Add files using upload-large-folder tool
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/code_metrics_step_final.json +136 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/dict_metrics_step_final.json +545 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed0/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed1/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/code_metrics_step_final.json +136 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/dict_metrics_step_final.json +545 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed2/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed2/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_64/code_metrics_step_final.json +136 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_64/dict_metrics_step_final.json +545 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed0/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed0/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed1/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed1/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed2/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed2/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_128/code_metrics_step_final.json +136 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_128/dict_metrics_step_final.json +545 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed0/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed0/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed1/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed1/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed2/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed2/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_32/code_metrics_step_final.json +136 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_32/dict_metrics_step_final.json +545 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed0/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed0/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed1/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed1/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed2/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed2/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_64/code_metrics_step_final.json +136 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_64/dict_metrics_step_final.json +545 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed0/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed0/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed1/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed1/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed2/config.json +12 -0
- vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed2/stats.imagenet.split_val.seed_no/metrics.json +15 -0
- vision/dinov2-b_layer10/batchtopk/x8/l0_128/code_metrics_step_final.json +136 -0
- vision/dinov2-b_layer10/batchtopk/x8/l0_128/dict_metrics_step_final.json +545 -0
- vision/dinov2-b_layer10/batchtopk/x8/l0_32/code_metrics_step_final.json +136 -0
