Add files using upload-large-folder tool
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/149_twt_emotion_happiness_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/150_twt_emotion_sadness_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/151_it_tick_HR Support_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/152_it_tick_Hardware_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/153_it_tick_Administrative rights_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/154_athlete_sport_football_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/155_athlete_sport_basketball_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/156_athlete_sport_baseball_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/157_amazon_5star_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/158_code_C_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/159_code_Python_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/160_code_HTML_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/161_agnews_0_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/162_agnews_1_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/163_agnews_2_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/21_headline_istrump_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/22_headline_isobama_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/23_headline_ischina_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/24_headline_isiran_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/26_headline_isfrontpage_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/36_sciq_tf_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/41_truthqa_tf_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/42_temp_sense_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/44_phys_tf_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/47_reasoning_tf_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/48_cm_correct_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/49_cm_isshort_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/50_deon_isvalid_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/51_just_is_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/52_virtue_is_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/54_cs_tf_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/56_wikidatasex_or_gender_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/57_wikidatais_alive_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/58_wikidatapolitical_party_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/59_wikidata_occupation_isjournalist_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/5_hist_fig_ismale_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/60_wikidata_occupation_isathlete_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/61_wikidata_occupation_isactor_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/62_wikidata_occupation_ispolitician_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/63_wikidata_occupation_issinger_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/64_wikidata_occupation_isresearcher_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/65_high-school_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/66_living-room_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/67_social-security_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/68_credit-card_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/69_blood-pressure_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/6_hist_fig_isamerican_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/70_prime-factors_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/71_social-media_blocks.12.hook_resid_post_l1.json +101 -0
- gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/72_gene-expression_blocks.12.hook_resid_post_l1.json +101 -0
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/149_twt_emotion_happiness_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6075800641892788,
|
| 4 |
+
"test_acc": 0.6081355932203389,
|
| 5 |
+
"test_auc": 0.6351933973885193,
|
| 6 |
+
"val_auc": 0.6308373590982287,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "149_twt_emotion_happiness",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
14530
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6113367780174659,
|
| 18 |
+
"test_acc": 0.6115254237288136,
|
| 19 |
+
"test_auc": 0.6330514831606933,
|
| 20 |
+
"val_auc": 0.6892109500805154,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "149_twt_emotion_happiness",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
14530,
|
| 28 |
+
17630
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.6222218114076764,
|
| 33 |
+
"test_acc": 0.6223728813559322,
|
| 34 |
+
"test_auc": 0.6505388798799793,
|
| 35 |
+
"val_auc": 0.7375201288244766,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "149_twt_emotion_happiness",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
14530,
|
| 43 |
+
17630,
|
| 44 |
+
31015,
|
| 45 |
+
7467,
|
| 46 |
+
27275
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.6408101058555727,
|
| 51 |
+
"test_acc": 0.6413559322033898,
|
| 52 |
+
"test_auc": 0.6763760649818167,
|
| 53 |
+
"val_auc": 0.7648953301127215,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "149_twt_emotion_happiness",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
14530,
|
| 61 |
+
17630,
|
| 62 |
+
31015,
|
| 63 |
+
7467,
|
| 64 |
+
27275,
|
| 65 |
+
20399,
|
| 66 |
+
2509,
|
| 67 |
+
13504,
|
| 68 |
+
16025,
|
| 69 |
+
5460
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.6450860385250187,
|
| 74 |
+
"test_acc": 0.6454237288135594,
|
| 75 |
+
"test_auc": 0.6769717561490405,
|
| 76 |
+
"val_auc": 0.7608695652173914,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "149_twt_emotion_happiness",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
14530,
|
| 84 |
+
17630,
|
| 85 |
+
31015,
|
| 86 |
+
7467,
|
| 87 |
+
27275,
|
| 88 |
+
20399,
|
| 89 |
+
2509,
|
| 90 |
+
13504,
|
| 91 |
+
16025,
|
| 92 |
+
5460,
|
| 93 |
+
14822,
|
| 94 |
+
4648,
|
| 95 |
+
19078,
|
| 96 |
+
29250,
|
| 97 |
+
7125,
|
| 98 |
+
5437
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/150_twt_emotion_sadness_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5695601396124615,
|
| 4 |
+
"test_acc": 0.5701694915254237,
|
| 5 |
+
"test_auc": 0.5770822531834545,
|
| 6 |
+
"val_auc": 0.5915861513687601,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "150_twt_emotion_sadness",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
2509
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5643123312527998,
|
| 18 |
+
"test_acc": 0.5654237288135593,
|
| 19 |
+
"test_auc": 0.5857850437391756,
|
| 20 |
+
"val_auc": 0.6201690821256038,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "150_twt_emotion_sadness",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
2509,
|
| 28 |
+
25315
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.5647277566952745,
|
| 33 |
+
"test_acc": 0.5647457627118644,
|
| 34 |
+
"test_auc": 0.6059043290568591,
|
| 35 |
+
"val_auc": 0.7033011272141707,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "150_twt_emotion_sadness",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
2509,
|
| 43 |
+
25315,
|
| 44 |
+
7467,
|
| 45 |
+
1833,
|
| 46 |
+
22739
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.5721994571697209,
|
| 51 |
+
"test_acc": 0.5722033898305084,
|
| 52 |
+
"test_auc": 0.601387004372079,
|
| 53 |
+
"val_auc": 0.7028985507246377,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "150_twt_emotion_sadness",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
2509,
|
| 61 |
+
25315,
|
| 62 |
+
7467,
|
| 63 |
+
1833,
|
| 64 |
+
22739,
|
| 65 |
+
14873,
|
| 66 |
+
14288,
|
| 67 |
+
17402,
|
| 68 |
+
27151,
|
| 69 |
+
9524
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.5769491525423729,
|
| 74 |
+
"test_acc": 0.5769491525423729,
|
| 75 |
+
"test_auc": 0.6028210757005806,
|
| 76 |
+
"val_auc": 0.716988727858293,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "150_twt_emotion_sadness",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
2509,
|
| 84 |
+
25315,
|
| 85 |
+
7467,
|
| 86 |
+
1833,
|
| 87 |
+
22739,
|
| 88 |
+
14873,
|
| 89 |
+
14288,
|
| 90 |
+
17402,
|
| 91 |
+
27151,
|
| 92 |
+
9524,
|
| 93 |
+
26568,
|
| 94 |
+
9773,
|
| 95 |
+
6399,
|
| 96 |
+
6201,
|
| 97 |
+
25415,
|
| 98 |
+
8950
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/151_it_tick_HR Support_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5754470989050432,
|
| 4 |
+
"test_acc": 0.575682382133995,
|
| 5 |
+
"test_auc": 0.627013447613418,
|
| 6 |
+
"val_auc": 0.5542216886754702,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "151_it_tick_HR Support",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
26510
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6243630110948751,
|
| 18 |
+
"test_acc": 0.6277915632754343,
|
| 19 |
+
"test_auc": 0.6455593320526083,
|
| 20 |
+
"val_auc": 0.6102440976390556,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "151_it_tick_HR Support",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
26510,
|
| 28 |
+
4981
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.604697629182029,
|
| 33 |
+
"test_acc": 0.607940446650124,
|
| 34 |
+
"test_auc": 0.6626767154327373,
|
| 35 |
+
"val_auc": 0.6246498599439776,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "151_it_tick_HR Support",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
26510,
|
| 43 |
+
4981,
|
| 44 |
+
32528,
|
| 45 |
+
18849,
|
| 46 |
+
10549
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.5959946125980281,
|
| 51 |
+
"test_acc": 0.598014888337469,
|
| 52 |
+
"test_auc": 0.648243928870499,
|
| 53 |
+
"val_auc": 0.6566626650660263,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "151_it_tick_HR Support",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
26510,
|
| 61 |
+
4981,
|
| 62 |
+
32528,
|
| 63 |
+
18849,
|
| 64 |
+
10549,
|
| 65 |
+
1765,
|
| 66 |
+
28723,
|
| 67 |
+
12462,
|
| 68 |
+
8489,
|
| 69 |
+
15544
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.5883652482425866,
|
| 74 |
+
"test_acc": 0.5905707196029777,
|
| 75 |
+
"test_auc": 0.6557066154376632,
|
| 76 |
+
"val_auc": 0.6910764305722289,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "151_it_tick_HR Support",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
26510,
|
| 84 |
+
4981,
|
| 85 |
+
32528,
|
| 86 |
+
18849,
|
| 87 |
+
10549,
|
| 88 |
+
1765,
|
| 89 |
+
28723,
|
| 90 |
+
12462,
|
| 91 |
+
8489,
|
| 92 |
+
15544,
|
| 93 |
+
13474,
|
| 94 |
+
14050,
|
| 95 |
+
17050,
|
| 96 |
+
9796,
|
| 97 |
+
2176,
|
| 98 |
+
11678
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/152_it_tick_Hardware_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5804126299120017,
|
| 4 |
+
"test_acc": 0.5806451612903226,
|
| 5 |
+
"test_auc": 0.6025072656519384,
|
| 6 |
+
"val_auc": 0.6938775510204083,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "152_it_tick_Hardware",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
26677
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5724849221757128,
|
| 18 |
+
"test_acc": 0.5732009925558312,
|
| 19 |
+
"test_auc": 0.5984064824392887,
|
| 20 |
+
"val_auc": 0.7248899559823929,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "152_it_tick_Hardware",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
26677,
|
| 28 |
+
17459
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.6050163893966636,
|
| 33 |
+
"test_acc": 0.6054590570719603,
|
| 34 |
+
"test_auc": 0.6499433525442097,
|
| 35 |
+
"val_auc": 0.8151260504201681,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "152_it_tick_Hardware",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
26677,
|
| 43 |
+
17459,
|
| 44 |
+
7862,
|
| 45 |
+
9325,
|
| 46 |
+
31614
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.5872199984123812,
|
| 51 |
+
"test_acc": 0.5880893300248139,
|
| 52 |
+
"test_auc": 0.6498940938870006,
|
| 53 |
+
"val_auc": 0.7739095638255302,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "152_it_tick_Hardware",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
26677,
|
| 61 |
+
17459,
|
| 62 |
+
7862,
|
| 63 |
+
9325,
|
| 64 |
+
31614,
|
| 65 |
+
26960,
|
| 66 |
+
3195,
|
| 67 |
+
28038,
|
| 68 |
+
21724,
|
| 69 |
+
19460
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.6321381888488691,
|
| 74 |
+
"test_acc": 0.6327543424317618,
|
| 75 |
+
"test_auc": 0.6649426136643515,
|
| 76 |
+
"val_auc": 0.7991196478591436,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "152_it_tick_Hardware",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
26677,
|
| 84 |
+
17459,
|
| 85 |
+
7862,
|
| 86 |
+
9325,
|
| 87 |
+
31614,
|
| 88 |
+
26960,
|
| 89 |
+
3195,
|
| 90 |
+
28038,
|
| 91 |
+
21724,
|
| 92 |
+
19460,
|
| 93 |
+
22932,
|
| 94 |
+
12747,
|
| 95 |
+
22687,
|
| 96 |
+
31570,
|
| 97 |
+
23507,
|
| 98 |
+
24193
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/153_it_tick_Administrative rights_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.605531478580386,
|
| 4 |
+
"test_acc": 0.607940446650124,
|
| 5 |
+
"test_auc": 0.6494753953007242,
|
| 6 |
+
"val_auc": 0.7156862745098039,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "153_it_tick_Administrative rights",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
26179
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6234277412829916,
|
| 18 |
+
"test_acc": 0.6253101736972705,
|
| 19 |
+
"test_auc": 0.6808654746071622,
|
| 20 |
+
"val_auc": 0.7631052420968387,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "153_it_tick_Administrative rights",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
26179,
|
| 28 |
+
14082
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.6499300640094215,
|
| 33 |
+
"test_acc": 0.6501240694789082,
|
| 34 |
+
"test_auc": 0.7181296487857741,
|
| 35 |
+
"val_auc": 0.668267306922769,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "153_it_tick_Administrative rights",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
26179,
|
| 43 |
+
14082,
|
| 44 |
+
2223,
|
| 45 |
+
4843,
|
| 46 |
+
28727
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.6872062361131884,
|
| 51 |
+
"test_acc": 0.6873449131513648,
|
| 52 |
+
"test_auc": 0.7441751637850351,
|
| 53 |
+
"val_auc": 0.6466586634653861,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "153_it_tick_Administrative rights",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
26179,
|
| 61 |
+
14082,
|
| 62 |
+
2223,
|
| 63 |
+
4843,
|
| 64 |
+
28727,
|
| 65 |
+
