Upload scr/gemma-2-2b-res-snap-matryoshka-dc_blocks.18.hook_resid_post_eval_results.json with huggingface_hub
Browse files
scr/gemma-2-2b-res-snap-matryoshka-dc_blocks.18.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": 1745611512777,
|
| 74 |
+
"eval_result_metrics": {
|
| 75 |
+
"scr_metrics": {
|
| 76 |
+
"scr_dir1_threshold_2": 0.1724539248915756,
|
| 77 |
+
"scr_metric_threshold_2": 0.21952789140115042,
|
| 78 |
+
"scr_dir2_threshold_2": 0.21952789140115042,
|
| 79 |
+
"scr_dir1_threshold_5": -0.24196847441781916,
|
| 80 |
+
"scr_metric_threshold_5": 0.30581959349616733,
|
| 81 |
+
"scr_dir2_threshold_5": 0.30581959349616733,
|
| 82 |
+
"scr_dir1_threshold_10": 0.21243247361360068,
|
| 83 |
+
"scr_metric_threshold_10": 0.4067231088043306,
|
| 84 |
+
"scr_dir2_threshold_10": 0.4067231088043306,
|
| 85 |
+
"scr_dir1_threshold_20": -0.6975860176887168,
|
| 86 |
+
"scr_metric_threshold_20": 0.40918204088276966,
|
| 87 |
+
"scr_dir2_threshold_20": 0.40918204088276966,
|
| 88 |
+
"scr_dir1_threshold_50": -1.1476383568340953,
|
| 89 |
+
"scr_metric_threshold_50": 0.4168619724004479,
|
| 90 |
+
"scr_dir2_threshold_50": 0.4168619724004479,
|
| 91 |
+
"scr_dir1_threshold_100": -1.1252178603246805,
|
| 92 |
+
"scr_metric_threshold_100": 0.27308591126466947,
|
| 93 |
+
"scr_dir2_threshold_100": 0.27308591126466947,
|
| 94 |
+
"scr_dir1_threshold_500": -1.5168723824484853,
|
| 95 |
+
"scr_metric_threshold_500": 0.36658464300639004,
|
| 96 |
+
"scr_dir2_threshold_500": 0.36658464300639004
|
| 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": 0.3,
|
| 103 |
+
"scr_metric_threshold_2": 0.04379563525787202,
|
| 104 |
+
"scr_dir2_threshold_2": 0.04379563525787202,
|
| 105 |
+
"scr_dir1_threshold_5": 0.1,
|
| 106 |
+
"scr_metric_threshold_5": 0.07299262874747876,
|
| 107 |
+
"scr_dir2_threshold_5": 0.07299262874747876,
|
| 108 |
+
"scr_dir1_threshold_10": 0.4199997615811935,
|
| 109 |
+
"scr_metric_threshold_10": 0.1727493373878026,
|
| 110 |
+
"scr_dir2_threshold_10": 0.1727493373878026,
|
| 111 |
+
"scr_dir1_threshold_20": -2.380002622606871,
|
| 112 |
+
"scr_metric_threshold_20": 0.29683692227728603,
|
| 113 |
+
"scr_dir2_threshold_20": 0.29683692227728603,
|
| 114 |
+
"scr_dir1_threshold_50": -2.4000023841880647,
|
| 115 |
+
"scr_metric_threshold_50": 0.35766425792364687,
|
| 116 |
+
"scr_dir2_threshold_50": 0.35766425792364687,
|
| 117 |
+
"scr_dir1_threshold_100": -2.7600028610256775,
|
| 118 |
+
"scr_metric_threshold_100": 0.11922132262557436,
|
| 119 |
+
"scr_dir2_threshold_100": 0.11922132262557436,
|
| 120 |
+
"scr_dir1_threshold_500": -2.3000023841880646,
|
| 121 |
+
"scr_metric_threshold_500": 0.09732350499663836,
|
| 122 |
+
"scr_dir2_threshold_500": 0.09732350499663836
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"dataset_name": "LabHC/bias_in_bios_class_set1_scr_architect_journalist_results",
|
| 126 |
+
"scr_dir1_threshold_2": 0.11607148083950977,
|
| 127 |
+
"scr_metric_threshold_2": 0.2499999556513093,
|
| 128 |
+
"scr_dir2_threshold_2": 0.2499999556513093,
|
| 129 |
+
"scr_dir1_threshold_5": -0.9285707823478792,
|
| 130 |
+