- vision/dinov2-b_layer10/batchtopk/x8/l0_32/dict_metrics_step_final.json +545 -0
vision/dinov2-b_layer10/batchtopk/x16/l0_128/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 771000,
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"l0": 128.1538625162127,
|
| 20 |
+
"recon_r2": 0.9792289382208147,
|
| 21 |
+
"l1_per_token": 162.50409510279087
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 127.6798495460441,
|
| 25 |
+
"recon_r2": 0.9791598322876908,
|
| 26 |
+
"l1_per_token": 162.82304904433195
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 127.74862645914396,
|
| 30 |
+
"recon_r2": 0.9792085799181947,
|
| 31 |
+
"l1_per_token": 162.2423636379021
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.4383927214980303,
|
| 38 |
+
"mcc_abs": 0.43846079941761795,
|
| 39 |
+
"act_iou": 0.3078236567819484,
|
| 40 |
+
"act_iou_n_pairs": 12288,
|
| 41 |
+
"input_r2": 0.9908053203685809,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.9224367655692263,
|
| 44 |
+
"0.7": 0.948947476349449,
|
| 45 |
+
"0.8": 0.9669434511045226,
|
| 46 |
+
"0.9": 0.9793213312153681
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.9223180852693832,
|
| 50 |
+
"0.7": 0.9488004815301805,
|
| 51 |
+
"0.8": 0.9670508834416345,
|
| 52 |
+
"0.9": 0.9792289268871128
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.43886298781297234,
|
| 58 |
+
"mcc_abs": 0.438927486110604,
|
| 59 |
+
"act_iou": 0.30667207757774845,
|
| 60 |
+
"act_iou_n_pairs": 12288,
|
| 61 |
+
"input_r2": 0.9906160842156211,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.9202458094872127,
|
| 64 |
+
"0.7": 0.9465365631113593,
|
| 65 |
+
"0.8": 0.9650861535100407,
|
| 66 |
+
"0.9": 0.9792062832332554
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.9206007805549167,
|
| 70 |
+
"0.7": 0.9463517434874102,
|
| 71 |
+
"0.8": 0.9650725378646438,
|
| 72 |
+
"0.9": 0.9793397613945602
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.4373350937108221,
|
| 78 |
+
"mcc_abs": 0.43740546804447816,
|
| 79 |
+
"act_iou": 0.3077153714829926,
|
| 80 |
+
"act_iou_n_pairs": 12288,
|
| 81 |
+
"input_r2": 0.9909636343148913,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.9234180827963175,
|
| 84 |
+
"0.7": 0.9473867409230992,
|
| 85 |
+
"0.8": 0.9659999638660963,
|
| 86 |
+
"0.9": 0.9781258235592208
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.9237045008569276,
|
| 90 |
+
"0.7": 0.9476299767750603,
|
| 91 |
+
"0.8": 0.9657021444006431,
|
| 92 |
+
"0.9": 0.9780212498613615
|
| 93 |
+
}
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"summary": {
|
| 97 |
+
"n_seeds": 3,
|
| 98 |
+
"n_pairs": 3,
|
| 99 |
+
"report_band": [
|
| 100 |
+
0.6,
|
| 101 |
+
0.7,
|
| 102 |
+
0.8,
|
| 103 |
+
0.9
|
| 104 |
+
],
|
| 105 |
+
"l0_mean": 127.86077950713359,
|
| 106 |
+
"l0_std": 0.2561362988762837,
|
| 107 |
+
"recon_r2_mean": 0.9791991168089001,
|
| 108 |
+
"recon_r2_std": 3.551155201302099e-05,
|
| 109 |
+
"l1_per_token_mean": 162.52316926167495,
|
| 110 |
+
"l1_per_token_std": 0.29081222973425,
