Upload scr/gemma-2-2b-res-matryoshka-dc_blocks.0.hook_resid_post_eval_results.json with huggingface_hub
Browse files
scr/gemma-2-2b-res-matryoshka-dc_blocks.0.hook_resid_post_eval_results.json
ADDED
|
@@ -0,0 +1,328 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"eval_type_id": "scr",
|
| 3 |
+
"eval_config": {
|
| 4 |
+
"random_seed": 42,
|
| 5 |
+
"dataset_names": [
|
| 6 |
+
"LabHC/bias_in_bios_class_set1",
|
| 7 |
+
"canrager/amazon_reviews_mcauley_1and5"
|
| 8 |
+
],
|
| 9 |
+
"perform_scr": true,
|
| 10 |
+
"early_stopping_patience": 20,
|
| 11 |
+
"train_set_size": 4000,
|
| 12 |
+
"test_set_size": 1000,
|
| 13 |
+
"context_length": 128,
|
| 14 |
+
"probe_train_batch_size": 16,
|
| 15 |
+
"probe_test_batch_size": 500,
|
| 16 |
+
"probe_epochs": 20,
|
| 17 |
+
"probe_lr": 0.001,
|
| 18 |
+
"probe_l1_penalty": 0.001,
|
| 19 |
+
"sae_batch_size": 125,
|
| 20 |
+
"llm_batch_size": 32,
|
| 21 |
+
"llm_dtype": "bfloat16",
|
| 22 |
+
"lower_vram_usage": false,
|
| 23 |
+
"model_name": "gemma-2-2b",
|
| 24 |
+
"n_values": [
|
| 25 |
+
2,
|
| 26 |
+
5,
|
| 27 |
+
10,
|
| 28 |
+
20,
|
| 29 |
+
50,
|
| 30 |
+
100,
|
| 31 |
+
500
|
| 32 |
+
],
|
| 33 |
+
"column1_vals_lookup": {
|
| 34 |
+
"LabHC/bias_in_bios_class_set1": [
|
| 35 |
+
[
|
| 36 |
+
"professor",
|
| 37 |
+
"nurse"
|
| 38 |
+
],
|
| 39 |
+
[
|
| 40 |
+
"architect",
|
| 41 |
+
"journalist"
|
| 42 |
+
],
|
| 43 |
+
[
|
| 44 |
+
"surgeon",
|
| 45 |
+
"psychologist"
|
| 46 |
+
],
|
| 47 |
+
[
|
| 48 |
+
"attorney",
|
| 49 |
+
"teacher"
|
| 50 |
+
]
|
| 51 |
+
],
|
| 52 |
+
"canrager/amazon_reviews_mcauley_1and5": [
|
| 53 |
+
[
|
| 54 |
+
"Books",
|
| 55 |
+
"CDs_and_Vinyl"
|
| 56 |
+
],
|
| 57 |
+
[
|
| 58 |
+
"Software",
|
| 59 |
+
"Electronics"
|
| 60 |
+
],
|
| 61 |
+
[
|
| 62 |
+
"Pet_Supplies",
|
| 63 |
+
"Office_Products"
|
| 64 |
+
],
|
| 65 |
+
[
|
| 66 |
+
"Industrial_and_Scientific",
|
| 67 |
+
"Toys_and_Games"
|
| 68 |
+
]
|
| 69 |
+
]
|
| 70 |
+
}
|
| 71 |
+
},
|
| 72 |
+
"eval_id": "570e36e0-86e6-40dc-b65a-13d96e4b9429",
|
| 73 |
+
"datetime_epoch_millis": 1745747427014,
|
| 74 |
+
"eval_result_metrics": {
|
| 75 |
+
"scr_metrics": {
|
| 76 |
+
"scr_dir1_threshold_2": -0.2773401137108581,
|
| 77 |
+
"scr_metric_threshold_2": 0.06683525174951971,
|
| 78 |
+
"scr_dir2_threshold_2": 0.06683525174951971,
|
| 79 |
+
"scr_dir1_threshold_5": 0.05575898647811038,
|
| 80 |
+
"scr_metric_threshold_5": 0.09524557167844104,
|
| 81 |
+
"scr_dir2_threshold_5": 0.09524557167844104,
|
| 82 |
+
"scr_dir1_threshold_10": 0.07924204292309966,
|
| 83 |
+
"scr_metric_threshold_10": 0.16685534471367303,
