
+
"std": 0.08103547245264053
|
| 1108 |
+
},
|
| 1109 |
+
{
|
| 1110 |
+
"name": "h.2.mlp.c_fc",
|
| 1111 |
+
"type": "Conv1D",
|
| 1112 |
+
"shape": [
|
| 1113 |
+
768,
|
| 1114 |
+
3072
|
| 1115 |
+
],
|
| 1116 |
+
"numel": 2359296,
|
| 1117 |
+
"mean_abs": 0.102850042283535,
|
| 1118 |
+
"std": 0.13352689146995544
|
| 1119 |
+
},
|
| 1120 |
+
{
|
| 1121 |
+
"name": "h.2.mlp.c_proj",
|
| 1122 |
+
"type": "Conv1D",
|
| 1123 |
+
"shape": [
|
| 1124 |
+
3072,
|
| 1125 |
+
768
|
| 1126 |
+
],
|
| 1127 |
+
"numel": 2359296,
|
| 1128 |
+
"mean_abs": 0.0694674625992775,
|
| 1129 |
+
"std": 0.09308750927448273
|
| 1130 |
+
},
|
| 1131 |
+
{
|
| 1132 |
+
"name": "h.3.attn.c_attn",
|
| 1133 |
+
"type": "Conv1D",
|
| 1134 |
+
"shape": [
|
| 1135 |
+
768,
|
| 1136 |
+
2304
|
| 1137 |
+
],
|
| 1138 |
+
"numel": 1769472,
|
| 1139 |
+
"mean_abs": 0.10865448415279388,
|
| 1140 |
+
"std": 0.141807422041893
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"name": "h.3.attn.c_proj",
|
| 1144 |
+
"type": "Conv1D",
|
| 1145 |
+
"shape": [
|
| 1146 |
+
768,
|
| 1147 |
+
768
|
| 1148 |
+
],
|
| 1149 |
+
"numel": 589824,
|
| 1150 |
+
"mean_abs": 0.06414998322725296,
|
| 1151 |
+
"std": 0.0841251015663147
|
| 1152 |
+
},
|
| 1153 |
+
{
|
| 1154 |
+
"name": "h.3.mlp.c_fc",
|
| 1155 |
+
"type": "Conv1D",
|
| 1156 |
+
"shape": [
|
| 1157 |
+
768,
|
| 1158 |
+
3072
|
| 1159 |
+
],
|
| 1160 |
+
"numel": 2359296,
|
| 1161 |
+
"mean_abs": 0.1019684374332428,
|
| 1162 |
+
"std": 0.12953029572963715
|
| 1163 |
+
},
|
| 1164 |
+
{
|
| 1165 |
+
"name": "h.3.mlp.c_proj",
|
| 1166 |
+
"type": "Conv1D",
|
| 1167 |
+
"shape": [
|
| 1168 |
+
3072,
|
| 1169 |
+
768
|
| 1170 |
+
],
|
| 1171 |
+
"numel": 2359296,
|
| 1172 |
+
"mean_abs": 0.06875843554735184,
|
| 1173 |
+
"std": 0.09180603921413422
|
| 1174 |
+
},
|
| 1175 |
+
{
|
| 1176 |
+
"name": "h.4.attn.c_attn",
|
| 1177 |
+
"type": "Conv1D",
|
| 1178 |
+
"shape": [
|
| 1179 |
+
768,
|
| 1180 |
+
2304
|
| 1181 |
+
],
|
| 1182 |
+
"numel": 1769472,
|
| 1183 |
+
"mean_abs": 0.11040088534355164,
|
| 1184 |
+
"std": 0.14643779397010803
|
| 1185 |
+
},
|
| 1186 |
+
{
|
| 1187 |
+
"name": "h.4.attn.c_proj",
|
| 1188 |
+
"type": "Conv1D",
|
| 1189 |
+
"shape": [
|
| 1190 |
+
768,
|
| 1191 |
+
768
|
| 1192 |
+
],
|
| 1193 |
+
"numel": 589824,
|
| 1194 |
+
"mean_abs": 0.07071730494499207,
|
| 1195 |
+