|
| 111 |
+
"input_r2_mean": 0.9907950129663644,
|
| 112 |
+
"input_r2_std": 0.0001740041659875606,
|
| 113 |
+
"mcc_signed_mean": 0.43819693434060825,
|
| 114 |
+
"mcc_signed_std": 0.000782537254771268,
|
| 115 |
+
"mcc_abs_mean": 0.4382645845242334,
|
| 116 |
+
"mcc_abs_std": 0.0007797499353592809,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.9220335526175855,
|
| 119 |
+
"0.7": 0.9476235934613024,
|
| 120 |
+
"0.8": 0.9660098561602198,
|
| 121 |
+
"0.9": 0.9788844793359481
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.9536378703937639,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.9222077888937426,
|
| 126 |
+
"0.7": 0.947594067264217,
|
| 127 |
+
"0.8": 0.9659418552356405,
|
| 128 |
+
"0.9": 0.9788633127143448
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.9536517560269862,
|
| 131 |
+
"act_iou_mean": 0.3074037019475631,
|
| 132 |
+
"act_iou_std": 0.0006359143734596319,
|
| 133 |
+
"act_iou_n_pairs": 12288.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,545 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 2,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"linkage": "complete",
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"d_sae": 12288,
|
| 20 |
+
"d_in": 768,
|
| 21 |
+
"l0": 128,
|
| 22 |
+
"l0_raw": 127.7952,
|
| 23 |
+
"recon_r2": 0.9792,
|
| 24 |
+
"l1_per_token": 162.512,
|
| 25 |
+
"stats_source": "stats",
|
| 26 |
+
"n_dead": 0,
|
| 27 |
+
"dead_rate": 0.0,
|
| 28 |
+
"alive_source": "stats",
|
| 29 |
+
"n_dense": 135,
|
| 30 |
+
"dense_rate": 0.010986328125,
|
| 31 |
+
"dense_threshold": 0.1,
|
| 32 |
+
"n_dup_signed": {
|
| 33 |
+
"0.5": 116,
|
| 34 |
+
"0.6": 45,
|
| 35 |
+
"0.7": 30,
|
| 36 |
+
"0.8": 20,
|
| 37 |
+
"0.9": 2
|
| 38 |
+
},
|
| 39 |
+
"n_dup_clusters_signed": {
|
| 40 |
+
"0.5": 56,
|
| 41 |
+
"0.6": 22,
|
| 42 |
+
"0.7": 15,
|
| 43 |
+
"0.8": 10,
|
| 44 |
+
"0.9": 1
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 3,
|
| 48 |
+
"0.6": 3,
|
| 49 |
+
"0.7": 2,
|
| 50 |
+
"0.8": 2,
|
| 51 |
+
"0.9": 2
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.18705104291439056,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 314,
|
| 56 |
+
"0.6": 217,
|
| 57 |
+
"0.7": 168,
|
| 58 |
+
"0.8": 142,
|
| 59 |
+
"0.9": 108
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 155,
|
| 63 |
+
"0.6": 108,
|
| 64 |
+
"0.7": 84,
|
| 65 |
+
"0.8": 71,
|
| 66 |
+
"0.9": 54
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 3,
|
| 70 |
+
"0.6": 3,
|
| 71 |
+
"0.7": 2,
|
| 72 |
+
"0.8": 2,
|
| 73 |
+
"0.9": 2
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.2123843878507614
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 12288,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 128,
|
| 81 |
+
"l0_raw": 127.7952,
|
| 82 |
+
"recon_r2": 0.9791,
|
| 83 |
+
"l1_per_token": 162.8207,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 138,
|
| 89 |
+
"dense_rate": 0.01123046875,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 114,
|
| 93 |
+
"0.6": 52,
|
| 94 |
+
"0.7": 46,
|
| 95 |