|
| 84 |
+
"scr_dir2_threshold_10": 0.16685534471367303,
|
| 85 |
+
"scr_dir1_threshold_20": 0.4045781571917739,
|
| 86 |
+
"scr_metric_threshold_20": 0.12314745821347976,
|
| 87 |
+
"scr_dir2_threshold_20": 0.12314745821347976,
|
| 88 |
+
"scr_dir1_threshold_50": -0.0036895977856749905,
|
| 89 |
+
"scr_metric_threshold_50": 0.18493945534512707,
|
| 90 |
+
"scr_dir2_threshold_50": 0.18493945534512707,
|
| 91 |
+
"scr_dir1_threshold_100": -0.34811777183225473,
|
| 92 |
+
"scr_metric_threshold_100": -0.09146590897818013,
|
| 93 |
+
"scr_dir2_threshold_100": -0.09146590897818013,
|
| 94 |
+
"scr_dir1_threshold_500": -1.1459595551348116,
|
| 95 |
+
"scr_metric_threshold_500": -0.3136031653830803,
|
| 96 |
+
"scr_dir2_threshold_500": -0.3136031653830803
|
| 97 |
+
}
|
| 98 |
+
},
|
| 99 |
+
"eval_result_details": [
|
| 100 |
+
{
|
| 101 |
+
"dataset_name": "LabHC/bias_in_bios_class_set1_scr_professor_nurse_results",
|
| 102 |
+
"scr_dir1_threshold_2": 0.3614461378705631,
|
| 103 |
+
"scr_metric_threshold_2": -0.12041884898446643,
|
| 104 |
+
"scr_dir2_threshold_2": -0.12041884898446643,
|
| 105 |
+
"scr_dir1_threshold_5": -0.20481878393604638,
|
| 106 |
+
"scr_metric_threshold_5": -0.05497373416985794,
|
| 107 |
+
"scr_dir2_threshold_5": -0.05497373416985794,
|
| 108 |
+
"scr_dir1_threshold_10": -0.32530130864504286,
|
| 109 |
+
"scr_metric_threshold_10": 0.06020942449223322,
|
| 110 |
+
"scr_dir2_threshold_10": 0.06020942449223322,
|
| 111 |
+
"scr_dir1_threshold_20": 0.8313253271612607,
|
| 112 |
+
"scr_metric_threshold_20": -0.06020942449223322,
|
| 113 |
+
"scr_dir2_threshold_20": -0.06020942449223322,
|
| 114 |
+
"scr_dir1_threshold_50": 0.5662649218066095,
|
| 115 |
+
"scr_metric_threshold_50": 0.07591618339317982,
|
| 116 |
+
"scr_dir2_threshold_50": 0.07591618339317982,
|
| 117 |
+
"scr_dir1_threshold_100": 0.30120523458017145,
|
| 118 |
+
"scr_metric_threshold_100": 0.15706805710873495,
|
| 119 |
+
"scr_dir2_threshold_100": 0.15706805710873495,
|
| 120 |
+
"scr_dir1_threshold_500": -3.2048187839360462,
|
| 121 |
+
"scr_metric_threshold_500": -0.12565438327375208,
|
| 122 |
+
"scr_dir2_threshold_500": -0.12565438327375208
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"dataset_name": "LabHC/bias_in_bios_class_set1_scr_architect_journalist_results",
|
| 126 |
+
"scr_dir1_threshold_2": -0.0506330164478648,
|
| 127 |
+
"scr_metric_threshold_2": 0.08474567709761362,
|
| 128 |
+
"scr_dir2_threshold_2": 0.08474567709761362,
|
| 129 |
+
"scr_dir1_threshold_5": -0.031645540968764145,
|
| 130 |
+
"scr_metric_threshold_5": -0.27796614415958365,
|
