Upload scr/gemma-2-2b-res-snap-matryoshka-dc_blocks.2.hook_resid_post_eval_results.json with huggingface_hub
Browse files
scr/gemma-2-2b-res-snap-matryoshka-dc_blocks.2.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": "404efc13-d102-4864-9ae1-43c4ff9dca0f",
|
| 73 |
+
"datetime_epoch_millis": 1745600122303,
|
| 74 |
+
"eval_result_metrics": {
|
| 75 |
+
"scr_metrics": {
|
| 76 |
+
"scr_dir1_threshold_2": -0.5205385964757221,
|
| 77 |
+
"scr_metric_threshold_2": 0.07698124768645498,
|
| 78 |
+
"scr_dir2_threshold_2": 0.07698124768645498,
|
| 79 |
+
"scr_dir1_threshold_5": -0.0628458750337362,
|
| 80 |
+
"scr_metric_threshold_5": 0.10034919410713425,
|
| 81 |
+
"scr_dir2_threshold_5": 0.10034919410713425,
|
| 82 |
+
"scr_dir1_threshold_10": 0.17638798270761677,
|
| 83 |
+
"scr_metric_threshold_10": 0.13616846183571646,
|
| 84 |
+
"scr_dir2_threshold_10": 0.13616846183571646,
|
| 85 |
+
"scr_dir1_threshold_20": 0.042470297586520316,
|
| 86 |
+
"scr_metric_threshold_20": 0.16325446167680585,
|
| 87 |
+
"scr_dir2_threshold_20": 0.16325446167680585,
|
| 88 |
+
"scr_dir1_threshold_50": -0.18862833898626105,
|
| 89 |
+
"scr_metric_threshold_50": 0.19565664865558605,
|
| 90 |
+
"scr_dir2_threshold_50": 0.19565664865558605,
|
| 91 |
+
"scr_dir1_threshold_100": -1.0415166508518798,
|
| 92 |
+
"scr_metric_threshold_100": 0.17805649946559163,
|
| 93 |
+
"scr_dir2_threshold_100": 0.17805649946559163,
|
| 94 |
+
"scr_dir1_threshold_500": -2.990358972449936,
|
| 95 |
+
"scr_metric_threshold_500": -0.003763628881926547,
|
| 96 |
+
"scr_dir2_threshold_500": -0.003763628881926547
|
| 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": -0.4782624470037007,
|
| 103 |
+
"scr_metric_threshold_2": 0.009280760107100267,
|
| 104 |
+
"scr_dir2_threshold_2": 0.009280760107100267,
|
| 105 |
+
"scr_dir1_threshold_5": -0.43478215799894265,
|
| 106 |
+
"scr_metric_threshold_5": 0.004640449200461182,
|
| 107 |
+
"scr_dir2_threshold_5": 0.004640449200461182,
|
| 108 |
+
"scr_dir1_threshold_10": 0.478259855497621,
|
| 109 |
+
"scr_metric_threshold_10": 0.011600915560419811,
|
| 110 |
+
"scr_dir2_threshold_10": 0.011600915560419811,
|
| 111 |
+
"scr_dir1_threshold_20": -0.7391312235018503,
|
| 112 |
+
"scr_metric_threshold_20": 0.03480274668125943,
|
| 113 |
+
"scr_dir2_threshold_20": 0.03480274668125943,
|
| 114 |
+
"scr_dir1_threshold_50": -0.04347769749867833,
|
| 115 |
+
"scr_metric_threshold_50": 0.13689096956554028,
|
| 116 |
+
"scr_dir2_threshold_50": 0.13689096956554028,
|
| 117 |
+
"scr_dir1_threshold_100": -1.2173910789994713,
|
| 118 |
+
"scr_metric_threshold_100": 0.04176335133504015,
|
| 119 |
+
"scr_dir2_threshold_100": 0.04176335133504015,
|
| 120 |
+
"scr_dir1_threshold_500": -7.173913381500793,
|
| 121 |
+
"scr_metric_threshold_500": 0.2320185877960404,
|
| 122 |
+
"scr_dir2_threshold_500": 0.2320185877960404
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"dataset_name": "LabHC/bias_in_bios_class_set1_scr_architect_journalist_results",
|
| 126 |
+
"scr_dir1_threshold_2": -0.47368404541661524,
|
| 127 |
+
"scr_metric_threshold_2": 0.025575421058761334,
|
| 128 |
+
"scr_dir2_threshold_2": 0.025575421058761334,
|