+
"0.8": 30,
|
| 96 |
+
"0.9": 4
|
| 97 |
+
},
|
| 98 |
+
"n_dup_clusters_signed": {
|
| 99 |
+
"0.5": 57,
|
| 100 |
+
"0.6": 26,
|
| 101 |
+
"0.7": 23,
|
| 102 |
+
"0.8": 15,
|
| 103 |
+
"0.9": 2
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 2,
|
| 107 |
+
"0.6": 2,
|
| 108 |
+
"0.7": 2,
|
| 109 |
+
"0.8": 2,
|
| 110 |
+
"0.9": 2
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.185121551156044,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 317,
|
| 115 |
+
"0.6": 225,
|
| 116 |
+
"0.7": 192,
|
| 117 |
+
"0.8": 162,
|
| 118 |
+
"0.9": 114
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 158,
|
| 122 |
+
"0.6": 112,
|
| 123 |
+
"0.7": 96,
|
| 124 |
+
"0.8": 81,
|
| 125 |
+
"0.9": 57
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 3,
|
| 129 |
+
"0.6": 3,
|
| 130 |
+
"0.7": 2,
|
| 131 |
+
"0.8": 2,
|
| 132 |
+
"0.9": 2
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.21026749908924103
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 12288,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 128,
|
| 140 |
+
"l0_raw": 127.7952,
|
| 141 |
+
"recon_r2": 0.9792,
|
| 142 |
+
"l1_per_token": 162.2623,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 138,
|
| 148 |
+
"dense_rate": 0.01123046875,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 106,
|
| 152 |
+
"0.6": 42,
|
| 153 |
+
"0.7": 24,
|
| 154 |
+
"0.8": 22,
|
| 155 |
+
"0.9": 2
|
| 156 |
+
},
|
| 157 |
+
"n_dup_clusters_signed": {
|
| 158 |
+
"0.5": 53,
|
| 159 |
+
"0.6": 21,
|
| 160 |
+
"0.7": 12,
|
| 161 |
+
"0.8": 11,
|
| 162 |
+
"0.9": 1
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 2,
|
| 166 |
+
"0.6": 2,
|
| 167 |
+
"0.7": 2,
|
| 168 |
+
"0.8": 2,
|
| 169 |
+
"0.9": 2
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.18652190268039703,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 322,
|
| 174 |
+
"0.6": 218,
|
| 175 |
+
"0.7": 172,
|
| 176 |
+
"0.8": 148,
|
| 177 |
+
"0.9": 118
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 160,
|
| 181 |
+
"0.6": 109,
|
| 182 |
+
"0.7": 86,
|
| 183 |
+
"0.8": 74,
|
| 184 |
+
"0.9": 59
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 3,
|
| 188 |
+
"0.6": 2,
|
| 189 |
+
"0.7": 2,
|
| 190 |
+
"0.8": 2,
|
| 191 |
+
"0.9": 2
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.2124355435371399
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.4814791977405548,
|
| 199 |
+
"dcs_abs": 0.4836738109588623,
|
| 200 |
+
"frac_sign_flipped": 0.14803059895833334,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"n_alive_a": 12288,
|
| 203 |
+
"n_alive_b": 12288,
|
| 204 |
+
"matched_cos_signed": {
|
| 205 |
+
"0.5": 0.4813920557498932,
|
| 206 |
+
"0.6": 0.48155900835990906,
|
| 207 |
+
"0.7": 0.4816385507583618,
|
| 208 |
+
"0.8": 0.48155462741851807,
|
| 209 |
+
"0.9": 0.4814692437648773
|
| 210 |
+
},
|
| 211 |
+
"unique_frac_signed": {
|
| 212 |
+
"0.5": 0.9952392578125,
|
| 213 |
+
"0.6": 0.9980061848958334,
|
| 214 |
+
"0.7": 0.9984537760416666,
|
| 215 |
+
"0.8": 0.9989827473958334,
|
| 216 |
+
