Add files using upload-large-folder tool
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/119_us_state_TX_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/121_us_timezone_New_York_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/124_world_country_United_States_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/125_world_country_Italy_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/126_art_type_book_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/127_art_type_song_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/128_art_type_movie_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/132_temp_cat_Event Ordering_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/133_context_type_Causality_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/135_context_type_Event_duration_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/137_glue_mnli_neutral_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/141_news_class_Entertainment_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/142_cancer_cat_Thyroid_Cancer_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/143_cancer_cat_Lung_Cancer_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/144_cancer_cat_Colon_Cancer_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/145_disease_class_digestive system diseases_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/147_disease_class_nervous system diseases_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/148_twt_emotion_worry_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/149_twt_emotion_happiness_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/150_twt_emotion_sadness_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/151_it_tick_HR Support_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/152_it_tick_Hardware_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/153_it_tick_Administrative rights_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/154_athlete_sport_football_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/155_athlete_sport_basketball_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/156_athlete_sport_baseball_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/157_amazon_5star_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/158_code_C_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/159_code_Python_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/160_code_HTML_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/161_agnews_0_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/162_agnews_1_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/163_agnews_2_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/21_headline_istrump_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/22_headline_isobama_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/23_headline_ischina_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/24_headline_isiran_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/26_headline_isfrontpage_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/36_sciq_tf_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/41_truthqa_tf_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/42_temp_sense_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/44_phys_tf_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/47_reasoning_tf_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/48_cm_correct_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/49_cm_isshort_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/50_deon_isvalid_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/51_just_is_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/52_virtue_is_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/54_cs_tf_blocks.12.hook_resid_post_l1.json +45 -0
- denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/56_wikidatasex_or_gender_blocks.12.hook_resid_post_l1.json +45 -0
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/119_us_state_TX_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8353300446323702,
|
| 4 |
+
"test_acc": 0.8383838383838383,
|
| 5 |
+
"test_auc": 0.8448979591836735,
|
| 6 |
+
"val_auc": 0.8159054487179487,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "119_us_state_TX",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
21609
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8584992077639138,
|
| 18 |
+
"test_acc": 0.8585858585858586,
|
| 19 |
+
"test_auc": 0.9569387755102041,
|
| 20 |
+
"val_auc": 0.9525240384615384,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "119_us_state_TX",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
21609,
|
| 28 |
+
31995,
|
| 29 |
+
21136,
|
| 30 |
+
12114,
|
| 31 |
+
22225,
|
| 32 |
+
23572,
|
| 33 |
+
19740,
|
| 34 |
+
21077,
|
| 35 |
+
21583,
|
| 36 |
+
16430,
|
| 37 |
+
16199,
|
| 38 |
+
19112,
|
| 39 |
+
22554,
|
| 40 |
+
2909,
|
| 41 |
+
6095,
|
| 42 |
+
19259
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/121_us_timezone_New_York_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6716551227542524,
|
| 4 |
+
"test_acc": 0.6762028608582574,
|
| 5 |
+
"test_auc": 0.7214274870317956,
|
| 6 |
+
"val_auc": 0.7428,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "121_us_timezone_New_York",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
8251
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.799704922579349,
|
| 18 |
+
"test_acc": 0.7997399219765929,
|
| 19 |
+
"test_auc": 0.8939230189949812,
|
| 20 |
+
"val_auc": 0.9088,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "121_us_timezone_New_York",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
8251,
|
| 28 |
+
12114,
|
| 29 |
+
30718,
|
| 30 |
+
16430,
|
| 31 |
+
19338,
|
| 32 |
+
21077,
|
| 33 |
+
23502,
|
| 34 |
+
23774,
|
| 35 |
+
29685,
|
| 36 |
+
21136,
|
| 37 |
+
6872,
|
| 38 |
+
28783,
|
| 39 |
+
18533,
|
| 40 |
+
2266,
|
| 41 |
+
19112,
|
| 42 |
+
13525
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/124_world_country_United_States_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9175297278060466,
|
| 4 |
+
"test_acc": 0.9179487179487179,
|
| 5 |
+
"test_auc": 0.920784663547312,
|
| 6 |
+
"val_auc": 0.9651162790697674,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "124_world_country_United_States",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
30233
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9620376248400001,
|
| 18 |
+
"test_acc": 0.9620512820512821,
|
| 19 |
+
"test_auc": 0.9923355606422729,
|
| 20 |
+