4297,
|
| 66 |
+
30058,
|
| 67 |
+
587,
|
| 68 |
+
29780,
|
| 69 |
+
18374
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.6674937965260546,
|
| 74 |
+
"test_acc": 0.6674937965260546,
|
| 75 |
+
"test_auc": 0.7585094330328557,
|
| 76 |
+
"val_auc": 0.6994797919167667,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "153_it_tick_Administrative rights",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
26179,
|
| 84 |
+
14082,
|
| 85 |
+
2223,
|
| 86 |
+
4843,
|
| 87 |
+
28727,
|
| 88 |
+
4297,
|
| 89 |
+
30058,
|
| 90 |
+
587,
|
| 91 |
+
29780,
|
| 92 |
+
18374,
|
| 93 |
+
21723,
|
| 94 |
+
1192,
|
| 95 |
+
4452,
|
| 96 |
+
17981,
|
| 97 |
+
28735,
|
| 98 |
+
27025
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/154_athlete_sport_football_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7102819956452633,
|
| 4 |
+
"test_acc": 0.7121820615796519,
|
| 5 |
+
"test_auc": 0.7741681122851285,
|
| 6 |
+
"val_auc": 0.7858585858585858,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "154_athlete_sport_football",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
16256
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.738777926617996,
|
| 18 |
+
"test_acc": 0.7389558232931727,
|
| 19 |
+
"test_auc": 0.8139596564923801,
|
| 20 |
+
"val_auc": 0.846060606060606,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "154_athlete_sport_football",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
16256,
|
| 28 |
+
15495
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7520538692122879,
|
| 33 |
+
"test_acc": 0.7523427041499331,
|
| 34 |
+
"test_auc": 0.8295974251265215,
|
| 35 |
+
"val_auc": 0.8937373737373737,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "154_athlete_sport_football",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
16256,
|
| 43 |
+
15495,
|
| 44 |
+
4069,
|
| 45 |
+
8890,
|
| 46 |
+
19080
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.7758743279124481,
|
| 51 |
+
"test_acc": 0.7764390896921017,
|
| 52 |
+
"test_auc": 0.8593783601668793,
|
| 53 |
+
"val_auc": 0.922020202020202,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "154_athlete_sport_football",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
16256,
|
| 61 |
+
15495,
|
| 62 |
+
4069,
|
| 63 |
+
8890,
|
| 64 |
+
19080,
|
| 65 |
+
29735,
|
| 66 |
+
15182,
|
| 67 |
+
22863,
|
| 68 |
+
638,
|
| 69 |
+
8917
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8097016640999816,
|
| 74 |
+
"test_acc": 0.8099062918340026,
|
| 75 |
+
"test_auc": 0.8992559246462416,
|
| 76 |
+
"val_auc": 0.9490909090909091,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "154_athlete_sport_football",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
16256,
|
| 84 |
+
15495,
|
| 85 |
+
4069,
|
| 86 |
+
8890,
|
| 87 |
+
19080,
|
| 88 |
+
29735,
|
| 89 |
+
15182,
|
| 90 |
+
22863,
|
| 91 |
+
638,
|
| 92 |
+
8917,
|
| 93 |
+
13054,
|
| 94 |
+
23830,
|
| 95 |
+
23533,
|
| 96 |
+
12157,
|
| 97 |
+
1177,
|
| 98 |
+
7662
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/155_athlete_sport_basketball_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6492067068935873,
|
| 4 |
+
"test_acc": 0.6532797858099063,
|
| 5 |
+
"test_auc": 0.6854023598227982,
|
| 6 |
+
"val_auc": 0.6844444444444444,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "155_athlete_sport_basketball",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
14503
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6773345383673495,
|
| 18 |
+
"test_acc": 0.677376171352075,
|
| 19 |
+
"test_auc": 0.7388424538716292,
|
| 20 |
+
"val_auc": 0.7729292929292929,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "155_athlete_sport_basketball",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
14503,
|
| 28 |
+
26815
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7456237356528174,
|
| 33 |
+
"test_acc": 0.7456492637215528,
|
| 34 |
+
"test_auc": 0.8182319966738829,
|
| 35 |
+
"val_auc": 0.8432323232323231,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "155_athlete_sport_basketball",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
14503,
|
| 43 |
+
26815,
|
| 44 |
+
28317,
|
| 45 |
+
11462,
|
| 46 |
+
23830
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8111529295578486,
|
| 51 |
+
"test_acc": 0.8112449799196787,
|
| 52 |
+
"test_auc": 0.8624392481828218,
|
| 53 |
+
"val_auc": 0.8767676767676769,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "155_athlete_sport_basketball",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
14503,
|
| 61 |
+
26815,
|
| 62 |
+
28317,
|
| 63 |
+
11462,
|
| 64 |
+
23830,
|
| 65 |
+
31848,
|
| 66 |
+
5574,
|
| 67 |
+
12196,
|
| 68 |
+
563,
|
| 69 |
+
26848
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.82297849034258,
|
| 74 |
+
"test_acc": 0.8232931726907631,
|
| 75 |
+
"test_auc": 0.8917649926165933,
|
| 76 |
+
"val_auc": 0.907070707070707,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "155_athlete_sport_basketball",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
14503,
|
| 84 |
+
26815,
|
| 85 |
+
28317,
|
| 86 |
+
11462,
|
| 87 |
+
23830,
|
| 88 |
+
31848,
|
| 89 |
+
5574,
|
| 90 |
+
12196,
|
| 91 |
+
563,
|
| 92 |
+
26848,
|
| 93 |
+
4093,
|
| 94 |
+
28801,
|
| 95 |
+
31715,
|
| 96 |
+
5690,
|
| 97 |
+
21568,
|
| 98 |
+
21604
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/156_athlete_sport_baseball_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6636124839923084,
|
| 4 |
+
"test_acc": 0.6693440428380187,
|
| 5 |
+
"test_auc": 0.7070507949706814,
|
| 6 |
+
"val_auc": 0.7668686868686869,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "156_athlete_sport_baseball",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
19907
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7165757524285254,
|
| 18 |
+
"test_acc": 0.7175368139223561,
|
| 19 |
+
"test_auc": 0.8061640693323392,
|
| 20 |
+
"val_auc": 0.8626262626262626,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "156_athlete_sport_baseball",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
19907,
|
| 28 |
+
20419
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7723781535840889,
|
| 33 |
+
"test_acc": 0.7724230254350736,
|
| 34 |
+
"test_auc": 0.8803458014938854,
|
| 35 |
+
"val_auc": 0.8917171717171717,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "156_athlete_sport_baseball",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
19907,
|
| 43 |
+
20419,
|
| 44 |
+
32272,
|
| 45 |
+
7565,
|
| 46 |
+
30535
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8458713241049357,
|
| 51 |
+
"test_acc": 0.8460508701472557,
|
| 52 |
+
"test_auc": 0.9177144413700163,
|
| 53 |
+
"val_auc": 0.915151515151515,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "156_athlete_sport_baseball",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
19907,
|
| 61 |
+
20419,
|
| 62 |
+
32272,
|
| 63 |
+
7565,
|
| 64 |
+
30535,
|
| 65 |
+
31886,
|
| 66 |
+
22646,
|
| 67 |
+
31295,
|
| 68 |
+
22246,
|
| 69 |
+
12522
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8647925033467202,
|
| 74 |
+
"test_acc": 0.8647925033467202,
|
| 75 |
+
"test_auc": 0.9412338174363093,
|
| 76 |
+
"val_auc": 0.9341414141414142,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "156_athlete_sport_baseball",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
19907,
|
| 84 |
+
20419,
|
| 85 |
+
32272,
|
| 86 |
+
7565,
|
| 87 |
+
30535,
|
| 88 |
+
31886,
|
| 89 |
+
22646,
|
| 90 |
+
31295,
|
| 91 |
+
22246,
|
| 92 |
+
12522,
|
| 93 |
+
12480,
|
| 94 |
+
31512,
|
| 95 |
+
10599,
|
| 96 |
+
18685,
|
| 97 |
+
1136,
|
| 98 |
+
2860
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/157_amazon_5star_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6601897537201162,
|
| 4 |
+
"test_acc": 0.6656410256410257,
|
| 5 |
+
"test_auc": 0.7061509071935906,
|
| 6 |
+
"val_auc": 0.7860057119543045,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "157_amazon_5star",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
21991
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6952865282441281,
|
| 18 |
+
"test_acc": 0.6953846153846154,
|
| 19 |
+
"test_auc": 0.7650196081731578,
|
| 20 |
+
"val_auc": 0.8535291717666259,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "157_amazon_5star",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
21991,
|
| 28 |
+
2436
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.752562296947444,
|
| 33 |
+
"test_acc": 0.7528205128205128,
|
| 34 |
+
"test_auc": 0.8233581310802168,
|
| 35 |
+
"val_auc": 0.868625050999592,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "157_amazon_5star",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
21991,
|
| 43 |
+
2436,
|
| 44 |
+
27396,
|
| 45 |
+
15985,
|
| 46 |
+
27711
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.7565235089454156,
|
| 51 |
+
"test_acc": 0.7569230769230769,
|
| 52 |
+
"test_auc": 0.8438457265964252,
|
| 53 |
+
"val_auc": 0.8763769889840882,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "157_amazon_5star",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
21991,
|
| 61 |
+
2436,
|
| 62 |
+
27396,
|
| 63 |
+
15985,
|
| 64 |
+
27711,
|
| 65 |
+
18248,
|
| 66 |
+
27233,
|
| 67 |
+
13859,
|
| 68 |
+
15664,
|
| 69 |
+
12103
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.7817785138457813,
|
| 74 |
+
"test_acc": 0.7825641025641026,
|
| 75 |
+
"test_auc": 0.8614720436260814,
|
| 76 |
+
"val_auc": 0.8698490412076703,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "157_amazon_5star",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
21991,
|
| 84 |
+
2436,
|
| 85 |
+
27396,
|
| 86 |
+
15985,
|
| 87 |
+
27711,
|
| 88 |
+
18248,
|
| 89 |
+
27233,
|
| 90 |
+
13859,
|
| 91 |
+
15664,
|
| 92 |
+
12103,
|
| 93 |
+
12663,
|
| 94 |
+
31432,
|
| 95 |
+
2179,
|
| 96 |
+
21946,
|
| 97 |
+
30131,
|
| 98 |
+
14101
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/158_code_C_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7199675524477226,
|
| 4 |
+
"test_acc": 0.7220168067226891,
|
| 5 |
+
"test_auc": 0.764518750316362,
|
| 6 |
+
"val_auc": 0.7996788438378161,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "158_code_C",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
19541
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.786688982814831,
|
| 18 |
+
"test_acc": 0.786890756302521,
|
| 19 |
+
"test_auc": 0.858696290792604,
|
| 20 |
+
"val_auc": 0.9221196306704135,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "158_code_C",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
19541,
|
| 28 |
+
25616
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.8263236162905293,
|
| 33 |
+
"test_acc": 0.8265546218487395,
|
| 34 |
+
"test_auc": 0.9046600556073786,
|
| 35 |
+
"val_auc": 0.9731031714171016,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "158_code_C",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
19541,
|
| 43 |
+
25616,
|
| 44 |
+
23529,
|
| 45 |
+
3974,
|
| 46 |
+
22117
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.843128976106703,
|
| 51 |
+
"test_acc": 0.8433613445378151,
|
| 52 |
+
"test_auc": 0.9233961356848962,
|
| 53 |
+
"val_auc": 0.967482938578884,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "158_code_C",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
19541,
|
| 61 |
+
25616,
|
| 62 |
+
23529,
|
| 63 |
+
3974,
|
| 64 |
+
22117,
|
| 65 |
+
385,
|
| 66 |
+
8067,
|
| 67 |
+
22719,
|
| 68 |
+
28670,
|
| 69 |
+
4049
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8472775155452953,
|
| 74 |
+
"test_acc": 0.8473949579831933,
|
| 75 |
+
"test_auc": 0.9269131758393533,
|
| 76 |
+
"val_auc": 0.9731031714171015,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "158_code_C",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
19541,
|
| 84 |
+
25616,
|
| 85 |
+
23529,
|
| 86 |
+
3974,
|
| 87 |
+
22117,
|
| 88 |
+
385,
|
| 89 |
+
8067,
|
| 90 |
+
22719,
|
| 91 |
+
28670,
|
| 92 |
+
4049,
|
| 93 |
+
4852,
|
| 94 |
+
7539,
|
| 95 |
+
16227,
|
| 96 |
+
19278,
|
| 97 |
+
8256,
|
| 98 |
+
27684
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/159_code_Python_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7901179443480568,
|
| 4 |
+
"test_acc": 0.7912605042016807,
|
| 5 |
+
"test_auc": 0.8456913320461925,
|
| 6 |
+
"val_auc": 0.8588920112404657,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "159_code_Python",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
23006
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8080051417117977,
|
| 18 |
+
"test_acc": 0.8080672268907563,
|
| 19 |
+
"test_auc": 0.8921599200237181,
|
| 20 |
+
"val_auc": 0.9277398635086311,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "159_code_Python",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
23006,
|
| 28 |
+
116
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.8112726943001087,
|
| 33 |
+
"test_acc": 0.8114285714285714,
|
| 34 |
+
"test_auc": 0.9000190269070294,
|
| 35 |
+
"val_auc": 0.9277398635086311,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "159_code_Python",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
23006,
|
| 43 |
+
116,
|
| 44 |
+
25843,
|
| 45 |
+
19593,
|
| 46 |
+
12034
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8409569737822774,
|
| 51 |
+
"test_acc": 0.8410084033613445,
|
| 52 |
+
"test_auc": 0.9209334844639202,
|
| 53 |
+
"val_auc": 0.9566439181051787,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "159_code_Python",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
23006,
|
| 61 |
+
116,
|
| 62 |
+
25843,
|
| 63 |
+
19593,
|
| 64 |
+
12034,
|
| 65 |
+
20787,
|
| 66 |
+
5522,
|
| 67 |
+
13405,
|
| 68 |
+
32705,
|
| 69 |
+
8835
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8396194693007528,
|
| 74 |
+
"test_acc": 0.8396638655462185,
|
| 75 |
+
"test_auc": 0.927649847061631,
|
| 76 |
+
"val_auc": 0.9618627057406663,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "159_code_Python",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
23006,
|
| 84 |
+
116,
|
| 85 |
+
25843,
|
| 86 |
+
19593,
|
| 87 |
+
12034,
|
| 88 |
+
20787,
|
| 89 |
+
5522,
|
| 90 |
+
13405,
|
| 91 |
+
32705,
|
| 92 |
+
8835,
|
| 93 |
+
3039,
|
| 94 |
+
3620,
|
| 95 |
+
7175,
|
| 96 |
+
3963,
|
| 97 |
+
17283,
|
| 98 |
+
786
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/160_code_HTML_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7582792541891996,
|
| 4 |
+
"test_acc": 0.7589915966386555,
|
| 5 |
+
"test_auc": 0.7943717866672453,
|
| 6 |
+
"val_auc": 0.8317944600562023,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "160_code_HTML",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