| 129 |
+
"scr_dir1_threshold_5": -1.245612787592204,
|
| 130 |
+
"scr_metric_threshold_5": 0.005115175676672726,
|
| 131 |
+
"scr_dir2_threshold_5": 0.005115175676672726,
|
| 132 |
+
"scr_dir1_threshold_10": 0.3508776516205132,
|
| 133 |
+
"scr_metric_threshold_10": 0.01534537458848408,
|
| 134 |
+
"scr_dir2_threshold_10": 0.01534537458848408,
|
| 135 |
+
"scr_dir1_threshold_20": -0.17543830296287155,
|
| 136 |
+
"scr_metric_threshold_20": -0.0025575116175693136,
|
| 137 |
+
"scr_dir2_threshold_20": -0.0025575116175693136,
|
| 138 |
+
"scr_dir1_threshold_50": 0.47368404541661524,
|
| 139 |
+
"scr_metric_threshold_50": 0.08439895047052604,
|
| 140 |
+
"scr_dir2_threshold_50": 0.08439895047052604,
|
| 141 |
+
"scr_dir1_threshold_100": -2.1403489692586652,
|
| 142 |
+
"scr_metric_threshold_100": -0.048593330499953354,
|
| 143 |
+
"scr_dir2_threshold_100": -0.048593330499953354,
|
| 144 |
+
"scr_dir1_threshold_500": -5.368417089998766,
|
| 145 |
+
"scr_metric_threshold_500": -0.09207163776476808,
|
| 146 |
+
"scr_dir2_threshold_500": -0.09207163776476808
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"dataset_name": "LabHC/bias_in_bios_class_set1_scr_surgeon_psychologist_results",
|
| 150 |
+
"scr_dir1_threshold_2": -3.2777774098480084,
|
| 151 |
+
"scr_metric_threshold_2": 0.00767268729424204,
|
| 152 |
+
"scr_dir2_threshold_2": 0.00767268729424204,
|
| 153 |
+
"scr_dir1_threshold_5": 0.30555481969601644,
|
| 154 |
+
"scr_metric_threshold_5": 0.02557557350029543,
|
| 155 |
+
"scr_dir2_threshold_5": 0.02557557350029543,
|
| 156 |
+
"scr_dir1_threshold_10": -0.22222259015199178,
|
| 157 |
+
"scr_metric_threshold_10": 0.08439910291206014,
|
| 158 |
+
"scr_dir2_threshold_10": 0.08439910291206014,
|
| 159 |
+
"scr_dir1_threshold_20": 0.0,
|
| 160 |
+
"scr_metric_threshold_20": 0.10485934829414875,
|
| 161 |
+
"scr_dir2_threshold_20": 0.10485934829414875,
|
| 162 |
+
"scr_dir1_threshold_50": 0.2777774098480082,
|
| 163 |
+
"scr_metric_threshold_50": 0.05626601779419539,
|
| 164 |
+
"scr_dir2_threshold_50": 0.05626601779419539,
|
| 165 |
+
"scr_dir1_threshold_100": -2.5555548196960163,
|
| 166 |
+
"scr_metric_threshold_100": 0.012787710529380667,
|
| 167 |
+
"scr_dir2_threshold_100": 0.012787710529380667,
|
| 168 |
+
"scr_dir1_threshold_500": -8.888885393556079,
|
| 169 |
+
"scr_metric_threshold_500": -0.08695646208809535,
|
| 170 |
+
"scr_dir2_threshold_500": -0.08695646208809535
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"dataset_name": "LabHC/bias_in_bios_class_set1_scr_attorney_teacher_results",
|
| 174 |
+
"scr_dir1_threshold_2": 0.5056179022792402,
|
| 175 |
+
"scr_metric_threshold_2": 0.04065036579320291,
|
| 176 |
+
"scr_dir2_threshold_2": 0.04065036579320291,
|
| 177 |
+
"scr_dir1_threshold_5": 0.6516853706837721,
|
| 178 |
+
"scr_metric_threshold_5": -0.018970267621611643,
|
| 179 |
+
"scr_dir2_threshold_5": -0.018970267621611643,
|
| 180 |
+
"scr_dir1_threshold_10": 0.26966332769210266,
|
| 181 |
+
"scr_metric_threshold_10": 0.1653116167831811,
|
| 182 |
+
"scr_dir2_threshold_10": 0.1653116167831811,
|
| 183 |
+
"scr_dir1_threshold_20": 0.6741569798007331,
|
| 184 |
+
"scr_metric_threshold_20": 0.2953930134637033,