"val_auc": 1.0,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "124_world_country_United_States",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
30233,
|
| 28 |
+
4321,
|
| 29 |
+
4693,
|
| 30 |
+
12292,
|
| 31 |
+
844,
|
| 32 |
+
9819,
|
| 33 |
+
10187,
|
| 34 |
+
26535,
|
| 35 |
+
5917,
|
| 36 |
+
15943,
|
| 37 |
+
22455,
|
| 38 |
+
31104,
|
| 39 |
+
27488,
|
| 40 |
+
26858,
|
| 41 |
+
4973,
|
| 42 |
+
8626
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/125_world_country_Italy_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9538170055520213,
|
| 4 |
+
"test_acc": 0.9538461538461539,
|
| 5 |
+
"test_auc": 0.9704194297640287,
|
| 6 |
+
"val_auc": 0.9873521011831906,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "125_world_country_Italy",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
15305
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9702405702322346,
|
| 18 |
+
"test_acc": 0.9702564102564103,
|
| 19 |
+
"test_auc": 0.9956533813579291,
|
| 20 |
+
"val_auc": 1.0,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "125_world_country_Italy",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
15305,
|
| 28 |
+
14424,
|
| 29 |
+
29979,
|
| 30 |
+
20799,
|
| 31 |
+
21077,
|
| 32 |
+
12875,
|
| 33 |
+
223,
|
| 34 |
+
11456,
|
| 35 |
+
13973,
|
| 36 |
+
30249,
|
| 37 |
+
8626,
|
| 38 |
+
12292,
|
| 39 |
+
74,
|
| 40 |
+
19719,
|
| 41 |
+
13869,
|
| 42 |
+
815
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/126_art_type_book_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5881676886896738,
|
| 4 |
+
"test_acc": 0.5899436497615952,
|
| 5 |
+
"test_auc": 0.6187513246282398,
|
| 6 |
+
"val_auc": 0.627,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "126_art_type_book",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
9917
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7007939028436188,
|
| 18 |
+
"test_acc": 0.7009102730819246,
|
| 19 |
+
"test_auc": 0.7863692184204871,
|
| 20 |
+
"val_auc": 0.8162,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "126_art_type_book",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
9917,
|
| 28 |
+
16223,
|
| 29 |
+
20721,
|
| 30 |
+
5359,
|
| 31 |
+
4333,
|
| 32 |
+
18547,
|
| 33 |
+
12290,
|
| 34 |
+
23681,
|
| 35 |
+
23991,
|
| 36 |
+
10262,
|
| 37 |
+
28800,
|
| 38 |
+
23339,
|
| 39 |
+
5513,
|
| 40 |
+
19963,
|
| 41 |
+
29096,
|
| 42 |
+
18463
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/127_art_type_song_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6743502561471805,
|
| 4 |
+
"test_acc": 0.6913740788903338,
|
| 5 |
+
"test_auc": 0.6980403769234355,
|
| 6 |
+
"val_auc": 0.7722,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "127_art_type_song",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
20721
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7416334025527475,
|
| 18 |
+
"test_acc": 0.741655830082358,
|
| 19 |
+
"test_auc": 0.8254451878228899,
|
| 20 |
+
"val_auc": 0.8366,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "127_art_type_song",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
20721,
|
| 28 |
+
23681,
|
| 29 |
+
4333,
|
| 30 |
+
16223,
|
| 31 |
+
12290,
|
| 32 |
+
10262,
|
| 33 |
+
28800,
|
| 34 |
+
30970,
|
| 35 |
+
13664,
|
| 36 |
+
10661,
|
| 37 |
+
19149,
|
| 38 |
+
5513,
|
| 39 |
+
2844,
|
| 40 |
+
5359,
|
| 41 |
+
9917,
|
| 42 |
+
14232
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/128_art_type_movie_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5390079562312783,
|
| 4 |
+
"test_acc": 0.5622019939315128,
|
| 5 |
+
"test_auc": 0.5714510109262101,
|
| 6 |
+
"val_auc": 0.5852,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "128_art_type_movie",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
20799
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6375527062496742,
|
| 18 |
+
"test_acc": 0.6376246207195492,
|
| 19 |
+
"test_auc": 0.7034215617160268,
|
| 20 |
+
"val_auc": 0.7365999999999999,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "128_art_type_movie",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
20799,
|
| 28 |
+
20721,
|
| 29 |
+
6298,
|
| 30 |
+
25342,
|
| 31 |
+
29979,
|
| 32 |
+
10661,
|
| 33 |
+
31104,
|
| 34 |
+
16414,
|
| 35 |
+
29096,
|
| 36 |
+
6551,
|
| 37 |
+
44,
|
| 38 |
+
18628,
|
| 39 |
+
28104,
|
| 40 |
+
17319,
|
| 41 |
+
19149,
|
| 42 |
+
28923
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/132_temp_cat_Event Ordering_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8686600618824769,
|
| 4 |
+
"test_acc": 0.8686868686868687,
|
| 5 |
+
"test_auc": 0.9310204081632654,
|
| 6 |
+
"val_auc": 0.910426731078905,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "132_temp_cat_Event Ordering",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
20495
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.98989898989899,
|
| 18 |
+
"test_acc": 0.98989898989899,
|
| 19 |
+
"test_auc": 0.9991836734693877,
|
| 20 |
+
"val_auc": 1.0,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "132_temp_cat_Event Ordering",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
20495,
|
| 28 |
+
14534,
|
| 29 |
+
206,
|
| 30 |
+
9078,
|
| 31 |
+
31124,
|
| 32 |
+
14809,
|
| 33 |
+
3404,
|
| 34 |
+
5075,
|
| 35 |
+
14184,
|
| 36 |
+
3176,
|
| 37 |
+
31864,
|
| 38 |
+
3669,
|
| 39 |
+
23701,
|
| 40 |
+
136,
|
| 41 |
+
12761,
|
| 42 |
+
24231
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/133_context_type_Causality_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9377630261840788,
|
| 4 |
+
"test_acc": 0.9377777777777778,
|
| 5 |
+
"test_auc": 0.9451248419721872,
|
| 6 |
+
"val_auc": 0.9389204545454544,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "133_context_type_Causality",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
15604
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.955553799690968,
|
| 18 |
+
"test_acc": 0.9555555555555556,
|
| 19 |
+
"test_auc": 0.9722661188369153,
|
| 20 |
+
"val_auc": 0.9590097402597402,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "133_context_type_Causality",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
15604,
|
| 28 |
+
31085,
|
| 29 |
+
14707,
|
| 30 |
+
20017,
|
| 31 |
+
12033,
|
| 32 |
+
12026,
|
| 33 |
+
7738,
|
| 34 |
+
30042,
|
| 35 |
+
2370,
|
| 36 |
+
3176,
|
| 37 |
+
4971,
|
| 38 |
+
17696,
|
| 39 |
+
8916,
|
| 40 |
+
24164,
|
| 41 |
+
5197,
|
| 42 |
+
15017
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/135_context_type_Event_duration_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9599430391012836,
|
| 4 |
+
"test_acc": 0.96,
|
| 5 |