22683
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8099742764605607,
|
| 18 |
+
"test_acc": 0.8100840336134454,
|
| 19 |
+
"test_auc": 0.8721256263965118,
|
| 20 |
+
"val_auc": 0.8703331995182657,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "160_code_HTML",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
22683,
|
| 28 |
+
191
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.8294757511820815,
|
| 33 |
+
"test_acc": 0.8295798319327731,
|
| 34 |
+
"test_auc": 0.9106535313216333,
|
| 35 |
+
"val_auc": 0.8900040144520274,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "160_code_HTML",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
22683,
|
| 43 |
+
191,
|
| 44 |
+
21984,
|
| 45 |
+
575,
|
| 46 |
+
29104
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8372836780846908,
|
| 51 |
+
"test_acc": 0.8373109243697479,
|
| 52 |
+
"test_auc": 0.9308577564700523,
|
| 53 |
+
"val_auc": 0.8948213568847851,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "160_code_HTML",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
22683,
|
| 61 |
+
191,
|
| 62 |
+
21984,
|
| 63 |
+
575,
|
| 64 |
+
29104,
|
| 65 |
+
1029,
|
| 66 |
+
25764,
|
| 67 |
+
12385,
|
| 68 |
+
20928,
|
| 69 |
+
16050
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8682349963644872,
|
| 74 |
+
"test_acc": 0.8682352941176471,
|
| 75 |
+
"test_auc": 0.9490218994728508,
|
| 76 |
+
"val_auc": 0.9253311922922521,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "160_code_HTML",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
22683,
|
| 84 |
+
191,
|
| 85 |
+
21984,
|
| 86 |
+
575,
|
| 87 |
+
29104,
|
| 88 |
+
1029,
|
| 89 |
+
25764,
|
| 90 |
+
12385,
|
| 91 |
+
20928,
|
| 92 |
+
16050,
|
| 93 |
+
2086,
|
| 94 |
+
882,
|
| 95 |
+
24733,
|
| 96 |
+
14017,
|
| 97 |
+
28725,
|
| 98 |
+
28948
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/161_agnews_0_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9213396319005319,
|
| 4 |
+
"test_acc": 0.921344537815126,
|
| 5 |
+
"test_auc": 0.9757106391594536,
|
| 6 |
+
"val_auc": 0.9817342432757927,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "161_agnews_0",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
1748
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9606642423511109,
|
| 18 |
+
"test_acc": 0.9606722689075631,
|
| 19 |
+
"test_auc": 0.9930587944985575,
|
| 20 |
+
"val_auc": 0.9823364110798877,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "161_agnews_0",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
1748,
|
| 28 |
+
32420
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9737781884598148,
|
| 33 |
+
"test_acc": 0.973781512605042,
|
| 34 |
+
"test_auc": 0.9962425248208487,
|
| 35 |
+
"val_auc": 0.9883580891208351,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "161_agnews_0",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
1748,
|
| 43 |
+
32420,
|
| 44 |
+
23362,
|
| 45 |
+
21505,
|
| 46 |
+
30222
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.9936134410485875,
|
| 51 |
+
"test_acc": 0.9936134453781512,
|
| 52 |
+
"test_auc": 0.9997324482431973,
|
| 53 |
+
"val_auc": 0.9983942191890808,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "161_agnews_0",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
1748,
|
| 61 |
+
32420,
|
| 62 |
+
23362,
|
| 63 |
+
21505,
|
| 64 |
+
30222,
|
| 65 |
+
19168,
|
| 66 |
+
19479,
|
| 67 |
+
3095,
|
| 68 |
+
22929,
|
| 69 |
+
4678
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.992604950107309,
|
| 74 |
+
"test_acc": 0.9926050420168068,
|
| 75 |
+
"test_auc": 0.9997505260646029,
|
| 76 |
+
"val_auc": 0.9987956643918106,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "161_agnews_0",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
1748,
|
| 84 |
+
32420,
|
| 85 |
+
23362,
|
| 86 |
+
21505,
|
| 87 |
+
30222,
|
| 88 |
+
19168,
|
| 89 |
+
19479,
|
| 90 |
+
3095,
|
| 91 |
+
22929,
|
| 92 |
+
4678,
|
| 93 |
+
75,
|
| 94 |
+
14476,
|
| 95 |
+
1719,
|
| 96 |
+
22446,
|
| 97 |
+
29926,
|
| 98 |
+
25590
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/162_agnews_1_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9374365366409826,
|
| 4 |
+
"test_acc": 0.9374789915966386,
|
| 5 |
+
"test_auc": 0.9830448113037002,
|
| 6 |
+
"val_auc": 0.9606583701324769,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "162_agnews_1",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
32420
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.941762476197666,
|
| 18 |
+
"test_acc": 0.9418487394957983,
|
| 19 |
+
"test_auc": 0.9891813277798266,
|
| 20 |
+
"val_auc": 0.9763147330389402,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "162_agnews_1",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
32420,
|
| 28 |
+
19168
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9704132483000207,
|
| 33 |
+
"test_acc": 0.9704201680672269,
|
| 34 |
+
"test_auc": 0.9971518392375498,
|
| 35 |
+
"val_auc": 0.9931754315535929,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "162_agnews_1",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
32420,
|
| 43 |
+
19168,
|
| 44 |
+
28257,
|
| 45 |
+
13386,
|
| 46 |
+
25463
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.9936133328077134,
|
| 51 |
+
"test_acc": 0.9936134453781512,
|
| 52 |
+
"test_auc": 0.9997451027181812,
|
| 53 |
+
"val_auc": 0.9979927739863508,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "162_agnews_1",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
32420,
|
| 61 |
+
19168,
|
| 62 |
+
28257,
|
| 63 |
+
13386,
|
| 64 |
+
25463,
|
| 65 |
+
23362,
|
| 66 |
+
12674,
|
| 67 |
+
13544,
|
| 68 |
+
15978,
|
| 69 |
+
20986
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9949579205278593,
|
| 74 |
+
"test_acc": 0.9949579831932773,
|
| 75 |
+
"test_auc": 0.9998223854046902,
|
| 76 |
+
"val_auc": 0.9983942191890807,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "162_agnews_1",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
32420,
|
| 84 |
+
19168,
|
| 85 |
+
28257,
|
| 86 |
+
13386,
|
| 87 |
+
25463,
|
| 88 |
+
23362,
|
| 89 |
+
12674,
|
| 90 |
+
13544,
|
| 91 |
+
15978,
|
| 92 |
+
20986,
|
| 93 |
+
25590,
|
| 94 |
+
30205,
|
| 95 |
+
24802,
|
| 96 |
+
1748,
|
| 97 |
+
4396,
|
| 98 |
+
22446
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/163_agnews_2_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9293358688376324,
|
| 4 |
+
"test_acc": 0.9294117647058824,
|
| 5 |
+
"test_auc": 0.9778786218915185,
|
| 6 |
+
"val_auc": 0.9614612605379367,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "163_agnews_2",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
32420
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9234155700887897,
|
| 18 |
+
"test_acc": 0.9236974789915966,
|
| 19 |
+
"test_auc": 0.9806876893651794,
|
| 20 |
+
"val_auc": 0.9702930549979927,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "163_agnews_2",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
32420,
|
| 28 |
+
1748
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9488653297535855,
|
| 33 |
+
"test_acc": 0.9489075630252101,
|
| 34 |
+
"test_auc": 0.9912205060343767,
|
| 35 |
+
"val_auc": 0.9971898835808912,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "163_agnews_2",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
32420,
|
| 43 |
+
1748,
|
| 44 |
+
75,
|
| 45 |
+
19168,
|
| 46 |
+
23362
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.9875627301097328,
|
| 51 |
+
"test_acc": 0.987563025210084,
|
| 52 |
+
"test_auc": 0.9995756231425039,
|
| 53 |
+
"val_auc": 0.9983942191890808,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "163_agnews_2",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
32420,
|
| 61 |
+
1748,
|
| 62 |
+
75,
|
| 63 |
+
19168,
|
| 64 |
+
23362,
|
| 65 |
+
13544,
|
| 66 |
+
24802,
|
| 67 |
+
13386,
|
| 68 |
+
26281,
|
| 69 |
+
16896
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9939494895935167,
|
| 74 |
+
"test_acc": 0.9939495798319328,
|
| 75 |
+
"test_auc": 0.9999227173134911,
|
| 76 |
+
"val_auc": 1.0,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "163_agnews_2",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
32420,
|
| 84 |
+
1748,
|
| 85 |
+
75,
|
| 86 |
+
19168,
|
| 87 |
+
23362,
|
| 88 |
+
13544,
|
| 89 |
+
24802,
|
| 90 |
+
13386,
|
| 91 |
+
26281,
|
| 92 |
+
16896,
|
| 93 |
+
29926,
|
| 94 |
+
29921,
|
| 95 |
+
19479,
|
| 96 |
+
20986,
|
| 97 |
+
30222,
|
| 98 |
+
15978
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/21_headline_istrump_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7327137290721708,
|
| 4 |
+
"test_acc": 0.7330817610062893,
|
| 5 |
+
"test_auc": 0.7986383322582703,
|
| 6 |
+
"val_auc": 0.7959183673469388,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "21_headline_istrump",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
7777
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7712616286635317,
|
| 18 |
+
"test_acc": 0.7715723270440251,
|
| 19 |
+
"test_auc": 0.8444310806965598,
|
| 20 |
+
"val_auc": 0.81312525010004,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "21_headline_istrump",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
7777,
|
| 28 |
+
18810
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.8206500113869484,
|
| 33 |
+
"test_acc": 0.8208805031446541,
|
| 34 |
+
"test_auc": 0.8941540536626907,
|
| 35 |
+
"val_auc": 0.876750700280112,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "21_headline_istrump",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
7777,
|
| 43 |
+
18810,
|
| 44 |
+
17023,
|
| 45 |
+
8090,
|
| 46 |
+
6110
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8444207785205643,
|
| 51 |
+
"test_acc": 0.8445283018867924,
|
| 52 |
+
"test_auc": 0.9198601270430838,
|
| 53 |
+
"val_auc": 0.9267707082833132,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "21_headline_istrump",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
7777,
|
| 61 |
+
18810,
|
| 62 |
+
17023,
|
| 63 |
+
8090,
|
| 64 |
+
6110,
|
| 65 |
+
24298,
|
| 66 |
+
22531,
|
| 67 |
+
5760,
|
| 68 |
+
28678,
|
| 69 |
+
14087
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8592563222570098,
|
| 74 |
+
"test_acc": 0.8593710691823899,
|
| 75 |
+
"test_auc": 0.9304617336631771,
|
| 76 |
+
"val_auc": 0.9395758303321329,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "21_headline_istrump",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
7777,
|
| 84 |
+
18810,
|
| 85 |
+
17023,
|
| 86 |
+
8090,
|
| 87 |
+
6110,
|
| 88 |
+
24298,
|
| 89 |
+
22531,
|
| 90 |
+
5760,
|
| 91 |
+
28678,
|
| 92 |
+
14087,
|
| 93 |
+
3714,
|
| 94 |
+
9041,
|
| 95 |
+
13597,
|
| 96 |
+
15272,
|
| 97 |
+
18882,
|
| 98 |
+
9958
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/22_headline_isobama_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7297786783688569,
|
| 4 |
+
"test_acc": 0.7305863708399366,
|
| 5 |
+
"test_auc": 0.7819469559975889,
|
| 6 |
+
"val_auc": 0.8483393357342937,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "22_headline_isobama",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
19983
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7458405350927172,
|
| 18 |
+
"test_acc": 0.7464342313787639,
|
| 19 |
+
"test_auc": 0.8503516174402249,
|
| 20 |
+
"val_auc": 0.8509403761504601,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "22_headline_isobama",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
19983,
|
| 28 |
+
24827
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.8080181119362585,
|
| 33 |
+
"test_acc": 0.8082408874801902,
|
| 34 |
+
"test_auc": 0.8881052843078159,
|
| 35 |
+
"val_auc": 0.8699479791916767,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "22_headline_isobama",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
19983,
|
| 43 |
+
24827,
|
| 44 |
+
5910,
|
| 45 |
+
4093,
|
| 46 |
+
5760
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8186593534661042,
|
| 51 |
+
"test_acc": 0.8193343898573693,
|
| 52 |
+
"test_auc": 0.9061482820976492,
|
| 53 |
+
"val_auc": 0.9179671868747499,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "22_headline_isobama",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
19983,
|
| 61 |
+
24827,
|
| 62 |
+
5910,
|
| 63 |
+
4093,
|
| 64 |
+
5760,
|
| 65 |
+
662,
|
| 66 |
+
4047,
|
| 67 |
+
20972,
|
| 68 |
+
21088,
|
| 69 |
+
30573
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8443273277442892,
|
| 74 |
+
"test_acc": 0.8446909667194928,
|
| 75 |
+
"test_auc": 0.9233172593932087,
|
| 76 |
+
"val_auc": 0.9367747098839536,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "22_headline_isobama",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
19983,
|
| 84 |
+
24827,
|
| 85 |
+
5910,
|
| 86 |
+
4093,
|
| 87 |
+
5760,
|
| 88 |
+
662,
|
| 89 |
+
4047,
|
| 90 |
+
20972,
|
| 91 |
+
21088,
|
| 92 |
+
30573,
|
| 93 |
+
23635,
|
| 94 |
+
3107,
|
| 95 |
+
2192,
|
| 96 |
+
27034,
|
| 97 |
+
21704,
|
| 98 |
+
10426
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/23_headline_ischina_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.639002678250414,
|
| 4 |
+
"test_acc": 0.6398559423769508,
|
| 5 |
+
"test_auc": 0.67043096292197,
|
| 6 |
+
"val_auc": 0.7020933977455717,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "23_headline_ischina",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
13472
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6737803210995694,
|
| 18 |
+
"test_acc": 0.6746698679471789,
|
| 19 |
+
"test_auc": 0.7291378435712969,
|
| 20 |
+
"val_auc": 0.7177938808373592,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "23_headline_ischina",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
13472,
|
| 28 |
+
9512
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7178578792883493,
|
| 33 |
+
"test_acc": 0.7178871548619448,
|
| 34 |
+
"test_auc": 0.7950562626821619,
|
| 35 |
+
"val_auc": 0.783011272141707,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "23_headline_ischina",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
13472,
|
| 43 |
+
9512,
|
| 44 |
+
28764,
|
| 45 |
+
31643,
|
| 46 |
+
21733
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.7442490661948856,
|
| 51 |
+
"test_acc": 0.7442977190876351,
|
| 52 |
+
"test_auc": 0.824669110865154,
|
| 53 |
+
"val_auc": 0.8152173913043478,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "23_headline_ischina",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
13472,
|
| 61 |
+
9512,
|
| 62 |
+
28764,
|
| 63 |
+
31643,
|
| 64 |
+
21733,
|
| 65 |
+
13595,
|
| 66 |
+
2435,
|
| 67 |
+
19825,