| 131 |
+
"scr_dir2_threshold_5": -0.27796614415958365,
|
| 132 |
+
"scr_dir1_threshold_10": 0.13291119661988834,
|
| 133 |
+
"scr_metric_threshold_10": -0.16610176957088094,
|
| 134 |
+
"scr_dir2_threshold_10": -0.16610176957088094,
|
| 135 |
+
"scr_dir1_threshold_20": 0.8607593933906745,
|
| 136 |
+
"scr_metric_threshold_20": 0.19661005168631632,
|
| 137 |
+
"scr_dir2_threshold_20": 0.19661005168631632,
|
| 138 |
+
"scr_dir1_threshold_50": 0.35443036064584277,
|
| 139 |
+
"scr_metric_threshold_50": -0.17627133164244732,
|
| 140 |
+
"scr_dir2_threshold_50": -0.17627133164244732,
|
| 141 |
+
"scr_dir1_threshold_100": -0.07594952467179719,
|
| 142 |
+
"scr_metric_threshold_100": -0.62033904332328,
|
| 143 |
+
"scr_dir2_threshold_100": -0.62033904332328,
|
| 144 |
+
"scr_dir1_threshold_500": -0.696202655802022,
|
| 145 |
+
"scr_metric_threshold_500": -0.6338983920688246,
|
| 146 |
+
"scr_dir2_threshold_500": -0.6338983920688246
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"dataset_name": "LabHC/bias_in_bios_class_set1_scr_surgeon_psychologist_results",
|
| 150 |
+
"scr_dir1_threshold_2": -3.0250011734666886,
|
| 151 |
+
"scr_metric_threshold_2": 0.10610070181707222,
|
| 152 |
+
"scr_dir2_threshold_2": 0.10610070181707222,
|
| 153 |
+
"scr_dir1_threshold_5": 0.599999850988357,
|
| 154 |
+
"scr_metric_threshold_5": 0.12997339134641683,
|
| 155 |
+
"scr_dir2_threshold_5": 0.12997339134641683,
|
| 156 |
+
"scr_dir1_threshold_10": 0.4249998323619016,
|
| 157 |
+
"scr_metric_threshold_10": 0.1803712563300295,
|
| 158 |
+
"scr_dir2_threshold_10": 0.1803712563300295,
|
| 159 |
+
"scr_dir1_threshold_20": 0.6124996926634863,
|
| 160 |
+
"scr_metric_threshold_20": 0.1087533458445125,
|
| 161 |
+
"scr_dir2_threshold_20": 0.1087533458445125,
|
| 162 |
+
"scr_dir1_threshold_50": -0.6875002328306923,
|
| 163 |
+
"scr_metric_threshold_50": 0.34217511118559496,
|
| 164 |
+
"scr_dir2_threshold_50": 0.34217511118559496,
|
| 165 |
+
"scr_dir1_threshold_100": -2.0500008568169474,
|
| 166 |
+
"scr_metric_threshold_100": 0.36870028663986304,
|
| 167 |
+
"scr_dir2_threshold_100": 0.36870028663986304,
|
| 168 |
+
"scr_dir1_threshold_500": -3.3375009406359966,
|
| 169 |
+
"scr_metric_threshold_500": -0.23342176534108247,
|
| 170 |
+
"scr_dir2_threshold_500": -0.23342176534108247
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"dataset_name": "LabHC/bias_in_bios_class_set1_scr_attorney_teacher_results",
|
| 174 |
+
"scr_dir1_threshold_2": 0.028571399377318833,
|
| 175 |
+
"scr_metric_threshold_2": 0.13167259107101267,
|
| 176 |
+
"scr_dir2_threshold_2": 0.13167259107101267,