"scr_metric_threshold_5": 0.3660714776717647,
|
| 131 |
+
"scr_dir2_threshold_5": 0.3660714776717647,
|
| 132 |
+
"scr_dir1_threshold_10": 0.1875001663075311,
|
| 133 |
+
"scr_metric_threshold_10": 0.6815476628683491,
|
| 134 |
+
"scr_dir2_threshold_10": 0.6815476628683491,
|
| 135 |
+
"scr_dir1_threshold_20": -1.2857136775038862,
|
| 136 |
+
"scr_metric_threshold_20": 0.3660714776717647,
|
| 137 |
+
"scr_dir2_threshold_20": 0.3660714776717647,
|
| 138 |
+
"scr_dir1_threshold_50": -1.2857136775038862,
|
| 139 |
+
"scr_metric_threshold_50": 0.14880954809571156,
|
| 140 |
+
"scr_dir2_threshold_50": 0.14880954809571156,
|
| 141 |
+
"scr_dir1_threshold_100": -0.19642808676090937,
|
| 142 |
+
"scr_metric_threshold_100": -0.0744046853504744,
|
| 143 |
+
"scr_dir2_threshold_100": -0.0744046853504744,
|
| 144 |
+
"scr_dir1_threshold_500": -2.1249988691087887,
|
| 145 |
+
"scr_metric_threshold_500": -0.04464281121028464,
|
| 146 |
+
"scr_dir2_threshold_500": -0.04464281121028464
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"dataset_name": "LabHC/bias_in_bios_class_set1_scr_surgeon_psychologist_results",
|
| 150 |
+
"scr_dir1_threshold_2": 0.37254890502759347,
|
| 151 |
+
"scr_metric_threshold_2": 0.1642157424904366,
|
| 152 |
+
"scr_dir2_threshold_2": 0.1642157424904366,
|
| 153 |
+
"scr_dir1_threshold_5": -0.11764671508278034,
|
| 154 |
+
"scr_metric_threshold_5": 0.20343135715230082,
|
| 155 |
+
"scr_dir2_threshold_5": 0.20343135715230082,
|
| 156 |
+
"scr_dir1_threshold_10": 0.3333337229061823,
|
| 157 |
+
"scr_metric_threshold_10": 0.21323529734021612,
|
| 158 |
+
"scr_dir2_threshold_10": 0.21323529734021612,
|
| 159 |
+
"scr_dir1_threshold_20": -0.9607848178785888,
|
| 160 |
+
"scr_metric_threshold_20": 0.3088236046050426,
|
| 161 |
+
"scr_dir2_threshold_20": 0.3088236046050426,
|
| 162 |
+
"scr_dir1_threshold_50": -3.1764706569834438,
|
| 163 |
+
"scr_metric_threshold_50": 0.2965687158925977,
|
| 164 |
+
"scr_dir2_threshold_50": 0.2965687158925977,
|
| 165 |
+
"scr_dir1_threshold_100": -4.176470656983444,
|
| 166 |
+
"scr_metric_threshold_100": 0.10784319597727135,
|
| 167 |
+
"scr_dir2_threshold_100": 0.10784319597727135,
|
| 168 |
+
"scr_dir1_threshold_500": -3.607843503911701,
|
| 169 |
+
"scr_metric_threshold_500": -0.1127450930263305,
|
| 170 |
+
"scr_dir2_threshold_500": -0.1127450930263305
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"dataset_name": "LabHC/bias_in_bios_class_set1_scr_attorney_teacher_results",
|
| 174 |
+
"scr_dir1_threshold_2": 0.18018000602490186,
|
| 175 |
+
"scr_metric_threshold_2": 0.23563219965771867,
|
| 176 |
+
"scr_dir2_threshold_2": 0.23563219965771867,
|
| 177 |
+
"scr_dir1_threshold_5": -1.9459461055882845,
|
| 178 |
+
"scr_metric_threshold_5": 0.456896598973156,
|
| 179 |
+
"scr_dir2_threshold_5": 0.456896598973156,
|
| 180 |
+
"scr_dir1_threshold_10": -0.5225227692425002,
|
| 181 |
+
"scr_metric_threshold_10": 0.5172413261551995,
|
| 182 |
+
"scr_dir2_threshold_10": 0.5172413261551995,
|
| 183 |
+
"scr_dir1_threshold_20": -1.9459461055882845,
|
| 184 |
+
"scr_metric_threshold_20": 0.5114942745293631,
|
| 185 |
+