"std": 0.09297914057970047
|
| 1196 |
+
},
|
| 1197 |
+
{
|
| 1198 |
+
"name": "h.4.mlp.c_fc",
|
| 1199 |
+
"type": "Conv1D",
|
| 1200 |
+
"shape": [
|
| 1201 |
+
768,
|
| 1202 |
+
3072
|
| 1203 |
+
],
|
| 1204 |
+
"numel": 2359296,
|
| 1205 |
+
"mean_abs": 0.10167700052261353,
|
| 1206 |
+
"std": 0.12971320748329163
|
| 1207 |
+
},
|
| 1208 |
+
{
|
| 1209 |
+
"name": "h.4.mlp.c_proj",
|
| 1210 |
+
"type": "Conv1D",
|
| 1211 |
+
"shape": [
|
| 1212 |
+
3072,
|
| 1213 |
+
768
|
| 1214 |
+
],
|
| 1215 |
+
"numel": 2359296,
|
| 1216 |
+
"mean_abs": 0.0693950429558754,
|
| 1217 |
+
"std": 0.09099913388490677
|
| 1218 |
+
},
|
| 1219 |
+
{
|
| 1220 |
+
"name": "h.5.attn.c_attn",
|
| 1221 |
+
"type": "Conv1D",
|
| 1222 |
+
"shape": [
|
| 1223 |
+
768,
|
| 1224 |
+
2304
|
| 1225 |
+
],
|
| 1226 |
+
"numel": 1769472,
|
| 1227 |
+
"mean_abs": 0.09946709871292114,
|
| 1228 |
+
"std": 0.12801270186901093
|
| 1229 |
+
},
|
| 1230 |
+
{
|
| 1231 |
+
"name": "h.5.attn.c_proj",
|
| 1232 |
+
"type": "Conv1D",
|
| 1233 |
+
"shape": [
|
| 1234 |
+
768,
|
| 1235 |
+
768
|
| 1236 |
+
],
|
| 1237 |
+
"numel": 589824,
|
| 1238 |
+
"mean_abs": 0.07261313498020172,
|
| 1239 |
+
"std": 0.09377486258745193
|
| 1240 |
+
},
|
| 1241 |
+
{
|
| 1242 |
+
"name": "h.5.mlp.c_fc",
|
| 1243 |
+
"type": "Conv1D",
|
| 1244 |
+
"shape": [
|
| 1245 |
+
768,
|
| 1246 |
+
3072
|
| 1247 |
+
],
|
| 1248 |
+
"numel": 2359296,
|
| 1249 |
+
"mean_abs": 0.09962155669927597,
|
| 1250 |
+
"std": 0.1267070472240448
|
| 1251 |
+
},
|
| 1252 |
+
{
|
| 1253 |
+
"name": "h.5.mlp.c_proj",
|
| 1254 |
+
"type": "Conv1D",
|
| 1255 |
+
"shape": [
|
| 1256 |
+
3072,
|
| 1257 |
+
768
|
| 1258 |
+
],
|
| 1259 |
+
"numel": 2359296,
|
| 1260 |
+
"mean_abs": 0.07507043331861496,
|
| 1261 |
+
"std": 0.09735694527626038
|
| 1262 |
+
},
|
| 1263 |
+
{
|
| 1264 |
+
"name": "h.6.attn.c_attn",
|
| 1265 |
+
"type": "Conv1D",
|
| 1266 |
+
"shape": [
|
| 1267 |
+
768,
|
| 1268 |
+
2304
|
| 1269 |
+
],
|
| 1270 |
+
"numel": 1769472,
|
| 1271 |
+
"mean_abs": 0.09937240183353424,
|
| 1272 |
+
"std": 0.1268601417541504
|
| 1273 |
+
},
|
| 1274 |
+
{
|
| 1275 |
+
"name": "h.6.attn.c_proj",
|
| 1276 |
+
"type": "Conv1D",
|
| 1277 |
+
"shape": [
|
| 1278 |
+
768,
|
| 1279 |
+
768
|
| 1280 |
+
],
|
| 1281 |
+
"numel": 589824,
|
| 1282 |
+
"mean_abs": 0.08774353563785553,
|
| 1283 |
+
"std": 0.11368992179632187