|
| 68 |
+
28334,
|
| 69 |
+
26354
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.7646814646805151,
|
| 74 |
+
"test_acc": 0.7647058823529411,
|
| 75 |
+
"test_auc": 0.847981230400295,
|
| 76 |
+
"val_auc": 0.8301127214170693,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "23_headline_ischina",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
13472,
|
| 84 |
+
9512,
|
| 85 |
+
28764,
|
| 86 |
+
31643,
|
| 87 |
+
21733,
|
| 88 |
+
13595,
|
| 89 |
+
2435,
|
| 90 |
+
19825,
|
| 91 |
+
28334,
|
| 92 |
+
26354,
|
| 93 |
+
6671,
|
| 94 |
+
22798,
|
| 95 |
+
29937,
|
| 96 |
+
23387,
|
| 97 |
+
7325,
|
| 98 |
+
23904
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/24_headline_isiran_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6166958787564848,
|
| 4 |
+
"test_acc": 0.6170798898071626,
|
| 5 |
+
"test_auc": 0.6691305931637423,
|
| 6 |
+
"val_auc": 0.755502200880352,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "24_headline_isiran",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
8923
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6519891866653652,
|
| 18 |
+
"test_acc": 0.6528925619834711,
|
| 19 |
+
"test_auc": 0.7105215226762187,
|
| 20 |
+
"val_auc": 0.7519007603041217,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "24_headline_isiran",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
8923,
|
| 28 |
+
31900
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7373710444891464,
|
| 33 |
+
"test_acc": 0.7382920110192838,
|
| 34 |
+
"test_auc": 0.8027745734928056,
|
| 35 |
+
"val_auc": 0.8255302120848339,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "24_headline_isiran",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
8923,
|
| 43 |
+
31900,
|
| 44 |
+
188,
|
| 45 |
+
10273,
|
| 46 |
+
11435
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.770688465336227,
|
| 51 |
+
"test_acc": 0.7713498622589532,
|
| 52 |
+
"test_auc": 0.8384736810151175,
|
| 53 |
+
"val_auc": 0.7899159663865546,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "24_headline_isiran",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
8923,
|
| 61 |
+
31900,
|
| 62 |
+
188,
|
| 63 |
+
10273,
|
| 64 |
+
11435,
|
| 65 |
+
2615,
|
| 66 |
+
570,
|
| 67 |
+
18041,
|
| 68 |
+
30957,
|
| 69 |
+
3835
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.7905191383992516,
|
| 74 |
+
"test_acc": 0.790633608815427,
|
| 75 |
+
"test_auc": 0.8471252504401675,
|
| 76 |
+
"val_auc": 0.7971188475390155,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "24_headline_isiran",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
8923,
|
| 84 |
+
31900,
|
| 85 |
+
188,
|
| 86 |
+
10273,
|
| 87 |
+
11435,
|
| 88 |
+
2615,
|
| 89 |
+
570,
|
| 90 |
+
18041,
|
| 91 |
+
30957,
|
| 92 |
+
3835,
|
| 93 |
+
15043,
|
| 94 |
+
15926,
|
| 95 |
+
27321,
|
| 96 |
+
16173,
|
| 97 |
+
29696,
|
| 98 |
+
1460
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/26_headline_isfrontpage_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6173725635151596,
|
| 4 |
+
"test_acc": 0.6188679245283019,
|
| 5 |
+
"test_auc": 0.647536325147665,
|
| 6 |
+
"val_auc": 0.6346538615446178,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "26_headline_isfrontpage",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
22138
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.621222153514086,
|
| 18 |
+
"test_acc": 0.6221383647798742,
|
| 19 |
+
"test_auc": 0.6608459260849444,
|
| 20 |
+
"val_auc": 0.6550620248099239,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "26_headline_isfrontpage",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
22138,
|
| 28 |
+
17430
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.6550288979860017,
|
| 33 |
+
"test_acc": 0.6553459119496855,
|
| 34 |
+
"test_auc": 0.703127927099588,
|
| 35 |
+
"val_auc": 0.7202881152460985,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "26_headline_isfrontpage",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
22138,
|
| 43 |
+
17430,
|
| 44 |
+
30980,
|
| 45 |
+
4281,
|
| 46 |
+
30088
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.6568742953686459,
|
| 51 |
+
"test_acc": 0.6571069182389937,
|
| 52 |
+
"test_auc": 0.7220016120882314,
|
| 53 |
+
"val_auc": 0.7062825130052021,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "26_headline_isfrontpage",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
22138,
|
| 61 |
+
17430,
|
| 62 |
+
30980,
|
| 63 |
+
4281,
|
| 64 |
+
30088,
|
| 65 |
+
8280,
|
| 66 |
+
7168,
|
| 67 |
+
21568,
|
| 68 |
+
15801,
|
| 69 |
+
31465
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.6713935082151044,
|
| 74 |
+
"test_acc": 0.6714465408805032,
|
| 75 |
+
"test_auc": 0.7382462869820837,
|
| 76 |
+
"val_auc": 0.7759103641456583,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "26_headline_isfrontpage",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
22138,
|
| 84 |
+
17430,
|
| 85 |
+
30980,
|
| 86 |
+
4281,
|
| 87 |
+
30088,
|
| 88 |
+
8280,
|
| 89 |
+
7168,
|
| 90 |
+
21568,
|
| 91 |
+
15801,
|
| 92 |
+
31465,
|
| 93 |
+
29657,
|
| 94 |
+
31293,
|
| 95 |
+
16542,
|
| 96 |
+
4116,
|
| 97 |
+
23084,
|
| 98 |
+
22537
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/36_sciq_tf_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6728105126120991,
|
| 4 |
+
"test_acc": 0.6737106918238994,
|
| 5 |
+
"test_auc": 0.7319721043928393,
|
| 6 |
+
"val_auc": 0.7114845938375349,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "36_sciq_tf",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
29312
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7363018704191072,
|
| 18 |
+
"test_acc": 0.7366037735849057,
|
| 19 |
+
"test_auc": 0.801022288739989,
|
| 20 |
+
"val_auc": 0.7879151660664265,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "36_sciq_tf",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
29312,
|
| 28 |
+
24512
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7859024376414564,
|
| 33 |
+
"test_acc": 0.7859119496855346,
|
| 34 |
+
"test_auc": 0.8566261686880214,
|
| 35 |
+
"val_auc": 0.8227290916366546,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "36_sciq_tf",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
29312,
|
| 43 |
+
24512,
|
| 44 |
+
2014,
|
| 45 |
+
27875,
|
| 46 |
+
7548
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.7994186320928678,
|
| 51 |
+
"test_acc": 0.799496855345912,
|
| 52 |
+
"test_auc": 0.8805566159918747,
|
| 53 |
+
"val_auc": 0.8867547018807523,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "36_sciq_tf",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
29312,
|
| 61 |
+
24512,
|
| 62 |
+
2014,
|
| 63 |
+
27875,
|
| 64 |
+
7548,
|
| 65 |
+
25212,
|
| 66 |
+
28591,
|
| 67 |
+
18596,
|
| 68 |
+
31010,
|
| 69 |
+
11791
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8111774713658202,
|
| 74 |
+
"test_acc": 0.8113207547169812,
|
| 75 |
+
"test_auc": 0.8996247236817989,
|
| 76 |
+
"val_auc": 0.9047619047619048,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "36_sciq_tf",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
29312,
|
| 84 |
+
24512,
|
| 85 |
+
2014,
|
| 86 |
+
27875,
|
| 87 |
+
7548,
|
| 88 |
+
25212,
|
| 89 |
+
28591,
|
| 90 |
+
18596,
|
| 91 |
+
31010,
|
| 92 |
+
11791,
|
| 93 |
+
14706,
|
| 94 |
+
12097,
|
| 95 |
+
32594,
|
| 96 |
+
32755,
|
| 97 |
+
28229,
|
| 98 |
+
8717
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/41_truthqa_tf_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5586401448470414,
|
| 4 |
+
"test_acc": 0.5656565656565656,
|
| 5 |
+
"test_auc": 0.6228571428571429,
|
| 6 |
+
"val_auc": 0.6463267763950221,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "41_truthqa_tf",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
29959
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5552832244008714,
|
| 18 |
+
"test_acc": 0.5555555555555556,
|
| 19 |
+
"test_auc": 0.6053061224489795,
|
| 20 |
+
"val_auc": 0.7224006423123244,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "41_truthqa_tf",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
29959,
|
| 28 |
+
26366
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.6565656565656566,
|
| 33 |
+
"test_acc": 0.6565656565656566,
|
| 34 |
+
"test_auc": 0.7020408163265306,
|
| 35 |
+
"val_auc": 0.8309915696507428,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "41_truthqa_tf",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
29959,
|
| 43 |
+
26366,
|
| 44 |
+
31591,
|
| 45 |
+
25985,
|
| 46 |
+
29262
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.6550865504926211,
|
| 51 |
+
"test_acc": 0.6565656565656566,
|
| 52 |
+
"test_auc": 0.7024489795918367,
|
| 53 |
+
"val_auc": 0.8896025692492975,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "41_truthqa_tf",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
29959,
|
| 61 |
+
26366,
|
| 62 |
+
31591,
|
| 63 |
+
25985,
|
| 64 |
+
29262,
|
| 65 |
+
24895,
|
| 66 |
+
3199,
|
| 67 |
+
20741,
|
| 68 |
+
31184,
|
| 69 |
+
8250
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.6647495806374312,
|
| 74 |
+
"test_acc": 0.6666666666666666,
|
| 75 |
+
"test_auc": 0.7338775510204082,
|
| 76 |
+
"val_auc": 0.8851866720192694,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "41_truthqa_tf",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
29959,
|
| 84 |
+
26366,
|
| 85 |
+
31591,
|
| 86 |
+
25985,
|
| 87 |
+
29262,
|
| 88 |
+
24895,
|
| 89 |
+
3199,
|
| 90 |
+
20741,
|
| 91 |
+
31184,
|
| 92 |
+
8250,
|
| 93 |
+
12799,
|
| 94 |
+
30887,
|
| 95 |
+
23695,
|
| 96 |
+
25652,
|
| 97 |
+
15018,
|
| 98 |
+
28965
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/42_temp_sense_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6787860249233824,
|
| 4 |
+
"test_acc": 0.6788990825688074,
|
| 5 |
+
"test_auc": 0.7317195325542571,
|
| 6 |
+
"val_auc": 0.7612721417069244,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "42_temp_sense",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
19232
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7255836733655877,
|
| 18 |
+
"test_acc": 0.725604670558799,
|
| 19 |
+
"test_auc": 0.7760948803561493,
|
| 20 |
+
"val_auc": 0.8146135265700483,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "42_temp_sense",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
19232,
|
| 28 |
+
8361
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7372708332039172,
|
| 33 |
+
"test_acc": 0.737281067556297,
|
| 34 |
+
"test_auc": 0.819794101279911,
|
| 35 |
+
"val_auc": 0.9045893719806763,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "42_temp_sense",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
19232,
|
| 43 |
+
8361,
|
| 44 |
+
27647,
|
| 45 |
+
4807,
|
| 46 |
+
27712
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.7556143917138742,
|
| 51 |
+
"test_acc": 0.755629691409508,
|
| 52 |
+
"test_auc": 0.838906510851419,
|
| 53 |
+
"val_auc": 0.8969404186795491,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "42_temp_sense",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
19232,
|
| 61 |
+
8361,
|
| 62 |
+
27647,
|
| 63 |
+
4807,
|
| 64 |
+
27712,
|
| 65 |
+
29325,
|
| 66 |
+
26813,
|
| 67 |
+
8846,
|
| 68 |
+
15281,
|
| 69 |
+
15230
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.7721530318748848,
|
| 74 |
+
"test_acc": 0.7723102585487907,
|
| 75 |
+
"test_auc": 0.8509766277128548,
|
| 76 |
+
"val_auc": 0.932367149758454,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "42_temp_sense",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
19232,
|
| 84 |
+
8361,
|
| 85 |
+
27647,
|
| 86 |
+
4807,
|
| 87 |
+
27712,
|
| 88 |
+
29325,
|
| 89 |
+
26813,
|
| 90 |
+
8846,
|
| 91 |
+
15281,
|
| 92 |
+
15230,
|
| 93 |
+
27660,
|
| 94 |
+
21432,
|
| 95 |
+
5997,
|
| 96 |
+
7326,
|
| 97 |
+
18177,
|
| 98 |
+
13920
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/44_phys_tf_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.3334731051125547,
|
| 4 |
+
"test_acc": 0.500125786163522,
|
| 5 |
+
"test_auc": 0.5,
|
| 6 |
+
"val_auc": 0.5,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "44_phys_tf",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
22397
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5512730566037737,
|
| 18 |
+
"test_acc": 0.5527044025157233,
|
| 19 |
+
"test_auc": 0.5691291685695452,
|
| 20 |
+
"val_auc": 0.5214085634253701,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "44_phys_tf",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
22397,
|
| 28 |
+
28588
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.5964867959141555,
|
| 33 |
+
"test_acc": 0.5967295597484277,
|
| 34 |
+
"test_auc": 0.6346660739474592,
|
| 35 |
+
"val_auc": 0.6234493797519007,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "44_phys_tf",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
22397,
|
| 43 |
+
28588,
|
| 44 |
+
15653,
|
| 45 |
+
15351,
|
| 46 |
+
5535
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.6117815900397015,
|
| 51 |
+
"test_acc": 0.6118238993710692,
|
| 52 |
+
"test_auc": 0.6550500790348532,
|
| 53 |
+
"val_auc": 0.6202480992396958,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "44_phys_tf",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
22397,
|
| 61 |
+
28588,
|
| 62 |
+
15653,
|
| 63 |
+
15351,
|
| 64 |
+
5535,
|
| 65 |
+
16611,
|
| 66 |
+
30087,
|
| 67 |
+
9388,
|
| 68 |
+
22952,
|
| 69 |
+
17748
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.6266474799064764,
|
| 74 |
+
"test_acc": 0.6266666666666667,
|
| 75 |
+
"test_auc": 0.6658357290193097,
|
| 76 |
+
"val_auc": 0.6390556222488996,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "44_phys_tf",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
22397,
|
| 84 |
+
28588,
|
| 85 |
+
15653,
|
| 86 |
+
15351,
|
| 87 |
+
5535,
|
| 88 |
+
16611,
|
| 89 |
+
30087,
|
| 90 |
+
9388,
|
| 91 |
+
22952,
|
| 92 |
+
17748,
|
| 93 |
+
26339,
|
| 94 |
+
12049,
|
| 95 |
+
12364,
|
| 96 |
+
649,
|
| 97 |
+
30430,
|
| 98 |
+
22382
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/47_reasoning_tf_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5115572776524306,
|
| 4 |
+
"test_acc": 0.5152603231597845,
|
| 5 |
+
"test_auc": 0.5145059163968714,
|
| 6 |
+
"val_auc": 0.5467683661180249,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "47_reasoning_tf",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
1285
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5208909313836534,
|
| 18 |
+
"test_acc": 0.5218432076600837,
|
| 19 |
+
"test_auc": 0.5251640260149557,
|
| 20 |
+