+
"test_auc": 0.9601769911504425,
|
| 6 |
+
"val_auc": 0.9545454545454546,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "135_context_type_Event_duration",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
3669
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9599430391012836,
|
| 18 |
+
"test_acc": 0.96,
|
| 19 |
+
"test_auc": 0.9883059418457649,
|
| 20 |
+
"val_auc": 0.9805194805194806,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "135_context_type_Event_duration",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
3669,
|
| 28 |
+
31124,
|
| 29 |
+
14809,
|
| 30 |
+
31864,
|
| 31 |
+
14954,
|
| 32 |
+
24231,
|
| 33 |
+
24473,
|
| 34 |
+
3286,
|
| 35 |
+
10898,
|
| 36 |
+
20495,
|
| 37 |
+
29595,
|
| 38 |
+
19694,
|
| 39 |
+
22745,
|
| 40 |
+
10850,
|
| 41 |
+
14167,
|
| 42 |
+
26524
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/137_glue_mnli_neutral_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6160375393433526,
|
| 4 |
+
"test_acc": 0.6194191590810576,
|
| 5 |
+
"test_auc": 0.6362818869019257,
|
| 6 |
+
"val_auc": 0.6654,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "137_glue_mnli_neutral",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
10455
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6966305012681863,
|
| 18 |
+
"test_acc": 0.6970091027308193,
|
| 19 |
+
"test_auc": 0.7671780796385288,
|
| 20 |
+
"val_auc": 0.8144,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "137_glue_mnli_neutral",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
10455,
|
| 28 |
+
7835,
|
| 29 |
+
127,
|
| 30 |
+
9703,
|
| 31 |
+
11060,
|
| 32 |
+
21005,
|
| 33 |
+
32584,
|
| 34 |
+
23877,
|
| 35 |
+
21531,
|
| 36 |
+
12220,
|
| 37 |
+
29195,
|
| 38 |
+
13811,
|
| 39 |
+
6485,
|
| 40 |
+
16914,
|
| 41 |
+
20849,
|
| 42 |
+
32531
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/141_news_class_Entertainment_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5860234401965285,
|
| 4 |
+
"test_acc": 0.6006779661016949,
|
| 5 |
+
"test_auc": 0.6060661217195618,
|
| 6 |
+
"val_auc": 0.7057165861513688,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "141_news_class_Entertainment",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
30820
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7768824921053477,
|
| 18 |
+
"test_acc": 0.7769491525423728,
|
| 19 |
+
"test_auc": 0.8695326030600876,
|
| 20 |
+
"val_auc": 0.9005636070853462,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "141_news_class_Entertainment",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
30820,
|
| 28 |
+
19499,
|
| 29 |
+
21868,
|
| 30 |
+
16223,
|
| 31 |
+
8540,
|
| 32 |
+
1539,
|
| 33 |
+
24479,
|
| 34 |
+
6390,
|
| 35 |
+
9156,
|
| 36 |
+
992,
|
| 37 |
+
17353,
|
| 38 |
+
3357,
|
| 39 |
+
20470,
|
| 40 |
+
5744,
|
| 41 |
+
20998,
|
| 42 |
+
28740
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/142_cancer_cat_Thyroid_Cancer_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6161730928024542,
|
| 4 |
+
"test_acc": 0.6297376093294461,
|
| 5 |
+
"test_auc": 0.6300489596083231,
|
| 6 |
+
"val_auc": 0.59343737494998,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "142_cancer_cat_Thyroid_Cancer",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
28302
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6849812578092461,
|
| 18 |
+
"test_acc": 0.685131195335277,
|
| 19 |
+
"test_auc": 0.7311981504147967,
|
| 20 |
+
"val_auc": 0.7949179671868747,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "142_cancer_cat_Thyroid_Cancer",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
28302,
|
| 28 |
+
22835,
|
| 29 |
+
19716,
|
| 30 |
+
7305,
|
| 31 |
+
1728,
|
| 32 |
+
22973,
|
| 33 |
+
32517,
|
| 34 |
+
5360,
|
| 35 |
+
30768,
|
| 36 |
+
29717,
|
| 37 |
+
7166,
|
| 38 |
+
13411,
|
| 39 |
+
26464,
|
| 40 |
+
8974,
|
| 41 |
+
13811,
|
| 42 |
+
30079
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/143_cancer_cat_Lung_Cancer_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5421962297672415,
|
| 4 |
+
"test_acc": 0.5422740524781341,
|
| 5 |
+
"test_auc": 0.5701414388684891,
|
| 6 |
+
"val_auc": 0.6778711484593838,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "143_cancer_cat_Lung_Cancer",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
28273
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8682873409070635,
|
| 18 |
+
"test_acc": 0.8688046647230321,
|
| 19 |
+
"test_auc": 0.9338705290357678,
|
| 20 |
+
"val_auc": 0.9227691076430572,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "143_cancer_cat_Lung_Cancer",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
28273,
|
| 28 |
+
26464,
|
| 29 |
+
10790,
|
| 30 |
+
22973,
|
| 31 |
+
1728,
|
| 32 |
+
13411,
|
| 33 |
+
17171,
|
| 34 |
+
4578,
|
| 35 |
+
19716,
|
| 36 |
+
830,
|
| 37 |
+
28302,
|
| 38 |
+
22835,
|
| 39 |
+
21038,
|
| 40 |
+
13787,
|
| 41 |
+
30079,
|
| 42 |
+
8974
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/144_cancer_cat_Colon_Cancer_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6354106077995634,
|
| 4 |
+
"test_acc": 0.6384839650145773,
|
| 5 |
+
"test_auc": 0.656483748130015,
|
| 6 |
+
"val_auc": 0.6182472989195678,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "144_cancer_cat_Colon_Cancer",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
17171
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.781329956073065,
|
| 18 |
+
"test_acc": 0.7813411078717201,
|
| 19 |
+
"test_auc": 0.8706990344077248,
|
| 20 |
+
"val_auc": 0.8811524609843937,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "144_cancer_cat_Colon_Cancer",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
17171,
|
| 28 |
+
28273,
|
| 29 |
+
8974,
|
| 30 |
+
30768,
|
| 31 |
+
830,
|
| 32 |
+
26464,
|
| 33 |
+
10790,
|
| 34 |
+
815,
|
| 35 |
+
9567,
|
| 36 |
+
22973,
|
| 37 |
+
21038,
|
| 38 |
+
4578,
|
| 39 |
+
22692,
|
| 40 |
+
13411,
|
| 41 |
+
16711,
|
| 42 |
+
4357
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/145_disease_class_digestive system diseases_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5185199452148604,
|
| 4 |
+
"test_acc": 0.5353535353535354,
|
| 5 |
+
"test_auc": 0.5783673469387756,
|
| 6 |
+
"val_auc": 0.6321212121212121,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "145_disease_class_digestive system diseases",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
733
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5870990607832713,
|
| 18 |
+
"test_acc": 0.5959595959595959,
|
| 19 |
+
"test_auc": 0.6334693877551021,
|
| 20 |