|
| 177 |
+
"scr_dir1_threshold_5": -0.5028571058779372,
|
| 178 |
+
"scr_metric_threshold_5": 0.30604968015004247,
|
| 179 |
+
"scr_dir2_threshold_5": 0.30604968015004247,
|
| 180 |
+
"scr_dir1_threshold_10": -0.10285717399752657,
|
| 181 |
+
"scr_metric_threshold_10": 0.4555160595265822,
|
| 182 |
+
"scr_dir2_threshold_10": 0.4555160595265822,
|
| 183 |
+
"scr_dir1_threshold_20": -0.04571403464494197,
|
| 184 |
+
"scr_metric_threshold_20": -0.17081862838054826,
|
| 185 |
+
"scr_dir2_threshold_20": -0.17081862838054826,
|
| 186 |
+
"scr_dir1_threshold_50": -0.9028570377583478,
|
| 187 |
+
"scr_metric_threshold_50": 0.3487543902742195,
|
| 188 |
+
"scr_dir2_threshold_50": 0.3487543902742195,
|
| 189 |
+
"scr_dir1_threshold_100": -1.6057142798754638,
|
| 190 |
+
"scr_metric_threshold_100": -0.4377224833372151,
|
| 191 |
+
"scr_dir2_threshold_100": -0.4377224833372151,
|
| 192 |
+
"scr_dir1_threshold_500": -0.9599998365129855,
|
| 193 |
+
"scr_metric_threshold_500": -0.6548042825404219,
|
| 194 |
+
"scr_dir2_threshold_500": -0.6548042825404219
|
| 195 |
+
},
|
| 196 |
+
{
|
| 197 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Books_CDs_and_Vinyl_results",
|
| 198 |
+
"scr_dir1_threshold_2": -0.07500002483526767,
|
| 199 |
+
"scr_metric_threshold_2": 0.21052640584935314,
|
| 200 |
+
"scr_dir2_threshold_2": 0.21052640584935314,
|
| 201 |
+
"scr_dir1_threshold_5": -0.09166647626294787,
|
| 202 |
+
"scr_metric_threshold_5": 0.2583732384906739,
|
| 203 |
+
"scr_dir2_threshold_5": 0.2583732384906739,
|
| 204 |
+
"scr_dir1_threshold_10": -0.3000000993410707,
|
| 205 |
+
"scr_metric_threshold_10": 0.3827752315098024,
|
| 206 |
+
"scr_dir2_threshold_10": 0.3827752315098024,
|
| 207 |
+
"scr_dir1_threshold_20": 0.10000019868214136,
|
| 208 |
+
"scr_metric_threshold_20": 0.4736842705662354,
|
| 209 |
+
"scr_dir2_threshold_20": 0.4736842705662354,
|
| 210 |
+
"scr_dir1_threshold_50": 0.10833342439598145,
|
| 211 |
+
"scr_metric_threshold_50": 0.6172247684901976,
|
| 212 |
+
"scr_dir2_threshold_50": 0.6172247684901976,
|
| 213 |
+
"scr_dir1_threshold_100": 0.12500037252901505,
|
| 214 |
+
"scr_metric_threshold_100": 0.30143544490578617,
|
| 215 |
+
"scr_dir2_threshold_100": 0.30143544490578617,
|
| 216 |
+
"scr_dir1_threshold_500": -2.1249998758236615,
|
| 217 |
+
"scr_metric_threshold_500": -0.3301432022630367,
|
| 218 |
+
"scr_dir2_threshold_500": -0.3301432022630367
|
| 219 |
+
},
|
| 220 |
+
{
|
| 221 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Software_Electronics_results",
|
| 222 |
+
"scr_dir1_threshold_2": 0.10752678523932548,
|
| 223 |
+