|
| 185 |
+
"scr_dir2_threshold_20": 0.2953930134637033,
|
| 186 |
+
"scr_dir1_threshold_50": -2.999997990855715,
|
| 187 |
+
"scr_metric_threshold_50": -0.03523038163285382,
|
| 188 |
+
"scr_dir2_threshold_50": -0.03523038163285382,
|
| 189 |
+
"scr_dir1_threshold_100": -3.494380088576474,
|
| 190 |
+
"scr_metric_threshold_100": -0.17886173871424874,
|
| 191 |
+
"scr_dir2_threshold_100": -0.17886173871424874,
|
| 192 |
+
"scr_dir1_threshold_500": -0.6516847009690103,
|
| 193 |
+
"scr_metric_threshold_500": -0.3252032494060131,
|
| 194 |
+
"scr_dir2_threshold_500": -0.3252032494060131
|
| 195 |
+
},
|
| 196 |
+
{
|
| 197 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Books_CDs_and_Vinyl_results",
|
| 198 |
+
"scr_dir1_threshold_2": -0.24000024795532227,
|
| 199 |
+
"scr_metric_threshold_2": 0.21524654807569202,
|
| 200 |
+
"scr_dir2_threshold_2": 0.21524654807569202,
|
| 201 |
+
"scr_dir1_threshold_5": -0.15199995040893555,
|
| 202 |
+
"scr_metric_threshold_5": 0.3228700893989526,
|
| 203 |
+
"scr_dir2_threshold_5": 0.3228700893989526,
|
| 204 |
+
"scr_dir1_threshold_10": -0.2239999771118164,
|
| 205 |
+
"scr_metric_threshold_10": 0.2556053092505611,
|
| 206 |
+
"scr_dir2_threshold_10": 0.2556053092505611,
|
| 207 |
+
"scr_dir1_threshold_20": -0.07999992370605469,
|
| 208 |
+
"scr_metric_threshold_20": 0.26008982393642455,
|
| 209 |
+
"scr_dir2_threshold_20": 0.26008982393642455,
|
| 210 |
+
"scr_dir1_threshold_50": -0.039999961853027344,
|
| 211 |
+
"scr_metric_threshold_50": 0.6636771683997007,
|
| 212 |
+
"scr_dir2_threshold_50": 0.6636771683997007,
|
| 213 |
+
"scr_dir1_threshold_100": -0.039999961853027344,
|
| 214 |
+
"scr_metric_threshold_100": 0.7533631855503367,
|
| 215 |
+
"scr_dir2_threshold_100": 0.7533631855503367,
|
| 216 |
+
"scr_dir1_threshold_500": -0.07999992370605469,
|
| 217 |
+
"scr_metric_threshold_500": 0.08968601715063591,
|
| 218 |
+
"scr_dir2_threshold_500": 0.08968601715063591
|
| 219 |
+
},
|
| 220 |
+
{
|
| 221 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Software_Electronics_results",
|
| 222 |
+
"scr_dir1_threshold_2": 0.06382961184686929,
|
| 223 |
+
"scr_metric_threshold_2": 0.1570249603025748,
|
| 224 |
+
"scr_dir2_threshold_2": 0.1570249603025748,
|
| 225 |
+
"scr_dir1_threshold_5": 0.22340395851008651,
|
| 226 |
+
"scr_metric_threshold_5": 0.19834722753477943,
|
| 227 |
+
"scr_dir2_threshold_5": 0.19834722753477943,
|
| 228 |
+
"scr_dir1_threshold_10": 0.28723420444904374,
|
| 229 |
+
"scr_metric_threshold_10": 0.23140489354045862,
|
| 230 |
+
"scr_dir2_threshold_10": 0.23140489354045862,
|
| 231 |
+
"scr_dir1_threshold_20": 0.20212763259182603,
|
| 232 |
+
"scr_metric_threshold_20": 0.2644628058462787,
|
| 233 |
+
"scr_dir2_threshold_20": 0.2644628058462787,
|
| 234 |
+
"scr_dir1_threshold_50": 0.27659540739782557,
|
| 235 |
+
"scr_metric_threshold_50": 0.318181851768201,
|
| 236 |
+
"scr_dir2_threshold_50": 0.318181851768201,
|
| 237 |
+
"scr_dir1_threshold_100": 0.26595724443869534,
|
| 238 |
+
"scr_metric_threshold_100": 0.30991749684181646,
|
| 239 |
+
"scr_dir2_threshold_100": 0.30991749684181646,
|
| 240 |
+
"scr_dir1_threshold_500": -0.3085105303673042,