+
"val_auc": 0.5963636363636363,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "145_disease_class_digestive system diseases",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
733,
|
| 28 |
+
23712,
|
| 29 |
+
26966,
|
| 30 |
+
6152,
|
| 31 |
+
28179,
|
| 32 |
+
9269,
|
| 33 |
+
29096,
|
| 34 |
+
7769,
|
| 35 |
+
28502,
|
| 36 |
+
21870,
|
| 37 |
+
2187,
|
| 38 |
+
24883,
|
| 39 |
+
15498,
|
| 40 |
+
26959,
|
| 41 |
+
16005,
|
| 42 |
+
28115
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/147_disease_class_nervous system diseases_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5554648526077097,
|
| 4 |
+
"test_acc": 0.5555555555555556,
|
| 5 |
+
"test_auc": 0.6028571428571428,
|
| 6 |
+
"val_auc": 0.6076767676767677,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "147_disease_class_nervous system diseases",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
29341
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5440580924451892,
|
| 18 |
+
"test_acc": 0.5454545454545454,
|
| 19 |
+
"test_auc": 0.5820408163265306,
|
| 20 |
+
"val_auc": 0.6731313131313131,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "147_disease_class_nervous system diseases",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
29341,
|
| 28 |
+
13208,
|
| 29 |
+
733,
|
| 30 |
+
14167,
|
| 31 |
+
26524,
|
| 32 |
+
29127,
|
| 33 |
+
19063,
|
| 34 |
+
26966,
|
| 35 |
+
12047,
|
| 36 |
+
8237,
|
| 37 |
+
10586,
|
| 38 |
+
21895,
|
| 39 |
+
32626,
|
| 40 |
+
18365,
|
| 41 |
+
7451,
|
| 42 |
+
32275
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/148_twt_emotion_worry_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5545791910623937,
|
| 4 |
+
"test_acc": 0.5735593220338983,
|
| 5 |
+
"test_auc": 0.5776227877611205,
|
| 6 |
+
"val_auc": 0.5897745571658615,
|
| 7 |
+
"k": 1,
|
| 8 |
+
"dataset": "148_twt_emotion_worry",
|
| 9 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 10 |
+
"reg_type": "l1",
|
| 11 |
+
"binarize": false,
|
| 12 |
+
"indices": [
|
| 13 |
+
24164
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5606167891050665,
|
| 18 |
+
"test_acc": 0.5613559322033899,
|
| 19 |
+
"test_auc": 0.6058271098314782,
|
| 20 |
+
"val_auc": 0.6316425120772946,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "148_twt_emotion_worry",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
24164,
|
| 28 |
+
26524,
|
| 29 |
+
17180,
|
| 30 |
+
26883,
|
| 31 |
+
15055,
|
| 32 |
+
22924,
|
| 33 |
+
22662,
|
| 34 |
+
16223,
|
| 35 |
+
22524,
|
| 36 |
+
27433,
|
| 37 |
+
12472,
|
| 38 |
+
13811,
|
| 39 |
+
8890,
|
| 40 |
+
30970,
|
| 41 |
+
9949,
|
| 42 |
+
13693
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/149_twt_emotion_happiness_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5835507942407561,
|
| 4 |
+
"test_acc": 0.583728813559322,
|
| 5 |
+
"test_auc": 0.6076417616279284,
|
| 6 |
+
"val_auc": 0.6223832528180355,
|
| 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 |
+
30356
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6763539998062809,
|
| 18 |
+
"test_acc": 0.6766101694915254,
|
| 19 |
+
"test_auc": 0.7438491209878177,
|
| 20 |
+
"val_auc": 0.751207729468599,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "149_twt_emotion_happiness",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
30356,
|
| 28 |
+
27433,
|
| 29 |
+
24164,
|
| 30 |
+
31382,
|
| 31 |
+
15055,
|
| 32 |
+
26524,
|
| 33 |
+
6109,
|
| 34 |
+
12894,
|
| 35 |
+
1412,
|
| 36 |
+
11988,
|
| 37 |
+
127,
|
| 38 |
+
22524,
|
| 39 |
+
5756,
|
| 40 |
+
30234,
|
| 41 |
+
24480,
|
| 42 |
+
7503
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/150_twt_emotion_sadness_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5563981796283983,
|
| 4 |
+
"test_acc": 0.5769491525423729,
|
| 5 |
+
"test_auc": 0.5766750136972197,
|
| 6 |
+
"val_auc": 0.6469404186795491,
|
| 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 |
+
24164
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6197466859011922,
|
| 18 |
+
"test_acc": 0.6230508474576271,
|
| 19 |
+
"test_auc": 0.6487683533551754,
|
| 20 |
+
"val_auc": 0.7805958132045089,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "150_twt_emotion_sadness",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
24164,
|
| 28 |
+
1412,
|
| 29 |
+
29986,
|
| 30 |
+
127,
|
| 31 |
+
11079,
|
| 32 |
+
7146,
|
| 33 |
+
15162,
|
| 34 |
+
3568,
|
| 35 |
+
26883,
|
| 36 |
+
5359,
|
| 37 |
+
6750,
|
| 38 |
+
30234,
|
| 39 |
+
22662,
|
| 40 |
+
25562,
|
| 41 |
+
25397,
|
| 42 |
+
17316
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/151_it_tick_HR Support_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6080894868756354,
|
| 4 |
+
"test_acc": 0.6253101736972705,
|
| 5 |
+
"test_auc": 0.6376779468991675,
|
| 6 |
+
"val_auc": 0.704481792717087,
|
| 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 |
+
9784
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6101578114483011,
|
| 18 |
+
"test_acc": 0.6104218362282878,
|
| 19 |
+
"test_auc": 0.6617408009457661,
|
| 20 |
+
"val_auc": 0.7390956382553021,
|
| 21 |
+
"k": 16,
|
| 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 |
+
9784,
|
| 28 |
+
21846,
|
| 29 |
+
26837,
|
| 30 |
+
29158,
|
| 31 |
+
793,
|
| 32 |
+
10328,
|
| 33 |
+
19664,
|
| 34 |
+
18709,
|
| 35 |
+
22659,
|
| 36 |
+
20975,
|
| 37 |
+
5360,
|
| 38 |
+
19892,
|
| 39 |
+
14202,
|
| 40 |
+
5714,
|
| 41 |
+
23328,
|
| 42 |
+
12152
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/152_it_tick_Hardware_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5877998263225331,
|
| 4 |
+
"test_acc": 0.5930521091811415,
|
| 5 |
+
"test_auc": 0.6093665336682922,
|
| 6 |
+
"val_auc": 0.5898359343737495,
|
| 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 |
+
564
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6050163893966636,
|
| 18 |
+
"test_acc": 0.6054590570719603,
|
| 19 |
+
"test_auc": 0.6413230875326338,
|
| 20 |
+
"val_auc": 0.7174869947979191,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "152_it_tick_Hardware",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
564,
|
| 28 |
+
5008,
|
| 29 |
+
23088,
|
| 30 |
+
27327,
|
| 31 |
+
14167,
|
| 32 |
+
26837,
|
| 33 |
+
8974,
|
| 34 |
+
12220,
|
| 35 |
+
16662,
|
| 36 |
+
29158,
|
| 37 |
+
31971,
|
| 38 |
+
9476,
|
| 39 |
+
11763,
|
| 40 |
+
11774,
|
| 41 |
+
24165,
|
| 42 |
+
22455
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/153_it_tick_Administrative rights_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5064061328127345,
|
| 4 |
+