|
| 1284 |
+
},
|
| 1285 |
+
{
|
| 1286 |
+
"name": "h.6.mlp.c_fc",
|
| 1287 |
+
"type": "Conv1D",
|
| 1288 |
+
"shape": [
|
| 1289 |
+
768,
|
| 1290 |
+
3072
|
| 1291 |
+
],
|
| 1292 |
+
"numel": 2359296,
|
| 1293 |
+
"mean_abs": 0.0995308980345726,
|
| 1294 |
+
"std": 0.12635649740695953
|
| 1295 |
+
},
|
| 1296 |
+
{
|
| 1297 |
+
"name": "h.6.mlp.c_proj",
|
| 1298 |
+
"type": "Conv1D",
|
| 1299 |
+
"shape": [
|
| 1300 |
+
3072,
|
| 1301 |
+
768
|
| 1302 |
+
],
|
| 1303 |
+
"numel": 2359296,
|
| 1304 |
+
"mean_abs": 0.08288692682981491,
|
| 1305 |
+
"std": 0.10733181238174438
|
| 1306 |
+
},
|
| 1307 |
+
{
|
| 1308 |
+
"name": "h.7.attn.c_attn",
|
| 1309 |
+
"type": "Conv1D",
|
| 1310 |
+
"shape": [
|
| 1311 |
+
768,
|
| 1312 |
+
2304
|
| 1313 |
+
],
|
| 1314 |
+
"numel": 1769472,
|
| 1315 |
+
"mean_abs": 0.10130105912685394,
|
| 1316 |
+
"std": 0.1289859414100647
|
| 1317 |
+
},
|
| 1318 |
+
{
|
| 1319 |
+
"name": "h.7.attn.c_proj",
|
| 1320 |
+
"type": "Conv1D",
|
| 1321 |
+
"shape": [
|
| 1322 |
+
768,
|
| 1323 |
+
768
|
| 1324 |
+
],
|
| 1325 |
+
"numel": 589824,
|
| 1326 |
+
"mean_abs": 0.08768072724342346,
|
| 1327 |
+
"std": 0.11391738802194595
|
| 1328 |
+
},
|
| 1329 |
+
{
|
| 1330 |
+
"name": "h.7.mlp.c_fc",
|
| 1331 |
+
"type": "Conv1D",
|
| 1332 |
+
"shape": [
|
| 1333 |
+
768,
|
| 1334 |
+
3072
|
| 1335 |
+
],
|
| 1336 |
+
"numel": 2359296,
|
| 1337 |
+
"mean_abs": 0.10009660571813583,
|
| 1338 |
+
"std": 0.12642338871955872
|
| 1339 |
+
},
|
| 1340 |
+
{
|
| 1341 |
+
"name": "h.7.mlp.c_proj",
|
| 1342 |
+
"type": "Conv1D",
|
| 1343 |
+
"shape": [
|
| 1344 |
+
3072,
|
| 1345 |
+
768
|
| 1346 |
+
],
|
| 1347 |
+
"numel": 2359296,
|
| 1348 |
+
"mean_abs": 0.09225291758775711,
|
| 1349 |
+
"std": 0.1187349334359169
|
| 1350 |
+
},
|
| 1351 |
+
{
|
| 1352 |
+
"name": "h.8.attn.c_attn",
|
| 1353 |
+
"type": "Conv1D",
|
| 1354 |
+
"shape": [
|
| 1355 |
+
768,
|
| 1356 |
+
2304
|
| 1357 |
+
],
|
| 1358 |
+
"numel": 1769472,
|
| 1359 |
+
"mean_abs": 0.10021781921386719,
|
| 1360 |
+
"std": 0.12693071365356445
|
| 1361 |
+
},
|
| 1362 |
+
{
|
| 1363 |
+
"name": "h.8.attn.c_proj",
|
| 1364 |
+
"type": "Conv1D",
|
| 1365 |
+
"shape": [
|
| 1366 |
+
768,
|
| 1367 |
+
768
|
| 1368 |
+
],
|
| 1369 |
+
"numel": 589824,
|
| 1370 |
+
"mean_abs": 0.09467821568250656,
|
| 1371 |
+
"std": 0.12236364185810089
|