|
| 241 |
+
"scr_metric_threshold_500": -0.26859498330947096,
|
| 242 |
+
"scr_dir2_threshold_500": -0.26859498330947096
|
| 243 |
+
},
|
| 244 |
+
{
|
| 245 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Pet_Supplies_Office_Products_results",
|
| 246 |
+
"scr_dir1_threshold_2": -0.31958780624189653,
|
| 247 |
+
"scr_metric_threshold_2": 0.06737606041288677,
|
| 248 |
+
"scr_dir2_threshold_2": 0.06737606041288677,
|
| 249 |
+
"scr_dir1_threshold_5": 0.03092805043658865,
|
| 250 |
+
"scr_metric_threshold_5": 0.14893633210854976,
|
| 251 |
+
"scr_dir2_threshold_5": 0.14893633210854976,
|
| 252 |
+
"scr_dir1_threshold_10": 0.4226804883893842,
|
| 253 |
+
"scr_metric_threshold_10": 0.14893633210854976,
|
| 254 |
+
"scr_dir2_threshold_10": 0.14893633210854976,
|
| 255 |
+
"scr_dir1_threshold_20": 0.3608250019970628,
|
| 256 |
+
"scr_metric_threshold_20": 0.2234042867988847,
|
| 257 |
+
"scr_dir2_threshold_20": 0.2234042867988847,
|
| 258 |
+
"scr_dir1_threshold_50": 0.546392075654883,
|
| 259 |
+
"scr_metric_threshold_50": 0.3085106113152417,
|
| 260 |
+
"scr_dir2_threshold_50": 0.3085106113152417,
|
| 261 |
+
"scr_dir1_threshold_100": 0.7731957305870135,
|
| 262 |
+
"scr_metric_threshold_100": 0.4503546264184634,
|
| 263 |
+
"scr_dir2_threshold_100": 0.4503546264184634,
|
| 264 |
+
"scr_dir1_threshold_500": -1.6597939031209483,
|
| 265 |
+
"scr_metric_threshold_500": 0.29078013584783147,
|
| 266 |
+
"scr_dir2_threshold_500": 0.29078013584783147
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Industrial_and_Scientific_Toys_and_Games_results",
|
| 270 |
+
"scr_dir1_threshold_2": 0.055555670533656135,
|
| 271 |
+
"scr_metric_threshold_2": 0.09302317844717961,
|
| 272 |
+
"scr_dir2_threshold_2": 0.09302317844717961,
|
| 273 |
+
"scr_dir1_threshold_5": 0.11805569640372877,
|
| 274 |
+
"scr_metric_threshold_5": 0.1162789730589745,
|
| 275 |
+
"scr_dir2_threshold_5": 0.1162789730589745,
|
| 276 |
+
"scr_dir1_threshold_10": 0.04861090127607755,
|
| 277 |
+
"scr_metric_threshold_10": 0.17674414994201715,
|
| 278 |
+
"scr_dir2_threshold_10": 0.17674414994201715,
|
| 279 |
+
"scr_dir1_threshold_20": 0.09722221647331719,
|
| 280 |
+
"scr_metric_threshold_20": 0.12558118001131657,
|
| 281 |
+
"scr_dir2_threshold_20": 0.12558118001131657,
|
| 282 |
+
"scr_dir1_threshold_50": 0.0,
|
| 283 |
+
"scr_metric_threshold_50": 0.03255800156413696,
|
| 284 |
+
"scr_dir2_threshold_50": 0.03255800156413696,
|
| 285 |
+
"scr_dir1_threshold_100": 0.07638873654290562,
|
| 286 |
+
"scr_metric_threshold_100": 0.08372069426389779,
|
| 287 |
+
"scr_dir2_threshold_100": 0.08372069426389779,
|
| 288 |
+
"scr_dir1_threshold_500": 0.20833314361946736,
|
| 289 |
+
"scr_metric_threshold_500": 0.13023256071842734,
|
| 290 |
+
"scr_dir2_threshold_500": 0.13023256071842734
|
| 291 |
+
}
|
| 292 |
+
],
|
| 293 |
+
"sae_bench_commit_hash": "Unknown",
|
| 294 |
+
"sae_lens_id": "blocks.2.hook_resid_post",
|
| 295 |
+
"sae_lens_release_id": "gemma-2-2b-res-snap-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.2.hook_resid_post",
|
| 307 |
+
"hook_layer": 2,
|
| 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 |
+
}
|