"test_acc": 0.5210918114143921,
|
| 5 |
+
"test_auc": 0.5619920200975321,
|
| 6 |
+
"val_auc": 0.5806322529011605,
|
| 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 |
+
26837
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6972145430113639,
|
| 18 |
+
"test_acc": 0.6972704714640199,
|
| 19 |
+
"test_auc": 0.7766858775429781,
|
| 20 |
+
"val_auc": 0.7266906762705082,
|
| 21 |
+
"k": 16,
|
| 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 |
+
26837,
|
| 28 |
+
23088,
|
| 29 |
+
6197,
|
| 30 |
+
8,
|
| 31 |
+
728,
|
| 32 |
+
24505,
|
| 33 |
+
12330,
|
| 34 |
+
21846,
|
| 35 |
+
1915,
|
| 36 |
+
18181,
|
| 37 |
+
564,
|
| 38 |
+
1828,
|
| 39 |
+
19808,
|
| 40 |
+
29717,
|
| 41 |
+
32521,
|
| 42 |
+
26720
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/154_athlete_sport_football_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8363916967617878,
|
| 4 |
+
"test_acc": 0.8380187416331994,
|
| 5 |
+
"test_auc": 0.8531311379048329,
|
| 6 |
+
"val_auc": 0.877979797979798,
|
| 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 |
+
29763
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8486028931053873,
|
| 18 |
+
"test_acc": 0.8487282463186078,
|
| 19 |
+
"test_auc": 0.9487785121360266,
|
| 20 |
+
"val_auc": 0.9775757575757575,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "154_athlete_sport_football",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
29763,
|
| 28 |
+
13989,
|
| 29 |
+
16062,
|
| 30 |
+
27248,
|
| 31 |
+
27501,
|
| 32 |
+
15619,
|
| 33 |
+
14278,
|
| 34 |
+
11157,
|
| 35 |
+
12795,
|
| 36 |
+
6582,
|
| 37 |
+
31377,
|
| 38 |
+
10505,
|
| 39 |
+
11299,
|
| 40 |
+
15820,
|
| 41 |
+
21943,
|
| 42 |
+
7334
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/155_athlete_sport_basketball_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8714769575016731,
|
| 4 |
+
"test_acc": 0.8728246318607764,
|
| 5 |
+
"test_auc": 0.8790304081661912,
|
| 6 |
+
"val_auc": 0.8727272727272726,
|
| 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 |
+
13989
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.879303516142668,
|
| 18 |
+
"test_acc": 0.8795180722891566,
|
| 19 |
+
"test_auc": 0.9634987312009863,
|
| 20 |
+
"val_auc": 0.9543434343434344,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "155_athlete_sport_basketball",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
13989,
|
| 28 |
+
27248,
|
| 29 |
+
29763,
|
| 30 |
+
16062,
|
| 31 |
+
30947,
|
| 32 |
+
27007,
|
| 33 |
+
29846,
|
| 34 |
+
15619,
|
| 35 |
+
30233,
|
| 36 |
+
25146,
|
| 37 |
+
4693,
|
| 38 |
+
21943,
|
| 39 |
+
18134,
|
| 40 |
+
31663,
|
| 41 |
+
9621,
|
| 42 |
+
21100
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/156_athlete_sport_baseball_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8726476607566184,
|
| 4 |
+
"test_acc": 0.8741633199464525,
|
| 5 |
+
"test_auc": 0.8798941950653036,
|
| 6 |
+
"val_auc": 0.8573737373737375,
|
| 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 |
+
16062
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8860988332655131,
|
| 18 |
+
"test_acc": 0.8862115127175368,
|
| 19 |
+
"test_auc": 0.9682800246591446,
|
| 20 |
+
"val_auc": 0.9535353535353536,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "156_athlete_sport_baseball",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
16062,
|
| 28 |
+
27007,
|
| 29 |
+
13989,
|
| 30 |
+
29763,
|
| 31 |
+
21077,
|
| 32 |
+
27501,
|
| 33 |
+
20711,
|
| 34 |
+
11299,
|
| 35 |
+
1441,
|
| 36 |
+
30233,
|
| 37 |
+
9819,
|
| 38 |
+
23892,
|
| 39 |
+
32396,
|
| 40 |
+
26059,
|
| 41 |
+
20773,
|
| 42 |
+
18533
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/157_amazon_5star_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8234545361975205,
|
| 4 |
+
"test_acc": 0.8256410256410256,
|
| 5 |
+
"test_auc": 0.8396905611472043,
|
| 6 |
+
"val_auc": 0.7935536515707874,
|
| 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 |
+
17751
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8386008458683296,
|
| 18 |
+
"test_acc": 0.841025641025641,
|
| 19 |
+
"test_auc": 0.9051380987646009,
|
| 20 |
+
"val_auc": 0.8751529987760098,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "157_amazon_5star",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
17751,
|
| 28 |
+
30356,
|
| 29 |
+
23556,
|
| 30 |
+
8957,
|
| 31 |
+
15162,
|
| 32 |
+
12089,
|
| 33 |
+
11988,
|
| 34 |
+
19102,
|
| 35 |
+
5235,
|
| 36 |
+
127,
|
| 37 |
+
5756,
|
| 38 |
+
24164,
|
| 39 |
+
19523,
|
| 40 |
+
19889,
|
| 41 |
+
15498,
|
| 42 |
+
12220
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/158_code_C_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7595409720006449,
|
| 4 |
+
"test_acc": 0.7714285714285715,
|
| 5 |
+
"test_auc": 0.7725706119704102,
|
| 6 |
+
"val_auc": 0.7924528301886793,
|
| 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 |
+
22058
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8741916371157472,
|
| 18 |
+
"test_acc": 0.8752941176470588,
|
| 19 |
+
"test_auc": 0.9591405080590929,
|
| 20 |
+
"val_auc": 0.9875551987153753,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "158_code_C",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
22058,
|
| 28 |
+
30738,
|
| 29 |
+
25759,
|
| 30 |
+
7392,
|
| 31 |
+
18945,
|
| 32 |
+
635,
|
| 33 |
+
28042,
|
| 34 |
+
30234,
|
| 35 |
+
31904,
|
| 36 |
+
4233,
|
| 37 |
+
23139,
|
| 38 |
+
10013,
|
| 39 |
+
9535,
|
| 40 |
+
26615,
|
| 41 |
+
32170,
|
| 42 |
+
24872
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/159_code_Python_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7127230137843076,
|
| 4 |
+
"test_acc": 0.7176470588235294,
|
| 5 |
+
"test_auc": 0.7460904451482745,
|
| 6 |
+
"val_auc": 0.7671617824167001,
|
| 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 |
+
6677
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8663129179880891,
|
| 18 |
+
"test_acc": 0.8665546218487395,
|
| 19 |
+
"test_auc": 0.9625423021020889,
|
| 20 |
+
"val_auc": 0.9841429144921718,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "159_code_Python",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
6677,
|
| 28 |
+
16223,
|
| 29 |
+
26651,
|
| 30 |
+
15935,
|
| 31 |
+
1222,
|
| 32 |
+
24872,
|
| 33 |
+
21931,
|
| 34 |
+
2110,
|
| 35 |
+
18945,
|
| 36 |
+
9955,
|
| 37 |
+
31403,
|
| 38 |
+
18229,
|
| 39 |
+
24026,
|
| 40 |
+
30738,
|
| 41 |
+
27822,
|
| 42 |
+
23066
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/160_code_HTML_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6629154013263857,
|
| 4 |
+
"test_acc": 0.666218487394958,