| 1372 |
+
},
|
| 1373 |
+
{
|
| 1374 |
+
"name": "h.8.mlp.c_fc",
|
| 1375 |
+
"type": "Conv1D",
|
| 1376 |
+
"shape": [
|
| 1377 |
+
768,
|
| 1378 |
+
3072
|
| 1379 |
+
],
|
| 1380 |
+
"numel": 2359296,
|
| 1381 |
+
"mean_abs": 0.10086178034543991,
|
| 1382 |
+
"std": 0.12728072702884674
|
| 1383 |
+
},
|
| 1384 |
+
{
|
| 1385 |
+
"name": "h.8.mlp.c_proj",
|
| 1386 |
+
"type": "Conv1D",
|
| 1387 |
+
"shape": [
|
| 1388 |
+
3072,
|
| 1389 |
+
768
|
| 1390 |
+
],
|
| 1391 |
+
"numel": 2359296,
|
| 1392 |
+
"mean_abs": 0.10524643957614899,
|
| 1393 |
+
"std": 0.13540396094322205
|
| 1394 |
+
},
|
| 1395 |
+
{
|
| 1396 |
+
"name": "h.9.attn.c_attn",
|
| 1397 |
+
"type": "Conv1D",
|
| 1398 |
+
"shape": [
|
| 1399 |
+
768,
|
| 1400 |
+
2304
|
| 1401 |
+
],
|
| 1402 |
+
"numel": 1769472,
|
| 1403 |
+
"mean_abs": 0.09957355260848999,
|
| 1404 |
+
"std": 0.1262170672416687
|
| 1405 |
+
},
|
| 1406 |
+
{
|
| 1407 |
+
"name": "h.9.attn.c_proj",
|
| 1408 |
+
"type": "Conv1D",
|
| 1409 |
+
"shape": [
|
| 1410 |
+
768,
|
| 1411 |
+
768
|
| 1412 |
+
],
|
| 1413 |
+
"numel": 589824,
|
| 1414 |
+
"mean_abs": 0.1055290699005127,
|
| 1415 |
+
"std": 0.13681966066360474
|
| 1416 |
+
},
|
| 1417 |
+
{
|
| 1418 |
+
"name": "h.9.mlp.c_fc",
|
| 1419 |
+
"type": "Conv1D",
|
| 1420 |
+
"shape": [
|
| 1421 |
+
768,
|
| 1422 |
+
3072
|
| 1423 |
+
],
|
| 1424 |
+
"numel": 2359296,
|
| 1425 |
+
"mean_abs": 0.10125003755092621,
|
| 1426 |
+
"std": 0.12761937081813812
|
| 1427 |
+
},
|
| 1428 |
+
{
|
| 1429 |
+
"name": "h.9.mlp.c_proj",
|
| 1430 |
+
"type": "Conv1D",
|
| 1431 |
+
"shape": [
|
| 1432 |
+
3072,
|
| 1433 |
+
768
|
| 1434 |
+
],
|
| 1435 |
+
"numel": 2359296,
|
| 1436 |
+
"mean_abs": 0.12123824656009674,
|
| 1437 |
+
"std": 0.1558738499879837
|
| 1438 |
+
},
|
| 1439 |
+
{
|
| 1440 |
+
"name": "h.10.attn.c_attn",
|
| 1441 |
+
"type": "Conv1D",
|
| 1442 |
+
"shape": [
|
| 1443 |
+
768,
|
| 1444 |
+
2304
|
| 1445 |
+
],
|
| 1446 |
+
"numel": 1769472,
|
| 1447 |
+
"mean_abs": 0.099522165954113,
|
| 1448 |
+
"std": 0.1266736388206482
|
| 1449 |
+
},
|
| 1450 |
+
{
|
| 1451 |
+
"name": "h.10.attn.c_proj",
|
| 1452 |
+
"type": "Conv1D",
|
| 1453 |
+
"shape": [
|
| 1454 |
+
768,
|
| 1455 |
+
768
|
| 1456 |
+
],
|
| 1457 |
+
"numel": 589824,
|
| 1458 |
+
"mean_abs": 0.1112886518239975,
|
| 1459 |
+
"std": 0.14662745594978333
|
| 1460 |
+
},