"scr_dir2_threshold_20": 0.5114942745293631,
|
| 186 |
+
"scr_dir1_threshold_50": -2.693693882361912,
|
| 187 |
+
"scr_metric_threshold_50": 0.42816099828859333,
|
| 188 |
+
"scr_dir2_threshold_50": 0.42816099828859333,
|
| 189 |
+
"scr_dir1_threshold_100": -1.7477477767736274,
|
| 190 |
+
"scr_metric_threshold_100": 0.4367815757273478,
|
| 191 |
+
"scr_dir2_threshold_100": 0.4367815757273478,
|
| 192 |
+
"scr_dir1_threshold_500": -2.891892211176569,
|
| 193 |
+
"scr_metric_threshold_500": 0.7902298752139258,
|
| 194 |
+
"scr_dir2_threshold_500": 0.7902298752139258
|
| 195 |
+
},
|
| 196 |
+
{
|
| 197 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Books_CDs_and_Vinyl_results",
|
| 198 |
+
"scr_dir1_threshold_2": 0.10827999603648109,
|
| 199 |
+
"scr_metric_threshold_2": 0.37024214744991146,
|
| 200 |
+
"scr_dir2_threshold_2": 0.37024214744991146,
|
| 201 |
+
"scr_dir1_threshold_5": 0.12101906475380056,
|
| 202 |
+
"scr_metric_threshold_5": 0.42214520597217847,
|
| 203 |
+
"scr_dir2_threshold_5": 0.42214520597217847,
|
| 204 |
+
"scr_dir1_threshold_10": 0.2738852318297904,
|
| 205 |
+
"scr_metric_threshold_10": 0.5570934056233778,
|
| 206 |
+
"scr_dir2_threshold_10": 0.5570934056233778,
|
| 207 |
+
"scr_dir1_threshold_20": 0.0764328937142918,
|
| 208 |
+
"scr_metric_threshold_20": 0.5051903471011109,
|
| 209 |
+
"scr_dir2_threshold_20": 0.5051903471011109,
|
| 210 |
+
"scr_dir1_threshold_50": -0.7070063549179335,
|
| 211 |
+
"scr_metric_threshold_50": 0.5363322234633553,
|
| 212 |
+
"scr_dir2_threshold_50": 0.5363322234633553,
|
| 213 |
+
"scr_dir1_threshold_100": -0.35031850519148844,
|
| 214 |
+
"scr_metric_threshold_100": 0.30449823207282245,
|
| 215 |
+
"scr_dir2_threshold_100": 0.30449823207282245,
|
| 216 |
+
"scr_dir1_threshold_500": -0.5923566346990896,
|
| 217 |
+
"scr_metric_threshold_500": 0.6539791973627114,
|
| 218 |
+
"scr_dir2_threshold_500": 0.6539791973627114
|
| 219 |
+
},
|
| 220 |
+
{
|
| 221 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Software_Electronics_results",
|
| 222 |
+
"scr_dir1_threshold_2": 0.09333333863152482,
|
| 223 |
+
"scr_metric_threshold_2": 0.143356666675003,
|
| 224 |
+
"scr_dir2_threshold_2": 0.143356666675003,
|
| 225 |
+
"scr_dir1_threshold_5": 0.33333346578812056,
|
| 226 |
+
"scr_metric_threshold_5": 0.16433562498593246,
|
| 227 |
+
"scr_dir2_threshold_5": 0.16433562498593246,
|
| 228 |
+
"scr_dir1_threshold_10": 0.25333329094780144,
|
| 229 |
+
"scr_metric_threshold_10": 0.21678312496717578,
|
| 230 |
+
"scr_dir2_threshold_10": 0.21678312496717578,
|
| 231 |
+
"scr_dir1_threshold_20": 0.026666724946773045,
|
| 232 |
+
"scr_metric_threshold_20": 0.29720270830155116,
|
| 233 |
+
"scr_dir2_threshold_20": 0.29720270830155116,
|
| 234 |
+
"scr_dir1_threshold_50": 0.19999984105425533,
|
| 235 |
+
"scr_metric_threshold_50": 0.31818187502031975,
|
| 236 |
+
"scr_dir2_threshold_50": 0.31818187502031975,
|
| 237 |
+
"scr_dir1_threshold_100": 0.26666645473900713,
|
| 238 |
+
"scr_metric_threshold_100": 0.17832166666249852,
|
| 239 |
+
"scr_dir2_threshold_100": 0.17832166666249852,
|
| 240 |
+
"scr_dir1_threshold_500": -0.26666685210336877,