|
| 5 |
+
"test_auc": 0.6993382613474486,
|
| 6 |
+
"val_auc": 0.6190285026093938,
|
| 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 |
+
24872
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.826104098041934,
|
| 18 |
+
"test_acc": 0.8275630252100841,
|
| 19 |
+
"test_auc": 0.9291923371730626,
|
| 20 |
+
"val_auc": 0.9261340826977118,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "160_code_HTML",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
24872,
|
| 28 |
+
4237,
|
| 29 |
+
31403,
|
| 30 |
+
22812,
|
| 31 |
+
4539,
|
| 32 |
+
12598,
|
| 33 |
+
26615,
|
| 34 |
+
21511,
|
| 35 |
+
959,
|
| 36 |
+
27498,
|
| 37 |
+
20975,
|
| 38 |
+
23491,
|
| 39 |
+
27822,
|
| 40 |
+
4906,
|
| 41 |
+
18662,
|
| 42 |
+
2514
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/161_agnews_0_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9248064995269547,
|
| 4 |
+
"test_acc": 0.9250420168067227,
|
| 5 |
+
"test_auc": 0.9371305345250234,
|
| 6 |
+
"val_auc": 0.9614612605379366,
|
| 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 |
+
32584
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9915963139264412,
|
| 18 |
+
"test_acc": 0.9915966386554622,
|
| 19 |
+
"test_auc": 0.9989939692387791,
|
| 20 |
+
"val_auc": 0.9995985547972701,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "161_agnews_0",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
32584,
|
| 28 |
+
5359,
|
| 29 |
+
21773,
|
| 30 |
+
6390,
|
| 31 |
+
25516,
|
| 32 |
+
20998,
|
| 33 |
+
9040,
|
| 34 |
+
17197,
|
| 35 |
+
24881,
|
| 36 |
+
22234,
|
| 37 |
+
3176,
|
| 38 |
+
12419,
|
| 39 |
+
5275,
|
| 40 |
+
18365,
|
| 41 |
+
23029,
|
| 42 |
+
30820
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/162_agnews_1_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9730900281614017,
|
| 4 |
+
"test_acc": 0.973109243697479,
|
| 5 |
+
"test_auc": 0.9731182795698925,
|
| 6 |
+
"val_auc": 0.9811320754716981,
|
| 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 |
+
21868
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9915963803934297,
|
| 18 |
+
"test_acc": 0.9915966386554622,
|
| 19 |
+
"test_auc": 0.9992371159366842,
|
| 20 |
+
"val_auc": 1.0,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "162_agnews_1",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
21868,
|
| 28 |
+
32584,
|
| 29 |
+
21773,
|
| 30 |
+
25516,
|
| 31 |
+
5359,
|
| 32 |
+
9040,
|
| 33 |
+
6390,
|
| 34 |
+
22234,
|
| 35 |
+
24881,
|
| 36 |
+
12419,
|
| 37 |
+
18365,
|
| 38 |
+
1071,
|
| 39 |
+
23029,
|
| 40 |
+
17197,
|
| 41 |
+
8540,
|
| 42 |
+
14167
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/163_agnews_2_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.915714465988716,
|
| 4 |
+
"test_acc": 0.9163025210084034,
|
| 5 |
+
"test_auc": 0.9163306451612903,
|
| 6 |
+
"val_auc": 0.9433962264150944,
|
| 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 |
+
6390
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.97882068178328,
|
| 18 |
+
"test_acc": 0.9788235294117648,
|
| 19 |
+
"test_auc": 0.9976923660975768,
|
| 20 |
+
"val_auc": 0.9927739863508631,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "163_agnews_2",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
6390,
|
| 28 |
+
21773,
|
| 29 |
+
5359,
|
| 30 |
+
32584,
|
| 31 |
+
25516,
|
| 32 |
+
20998,
|
| 33 |
+
24881,
|
| 34 |
+
9040,
|
| 35 |
+
22234,
|
| 36 |
+
17772,
|
| 37 |
+
12419,
|
| 38 |
+
3176,
|
| 39 |
+
18365,
|
| 40 |
+
23183,
|
| 41 |
+
23029,
|
| 42 |
+
17197
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/21_headline_istrump_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8231359081587879,
|
| 4 |
+
"test_acc": 0.8233962264150944,
|
| 5 |
+
"test_auc": 0.8942948075974719,
|
| 6 |
+
"val_auc": 0.9423769507803121,
|
| 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 |
+
11555
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8784443611077933,
|
| 18 |
+
"test_acc": 0.8784905660377359,
|
| 19 |
+
"test_auc": 0.9449188842162183,
|
| 20 |
+
"val_auc": 0.9635854341736695,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "21_headline_istrump",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
11555,
|
| 28 |
+
23152,
|
| 29 |
+
11260,
|
| 30 |
+
15997,
|
| 31 |
+
32389,
|
| 32 |
+
5038,
|
| 33 |
+
19025,
|
| 34 |
+
3936,
|
| 35 |
+
29979,
|
| 36 |
+
18628,
|
| 37 |
+
3083,
|
| 38 |
+
24736,
|
| 39 |
+
1402,
|
| 40 |
+
4354,
|
| 41 |
+
1580,
|
| 42 |
+
26747
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/22_headline_isobama_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.8002718118283132,
|
| 4 |
+
"test_acc": 0.8003169572107766,
|
| 5 |
+
"test_auc": 0.8303345388788426,
|
| 6 |
+
"val_auc": 0.7971188475390156,
|
| 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 |
+
11260
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8621215355933128,
|
| 18 |
+
"test_acc": 0.8621236133122029,
|
| 19 |
+
"test_auc": 0.9320474181233674,
|
| 20 |
+
"val_auc": 0.9591836734693877,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "22_headline_isobama",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
11260,
|
| 28 |
+
11555,
|
| 29 |
+
26747,
|
| 30 |
+
4354,
|
| 31 |
+
29979,
|
| 32 |
+
10850,
|
| 33 |
+
25722,
|
| 34 |
+
24736,
|
| 35 |
+
23152,
|
| 36 |
+
27103,
|
| 37 |
+
18627,
|
| 38 |
+
15997,
|
| 39 |
+
13944,
|
| 40 |
+
27678,
|
| 41 |
+
14562,
|
| 42 |
+
7246
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/23_headline_ischina_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6923170002149143,
|
| 4 |
+
"test_acc": 0.6926770708283313,
|
| 5 |
+
"test_auc": 0.7112675244419848,
|
| 6 |
+
"val_auc": 0.7711352657004831,
|
| 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 |
+
29979
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7646957091438839,
|
| 18 |
+
"test_acc": 0.7647058823529411,
|
| 19 |
+
"test_auc": 0.854103255856853,
|
| 20 |
+
"val_auc": 0.8804347826086958,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "23_headline_ischina",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
29979,
|
| 28 |
+
23020,
|
| 29 |
+
30970,
|
| 30 |
+
32480,
|
| 31 |
+
25829,
|
| 32 |
+
26612,
|
| 33 |
+
21943,
|
| 34 |
+
30233,
|
| 35 |
+
23152,
|
| 36 |
+
10850,
|
| 37 |
+
1857,
|
| 38 |
+
3357,
|
| 39 |
+
722,
|
| 40 |
+
19499,
|
| 41 |
+
29207,
|
| 42 |
+
27003
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/24_headline_isiran_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7071403181373855,
|
| 4 |
+
"test_acc": 0.7134986225895317,
|
| 5 |
+