"val_auc": 0.5724608590927338,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "47_reasoning_tf",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
1285,
|
| 28 |
+
27899
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.5252993066636028,
|
| 33 |
+
"test_acc": 0.5266307600239377,
|
| 34 |
+
"test_auc": 0.536958141134,
|
| 35 |
+
"val_auc": 0.5503813729425934,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "47_reasoning_tf",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
1285,
|
| 43 |
+
27899,
|
| 44 |
+
26653,
|
| 45 |
+
16357,
|
| 46 |
+
24507
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.5379174592409501,
|
| 51 |
+
"test_acc": 0.538001196888091,
|
| 52 |
+
"test_auc": 0.5511188150015758,
|
| 53 |
+
"val_auc": 0.6138097149739061,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "47_reasoning_tf",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
1285,
|
| 61 |
+
27899,
|
| 62 |
+
26653,
|
| 63 |
+
16357,
|
| 64 |
+
24507,
|
| 65 |
+
16461,
|
| 66 |
+
18077,
|
| 67 |
+
19993,
|
| 68 |
+
4140,
|
| 69 |
+
3865
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.5439601586874245,
|
| 74 |
+
"test_acc": 0.5439856373429084,
|
| 75 |
+
"test_auc": 0.5577701057215713,
|
| 76 |
+
"val_auc": 0.6407065435568046,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "47_reasoning_tf",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
1285,
|
| 84 |
+
27899,
|
| 85 |
+
26653,
|
| 86 |
+
16357,
|
| 87 |
+
24507,
|
| 88 |
+
16461,
|
| 89 |
+
18077,
|
| 90 |
+
19993,
|
| 91 |
+
4140,
|
| 92 |
+
3865,
|
| 93 |
+
26487,
|
| 94 |
+
9365,
|
| 95 |
+
22562,
|
| 96 |
+
2784,
|
| 97 |
+
12866,
|
| 98 |
+
26466
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/48_cm_correct_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5746440076331454,
|
| 4 |
+
"test_acc": 0.580377358490566,
|
| 5 |
+
"test_auc": 0.615825678783319,
|
| 6 |
+
"val_auc": 0.6578631452581032,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "48_cm_correct",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
5472
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5863574339622193,
|
| 18 |
+
"test_acc": 0.5891823899371069,
|
| 19 |
+
"test_auc": 0.6357547651282633,
|
| 20 |
+
"val_auc": 0.6778711484593838,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "48_cm_correct",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
5472,
|
| 28 |
+
13985
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.6180949348689766,
|
| 33 |
+
"test_acc": 0.6181132075471698,
|
| 34 |
+
"test_auc": 0.6764553096130886,
|
| 35 |
+
"val_auc": 0.7446978791516606,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "48_cm_correct",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
5472,
|
| 43 |
+
13985,
|
| 44 |
+
11157,
|
| 45 |
+
24380,
|
| 46 |
+
14345
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.6389911974404083,
|
| 51 |
+
"test_acc": 0.6389937106918239,
|
| 52 |
+
"test_auc": 0.706341217916457,
|
| 53 |
+
"val_auc": 0.7727090836334534,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "48_cm_correct",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
5472,
|
| 61 |
+
13985,
|
| 62 |
+
11157,
|
| 63 |
+
24380,
|
| 64 |
+
14345,
|
| 65 |
+
30805,
|
| 66 |
+
7125,
|
| 67 |
+
5127,
|
| 68 |
+
2586,
|
| 69 |
+
31578
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.6356344254646761,
|
| 74 |
+
"test_acc": 0.6359748427672955,
|
| 75 |
+
"test_auc": 0.7070328361715335,
|
| 76 |
+
"val_auc": 0.8051220488195278,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "48_cm_correct",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
5472,
|
| 84 |
+
13985,
|
| 85 |
+
11157,
|
| 86 |
+
24380,
|
| 87 |
+
14345,
|
| 88 |
+
30805,
|
| 89 |
+
7125,
|
| 90 |
+
5127,
|
| 91 |
+
2586,
|
| 92 |
+
31578,
|
| 93 |
+
8196,
|
| 94 |
+
30710,
|
| 95 |
+
29072,
|
| 96 |
+
7918,
|
| 97 |
+
28425,
|
| 98 |
+
20556
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/49_cm_isshort_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9443912818385808,
|
| 4 |
+
"test_acc": 0.9444025157232704,
|
| 5 |
+
"test_auc": 0.9841003747700091,
|
| 6 |
+
"val_auc": 0.9935974389755902,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "49_cm_isshort",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
7199
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9738356400974403,
|
| 18 |
+
"test_acc": 0.9738364779874213,
|
| 19 |
+
"test_auc": 0.9957414340091886,
|
| 20 |
+
"val_auc": 1.0,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "49_cm_isshort",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
7199,
|
| 28 |
+
12876
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9874213119999193,
|
| 33 |
+
"test_acc": 0.9874213836477987,
|
| 34 |
+
"test_auc": 0.9989499148894373,
|
| 35 |
+
"val_auc": 0.9995998399359743,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "49_cm_isshort",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
7199,
|
| 43 |
+
12876,
|
| 44 |
+
2192,
|
| 45 |
+
12282,
|
| 46 |
+
25406
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.9904401971199387,
|
| 51 |
+
"test_acc": 0.9904402515723271,
|
| 52 |
+
"test_auc": 0.9996230528617097,
|
| 53 |
+
"val_auc": 0.9987995198079233,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "49_cm_isshort",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
7199,
|
| 61 |
+
12876,
|
| 62 |
+
2192,
|
| 63 |
+
12282,
|
| 64 |
+
25406,
|
| 65 |
+
27704,
|
| 66 |
+
9580,
|
| 67 |
+
18920,
|
| 68 |
+
20872,
|
| 69 |
+
6476
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9939621724408219,
|
| 74 |
+
"test_acc": 0.9939622641509434,
|
| 75 |
+
"test_auc": 0.9997800086882646,
|
| 76 |
+
"val_auc": 0.9991996798719487,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "49_cm_isshort",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
7199,
|
| 84 |
+
12876,
|
| 85 |
+
2192,
|
| 86 |
+
12282,
|
| 87 |
+
25406,
|
| 88 |
+
27704,
|
| 89 |
+
9580,
|
| 90 |
+
18920,
|
| 91 |
+
20872,
|
| 92 |
+
6476,
|
| 93 |
+
10719,
|
| 94 |
+
17985,
|
| 95 |
+
7542,
|
| 96 |
+
21692,
|
| 97 |
+
25335,
|
| 98 |
+
26307
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/50_deon_isvalid_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6084261515945389,
|
| 4 |
+
"test_acc": 0.609056603773585,
|
| 5 |
+
"test_auc": 0.646568768423323,
|
| 6 |
+
"val_auc": 0.6634653861544617,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "50_deon_isvalid",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
13509
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.632085701342505,
|
| 18 |
+
"test_acc": 0.6332075471698113,
|
| 19 |
+
"test_auc": 0.6856759581140592,
|
| 20 |
+
"val_auc": 0.7264905962384953,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "50_deon_isvalid",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
13509,
|
| 28 |
+
15465
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.6709587778644674,
|
| 33 |
+
"test_acc": 0.6711949685534592,
|
| 34 |
+
"test_auc": 0.7390738492353214,
|
| 35 |
+
"val_auc": 0.760704281712685,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "50_deon_isvalid",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
13509,
|
| 43 |
+
15465,
|
| 44 |
+
30350,
|
| 45 |
+
9496,
|
| 46 |
+
26403
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.7084243144292207,
|
| 51 |
+
"test_acc": 0.7084276729559749,
|
| 52 |
+
"test_auc": 0.77924821196935,
|
| 53 |
+
"val_auc": 0.8147258903561425,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "50_deon_isvalid",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
13509,
|
| 61 |
+
15465,
|
| 62 |
+
30350,
|
| 63 |
+
9496,
|
| 64 |
+
26403,
|
| 65 |
+
11937,
|
| 66 |
+
8793,
|
| 67 |
+
13099,
|
| 68 |
+
8638,
|
| 69 |
+
12896
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.7174823099132951,
|
| 74 |
+
"test_acc": 0.7174842767295597,
|
| 75 |
+
"test_auc": 0.7927086930237692,
|
| 76 |
+
"val_auc": 0.8523409363745499,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "50_deon_isvalid",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
13509,
|
| 84 |
+
15465,
|
| 85 |
+
30350,
|
| 86 |
+
9496,
|
| 87 |
+
26403,
|
| 88 |
+
11937,
|
| 89 |
+
8793,
|
| 90 |
+
13099,
|
| 91 |
+
8638,
|
| 92 |
+
12896,
|
| 93 |
+
31979,
|
| 94 |
+
30184,
|
| 95 |
+
12485,
|
| 96 |
+
26943,
|
| 97 |
+
14448,
|
| 98 |
+
24919
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/51_just_is_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.642203091269332,
|
| 4 |
+
"test_acc": 0.6425157232704403,
|
| 5 |
+
"test_auc": 0.6783362479861554,
|
| 6 |
+
"val_auc": 0.7058823529411765,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "51_just_is",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
27345
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6477986975576877,
|
| 18 |
+
"test_acc": 0.6477987421383647,
|
| 19 |
+
"test_auc": 0.7017900052554886,
|
| 20 |
+
"val_auc": 0.7611044417767107,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "51_just_is",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
27345,
|
| 28 |
+
13193
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.6684272532601063,
|
| 33 |
+
"test_acc": 0.6684276729559748,
|
| 34 |
+
"test_auc": 0.7514012104838391,
|
| 35 |
+
"val_auc": 0.7803121248499401,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "51_just_is",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
27345,
|
| 43 |
+
13193,
|
| 44 |
+
12357,
|
| 45 |
+
8497,
|
| 46 |
+
30472
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.6948392522887357,
|
| 51 |
+
"test_acc": 0.6948427672955975,
|
| 52 |
+
"test_auc": 0.7781406607739036,
|
| 53 |
+
"val_auc": 0.7975190076030412,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "51_just_is",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
27345,
|
| 61 |
+
13193,
|
| 62 |
+
12357,
|
| 63 |
+
8497,
|
| 64 |
+
30472,
|
| 65 |
+
19033,
|
| 66 |
+
436,
|
| 67 |
+
28059,
|
| 68 |
+
10270,
|
| 69 |
+
9065
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.7053825887851659,
|
| 74 |
+
"test_acc": 0.7054088050314465,
|
| 75 |
+
"test_auc": 0.7901855015346229,
|
| 76 |
+
"val_auc": 0.8319327731092436,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "51_just_is",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
27345,
|
| 84 |
+
13193,
|
| 85 |
+
12357,
|
| 86 |
+
8497,
|
| 87 |
+
30472,
|
| 88 |
+
19033,
|
| 89 |
+
436,
|
| 90 |
+
28059,
|
| 91 |
+
10270,
|
| 92 |
+
9065,
|
| 93 |
+
31939,
|
| 94 |
+
22913,
|
| 95 |
+
13503,
|
| 96 |
+
7582,
|
| 97 |
+
16551,
|
| 98 |
+
24127
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/52_virtue_is_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7002967213903285,
|
| 4 |
+
"test_acc": 0.7009198423127464,
|
| 5 |
+
"test_auc": 0.7639553574438058,
|
| 6 |
+
"val_auc": 0.7203999999999999,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "52_virtue_is",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
2123
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.739340590265011,
|
| 18 |
+
"test_acc": 0.7400788436268069,
|
| 19 |
+
"test_auc": 0.8039644084026937,
|
| 20 |
+
"val_auc": 0.7869999999999999,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "52_virtue_is",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
2123,
|
| 28 |
+
23112
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7709112046640785,
|
| 33 |
+
"test_acc": 0.7710906701708279,
|
| 34 |
+
"test_auc": 0.8521314787156038,
|
| 35 |
+
"val_auc": 0.8412000000000001,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "52_virtue_is",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
2123,
|
| 43 |
+
23112,
|
| 44 |
+
9968,
|
| 45 |
+
32138,
|
| 46 |
+
21873
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8262506067351926,
|
| 51 |
+
"test_acc": 0.8262812089356111,
|
| 52 |
+
"test_auc": 0.9048383950738029,
|
| 53 |
+
"val_auc": 0.9104000000000001,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "52_virtue_is",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
2123,
|
| 61 |
+
23112,
|
| 62 |
+
9968,
|
| 63 |
+
32138,
|
| 64 |
+
21873,
|
| 65 |
+
7981,
|
| 66 |
+
18333,
|
| 67 |
+
23155,
|
| 68 |
+
603,
|
| 69 |
+
19878
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8383667469489923,
|
| 74 |
+
"test_acc": 0.8383705650459922,
|
| 75 |
+
"test_auc": 0.9194072340258588,
|
| 76 |
+
"val_auc": 0.9276000000000001,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "52_virtue_is",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
2123,
|
| 84 |
+
23112,
|
| 85 |
+
9968,
|
| 86 |
+
32138,
|
| 87 |
+
21873,
|
| 88 |
+
7981,
|
| 89 |
+
18333,
|
| 90 |
+
23155,
|
| 91 |
+
603,
|
| 92 |
+
19878,
|
| 93 |
+
29205,
|
| 94 |
+
9151,
|
| 95 |
+
29748,
|
| 96 |
+
5634,
|
| 97 |
+
13854,
|
| 98 |
+
4481
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/54_cs_tf_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5349594626591904,
|
| 4 |
+
"test_acc": 0.5373584905660377,
|
| 5 |
+
"test_auc": 0.5510086183938052,
|
| 6 |
+
"val_auc": 0.6262505002000801,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "54_cs_tf",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
18645
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5519980432599746,
|
| 18 |
+
"test_acc": 0.5567295597484276,
|
| 19 |
+
"test_auc": 0.5719705753393031,
|
| 20 |
+
"val_auc": 0.7006802721088435,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "54_cs_tf",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
18645,
|
| 28 |
+
25246
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.5319427535043354,
|
| 33 |
+
"test_acc": 0.5388679245283019,
|
| 34 |
+
"test_auc": 0.5720624704442052,
|
| 35 |
+
"val_auc": 0.6678671468587435,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "54_cs_tf",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
18645,
|
| 43 |
+
25246,
|
| 44 |
+
1217,
|
| 45 |
+
10948,
|
| 46 |
+
25544
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.5361955579012182,
|
| 51 |
+
"test_acc": 0.5408805031446541,
|
| 52 |
+
"test_auc": 0.5711842772791758,
|
| 53 |
+
"val_auc": 0.6598639455782314,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "54_cs_tf",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
18645,
|
| 61 |
+
25246,
|
| 62 |
+
1217,
|
| 63 |
+
10948,
|
| 64 |
+
25544,
|
| 65 |
+
10298,
|
| 66 |
+
19411,
|
| 67 |
+
24108,
|
| 68 |
+
20742,
|
| 69 |
+
13631
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.544265866202867,
|
| 74 |
+
"test_acc": 0.5461635220125786,
|
| 75 |
+
"test_auc": 0.5669034336871759,
|
| 76 |
+
"val_auc": 0.6634653861544618,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "54_cs_tf",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
18645,
|