|
| 241 |
+
"scr_metric_threshold_500": 0.5034964583171817,
|
| 242 |
+
"scr_dir2_threshold_500": 0.5034964583171817
|
| 243 |
+
},
|
| 244 |
+
{
|
| 245 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Pet_Supplies_Office_Products_results",
|
| 246 |
+
"scr_dir1_threshold_2": 0.14150929073288404,
|
| 247 |
+
"scr_metric_threshold_2": 0.3994082563719217,
|
| 248 |
+
"scr_dir2_threshold_2": 0.3994082563719217,
|
| 249 |
+
"scr_dir1_threshold_5": 0.3301883450433961,
|
| 250 |
+
"scr_metric_threshold_5": 0.5384616605466098,
|
| 251 |
+
"scr_dir2_threshold_5": 0.5384616605466098,
|
| 252 |
+
"scr_dir1_threshold_10": 0.4622637955223419,
|
| 253 |
+
"scr_metric_threshold_10": 0.6301775142711683,
|
| 254 |
+
"scr_dir2_threshold_10": 0.6301775142711683,
|
| 255 |
+
"scr_dir1_threshold_20": 0.5188679616618449,
|
| 256 |
+
"scr_metric_threshold_20": 0.671597628440515,
|
| 257 |
+
"scr_dir2_threshold_20": 0.671597628440515,
|
| 258 |
+
"scr_dir1_threshold_50": 0.5849054057473494,
|
| 259 |
+
"scr_metric_threshold_50": 0.8047337726795233,
|
| 260 |
+
"scr_dir2_threshold_50": 0.8047337726795233,
|
| 261 |
+
"scr_dir1_threshold_100": -0.4905664409000301,
|
| 262 |
+
"scr_metric_threshold_100": 0.8047337726795233,
|
| 263 |
+
"scr_dir2_threshold_100": 0.8047337726795233,
|
| 264 |
+
"scr_dir1_threshold_500": -0.5283018019157831,
|
| 265 |
+
"scr_metric_threshold_500": 0.7544378567792084,
|
| 266 |
+
"scr_dir2_threshold_500": 0.7544378567792084
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"dataset_name": "canrager/amazon_reviews_mcauley_1and5_scr_Industrial_and_Scientific_Toys_and_Games_results",
|
| 270 |
+
"scr_dir1_threshold_2": 0.06770838183970948,
|
| 271 |
+
"scr_metric_threshold_2": 0.14957252765503068,
|
| 272 |
+
"scr_dir2_threshold_2": 0.14957252765503068,
|
| 273 |
+
"scr_dir1_threshold_5": 0.1718749320910734,
|
| 274 |
+
"scr_metric_threshold_5": 0.22222219391991785,
|
| 275 |
+
"scr_dir2_threshold_5": 0.22222219391991785,
|
| 276 |
+
"scr_dir1_threshold_10": 0.2916665890564648,
|
| 277 |
+
"scr_metric_threshold_10": 0.2649572018213552,
|
| 278 |
+
"scr_dir2_threshold_10": 0.2649572018213552,
|
| 279 |
+
"scr_dir1_threshold_20": 0.36979150174498776,
|
| 280 |
+
"scr_metric_threshold_20": 0.31623936413552367,
|
| 281 |
+
"scr_dir2_threshold_20": 0.31623936413552367,
|
| 282 |
+
"scr_dir1_threshold_50": 0.2968748544808716,
|
| 283 |
+
"scr_metric_threshold_50": 0.4444443878398357,
|
| 284 |
+
"scr_dir2_threshold_50": 0.4444443878398357,
|
| 285 |
+
"scr_dir1_threshold_100": 0.4531249902987248,
|
| 286 |
+
"scr_metric_threshold_100": 0.3076922097227925,
|
| 287 |
+
"scr_dir2_threshold_100": 0.3076922097227925,
|
| 288 |
+
"scr_dir1_threshold_500": 0.17708319751548013,
|
| 289 |
+
"scr_metric_threshold_500": 0.2905981556180697,
|
| 290 |
+
"scr_dir2_threshold_500": 0.2905981556180697
|
| 291 |
+
}
|
| 292 |
+
],
|
| 293 |
+
"sae_bench_commit_hash": "Unknown",
|
| 294 |
+
"sae_lens_id": "blocks.18.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.18.hook_resid_post",
|
| 307 |
+
"hook_layer": 18,
|
| 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 |
+
}
|