|
| 1461 |
+
{
|
| 1462 |
+
"name": "h.10.mlp.c_fc",
|
| 1463 |
+
"type": "Conv1D",
|
| 1464 |
+
"shape": [
|
| 1465 |
+
768,
|
| 1466 |
+
3072
|
| 1467 |
+
],
|
| 1468 |
+
"numel": 2359296,
|
| 1469 |
+
"mean_abs": 0.10126420110464096,
|
| 1470 |
+
"std": 0.12764813005924225
|
| 1471 |
+
},
|
| 1472 |
+
{
|
| 1473 |
+
"name": "h.10.mlp.c_proj",
|
| 1474 |
+
"type": "Conv1D",
|
| 1475 |
+
"shape": [
|
| 1476 |
+
3072,
|
| 1477 |
+
768
|
| 1478 |
+
],
|
| 1479 |
+
"numel": 2359296,
|
| 1480 |
+
"mean_abs": 0.13795694708824158,
|
| 1481 |
+
"std": 0.17814528942108154
|
| 1482 |
+
},
|
| 1483 |
+
{
|
| 1484 |
+
"name": "h.11.attn.c_attn",
|
| 1485 |
+
"type": "Conv1D",
|
| 1486 |
+
"shape": [
|
| 1487 |
+
768,
|
| 1488 |
+
2304
|
| 1489 |
+
],
|
| 1490 |
+
"numel": 1769472,
|
| 1491 |
+
"mean_abs": 0.09929489344358444,
|
| 1492 |
+
"std": 0.12847475707530975
|
| 1493 |
+
},
|
| 1494 |
+
{
|
| 1495 |
+
"name": "h.11.attn.c_proj",
|
| 1496 |
+
"type": "Conv1D",
|
| 1497 |
+
"shape": [
|
| 1498 |
+
768,
|
| 1499 |
+
768
|
| 1500 |
+
],
|
| 1501 |
+
"numel": 589824,
|
| 1502 |
+
"mean_abs": 0.1202564612030983,
|
| 1503 |
+
"std": 0.1819266527891159
|
| 1504 |
+
},
|
| 1505 |
+
{
|
| 1506 |
+
"name": "h.11.mlp.c_fc",
|
| 1507 |
+
"type": "Conv1D",
|
| 1508 |
+
"shape": [
|
| 1509 |
+
768,
|
| 1510 |
+
3072
|
| 1511 |
+
],
|
| 1512 |
+
"numel": 2359296,
|
| 1513 |
+
"mean_abs": 0.10254928469657898,
|
| 1514 |
+
"std": 0.13000451028347015
|
| 1515 |
+
},
|
| 1516 |
+
{
|
| 1517 |
+
"name": "h.11.mlp.c_proj",
|
| 1518 |
+
"type": "Conv1D",
|
| 1519 |
+
"shape": [
|
| 1520 |
+
3072,
|
| 1521 |
+
768
|
| 1522 |
+
],
|
| 1523 |
+
"numel": 2359296,
|
| 1524 |
+
"mean_abs": 0.15085573494434357,
|
| 1525 |
+
"std": 0.19821906089782715
|
| 1526 |
+
}
|
| 1527 |
+
],
|
| 1528 |
+
"beta_sweep": [
|
| 1529 |
+
{
|
| 1530 |
+
"beta": 0.25,
|
| 1531 |
+
"N": 84934656,
|
| 1532 |
+
"n_eff": 47579784.0,
|
| 1533 |
+
"nu": 11894946,
|
| 1534 |
+
"sparsity": 0.8599517963550709,
|
| 1535 |
+
"s_eff": 0.372226744890213,
|
| 1536 |
+
"actual_1_minus_seff": 0.627773255109787,
|
| 1537 |
+
"theoretical_bound_1_minus_seff": 0.8597205739543827,
|
| 1538 |
+
"bound_holds": true,
|
| 1539 |
+
"bound_margin": 0.23194731884459574,
|
| 1540 |
+
"l2_preserved_fraction": 0.7956477600279875
|
| 1541 |
+
},
|
| 1542 |
+
{
|
| 1543 |
+
"beta": 0.5,
|
| 1544 |