| 84 |
+
25246,
|
| 85 |
+
1217,
|
| 86 |
+
10948,
|
| 87 |
+
25544,
|
| 88 |
+
10298,
|
| 89 |
+
19411,
|
| 90 |
+
24108,
|
| 91 |
+
20742,
|
| 92 |
+
13631,
|
| 93 |
+
22713,
|
| 94 |
+
26028,
|
| 95 |
+
406,
|
| 96 |
+
12686,
|
| 97 |
+
29676,
|
| 98 |
+
14559
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/56_wikidatasex_or_gender_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7100637907547411,
|
| 4 |
+
"test_acc": 0.7101886792452831,
|
| 5 |
+
"test_auc": 0.7797717355972776,
|
| 6 |
+
"val_auc": 0.74749899959984,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "56_wikidatasex_or_gender",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
8372
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7644902937896919,
|
| 18 |
+
"test_acc": 0.7645283018867924,
|
| 19 |
+
"test_auc": 0.8337025170651488,
|
| 20 |
+
"val_auc": 0.8095238095238095,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "56_wikidatasex_or_gender",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
8372,
|
| 28 |
+
17875
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.8269044357468237,
|
| 33 |
+
"test_acc": 0.8269182389937106,
|
| 34 |
+
"test_auc": 0.9087825898521475,
|
| 35 |
+
"val_auc": 0.9027611044417767,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "56_wikidatasex_or_gender",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
8372,
|
| 43 |
+
17875,
|
| 44 |
+
32056,
|
| 45 |
+
24191,
|
| 46 |
+
28208
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8515723082564541,
|
| 51 |
+
"test_acc": 0.8515723270440252,
|
| 52 |
+
"test_auc": 0.9302171863592222,
|
| 53 |
+
"val_auc": 0.9291716686674669,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "56_wikidatasex_or_gender",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
8372,
|
| 61 |
+
17875,
|
| 62 |
+
32056,
|
| 63 |
+
24191,
|
| 64 |
+
28208,
|
| 65 |
+
6677,
|
| 66 |
+
32649,
|
| 67 |
+
29475,
|
| 68 |
+
372,
|
| 69 |
+
11751
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8706893196297222,
|
| 74 |
+
"test_acc": 0.870691823899371,
|
| 75 |
+
"test_auc": 0.9424379695384182,
|
| 76 |
+
"val_auc": 0.9535814325730292,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "56_wikidatasex_or_gender",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
8372,
|
| 84 |
+
17875,
|
| 85 |
+
32056,
|
| 86 |
+
24191,
|
| 87 |
+
28208,
|
| 88 |
+
6677,
|
| 89 |
+
32649,
|
| 90 |
+
29475,
|
| 91 |
+
372,
|
| 92 |
+
11751,
|
| 93 |
+
9484,
|
| 94 |
+
10320,
|
| 95 |
+
11746,
|
| 96 |
+
446,
|
| 97 |
+
24588,
|
| 98 |
+
28087
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/57_wikidatais_alive_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7567055391593854,
|
| 4 |
+
"test_acc": 0.7567295597484277,
|
| 5 |
+
"test_auc": 0.8447709406919625,
|
| 6 |
+
"val_auc": 0.8835534213685474,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "57_wikidatais_alive",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
12638
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8191074073214824,
|
| 18 |
+
"test_acc": 0.8191194968553459,
|
| 19 |
+
"test_auc": 0.8962909312948653,
|
| 20 |
+
"val_auc": 0.9295718287314927,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "57_wikidatais_alive",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
12638,
|
| 28 |
+
1960
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.868653642158207,
|
| 33 |
+
"test_acc": 0.8686792452830189,
|
| 34 |
+
"test_auc": 0.9467393186496939,
|
| 35 |
+
"val_auc": 0.9799919967987195,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "57_wikidatais_alive",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
12638,
|
| 43 |
+
1960,
|
| 44 |
+
5209,
|
| 45 |
+
25466,
|
| 46 |
+
22771
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8995873715452428,
|
| 51 |
+
"test_acc": 0.8996226415094339,
|
| 52 |
+
"test_auc": 0.9646234224673658,
|
| 53 |
+
"val_auc": 0.9795918367346939,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "57_wikidatais_alive",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
12638,
|
| 61 |
+
1960,
|
| 62 |
+
5209,
|
| 63 |
+
25466,
|
| 64 |
+
22771,
|
| 65 |
+
9484,
|
| 66 |
+
2984,
|
| 67 |
+
2600,
|
| 68 |
+
3792,
|
| 69 |
+
15069
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9094210295130997,
|
| 74 |
+
"test_acc": 0.909433962264151,
|
| 75 |
+
"test_auc": 0.9704492683326936,
|
| 76 |
+
"val_auc": 0.9863945578231292,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "57_wikidatais_alive",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
12638,
|
| 84 |
+
1960,
|
| 85 |
+
5209,
|
| 86 |
+
25466,
|
| 87 |
+
22771,
|
| 88 |
+
9484,
|
| 89 |
+
2984,
|
| 90 |
+
2600,
|
| 91 |
+
3792,
|
| 92 |
+
15069,
|
| 93 |
+
17548,
|
| 94 |
+
2723,
|
| 95 |
+
4618,
|
| 96 |
+
6347,
|
| 97 |
+
15985,
|
| 98 |
+
27308
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/58_wikidatapolitical_party_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6281755205792489,
|
| 4 |
+
"test_acc": 0.6334177215189873,
|
| 5 |
+
"test_auc": 0.6865086201592361,
|
| 6 |
+
"val_auc": 0.6514605842336935,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "58_wikidatapolitical_party",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
22372
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6706260002909937,
|
| 18 |
+
"test_acc": 0.6708860759493671,
|
| 19 |
+
"test_auc": 0.72362422012478,
|
| 20 |
+
"val_auc": 0.7715086034413765,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "58_wikidatapolitical_party",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
22372,
|
| 28 |
+
22185
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7184264398300215,
|
| 33 |
+
"test_acc": 0.7184810126582278,
|
| 34 |
+
"test_auc": 0.7882215768553955,
|
| 35 |
+
"val_auc": 0.847939175670268,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "58_wikidatapolitical_party",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
22372,
|
| 43 |
+
22185,
|
| 44 |
+
24575,
|
| 45 |
+
8929,
|
| 46 |
+
17284
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.7964217721518987,
|
| 51 |
+
"test_acc": 0.7964556962025316,
|
| 52 |
+
"test_auc": 0.8687010078387458,
|
| 53 |
+
"val_auc": 0.9327731092436976,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "58_wikidatapolitical_party",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
22372,
|
| 61 |
+
22185,
|
| 62 |
+
24575,
|
| 63 |
+
8929,
|
| 64 |
+
17284,
|
| 65 |
+
28616,
|
| 66 |
+
26627,
|
| 67 |
+
727,
|
| 68 |
+
10603,
|
| 69 |
+
14125
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8202526115761262,
|
| 74 |
+
"test_acc": 0.8202531645569621,
|
| 75 |
+
"test_auc": 0.9026073776913643,
|
| 76 |
+
"val_auc": 0.9575830332132853,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "58_wikidatapolitical_party",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
22372,
|
| 84 |
+
22185,
|
| 85 |
+
24575,
|
| 86 |
+
8929,
|
| 87 |
+
17284,
|
| 88 |
+
28616,
|
| 89 |
+
26627,
|
| 90 |
+
727,
|
| 91 |
+
10603,
|
| 92 |
+
14125,
|
| 93 |
+
11956,
|
| 94 |
+
5192,
|
| 95 |
+
18018,
|
| 96 |
+
31241,
|
| 97 |
+
15986,
|
| 98 |
+
22046
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/59_wikidata_occupation_isjournalist_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.69007645111382,
|
| 4 |
+
"test_acc": 0.6902564102564103,
|
| 5 |
+
"test_auc": 0.7619963308311172,
|
| 6 |
+
"val_auc": 0.6509587923296614,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "59_wikidata_occupation_isjournalist",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
26907
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7321651320901117,
|
| 18 |
+
"test_acc": 0.7323076923076923,
|
| 19 |
+
"test_auc": 0.8038888140842226,
|
| 20 |
+
"val_auc": 0.7376580987352102,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "59_wikidata_occupation_isjournalist",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
26907,
|
| 28 |
+
31065
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7825613577851706,
|
| 33 |
+
"test_acc": 0.7825641025641026,
|
| 34 |
+
"test_auc": 0.857003399872084,
|
| 35 |
+
"val_auc": 0.8241534067727458,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "59_wikidata_occupation_isjournalist",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
26907,
|
| 43 |
+
31065,
|
| 44 |
+
5384,
|
| 45 |
+
19340,
|
| 46 |
+
4100
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8102564102564103,
|
| 51 |
+
"test_acc": 0.8102564102564103,
|
| 52 |
+
"test_auc": 0.8790689063183762,
|
| 53 |
+
"val_auc": 0.8718890248878008,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "59_wikidata_occupation_isjournalist",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
26907,
|
| 61 |
+
31065,
|
| 62 |
+
5384,
|
| 63 |
+
19340,
|
| 64 |
+
4100,
|
| 65 |
+
26738,
|
| 66 |
+
25282,
|
| 67 |
+
1994,
|
| 68 |
+
21934,
|
| 69 |
+
22976
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8225349804669749,
|
| 74 |
+
"test_acc": 0.8225641025641026,
|
| 75 |
+
"test_auc": 0.8889992257717035,
|
| 76 |
+
"val_auc": 0.8669930640554876,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "59_wikidata_occupation_isjournalist",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
26907,
|
| 84 |
+
31065,
|
| 85 |
+
5384,
|
| 86 |
+
19340,
|
| 87 |
+
4100,
|
| 88 |
+
26738,
|
| 89 |
+
25282,
|
| 90 |
+
1994,
|
| 91 |
+
21934,
|
| 92 |
+
22976,
|
| 93 |
+
10152,
|
| 94 |
+
9884,
|
| 95 |
+
18132,
|
| 96 |
+
20484,
|
| 97 |
+
3112,
|
| 98 |
+
31496
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/5_hist_fig_ismale_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6783553591865811,
|
| 4 |
+
"test_acc": 0.6784905660377358,
|
| 5 |
+
"test_auc": 0.7424823222171478,
|
| 6 |
+
"val_auc": 0.7242897158863546,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "5_hist_fig_ismale",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
14903
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7154239342931071,
|
| 18 |
+
"test_acc": 0.7154716981132075,
|
| 19 |
+
"test_auc": 0.7886729030448417,
|
| 20 |
+
"val_auc": 0.7783113245298119,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "5_hist_fig_ismale",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
14903,
|
| 28 |
+
25892
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7448298872103124,
|
| 33 |
+
"test_acc": 0.7449056603773585,
|
| 34 |
+
"test_auc": 0.8288571388066699,
|
| 35 |
+
"val_auc": 0.8299319727891157,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "5_hist_fig_ismale",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
14903,
|
| 43 |
+
25892,
|
| 44 |
+
14220,
|
| 45 |
+
23967,
|
| 46 |
+
8877
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.7546503664048292,
|
| 51 |
+
"test_acc": 0.7547169811320755,
|
| 52 |
+
"test_auc": 0.8410042540092089,
|
| 53 |
+
"val_auc": 0.8519407763105241,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "5_hist_fig_ismale",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
14903,
|
| 61 |
+
25892,
|
| 62 |
+
14220,
|
| 63 |
+
23967,
|
| 64 |
+
8877,
|
| 65 |
+
24869,
|
| 66 |
+
19415,
|
| 67 |
+
22481,
|
| 68 |
+
25020,
|
| 69 |
+
26167
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.7791064953846866,
|
| 74 |
+
"test_acc": 0.7791194968553459,
|
| 75 |
+
"test_auc": 0.8628859214674053,
|
| 76 |
+
"val_auc": 0.8467386954781914,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "5_hist_fig_ismale",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
14903,
|
| 84 |
+
25892,
|
| 85 |
+
14220,
|
| 86 |
+
23967,
|
| 87 |
+
8877,
|
| 88 |
+
24869,
|
| 89 |
+
19415,
|
| 90 |
+
22481,
|
| 91 |
+
25020,
|
| 92 |
+
26167,
|
| 93 |
+
9615,
|
| 94 |
+
1661,
|
| 95 |
+
18130,
|
| 96 |
+
4506,
|
| 97 |
+
11011,
|
| 98 |
+
31563
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/60_wikidata_occupation_isathlete_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6888851795486899,
|
| 4 |
+
"test_acc": 0.6892307692307692,
|
| 5 |
+
"test_auc": 0.7562064496583297,
|
| 6 |
+
"val_auc": 0.8376172990616075,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "60_wikidata_occupation_isathlete",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
10949
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7126656095354029,
|
| 18 |
+
"test_acc": 0.7138461538461538,
|
| 19 |
+
"test_auc": 0.7788063419396103,
|
| 20 |
+
"val_auc": 0.8551611587107303,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "60_wikidata_occupation_isathlete",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
10949,
|
| 28 |
+
8496
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7762054487098496,
|
| 33 |
+
"test_acc": 0.7764102564102564,
|
| 34 |
+
"test_auc": 0.8479525027771233,
|
| 35 |
+
"val_auc": 0.9106487148102815,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "60_wikidata_occupation_isathlete",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
10949,
|
| 43 |
+
8496,
|
| 44 |
+
21895,
|
| 45 |
+
28768,
|
| 46 |
+
28602
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8102524182136306,
|
| 51 |
+
"test_acc": 0.8102564102564103,
|
| 52 |
+
"test_auc": 0.8805374154239742,
|
| 53 |
+
"val_auc": 0.9298245614035088,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "60_wikidata_occupation_isathlete",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
10949,
|
| 61 |
+
8496,
|
| 62 |
+
21895,
|
| 63 |
+
28768,
|
| 64 |
+
28602,
|
| 65 |
+
24020,
|
| 66 |
+
21255,
|
| 67 |
+
26449,
|
| 68 |
+
32534,
|
| 69 |
+
12943
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8255397241369186,
|
| 74 |
+
"test_acc": 0.8256410256410256,
|
| 75 |
+
"test_auc": 0.8923990978557244,
|
| 76 |
+
"val_auc": 0.9298245614035089,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "60_wikidata_occupation_isathlete",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
10949,
|
| 84 |
+
8496,
|
| 85 |
+
21895,
|
| 86 |
+
28768,
|
| 87 |
+
28602,
|
| 88 |
+
24020,
|
| 89 |
+
21255,
|
| 90 |
+
26449,
|
| 91 |
+
32534,
|
| 92 |
+
12943,
|
| 93 |
+
1628,
|
| 94 |
+
27076,
|
| 95 |
+
25266,
|
| 96 |
+
12770,
|
| 97 |
+
18900,
|
| 98 |
+
15573
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/61_wikidata_occupation_isactor_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6727472416118107,
|
| 4 |
+
"test_acc": 0.6738461538461539,
|
| 5 |
+
"test_auc": 0.7172636079038611,
|
| 6 |
+
"val_auc": 0.7103223174214606,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "61_wikidata_occupation_isactor",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
26723
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.698558700155,
|
| 18 |
+
"test_acc": 0.7005128205128205,
|
| 19 |
+
"test_auc": 0.7810112094792473,