"test_auc": 0.7259729221055188,
|
| 6 |
+
"val_auc": 0.7018807523009204,
|
| 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 |
+
23020
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7654664844845479,
|
| 18 |
+
"test_acc": 0.7658402203856749,
|
| 19 |
+
"test_auc": 0.8256632869892538,
|
| 20 |
+
"val_auc": 0.7975190076030413,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "24_headline_isiran",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
23020,
|
| 28 |
+
29979,
|
| 29 |
+
11651,
|
| 30 |
+
3936,
|
| 31 |
+
30964,
|
| 32 |
+
21943,
|
| 33 |
+
9793,
|
| 34 |
+
7327,
|
| 35 |
+
17047,
|
| 36 |
+
5359,
|
| 37 |
+
29357,
|
| 38 |
+
26814,
|
| 39 |
+
24881,
|
| 40 |
+
22210,
|
| 41 |
+
2278,
|
| 42 |
+
3407
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/26_headline_isfrontpage_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5988520484884389,
|
| 4 |
+
"test_acc": 0.6017610062893082,
|
| 5 |
+
"test_auc": 0.6228879315146036,
|
| 6 |
+
"val_auc": 0.7010804321728691,
|
| 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 |
+
11555
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6792167384296215,
|
| 18 |
+
"test_acc": 0.6792452830188679,
|
| 19 |
+
"test_auc": 0.7452262644817066,
|
| 20 |
+
"val_auc": 0.7338935574229691,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "26_headline_isfrontpage",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
11555,
|
| 28 |
+
29979,
|
| 29 |
+
32389,
|
| 30 |
+
10850,
|
| 31 |
+
3936,
|
| 32 |
+
12677,
|
| 33 |
+
17339,
|
| 34 |
+
24479,
|
| 35 |
+
22210,
|
| 36 |
+
18628,
|
| 37 |
+
13032,
|
| 38 |
+
6390,
|
| 39 |
+
25829,
|
| 40 |
+
2512,
|
| 41 |
+
17993,
|
| 42 |
+
8626
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/36_sciq_tf_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.7144343047433973,
|
| 4 |
+
"test_acc": 0.7172327044025157,
|
| 5 |
+
"test_auc": 0.7580647447847629,
|
| 6 |
+
"val_auc": 0.8321328531412565,
|
| 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 |
+
17496
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8382230213851963,
|
| 18 |
+
"test_acc": 0.8382389937106918,
|
| 19 |
+
"test_auc": 0.9159843307454187,
|
| 20 |
+
"val_auc": 0.961984793917567,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "36_sciq_tf",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
17496,
|
| 28 |
+
255,
|
| 29 |
+
17249,
|
| 30 |
+
8230,
|
| 31 |
+
23543,
|
| 32 |
+
30234,
|
| 33 |
+
14202,
|
| 34 |
+
22924,
|
| 35 |
+
16324,
|
| 36 |
+
10819,
|
| 37 |
+
4973,
|
| 38 |
+
30610,
|
| 39 |
+
14836,
|
| 40 |
+
7344,
|
| 41 |
+
9917,
|
| 42 |
+
31154
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/41_truthqa_tf_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5356125356125355,
|
| 4 |
+
"test_acc": 0.5555555555555556,
|
| 5 |
+
"test_auc": 0.5485714285714286,
|
| 6 |
+
"val_auc": 0.5845042151746287,
|
| 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 |
+
12798
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5813216914911831,
|
| 18 |
+
"test_acc": 0.5959595959595959,
|
| 19 |
+
"test_auc": 0.6946938775510204,
|
| 20 |
+
"val_auc": 0.706945002007226,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "41_truthqa_tf",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
12798,
|
| 28 |
+
16780,
|
| 29 |
+
13664,
|
| 30 |
+
20849,
|
| 31 |
+
127,
|
| 32 |
+
21201,
|
| 33 |
+
5267,
|
| 34 |
+
32328,
|
| 35 |
+
26239,
|
| 36 |
+
31006,
|
| 37 |
+
9177,
|
| 38 |
+
8230,
|
| 39 |
+
29508,
|
| 40 |
+
18556,
|
| 41 |
+
25612,
|
| 42 |
+
12220
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/42_temp_sense_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6657650207365646,
|
| 4 |
+
"test_acc": 0.6697247706422018,
|
| 5 |
+
"test_auc": 0.6891360601001668,
|
| 6 |
+
"val_auc": 0.7962962962962963,
|
| 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 |
+
24909
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7803938082095264,
|
| 18 |
+
"test_acc": 0.7814845704753962,
|
| 19 |
+
"test_auc": 0.8533333333333333,
|
| 20 |
+
"val_auc": 0.9066022544283414,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "42_temp_sense",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
24909,
|
| 28 |
+
30082,
|
| 29 |
+
22794,
|
| 30 |
+
23979,
|
| 31 |
+
23088,
|
| 32 |
+
23543,
|
| 33 |
+
4973,
|
| 34 |
+
18235,
|
| 35 |
+
457,
|
| 36 |
+
24154,
|
| 37 |
+
17249,
|
| 38 |
+
16852,
|
| 39 |
+
27240,
|
| 40 |
+
4878,
|
| 41 |
+
18249,
|
| 42 |
+
30124
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/44_phys_tf_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5826861550393373,
|
| 4 |
+
"test_acc": 0.5841509433962264,
|
| 5 |
+
"test_auc": 0.601480549122617,
|
| 6 |
+
"val_auc": 0.5418167266906763,
|
| 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 |
+
23979
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6369732744749849,
|
| 18 |
+
"test_acc": 0.6369811320754717,
|
| 19 |
+
"test_auc": 0.6962418699413391,
|
| 20 |
+
"val_auc": 0.6676670668267307,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "44_phys_tf",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
23979,
|
| 28 |
+
24909,
|
| 29 |
+
30082,
|
| 30 |
+
27235,
|
| 31 |
+
457,
|
| 32 |
+
12544,
|
| 33 |
+
31824,
|
| 34 |
+
18709,
|
| 35 |
+
16703,
|
| 36 |
+
22424,
|
| 37 |
+
20486,
|
| 38 |
+
4114,
|
| 39 |
+
20052,
|
| 40 |
+
25612,
|
| 41 |
+
3083,
|
| 42 |
+
1255
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/47_reasoning_tf_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5356070880753162,
|
| 4 |
+
"test_acc": 0.535607420706164,
|
| 5 |
+
"test_auc": 0.5466077414548893,
|
| 6 |
+
"val_auc": 0.5240867121637897,
|
| 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 |
+
27235
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.5696560100861678,
|
| 18 |
+
"test_acc": 0.5703171753441053,
|
| 19 |
+
"test_auc": 0.5825337363550411,
|
| 20 |
+
"val_auc": 0.5411481332798074,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "47_reasoning_tf",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
27235,
|
| 28 |
+
30464,
|
| 29 |
+
30057,
|
| 30 |
+
26156,
|
| 31 |
+
24559,
|
| 32 |
+
1410,
|
| 33 |
+
7172,
|
| 34 |
+
1895,
|
| 35 |
+
12794,
|
| 36 |
+
13063,
|
| 37 |
+
5127,
|
| 38 |
+
29827,
|
| 39 |
+
127,
|
| 40 |
+
1453,
|
| 41 |
+
19800,
|
| 42 |
+
7461
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/48_cm_correct_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5858317096150064,
|
| 4 |
+
"test_acc": 0.610062893081761,
|
| 5 |
+
"test_auc": 0.6206028318881585,
|
| 6 |
+