+
"N": 84934656,
|
| 1545 |
+
"n_eff": 47579784.0,
|
| 1546 |
+
"nu": 23789892,
|
| 1547 |
+
"sparsity": 0.7199035927101418,
|
| 1548 |
+
"s_eff": 0.5907898545265198,
|
| 1549 |
+
"actual_1_minus_seff": 0.4092101454734802,
|
| 1550 |
+
"theoretical_bound_1_minus_seff": 0.7197783607246819,
|
| 1551 |
+
"bound_holds": true,
|
| 1552 |
+
"bound_margin": 0.3105682152512017,
|
| 1553 |
+
"l2_preserved_fraction": 0.9098612773205677
|
| 1554 |
+
},
|
| 1555 |
+
{
|
| 1556 |
+
"beta": 0.75,
|
| 1557 |
+
"N": 84934656,
|
| 1558 |
+
"n_eff": 47579784.0,
|
| 1559 |
+
"nu": 35684840,
|
| 1560 |
+
"sparsity": 0.5798553655176987,
|
| 1561 |
+
"s_eff": 0.7472900152206421,
|
| 1562 |
+
"actual_1_minus_seff": 0.2527099847793579,
|
| 1563 |
+
"theoretical_bound_1_minus_seff": 0.5797814496575094,
|
| 1564 |
+
"bound_holds": true,
|
| 1565 |
+
"bound_margin": 0.32707146487815153,
|
| 1566 |
+
"l2_preserved_fraction": 0.964834627084236
|
| 1567 |
+
},
|
| 1568 |
+
{
|
| 1569 |
+
"beta": 1.0,
|
| 1570 |
+
"N": 84934656,
|
| 1571 |
+
"n_eff": 47579784.0,
|
| 1572 |
+
"nu": 47579784,
|
| 1573 |
+
"sparsity": 0.4398071854202835,
|
| 1574 |
+
"s_eff": 0.8598938584327698,
|
| 1575 |
+
"actual_1_minus_seff": 0.14010614156723022,
|
| 1576 |
+
"theoretical_bound_1_minus_seff": 0.4397649008190052,
|
| 1577 |
+
"bound_holds": true,
|
| 1578 |
+
"bound_margin": 0.299658759251775,
|
| 1579 |
+
"l2_preserved_fraction": 0.9918765096187878
|
| 1580 |
+
},
|
| 1581 |
+
{
|
| 1582 |
+
"beta": 1.25,
|
| 1583 |
+
"N": 84934656,
|
| 1584 |
+
"n_eff": 47579784.0,
|
| 1585 |
+
"nu": 59474728,
|
| 1586 |
+
"sparsity": 0.2997590053228685,
|
| 1587 |
+
"s_eff": 0.9368233680725098,
|
| 1588 |
+
"actual_1_minus_seff": 0.06317663192749023,
|
| 1589 |
+
"theoretical_bound_1_minus_seff": 0.2997377243187093,
|
| 1590 |
+
"bound_holds": true,
|
| 1591 |
+
"bound_margin": 0.23656109239121909,
|
| 1592 |
+
"l2_preserved_fraction": 0.9991645834911109
|
| 1593 |
+
},
|
| 1594 |
+
{
|
| 1595 |
+
"beta": 1.5,
|
| 1596 |
+
"N": 84934656,
|
| 1597 |
+
"n_eff": 47579784.0,
|
| 1598 |
+
"nu": 71369680,
|
| 1599 |
+
"sparsity": 0.15971073103539735,
|
| 1600 |
+
"s_eff": 0.9825542569160461,
|
| 1601 |
+
"actual_1_minus_seff": 0.017445743083953857,
|
| 1602 |
+
"theoretical_bound_1_minus_seff": 0.15970317586515073,
|
| 1603 |
+
"bound_holds": true,
|
| 1604 |
+
"bound_margin": 0.14225743278119687,