"scr_metric_threshold_2": 0.090909263319198,
|
| 224 |
+
"scr_dir2_threshold_2": 0.090909263319198,
|
| 225 |
+
"scr_dir1_threshold_5": 0.1290325268333076,
|
| 226 |
+
"scr_metric_threshold_5": 0.15454561217755247,
|
| 227 |
+
"scr_dir2_threshold_5": 0.15454561217755247,
|
| 228 |
+
"scr_dir1_threshold_10": 0.16129017785898822,
|
| 229 |
+
"scr_metric_threshold_10": 0.22272735646932473,
|
| 230 |
+
"scr_dir2_threshold_10": 0.22272735646932473,
|
| 231 |
+
"scr_dir1_threshold_20": 0.23655931207263306,
|
| 232 |
+
"scr_metric_threshold_20": 0.24090920913316435,
|
| 233 |
+
"scr_dir2_threshold_20": 0.24090920913316435,
|
| 234 |
+
"scr_dir1_threshold_50": 0.16129017785898822,
|
| 235 |
+
"scr_metric_threshold_50": 0.27727264353067527,
|
| 236 |
+
"scr_dir2_threshold_50": 0.27727264353067527,
|
| 237 |
+
"scr_dir1_threshold_100": 0.34408609731195855,
|
| 238 |
+
"scr_metric_threshold_100": 0.2545453954334178,
|
| 239 |
+
"scr_dir2_threshold_100": 0.2545453954334178,
|
| 240 |
+
"scr_dir1_threshold_500": 0.5591396677906095,
|
| 241 |
+
"scr_metric_threshold_500": 0.09545465875261583,
|
| 242 |
+
"scr_dir2_threshold_500": 0.09545465875261583
|
| 243 |
+
},
|
| 244 |
+
{
|
| 245 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Pet_Supplies_Office_Products_results",
|
| 246 |
+
"scr_dir1_threshold_2": 0.28205111223754226,
|
| 247 |
+
"scr_metric_threshold_2": 0.07114604671544196,
|
| 248 |
+
"scr_dir2_threshold_2": 0.07114604671544196,
|
| 249 |
+
"scr_dir1_threshold_5": 0.31623920303015635,
|
| 250 |
+
"scr_metric_threshold_5": 0.21739113021496548,
|
| 251 |
+
"scr_dir2_threshold_5": 0.21739113021496548,
|
| 252 |
+
"scr_dir1_threshold_10": 0.38461538461538464,
|
| 253 |
+
"scr_metric_threshold_10": 0.2450592335972351,
|
| 254 |
+
"scr_dir2_threshold_10": 0.2450592335972351,
|
| 255 |
+
"scr_dir1_threshold_20": 0.4358972660836961,
|
| 256 |
+
"scr_metric_threshold_20": 0.2648220636167774,
|
| 257 |
+
"scr_dir2_threshold_20": 0.2648220636167774,
|
| 258 |
+
"scr_dir1_threshold_50": 0.41025658007015003,
|
| 259 |
+
"scr_metric_threshold_50": 0.3201580347897994,
|
| 260 |
+
"scr_dir2_threshold_50": 0.3201580347897994,
|
| 261 |
+
"scr_dir1_threshold_100": -0.1282049583913884,
|
| 262 |
+
"scr_metric_threshold_100": 0.2845848936363197,
|
| 263 |
+
"scr_dir2_threshold_100": 0.2845848936363197,
|
| 264 |
+
"scr_dir1_threshold_500": 0.17948734930091928,
|
| 265 |
+
"scr_metric_threshold_500": 0.4822134294232602,
|
| 266 |
+
"scr_dir2_threshold_500": 0.4822134294232602
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Industrial_and_Scientific_Toys_and_Games_results",