"0.9": 0.9998779296875
|
| 217 |
+
},
|
| 218 |
+
"dedup_dcs_signed": {
|
| 219 |
+
"0.5": 0.4791002722813573,
|
| 220 |
+
"0.6": 0.48059886873549357,
|
| 221 |
+
"0.7": 0.4808938296919223,
|
| 222 |
+
"0.8": 0.4810647647197281,
|
| 223 |
+
"0.9": 0.4814104706638318
|
| 224 |
+
},
|
| 225 |
+
"matched_frac_signed": {
|
| 226 |
+
"0.5": 0.4149576822916667,
|
| 227 |
+
"0.6": 0.3750813802083333,
|
| 228 |
+
"0.7": 0.3384602864583333,
|
| 229 |
+
"0.8": 0.3080240885416667,
|
| 230 |
+
"0.9": 0.2722981770833333
|
| 231 |
+
},
|
| 232 |
+
"matched_cos_abs": {
|
| 233 |
+
"0.5": 0.48271414637565613,
|
| 234 |
+
"0.6": 0.4830278754234314,
|
| 235 |
+
"0.7": 0.48328208923339844,
|
| 236 |
+
"0.8": 0.4832307696342468,
|
| 237 |
+
"0.9": 0.48328083753585815
|
| 238 |
+
},
|
| 239 |
+
"unique_frac_abs": {
|
| 240 |
+
"0.5": 0.987060546875,
|
| 241 |
+
"0.6": 0.990966796875,
|
| 242 |
+
"0.7": 0.99267578125,
|
| 243 |
+
"0.8": 0.9938151041666666,
|
| 244 |
+
"0.9": 0.9954833984375
|
| 245 |
+
},
|
| 246 |
+
"dedup_dcs_abs": {
|
| 247 |
+
"0.5": 0.47646808930585394,
|
| 248 |
+
"0.6": 0.47866458650969435,
|
| 249 |
+
"0.7": 0.479742425493896,
|
| 250 |
+
"0.8": 0.4802420376605975,
|
| 251 |
+
"0.9": 0.4810980505499174
|
| 252 |
+
},
|
| 253 |
+
"matched_frac_abs": {
|
| 254 |
+
"0.5": 0.4111328125,
|
| 255 |
+
"0.6": 0.3722330729166667,
|
| 256 |
+
"0.7": 0.3365071614583333,
|
| 257 |
+
"0.8": 0.3065592447916667,
|
| 258 |
+
"0.9": 0.271728515625
|
| 259 |
+
}
|
| 260 |
+
},
|
| 261 |
+
"seed0\u2194seed2": {
|
| 262 |
+
"dcs_signed": 0.48221930861473083,
|
| 263 |
+
"dcs_abs": 0.48423898220062256,
|
| 264 |
+
"frac_sign_flipped": 0.146240234375,
|
| 265 |
+
"alive_source": "stats",
|
| 266 |
+
"n_alive_a": 12288,
|
| 267 |
+
"n_alive_b": 12288,
|
| 268 |
+
"matched_cos_signed": {
|
| 269 |
+
"0.5": 0.4822535216808319,
|
| 270 |
+
"0.6": 0.48218315839767456,
|
| 271 |
+
"0.7": 0.4823228418827057,
|
| 272 |
+
"0.8": 0.4822227656841278,
|
| 273 |
+
"0.9": 0.48224908113479614
|
| 274 |
+
},
|
| 275 |
+
"unique_frac_signed": {
|
| 276 |
+
"0.5": 0.9954020182291666,
|
| 277 |
+
"0.6": 0.9982096354166666,
|
| 278 |
+
"0.7": 0.9989013671875,
|
| 279 |
+
"0.8": 0.9991455078125,
|
| 280 |
+
"0.9": 0.9999186197916666
|
| 281 |
+
},
|
| 282 |
+
"dedup_dcs_signed": {
|
| 283 |
+
"0.5": 0.48003612877922325,
|
| 284 |
+
"0.6": 0.48131987474819954,
|
| 285 |
+
"0.7": 0.4817929461823951,
|
| 286 |
+
"0.8": 0.4818107100982161,
|
| 287 |
+
"0.9": 0.4822098356041048
|
| 288 |
+
},
|
| 289 |
+
"matched_frac_signed": {
|
| 290 |
+
"0.5": 0.4169921875,
|
| 291 |
+
"0.6": 0.3743489583333333,
|
| 292 |
+
"0.7": 0.3388671875,
|
| 293 |
+
"0.8": 0.306884765625,
|
| 294 |
+
"0.9": 0.2711588541666667
|
| 295 |
+
},
|
| 296 |
+
"matched_cos_abs": {
|
| 297 |
+
"0.5": 0.4835225045681,
|
| 298 |
+
"0.6": 0.4834429919719696,
|
| 299 |
+
"0.7": 0.483679860830307,
|
| 300 |
+
"0.8": 0.48371192812919617,
|
| 301 |
+
"0.9": 0.48393094539642334
|
| 302 |
+
},
|
| 303 |
+
"unique_frac_abs": {
|
| 304 |
+
"0.5": 0.9869384765625,
|