|
| 20 |
+
"val_auc": 0.7723378212974296,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "61_wikidata_occupation_isactor",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
26723,
|
| 28 |
+
14135
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7804421475330291,
|
| 33 |
+
"test_acc": 0.7805128205128206,
|
| 34 |
+
"test_auc": 0.8511335712121721,
|
| 35 |
+
"val_auc": 0.835577315381477,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "61_wikidata_occupation_isactor",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
26723,
|
| 43 |
+
14135,
|
| 44 |
+
31625,
|
| 45 |
+
952,
|
| 46 |
+
20126
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8437916815035966,
|
| 51 |
+
"test_acc": 0.8441025641025641,
|
| 52 |
+
"test_auc": 0.9041345474130676,
|
| 53 |
+
"val_auc": 0.8780089759281925,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "61_wikidata_occupation_isactor",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
26723,
|
| 61 |
+
14135,
|
| 62 |
+
31625,
|
| 63 |
+
952,
|
| 64 |
+
20126,
|
| 65 |
+
14332,
|
| 66 |
+
21635,
|
| 67 |
+
6437,
|
| 68 |
+
19351,
|
| 69 |
+
8081
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8376255636682123,
|
| 74 |
+
"test_acc": 0.837948717948718,
|
| 75 |
+
"test_auc": 0.9033771501666275,
|
| 76 |
+
"val_auc": 0.8935128518971848,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "61_wikidata_occupation_isactor",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
26723,
|
| 84 |
+
14135,
|
| 85 |
+
31625,
|
| 86 |
+
952,
|
| 87 |
+
20126,
|
| 88 |
+
14332,
|
| 89 |
+
21635,
|
| 90 |
+
6437,
|
| 91 |
+
19351,
|
| 92 |
+
8081,
|
| 93 |
+
31432,
|
| 94 |
+
2600,
|
| 95 |
+
17221,
|
| 96 |
+
18848,
|
| 97 |
+
9712,
|
| 98 |
+
3348
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/62_wikidata_occupation_ispolitician_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7062845355587118,
|
| 4 |
+
"test_acc": 0.7076923076923077,
|
| 5 |
+
"test_auc": 0.7654277611337394,
|
| 6 |
+
"val_auc": 0.6852305181558548,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "62_wikidata_occupation_ispolitician",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
2984
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7445025044675654,
|
| 18 |
+
"test_acc": 0.7446153846153846,
|
| 19 |
+
"test_auc": 0.8232950146430135,
|
| 20 |
+
"val_auc": 0.8325173398612811,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "62_wikidata_occupation_ispolitician",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
2984,
|
| 28 |
+
25967
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.8287078580505725,
|
| 33 |
+
"test_acc": 0.8287179487179487,
|
| 34 |
+
"test_auc": 0.898437237014845,
|
| 35 |
+
"val_auc": 0.860873113015096,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "62_wikidata_occupation_ispolitician",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
2984,
|
| 43 |
+
25967,
|
| 44 |
+
8639,
|
| 45 |
+
27249,
|
| 46 |
+
5689
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8635854385665503,
|
| 51 |
+
"test_acc": 0.8635897435897436,
|
| 52 |
+
"test_auc": 0.9303741542397415,
|
| 53 |
+
"val_auc": 0.9184006527947777,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "62_wikidata_occupation_ispolitician",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
2984,
|
| 61 |
+
25967,
|
| 62 |
+
8639,
|
| 63 |
+
27249,
|
| 64 |
+
5689,
|
| 65 |
+
9719,
|
| 66 |
+
12075,
|
| 67 |
+
19811,
|
| 68 |
+
2816,
|
| 69 |
+
15985
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8738342090302322,
|
| 74 |
+
"test_acc": 0.8738461538461538,
|
| 75 |
+
"test_auc": 0.9405190695795604,
|
| 76 |
+
"val_auc": 0.9192166462668299,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "62_wikidata_occupation_ispolitician",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
2984,
|
| 84 |
+
25967,
|
| 85 |
+
8639,
|
| 86 |
+
27249,
|
| 87 |
+
5689,
|
| 88 |
+
9719,
|
| 89 |
+
12075,
|
| 90 |
+
19811,
|
| 91 |
+
2816,
|
| 92 |
+
15985,
|
| 93 |
+
10924,
|
| 94 |
+
9328,
|
| 95 |
+
17548,
|
| 96 |
+
1960,
|
| 97 |
+
25360,
|
| 98 |
+
23504
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/63_wikidata_occupation_issinger_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6763647711315054,
|
| 4 |
+
"test_acc": 0.676923076923077,
|
| 5 |
+
"test_auc": 0.741752785538762,
|
| 6 |
+
"val_auc": 0.8002855977152183,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "63_wikidata_occupation_issinger",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
914
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7507141625194841,
|
| 18 |
+
"test_acc": 0.7507692307692307,
|
| 19 |
+
"test_auc": 0.7998788164405696,
|
| 20 |
+
"val_auc": 0.8392492860057119,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "63_wikidata_occupation_issinger",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
914,
|
| 28 |
+
25374
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.7815370826827127,
|
| 33 |
+
"test_acc": 0.7815384615384615,
|
| 34 |
+
"test_auc": 0.8565615848116606,
|
| 35 |
+
"val_auc": 0.8331293349653203,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "63_wikidata_occupation_issinger",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
914,
|
| 43 |
+
25374,
|
| 44 |
+
31659,
|
| 45 |
+
7810,
|
| 46 |
+
9642
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8071754303624701,
|
| 51 |
+
"test_acc": 0.8071794871794872,
|
| 52 |
+
"test_auc": 0.8860032988857846,
|
| 53 |
+
"val_auc": 0.8882088943288454,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "63_wikidata_occupation_issinger",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
914,
|
| 61 |
+
25374,
|
| 62 |
+
31659,
|
| 63 |
+
7810,
|
| 64 |
+
9642,
|
| 65 |
+
664,
|
| 66 |
+
12641,
|
| 67 |
+
8922,
|
| 68 |
+
27100,
|
| 69 |
+
21727
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8133321551499237,
|
| 74 |
+
"test_acc": 0.8133333333333334,
|
| 75 |
+
"test_auc": 0.8926726024169389,
|
| 76 |
+
"val_auc": 0.8812729498164015,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "63_wikidata_occupation_issinger",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
914,
|
| 84 |
+
25374,
|
| 85 |
+
31659,
|
| 86 |
+
7810,
|
| 87 |
+
9642,
|
| 88 |
+
664,
|
| 89 |
+
12641,
|
| 90 |
+
8922,
|
| 91 |
+
27100,
|
| 92 |
+
21727,
|
| 93 |
+
21437,
|
| 94 |
+
9363,
|
| 95 |
+
17593,
|
| 96 |
+
30573,
|
| 97 |
+
27240,
|
| 98 |
+
22950
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/64_wikidata_occupation_isresearcher_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7817233053947233,
|
| 4 |
+
"test_acc": 0.7818181818181819,
|
| 5 |
+
"test_auc": 0.8543348157909186,
|
| 6 |
+
"val_auc": 0.868080808080808,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "64_wikidata_occupation_isresearcher",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
22124
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8405793465745001,
|
| 18 |
+
"test_acc": 0.8406417112299466,
|
| 19 |
+
"test_auc": 0.896996650743974,
|
| 20 |
+
"val_auc": 0.9002020202020202,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "64_wikidata_occupation_isresearcher",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
22124,
|
| 28 |
+
14502
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.8554989428060759,
|
| 33 |
+
"test_acc": 0.8556149732620321,
|
| 34 |
+
"test_auc": 0.9212490162704297,
|
| 35 |
+
"val_auc": 0.9373737373737374,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "64_wikidata_occupation_isresearcher",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
22124,
|
| 43 |
+
14502,
|
| 44 |
+
23590,
|
| 45 |
+
30474,
|
| 46 |
+
29233
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8715119613202587,
|
| 51 |
+
"test_acc": 0.8716577540106952,
|
| 52 |
+
"test_auc": 0.9325939347352625,
|
| 53 |
+
"val_auc": 0.945858585858586,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "64_wikidata_occupation_isresearcher",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
22124,
|
| 61 |
+
14502,
|
| 62 |
+
23590,
|
| 63 |
+
30474,
|
| 64 |
+
29233,
|
| 65 |
+
25575,
|
| 66 |
+
10867,
|
| 67 |
+
25410,
|
| 68 |
+
10200,
|
| 69 |
+
16795
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8780104002501677,
|
| 74 |
+
"test_acc": 0.8780748663101604,
|
| 75 |
+
"test_auc": 0.9391734841413641,
|
| 76 |
+
"val_auc": 0.9454545454545455,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "64_wikidata_occupation_isresearcher",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
22124,
|
| 84 |
+
14502,
|
| 85 |
+
23590,
|
| 86 |
+
30474,
|
| 87 |
+
29233,
|
| 88 |
+
25575,
|
| 89 |
+
10867,
|
| 90 |
+
25410,
|
| 91 |
+
10200,
|
| 92 |
+
16795,
|
| 93 |
+
3604,
|
| 94 |
+
31458,
|
| 95 |
+
11078,
|
| 96 |
+
6946,
|
| 97 |
+
23058,
|
| 98 |
+
28627
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/65_high-school_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8333979742408861,
|
| 4 |
+
"test_acc": 0.8335632183908046,
|
| 5 |
+
"test_auc": 0.9031189120082256,
|
| 6 |
+
"val_auc": 0.882711038961039,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "65_high-school",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
22854
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8776771016428161,
|
| 18 |
+
"test_acc": 0.8777011494252873,
|
| 19 |
+
"test_auc": 0.9416871854537583,
|
| 20 |
+
"val_auc": 0.9543425324675325,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "65_high-school",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
22854,
|
| 28 |
+
24800
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9158601124866035,
|
| 33 |
+
"test_acc": 0.9158620689655173,
|
| 34 |
+
"test_auc": 0.9704136283078089,
|
| 35 |
+
"val_auc": 0.9744318181818182,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "65_high-school",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
22854,
|
| 43 |
+
24800,
|
| 44 |
+
18373,
|
| 45 |
+
17819,
|
| 46 |
+
3827
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.949423876119365,
|
| 51 |
+
"test_acc": 0.9494252873563218,
|
| 52 |
+
"test_auc": 0.9848290627198442,
|
| 53 |
+
"val_auc": 1.0,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "65_high-school",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
22854,
|
| 61 |
+
24800,
|
| 62 |
+
18373,
|
| 63 |
+
17819,
|
| 64 |
+
3827,
|
| 65 |
+
28039,
|
| 66 |
+
10409,
|
| 67 |
+
25524,
|
| 68 |
+
27732,
|
| 69 |
+
14903
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.95678060422363,
|
| 74 |
+
"test_acc": 0.9567816091954023,
|
| 75 |
+
"test_auc": 0.9908857689810054,
|
| 76 |
+
"val_auc": 1.0,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "65_high-school",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
22854,
|
| 84 |
+
24800,
|
| 85 |
+
18373,
|
| 86 |
+
17819,
|
| 87 |
+
3827,
|
| 88 |
+
28039,
|
| 89 |
+
10409,
|
| 90 |
+
25524,
|
| 91 |
+
27732,
|
| 92 |
+
14903,
|
| 93 |
+
16260,
|
| 94 |
+
13625,
|
| 95 |
+
23839,
|
| 96 |
+
17256,
|
| 97 |
+
29120,
|
| 98 |
+
3672
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/66_living-room_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8431427765087466,
|
| 4 |
+
"test_acc": 0.8432871476349737,
|
| 5 |
+
"test_auc": 0.9158060633951871,
|
| 6 |
+
"val_auc": 0.9081939081939081,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "66_living-room",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
16345
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9383594809666759,
|
| 18 |
+
"test_acc": 0.9383659818442427,
|
| 19 |
+
"test_auc": 0.9833011006590805,
|
| 20 |
+
"val_auc": 0.9982839982839983,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "66_living-room",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
16345,
|
| 28 |
+
27140
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9708550321310305,
|
| 33 |
+
"test_acc": 0.9708552317247969,
|
| 34 |
+
"test_auc": 0.9934174122184664,
|
| 35 |
+
"val_auc": 1.0,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "66_living-room",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
16345,
|
| 43 |
+
27140,
|
| 44 |
+
9907,
|
| 45 |
+
17605,
|
| 46 |
+
3615
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.9765884043013552,
|
| 51 |
+
"test_acc": 0.9765886287625418,
|
| 52 |
+
"test_auc": 0.9953824183088895,
|
| 53 |
+
"val_auc": 0.9982839982839983,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "66_living-room",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
16345,
|
| 61 |
+
27140,
|
| 62 |
+
9907,
|
| 63 |
+
17605,
|
| 64 |
+
3615,
|
| 65 |
+
26578,
|
| 66 |
+
4395,
|
| 67 |
+
29567,
|
| 68 |
+
9656,
|
| 69 |
+
27282
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9827988665017662,
|
| 74 |
+
"test_acc": 0.9827998088867654,
|
| 75 |
+
"test_auc": 0.9964197077692616,
|
| 76 |
+
"val_auc": 0.9991419991419992,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "66_living-room",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
16345,
|
| 84 |
+
27140,
|
| 85 |
+
9907,
|
| 86 |
+
17605,
|
| 87 |
+
3615,
|
| 88 |
+
26578,
|
| 89 |
+
4395,
|
| 90 |
+
29567,
|
| 91 |
+
9656,
|
| 92 |
+
27282,
|
| 93 |
+
10020,
|
| 94 |
+
25627,
|
| 95 |
+
19064,
|
| 96 |
+
3347,
|
| 97 |
+
16833,
|
| 98 |
+
17343
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/67_social-security_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9512608442101776,
|
| 4 |
+
"test_acc": 0.9512643678160919,
|
| 5 |
+
"test_auc": 0.9883279668813247,
|
| 6 |
+
"val_auc": 0.9756493506493505,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "67_social-security",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
19157
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9641373852052999,
|
| 18 |
+
"test_acc": 0.9641379310344828,
|
| 19 |
+
"test_auc": 0.995671606959251,
|
| 20 |
+
"val_auc": 0.9902597402597403,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "67_social-security",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
19157,
|
| 28 |
+
13068
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9848275797923417,
|
| 33 |
+
"test_acc": 0.9848275862068966,
|
| 34 |
+
"test_auc": 0.9989937902483901,
|
| 35 |
+
"val_auc": 0.9983766233766234,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "67_social-security",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
19157,
|
| 43 |
+
13068,
|
| 44 |
+
10563,
|
| 45 |
+
8322,
|
| 46 |
+
7706
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.987586059943069,
|
| 51 |
+
"test_acc": 0.9875862068965517,
|
| 52 |
+
"test_auc": 0.999561157259592,
|
| 53 |
+
"val_auc": 1.0,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "67_social-security",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