|
| 270 |
+
"scr_dir1_threshold_2": 0.15231787033820696,
|
| 271 |
+
"scr_metric_threshold_2": -0.03999982288906759,
|
| 272 |
+
"scr_dir2_threshold_2": -0.03999982288906759,
|
| 273 |
+
"scr_dir1_threshold_5": 0.2317882180187577,
|
| 274 |
+
"scr_metric_threshold_5": 0.028571399377318833,
|
| 275 |
+
"scr_dir2_threshold_5": 0.028571399377318833,
|
| 276 |
+
"scr_dir1_threshold_10": 0.2582783339122746,
|
| 277 |
+
"scr_metric_threshold_10": -0.04571403464494197,
|
| 278 |
+
"scr_dir2_threshold_10": -0.04571403464494197,
|
| 279 |
+
"scr_dir1_threshold_20": 0.2052981021252408,
|
| 280 |
+
"scr_metric_threshold_20": -0.06857122226638643,
|
| 281 |
+
"scr_dir2_threshold_20": -0.06857122226638643,
|
| 282 |
+
"scr_dir1_threshold_50": -0.03973497647393185,
|
| 283 |
+
"scr_metric_threshold_50": -0.32571415726020286,
|
| 284 |
+
"scr_dir2_threshold_50": -0.32571415726020286,
|
| 285 |
+
"scr_dir1_threshold_100": 0.3046357406764139,
|
| 286 |
+
"scr_metric_threshold_100": -1.0399998228890677,
|
| 287 |
+
"scr_dir2_threshold_100": -1.0399998228890677,
|
| 288 |
+
"scr_dir1_threshold_500": 0.417218634540689,
|
| 289 |
+
"scr_metric_threshold_500": -1.108571385753401,
|
| 290 |
+
"scr_dir2_threshold_500": -1.108571385753401
|
| 291 |
+
}
|
| 292 |
+
],
|
| 293 |
+
"sae_bench_commit_hash": "Unknown",
|
| 294 |
+
"sae_lens_id": "blocks.0.hook_resid_post",
|
| 295 |
+
"sae_lens_release_id": "gemma-2-2b-res-matryoshka-dc",
|
| 296 |
+
"sae_lens_version": "5.9.1",
|
| 297 |
+
"sae_cfg_dict": {
|
| 298 |
+
"architecture": "jumprelu",
|
| 299 |
+
"d_in": 2304,
|
| 300 |
+
"d_sae": 32768,
|
| 301 |
+
"activation_fn_str": "relu",
|
| 302 |
+
"apply_b_dec_to_input": true,
|
| 303 |
+
"finetuning_scaling_factor": false,
|
| 304 |
+
"context_size": 1024,
|
| 305 |
+
"model_name": "gemma-2-2b",
|
| 306 |
+
"hook_name": "blocks.0.hook_resid_post",
|
| 307 |
+
"hook_layer": 0,
|
| 308 |
+
"hook_head_index": null,
|
| 309 |
+
"prepend_bos": true,
|
| 310 |
+
"dataset_path": "chanind/pile-uncopyrighted-gemma-1024-abbrv-1B",
|
| 311 |
+
"dataset_trust_remote_code": true,
|
| 312 |
+
"normalize_activations": "none",
|
| 313 |
+
"dtype": "torch.bfloat16",
|
| 314 |
+
"device": "cuda",
|
| 315 |
+
"sae_lens_training_version": "5.5.1",
|
| 316 |
+
"activation_fn_kwargs": {
|
| 317 |
+
"k": 40
|
| 318 |
+
},
|
| 319 |
+
"neuronpedia_id": null,
|
| 320 |
+
"model_from_pretrained_kwargs": {
|
| 321 |
+
"center_writing_weights": false
|
| 322 |
+
},
|
| 323 |
+
"seqpos_slice": [
|
| 324 |
+
null
|
| 325 |
+
]
|
| 326 |
+
},
|
| 327 |
+
"eval_result_unstructured": null
|
| 328 |
+
}
|