| 305 |
+
"0.6": 0.9911295572916666,
|
| 306 |
+
"0.7": 0.9930826822916666,
|
| 307 |
+
"0.8": 0.9940999348958334,
|
| 308 |
+
"0.9": 0.9954020182291666
|
| 309 |
+
},
|
| 310 |
+
"dedup_dcs_abs": {
|
| 311 |
+
"0.5": 0.47720696404212504,
|
| 312 |
+
"0.6": 0.479154638608937,
|
| 313 |
+
"0.7": 0.4803340935638213,
|
| 314 |
+
"0.8": 0.48085799626157194,
|
| 315 |
+
"0.9": 0.48170583973114844
|
| 316 |
+
},
|
| 317 |
+
"matched_frac_abs": {
|
| 318 |
+
"0.5": 0.4131673177083333,
|
| 319 |
+
"0.6": 0.3711751302083333,
|
| 320 |
+
"0.7": 0.336669921875,
|
| 321 |
+
"0.8": 0.305419921875,
|
| 322 |
+
"0.9": 0.2705078125
|
| 323 |
+
}
|
| 324 |
+
},
|
| 325 |
+
"seed1\u2194seed2": {
|
| 326 |
+
"dcs_signed": 0.4816232919692993,
|
| 327 |
+
"dcs_abs": 0.4837757647037506,
|
| 328 |
+
"frac_sign_flipped": 0.14689127604166666,
|
| 329 |
+
"alive_source": "stats",
|
| 330 |
+
"n_alive_a": 12288,
|
| 331 |
+
"n_alive_b": 12288,
|
| 332 |
+
"matched_cos_signed": {
|
| 333 |
+
"0.5": 0.4818921983242035,
|
| 334 |
+
"0.6": 0.48198845982551575,
|
| 335 |
+
"0.7": 0.4820137619972229,
|
| 336 |
+
"0.8": 0.48179060220718384,
|
| 337 |
+
"0.9": 0.481672078371048
|
| 338 |
+
},
|
| 339 |
+
"unique_frac_signed": {
|
| 340 |
+
"0.5": 0.9955240885416666,
|
| 341 |
+
"0.6": 0.9980875651041666,
|
| 342 |
+
"0.7": 0.9985758463541666,
|
| 343 |
+
"0.8": 0.9989420572916666,
|
| 344 |
+
"0.9": 0.9998779296875
|
| 345 |
+
},
|
| 346 |
+
"dedup_dcs_signed": {
|
| 347 |
+
"0.5": 0.47973529151204275,
|
| 348 |
+
"0.6": 0.48106668827555643,
|
| 349 |
+
"0.7": 0.4813273003407327,
|
| 350 |
+
"0.8": 0.4812808953526352,
|
| 351 |
+
"0.9": 0.4816132805099187
|
| 352 |
+
},
|
| 353 |
+
"matched_frac_signed": {
|
| 354 |
+
"0.5": 0.4130859375,
|
| 355 |
+
"0.6": 0.374267578125,
|
| 356 |
+
"0.7": 0.341064453125,
|
| 357 |
+
"0.8": 0.3095703125,
|
| 358 |
+
"0.9": 0.2753092447916667
|
| 359 |
+
},
|
| 360 |
+
"matched_cos_abs": {
|
| 361 |
+
"0.5": 0.4832015335559845,
|
| 362 |
+
"0.6": 0.4832523763179779,
|
| 363 |
+
"0.7": 0.4833603799343109,
|
| 364 |
+
"0.8": 0.483210951089859,
|
| 365 |
+
"0.9": 0.4832175672054291
|
| 366 |
+
},
|
| 367 |
+
"unique_frac_abs": {
|
| 368 |
+
"0.5": 0.9869384765625,
|
| 369 |
+
"0.6": 0.990966796875,
|
| 370 |
+
"0.7": 0.9925944010416666,
|
| 371 |
+
"0.8": 0.9936930338541666,
|
| 372 |
+
"0.9": 0.9952799479166666
|
| 373 |
+
},
|
| 374 |
+
"dedup_dcs_abs": {
|
| 375 |
+
"0.5": 0.47689018540040706,
|
| 376 |
+
"0.6": 0.47888705944205867,
|
| 377 |
+
"0.7": 0.47978080680816976,
|
| 378 |
+
"0.8": 0.48016335598003934,
|
| 379 |
+
"0.9": 0.4809367551206378
|
| 380 |
+
},
|
| 381 |
+
"matched_frac_abs": {
|
| 382 |
+
"0.5": 0.4092610677083333,
|
| 383 |
+
"0.6": 0.3709309895833333,
|
| 384 |
+
"0.7": 0.338623046875,
|
| 385 |
+
"0.8": 0.307861328125,
|
| 386 |
+
"0.9": 0.2744954427083333
|
| 387 |
+
}
|
| 388 |
+
}
|
| 389 |
+
},
|
| 390 |
+
"summary": {
|
| 391 |
+
"n_seeds": 3,
|
| 392 |
+
"n_pairs": 3,
|
| 393 |
+
"report_band": [
|
| 394 |
+
0.6,
|
| 395 |
+
0.7,
|
| 396 |
+
0.8,
|