|
| 1605 |
+
"l2_preserved_fraction": 1.000013490944357
|
| 1606 |
+
},
|
| 1607 |
+
{
|
| 1608 |
+
"beta": 2.0,
|
| 1609 |
+
"N": 84934656,
|
| 1610 |
+
"n_eff": 47579784.0,
|
| 1611 |
+
"nu": 84934656,
|
| 1612 |
+
"sparsity": 0.0,
|
| 1613 |
+
"s_eff": 1.0,
|
| 1614 |
+
"actual_1_minus_seff": 0.0,
|
| 1615 |
+
"theoretical_bound_1_minus_seff": 0.0,
|
| 1616 |
+
"bound_holds": true,
|
| 1617 |
+
"bound_margin": 0.0,
|
| 1618 |
+
"l2_preserved_fraction": 1.0
|
| 1619 |
+
}
|
| 1620 |
+
],
|
| 1621 |
+
"uniform_comparison": [
|
| 1622 |
+
{
|
| 1623 |
+
"target_sparsity": 0.8599517963550709,
|
| 1624 |
+
"actual_nu": 11894945,
|
| 1625 |
+
"s_eff": 0.3722267150878906,
|
| 1626 |
+
"l2_preserved": 0.7956480666403594,
|
| 1627 |
+
"beta": 0.25
|
| 1628 |
+
},
|
| 1629 |
+
{
|
| 1630 |
+
"target_sparsity": 0.7199035927101418,
|
| 1631 |
+
"actual_nu": 23789891,
|
| 1632 |
+
"s_eff": 0.5907898545265198,
|
| 1633 |
+
"l2_preserved": 0.9098612773205677,
|
| 1634 |
+
"beta": 0.5
|
| 1635 |
+
},
|
| 1636 |
+
{
|
| 1637 |
+
"target_sparsity": 0.5798553655176987,
|
| 1638 |
+
"actual_nu": 35684840,
|
| 1639 |
+
"s_eff": 0.7472900152206421,
|
| 1640 |
+
"l2_preserved": 0.964834627084236,
|
| 1641 |
+
"beta": 0.75
|
| 1642 |
+
},
|
| 1643 |
+
{
|
| 1644 |
+
"target_sparsity": 0.4398071854202835,
|
| 1645 |
+
"actual_nu": 47579784,
|
| 1646 |
+
"s_eff": 0.8598938584327698,
|
| 1647 |
+
"l2_preserved": 0.9918765096187878,
|
| 1648 |
+
"beta": 1.0
|
| 1649 |
+
},
|
| 1650 |
+
{
|
| 1651 |
+
"target_sparsity": 0.2997590053228685,
|
| 1652 |
+
"actual_nu": 59474728,
|
| 1653 |
+
"s_eff": 0.9368233680725098,
|
| 1654 |
+
"l2_preserved": 0.9991645834911109,
|
| 1655 |
+
"beta": 1.25
|
| 1656 |
+
},
|
| 1657 |
+
{
|
| 1658 |
+
"target_sparsity": 0.15971073103539735,
|
| 1659 |
+
"actual_nu": 71369680,
|
| 1660 |
+
"s_eff": 0.9825542569160461,
|
| 1661 |
+
"l2_preserved": 1.000013490944357,
|
| 1662 |
+
"beta": 1.5
|
| 1663 |
+
},
|
| 1664 |
+
{
|
| 1665 |
+
"target_sparsity": 0.0,
|
| 1666 |
+
"actual_nu": 84934656,
|
| 1667 |
+
"s_eff": 1.0,
|
| 1668 |
+
"l2_preserved": 1.0,
|
| 1669 |
+
"beta": 2.0
|
| 1670 |
+
}
|
| 1671 |
+
]
|
| 1672 |
+
}
|
| 1673 |
+
},
|
| 1674 |
+
"wall_time_seconds": 143.50027084350586,
|
| 1675 |
+
"sha256": "b57c6935d7af5ed2ee3629144152e62d36df167afe14ca314aa8318baa1d5169"
|
| 1676 |
+
}
|