"val_auc": 0.6690676270508203,
|
| 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 |
+
11687
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6393126804450052,
|
| 18 |
+
"test_acc": 0.6417610062893082,
|
| 19 |
+
"test_auc": 0.7215685912151317,
|
| 20 |
+
"val_auc": 0.771908763505402,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "48_cm_correct",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
11687,
|
| 28 |
+
22424,
|
| 29 |
+
15374,
|
| 30 |
+
23979,
|
| 31 |
+
21092,
|
| 32 |
+
32275,
|
| 33 |
+
21472,
|
| 34 |
+
24909,
|
| 35 |
+
14380,
|
| 36 |
+
24164,
|
| 37 |
+
22794,
|
| 38 |
+
12544,
|
| 39 |
+
25612,
|
| 40 |
+
22630,
|
| 41 |
+
21140,
|
| 42 |
+
32661
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/49_cm_isshort_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9984905620254628,
|
| 4 |
+
"test_acc": 0.9984905660377359,
|
| 5 |
+
"test_auc": 0.9999954432179388,
|
| 6 |
+
"val_auc": 1.0,
|
| 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 |
+
5008
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.9992452824456872,
|
| 18 |
+
"test_acc": 0.999245283018868,
|
| 19 |
+
"test_auc": 1.0,
|
| 20 |
+
"val_auc": 1.0,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "49_cm_isshort",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
5008,
|
| 28 |
+
12220,
|
| 29 |
+
17171,
|
| 30 |
+
25342,
|
| 31 |
+
23877,
|
| 32 |
+
32584,
|
| 33 |
+
31617,
|
| 34 |
+
27003,
|
| 35 |
+
28521,
|
| 36 |
+
1367,
|
| 37 |
+
17120,
|
| 38 |
+
18709,
|
| 39 |
+
6197,
|
| 40 |
+
25735,
|
| 41 |
+
23088,
|
| 42 |
+
8235
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/50_deon_isvalid_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6111491740123213,
|
| 4 |
+
"test_acc": 0.6113207547169811,
|
| 5 |
+
"test_auc": 0.6419726208281396,
|
| 6 |
+
"val_auc": 0.7076830732292917,
|
| 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 |
+
12544
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.6982861632112024,
|
| 18 |
+
"test_acc": 0.6986163522012578,
|
| 19 |
+
"test_auc": 0.7830348978622617,
|
| 20 |
+
"val_auc": 0.8303321328531412,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "50_deon_isvalid",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
12544,
|
| 28 |
+
24909,
|
| 29 |
+
23979,
|
| 30 |
+
19889,
|
| 31 |
+
30234,
|
| 32 |
+
17120,
|
| 33 |
+
5267,
|
| 34 |
+
21005,
|
| 35 |
+
30970,
|
| 36 |
+
21472,
|
| 37 |
+
12220,
|
| 38 |
+
27975,
|
| 39 |
+
21870,
|
| 40 |
+
10364,
|
| 41 |
+
27240,
|
| 42 |
+
15374
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/51_just_is_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.6551848973526271,
|
| 4 |
+
"test_acc": 0.6555974842767296,
|
| 5 |
+
"test_auc": 0.6999966583598218,
|
| 6 |
+
"val_auc": 0.7034813925570228,
|
| 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 |
+
24909
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.7206284736566576,
|
| 18 |
+
"test_acc": 0.720754716981132,
|
| 19 |
+
"test_auc": 0.8060528495583466,
|
| 20 |
+
"val_auc": 0.8227290916366546,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "51_just_is",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
24909,
|
| 28 |
+
12544,
|
| 29 |
+
23979,
|
| 30 |
+
21472,
|
| 31 |
+
19889,
|
| 32 |
+
21005,
|
| 33 |
+
30970,
|
| 34 |
+
457,
|
| 35 |
+
10364,
|
| 36 |
+
30234,
|
| 37 |
+
22424,
|
| 38 |
+
27975,
|
| 39 |
+
16139,
|
| 40 |
+
24219,
|
| 41 |
+
21140,
|
| 42 |
+
18267
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/52_virtue_is_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.699315306263416,
|
| 4 |
+
"test_acc": 0.6993429697766097,
|
| 5 |
+
"test_auc": 0.7378692009351552,
|
| 6 |
+
"val_auc": 0.6656,
|
| 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 |
+
11020
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.8196791468591011,
|
| 18 |
+
"test_acc": 0.8197109067017083,
|
| 19 |
+
"test_auc": 0.9041872012368539,
|
| 20 |
+
"val_auc": 0.9032,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "52_virtue_is",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
11020,
|
| 28 |
+
7868,
|
| 29 |
+
32715,
|
| 30 |
+
127,
|
| 31 |
+
17134,
|
| 32 |
+
2130,
|
| 33 |
+
2340,
|
| 34 |
+
21820,
|
| 35 |
+
6007,
|
| 36 |
+
24016,
|
| 37 |
+
28195,
|
| 38 |
+
9566,
|
| 39 |
+
14836,
|
| 40 |
+
32633,
|
| 41 |
+
610,
|
| 42 |
+
25886
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/54_cs_tf_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.5285309011491846,
|
| 4 |
+
"test_acc": 0.5285534591194968,
|
| 5 |
+
"test_auc": 0.5387316349025203,
|
| 6 |
+
"val_auc": 0.6274509803921569,
|
| 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 |
+
32584
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.539064303597846,
|
| 18 |
+
"test_acc": 0.5391194968553459,
|
| 19 |
+
"test_auc": 0.5801959466917257,
|
| 20 |
+
"val_auc": 0.663065226090436,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "54_cs_tf",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
32584,
|
| 28 |
+
25088,
|
| 29 |
+
28740,
|
| 30 |
+
21773,
|
| 31 |
+
16223,
|
| 32 |
+
32531,
|
| 33 |
+
25735,
|
| 34 |
+
9396,
|
| 35 |
+
32275,
|
| 36 |
+
5793,
|
| 37 |
+
21687,
|
| 38 |
+
31617,
|
| 39 |
+
26810,
|
| 40 |
+
23712,
|
| 41 |
+
31720,
|
| 42 |
+
3055
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|
denoised/gemma-2-2b-btk-softmax_topp_zero-p0.8/k-200/seed-0/model.layers.12/w-32768/t-300M/l0-193.4/denoised/sae_probes/sae_probes_gemma-2-2b--softmax_topp_zero-p0.8/normal_setting/56_wikidatasex_or_gender_blocks.12.hook_resid_post_l1.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"test_f1": 0.9901882097874898,
|
| 4 |
+
"test_acc": 0.990188679245283,
|
| 5 |
+
"test_auc": 0.9977550253711499,
|
| 6 |
+
"val_auc": 1.0,
|
| 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 |
+
818
|
| 14 |
+
]
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"test_f1": 0.989685498034894,
|
| 18 |
+
"test_acc": 0.989685534591195,
|
| 19 |
+
"test_auc": 0.9991337050992416,
|
| 20 |
+
"val_auc": 1.0,
|
| 21 |
+
"k": 16,
|
| 22 |
+
"dataset": "56_wikidatasex_or_gender",
|
| 23 |
+
"hook_name": "blocks.12.hook_resid_post",
|
| 24 |
+
"reg_type": "l1",
|
| 25 |
+
"binarize": false,
|
| 26 |
+
"indices": [
|
| 27 |
+
818,
|
| 28 |
+
13463,
|
| 29 |
+
21870,
|
| 30 |
+
26369,
|
| 31 |
+
842,
|
| 32 |
+
733,
|
| 33 |
+
4973,
|
| 34 |
+
16223,
|
| 35 |
+
30356,
|
| 36 |
+
22554,
|
| 37 |
+
2249,
|
| 38 |
+
12774,
|
| 39 |
+
9481,
|
| 40 |
+
30942,
|
| 41 |
+
23991,
|
| 42 |
+
10172
|
| 43 |
+
]
|
| 44 |
+
}
|
| 45 |
+
]
|