19157,
|
| 61 |
+
13068,
|
| 62 |
+
10563,
|
| 63 |
+
8322,
|
| 64 |
+
7706,
|
| 65 |
+
2814,
|
| 66 |
+
16865,
|
| 67 |
+
22232,
|
| 68 |
+
14397,
|
| 69 |
+
25577
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9926436750508345,
|
| 74 |
+
"test_acc": 0.9926436781609196,
|
| 75 |
+
"test_auc": 0.9998029858217437,
|
| 76 |
+
"val_auc": 1.0,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "67_social-security",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
19157,
|
| 84 |
+
13068,
|
| 85 |
+
10563,
|
| 86 |
+
8322,
|
| 87 |
+
7706,
|
| 88 |
+
2814,
|
| 89 |
+
16865,
|
| 90 |
+
22232,
|
| 91 |
+
14397,
|
| 92 |
+
25577,
|
| 93 |
+
22140,
|
| 94 |
+
21398,
|
| 95 |
+
26196,
|
| 96 |
+
17658,
|
| 97 |
+
18460,
|
| 98 |
+
12515
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/68_credit-card_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8537908781446845,
|
| 4 |
+
"test_acc": 0.8537931034482759,
|
| 5 |
+
"test_auc": 0.9330261716001947,
|
| 6 |
+
"val_auc": 0.939935064935065,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "68_credit-card",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
30753
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8960559186638465,
|
| 18 |
+
"test_acc": 0.8960919540229885,
|
| 19 |
+
"test_auc": 0.9662657611342607,
|
| 20 |
+
"val_auc": 0.9614448051948052,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "68_credit-card",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
30753,
|
| 28 |
+
17029
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9526436180973611,
|
| 33 |
+
"test_acc": 0.9526436781609195,
|
| 34 |
+
"test_auc": 0.9899793346501433,
|
| 35 |
+
"val_auc": 0.976461038961039,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "68_credit-card",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
30753,
|
| 43 |
+
17029,
|
| 44 |
+
18293,
|
| 45 |
+
12419,
|
| 46 |
+
16539
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.9595402298850575,
|
| 51 |
+
"test_acc": 0.9595402298850575,
|
| 52 |
+
"test_auc": 0.9921363439580063,
|
| 53 |
+
"val_auc": 0.9801136363636364,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "68_credit-card",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
30753,
|
| 61 |
+
17029,
|
| 62 |
+
18293,
|
| 63 |
+
12419,
|
| 64 |
+
16539,
|
| 65 |
+
30260,
|
| 66 |
+
10996,
|
| 67 |
+
5332,
|
| 68 |
+
7492,
|
| 69 |
+
32549
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9613792613609029,
|
| 74 |
+
"test_acc": 0.9613793103448276,
|
| 75 |
+
"test_auc": 0.9933522512040696,
|
| 76 |
+
"val_auc": 0.9809253246753247,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "68_credit-card",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
30753,
|
| 84 |
+
17029,
|
| 85 |
+
18293,
|
| 86 |
+
12419,
|
| 87 |
+
16539,
|
| 88 |
+
30260,
|
| 89 |
+
10996,
|
| 90 |
+
5332,
|
| 91 |
+
7492,
|
| 92 |
+
32549,
|
| 93 |
+
8156,
|
| 94 |
+
21159,
|
| 95 |
+
16356,
|
| 96 |
+
7677,
|
| 97 |
+
27628,
|
| 98 |
+
5256
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/69_blood-pressure_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9103334549548772,
|
| 4 |
+
"test_acc": 0.9103448275862069,
|
| 5 |
+
"test_auc": 0.9637172601331241,
|
| 6 |
+
"val_auc": 0.9809253246753247,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "69_blood-pressure",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
3947
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.926430871997599,
|
| 18 |
+
"test_acc": 0.9264367816091954,
|
| 19 |
+
"test_auc": 0.9771615753016938,
|
| 20 |
+
"val_auc": 0.9853896103896105,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "69_blood-pressure",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
3947,
|
| 28 |
+
2421
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9737903000033774,
|
| 33 |
+
"test_acc": 0.9737931034482759,
|
| 34 |
+
"test_auc": 0.9953223929866335,
|
| 35 |
+
"val_auc": 0.987012987012987,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "69_blood-pressure",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
3947,
|
| 43 |
+
2421,
|
| 44 |
+
9398,
|
| 45 |
+
25886,
|
| 46 |
+
29768
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.9797700636260362,
|
| 51 |
+
"test_acc": 0.9797701149425287,
|
| 52 |
+
"test_auc": 0.9973356580442665,
|
| 53 |
+
"val_auc": 0.9931006493506493,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "69_blood-pressure",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
3947,
|
| 61 |
+
2421,
|
| 62 |
+
9398,
|
| 63 |
+
25886,
|
| 64 |
+
29768,
|
| 65 |
+
5161,
|
| 66 |
+
21362,
|
| 67 |
+
15413,
|
| 68 |
+
2088,
|
| 69 |
+
26434
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9908045938135429,
|
| 74 |
+
"test_acc": 0.9908045977011494,
|
| 75 |
+
"test_auc": 0.9992280088208236,
|
| 76 |
+
"val_auc": 0.994724025974026,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "69_blood-pressure",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
3947,
|
| 84 |
+
2421,
|
| 85 |
+
9398,
|
| 86 |
+
25886,
|
| 87 |
+
29768,
|
| 88 |
+
5161,
|
| 89 |
+
21362,
|
| 90 |
+
15413,
|
| 91 |
+
2088,
|
| 92 |
+
26434,
|
| 93 |
+
19325,
|
| 94 |
+
460,
|
| 95 |
+
30603,
|
| 96 |
+
7233,
|
| 97 |
+
193,
|
| 98 |
+
10333
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/6_hist_fig_isamerican_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8037869561656528,
|
| 4 |
+
"test_acc": 0.8040251572327044,
|
| 5 |
+
"test_auc": 0.8752785965921346,
|
| 6 |
+
"val_auc": 0.8871548619447778,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "6_hist_fig_isamerican",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
6588
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8148216014810898,
|
| 18 |
+
"test_acc": 0.8148427672955975,
|
| 19 |
+
"test_auc": 0.8957815843222393,
|
| 20 |
+
"val_auc": 0.9279711884753902,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "6_hist_fig_isamerican",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
6588,
|
| 28 |
+
17771
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.8572161630717212,
|
| 33 |
+
"test_acc": 0.8573584905660377,
|
| 34 |
+
"test_auc": 0.9345856214286221,
|
| 35 |
+
"val_auc": 0.9683873549419768,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "6_hist_fig_isamerican",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
6588,
|
| 43 |
+
17771,
|
| 44 |
+
24464,
|
| 45 |
+
13633,
|
| 46 |
+
398
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.8711394625576627,
|
| 51 |
+
"test_acc": 0.8711949685534591,
|
| 52 |
+
"test_auc": 0.9426660617960405,
|
| 53 |
+
"val_auc": 0.9627851140456183,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "6_hist_fig_isamerican",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
6588,
|
| 61 |
+
17771,
|
| 62 |
+
24464,
|
| 63 |
+
13633,
|
| 64 |
+
398,
|
| 65 |
+
16455,
|
| 66 |
+
12692,
|
| 67 |
+
14074,
|
| 68 |
+
19678,
|
| 69 |
+
29051
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.8937513742510691,
|
| 74 |
+
"test_acc": 0.8938364779874214,
|
| 75 |
+
"test_auc": 0.953089953915744,
|
| 76 |
+
"val_auc": 0.9771908763505402,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "6_hist_fig_isamerican",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
6588,
|
| 84 |
+
17771,
|
| 85 |
+
24464,
|
| 86 |
+
13633,
|
| 87 |
+
398,
|
| 88 |
+
16455,
|
| 89 |
+
12692,
|
| 90 |
+
14074,
|
| 91 |
+
19678,
|
| 92 |
+
29051,
|
| 93 |
+
9070,
|
| 94 |
+
26677,
|
| 95 |
+
24721,
|
| 96 |
+
16352,
|
| 97 |
+
14006,
|
| 98 |
+
20838
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/70_prime-factors_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9475737726374909,
|
| 4 |
+
"test_acc": 0.9475862068965517,
|
| 5 |
+
"test_auc": 0.9824454448292657,
|
| 6 |
+
"val_auc": 0.9823457792207791,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "70_prime-factors",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
7072
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9742487582871985,
|
| 18 |
+
"test_acc": 0.9742528735632184,
|
| 19 |
+
"test_auc": 0.996217835110125,
|
| 20 |
+
"val_auc": 0.9983766233766234,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "70_prime-factors",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
7072,
|
| 28 |
+
23306
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9802249106223324,
|
| 33 |
+
"test_acc": 0.9802298850574712,
|
| 34 |
+
"test_auc": 0.9993353942312896,
|
| 35 |
+
"val_auc": 1.0,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "70_prime-factors",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
7072,
|
| 43 |
+
23306,
|
| 44 |
+
25059,
|
| 45 |
+
21598,
|
| 46 |
+
15824
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.9917240854480806,
|
| 51 |
+
"test_acc": 0.9917241379310345,
|
| 52 |
+
"test_auc": 0.9996896815303857,
|
| 53 |
+
"val_auc": 1.0,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "70_prime-factors",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
7072,
|
| 61 |
+
23306,
|
| 62 |
+
25059,
|
| 63 |
+
21598,
|
| 64 |
+
15824,
|
| 65 |
+
1554,
|
| 66 |
+
6365,
|
| 67 |
+
532,
|
| 68 |
+
827,
|
| 69 |
+
24058
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9917240854480806,
|
| 74 |
+
"test_acc": 0.9917241379310345,
|
| 75 |
+
"test_auc": 0.9997793102981763,
|
| 76 |
+
"val_auc": 1.0,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "70_prime-factors",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
7072,
|
| 84 |
+
23306,
|
| 85 |
+
25059,
|
| 86 |
+
21598,
|
| 87 |
+
15824,
|
| 88 |
+
1554,
|
| 89 |
+
6365,
|
| 90 |
+
532,
|
| 91 |
+
827,
|
| 92 |
+
24058,
|
| 93 |
+
14701,
|
| 94 |
+
18975,
|
| 95 |
+
19679,
|
| 96 |
+
14876,
|
| 97 |
+
17981,
|
| 98 |
+
22736
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/71_social-media_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8333729463963617,
|
| 4 |
+
"test_acc": 0.8335632183908046,
|
| 5 |
+
"test_auc": 0.8985326248714758,
|
| 6 |
+
"val_auc": 0.9062500000000001,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "71_social-media",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
26284
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9217805483619173,
|
| 18 |
+
"test_acc": 0.9218390804597701,
|
| 19 |
+
"test_auc": 0.9736013684452621,
|
| 20 |
+
"val_auc": 0.9914772727272727,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "71_social-media",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
26284,
|
| 28 |
+
2003
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9742525687717765,
|
| 33 |
+
"test_acc": 0.9742528735632184,
|
| 34 |
+
"test_auc": 0.9941741300936198,
|
| 35 |
+
"val_auc": 0.9995941558441559,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "71_social-media",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
26284,
|
| 43 |
+
2003,
|
| 44 |
+
8213,
|
| 45 |
+
2042,
|
| 46 |
+
4694
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.9894252873563218,
|
| 51 |
+
"test_acc": 0.9894252873563218,
|
| 52 |
+
"test_auc": 0.9968815953244222,
|
| 53 |
+
"val_auc": 1.0,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "71_social-media",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
26284,
|
| 61 |
+
2003,
|
| 62 |
+
8213,
|
| 63 |
+
2042,
|
| 64 |
+
4694,
|
| 65 |
+
9438,
|
| 66 |
+
17692,
|
| 67 |
+
26298,
|
| 68 |
+
26033,
|
| 69 |
+
20532
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9903448030942082,
|
| 74 |
+
"test_acc": 0.9903448275862069,
|
| 75 |
+
"test_auc": 0.9970574706423507,
|
| 76 |
+
"val_auc": 1.0,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "71_social-media",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
26284,
|
| 84 |
+
2003,
|
| 85 |
+
8213,
|
| 86 |
+
2042,
|
| 87 |
+
4694,
|
| 88 |
+
9438,
|
| 89 |
+
17692,
|
| 90 |
+
26298,
|
| 91 |
+
26033,
|
| 92 |
+
20532,
|
| 93 |
+
12534,
|
| 94 |
+
4175,
|
| 95 |
+
9658,
|
| 96 |
+
6859,
|
| 97 |
+
1400,
|
| 98 |
+
3635
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|
gemma-2-2b-btk-mat/k-100/seed-0/snapshots/step-1-tokens-0/sae_probes/sae_probes_gemma-2-2b/normal_setting/72_gene-expression_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7565934088832684,
|
| 4 |
+
"test_acc": 0.7586206896551724,
|
| 5 |
+
"test_auc": 0.8242126197305049,
|
| 6 |
+
"val_auc": 0.8498376623376624,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "72_gene-expression",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
25233
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8592966035818567,
|
| 18 |
+
"test_acc": 0.8593103448275862,
|
| 19 |
+
"test_auc": 0.9221523418475026,
|
| 20 |
+
"val_auc": 0.9285714285714286,
|
| 21 |
+
"k": 2,
|
| 22 |
+
"dataset": "72_gene-expression",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
25233,
|
| 28 |
+
21252
|
| 29 |
+
]
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"test_f1": 0.9144758886972089,
|
| 33 |
+
"test_acc": 0.9144827586206896,
|
| 34 |
+
"test_auc": 0.969383320255425,
|
| 35 |
+
"val_auc": 0.9898538961038961,
|
| 36 |
+
"k": 5,
|
| 37 |
+
"dataset": "72_gene-expression",
|
| 38 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 39 |
+
"reg_type": "l1",
|
| 40 |
+
"binarize": false,
|
| 41 |
+
"indices": [
|
| 42 |
+
25233,
|
| 43 |
+
21252,
|
| 44 |
+
20834,
|
| 45 |
+
10079,
|
| 46 |
+
15413
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"test_f1": 0.9411436773152865,
|
| 51 |
+
"test_acc": 0.9411494252873563,
|
| 52 |
+
"test_auc": 0.984878950430218,
|
| 53 |
+
"val_auc": 0.994724025974026,
|
| 54 |
+
"k": 10,
|
| 55 |
+
"dataset": "72_gene-expression",
|
| 56 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 57 |
+
"reg_type": "l1",
|
| 58 |
+
"binarize": false,
|
| 59 |
+
"indices": [
|
| 60 |
+
25233,
|
| 61 |
+
21252,
|
| 62 |
+
20834,
|
| 63 |
+
10079,
|
| 64 |
+
15413,
|
| 65 |
+
4784,
|
| 66 |
+
4628,
|
| 67 |
+
17996,
|
| 68 |
+
4375,
|
| 69 |
+
19746
|
| 70 |
+
]
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"test_f1": 0.9526402543064824,
|
| 74 |
+
"test_acc": 0.9526436781609195,
|
| 75 |
+
"test_auc": 0.9919765341739273,
|
| 76 |
+
"val_auc": 0.9987824675324676,
|
| 77 |
+
"k": 16,
|
| 78 |
+
"dataset": "72_gene-expression",
|
| 79 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 80 |
+
"reg_type": "l1",
|
| 81 |
+
"binarize": false,
|
| 82 |
+
"indices": [
|
| 83 |
+
25233,
|
| 84 |
+
21252,
|
| 85 |
+
20834,
|
| 86 |
+
10079,
|
| 87 |
+
15413,
|
| 88 |
+
4784,
|
| 89 |
+
4628,
|
| 90 |
+
17996,
|
| 91 |
+
4375,
|
| 92 |
+
19746,
|
| 93 |
+
30345,
|
| 94 |
+
23641,
|
| 95 |
+
853,
|
| 96 |
+
18532,
|
| 97 |
+
31591,
|
| 98 |
+
26634
|
| 99 |
+
]
|
| 100 |
+
}
|
| 101 |
+
]
|