| 397 |
+
0.9
|
| 398 |
+
],
|
| 399 |
+
"n_dead_mean": 0,
|
| 400 |
+
"n_dead_std": 0.0,
|
| 401 |
+
"n_dense_mean": 137,
|
| 402 |
+
"n_dense_std": 1.7320508075688772,
|
| 403 |
+
"l0_mean": 128,
|
| 404 |
+
"l0_std": 1.7404671430534633e-14,
|
| 405 |
+
"recon_r2_mean": 0.9791666666666666,
|
| 406 |
+
"recon_r2_std": 5.773502691895622e-05,
|
| 407 |
+
"l1_per_token_mean": 162.53166666666667,
|
| 408 |
+
"l1_per_token_std": 0.2777708828033249,
|
| 409 |
+
"babel_mu1_signed_mean": 0.18623149891694388,
|
| 410 |
+
"babel_mu1_signed_std": 0.0009969881497553742,
|
| 411 |
+
"n_dup_signed_per_threshold": {
|
| 412 |
+
"0.5": 112.0,
|
| 413 |
+
"0.6": 46.333333333333336,
|
| 414 |
+
"0.7": 33.333333333333336,
|
| 415 |
+
"0.8": 24.0,
|
| 416 |
+
"0.9": 2.6666666666666665
|
| 417 |
+
},
|
| 418 |
+
"n_dup_signed_band": 26.583333333333336,
|
| 419 |
+
"n_dup_signed_band_std": 5.625462943912558,
|
| 420 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 421 |
+
"0.5": 55.333333333333336,
|
| 422 |
+
"0.6": 23.0,
|
| 423 |
+
"0.7": 16.666666666666668,
|
| 424 |
+
"0.8": 12.0,
|
| 425 |
+
"0.9": 1.3333333333333333
|
| 426 |
+
},
|
| 427 |
+
"n_dup_clusters_signed_band": 13.250000000000002,
|
| 428 |
+
"n_dup_clusters_signed_band_std": 2.839454172900137,
|
| 429 |
+
"largest_dup_signed_per_threshold": {
|
| 430 |
+
"0.5": 2.3333333333333335,
|
| 431 |
+
"0.6": 2.3333333333333335,
|
| 432 |
+
"0.7": 2.0,
|
| 433 |
+
"0.8": 2.0,
|
| 434 |
+
"0.9": 2.0
|
| 435 |
+
},
|
| 436 |
+
"largest_dup_signed_band": 2.0833333333333335,
|
| 437 |
+
"largest_dup_signed_band_std": 0.14433756729740643,
|
| 438 |
+
"dcs_signed_mean": 0.48177393277486164,
|
| 439 |
+
"dcs_signed_std": 0.00039237802658119447,
|
| 440 |
+
"dedup_dcs_signed_per_threshold": {
|
| 441 |
+
"0.5": 0.47962389752420775,
|
| 442 |
+
"0.6": 0.4809951439197498,
|
| 443 |
+
"0.7": 0.48133802540501663,
|
| 444 |
+
"0.8": 0.4813854567235265,
|
| 445 |
+
"0.9": 0.4817445289259517
|
| 446 |
+
},
|
| 447 |
+
"dedup_dcs_signed_band": 0.4813657887435611,
|
| 448 |
+
"dedup_dcs_signed_band_std": 0.00039748879643519545,
|
| 449 |
+
"matched_cos_signed_per_threshold": {
|
| 450 |
+
"0.5": 0.4818459252516429,
|
| 451 |
+
"0.6": 0.48191020886103314,
|
| 452 |
+
"0.7": 0.4819917182127635,
|
| 453 |
+
"0.8": 0.4818559984366099,
|
| 454 |
+
"0.9": 0.4817968010902405
|
| 455 |
+
},
|
| 456 |
+
"matched_cos_signed_band": 0.48188868165016174,
|
| 457 |
+
"matched_cos_signed_band_std": 0.0003451004999191171,
|
| 458 |
+
"unique_frac_signed_per_threshold": {
|
| 459 |
+
"0.5": 0.995388454861111,
|
| 460 |
+
"0.6": 0.9981011284722222,
|
| 461 |
+
"0.7": 0.9986436631944443,
|
| 462 |
+
"0.8": 0.9990234375,
|
| 463 |
+
"0.9": 0.9998914930555555
|
| 464 |
+
},
|
| 465 |
+
"unique_frac_signed_band": 0.9989149305555555,
|
| 466 |
+
"unique_frac_signed_band_std": 0.00011342860706698603,
|
| 467 |
+
"matched_frac_signed_per_threshold": {
|
| 468 |
+
"0.5": 0.4150119357638889,
|
| 469 |
+
"0.6": 0.37456597222222215,
|
| 470 |
+
"0.7": 0.3394639756944444,
|
| 471 |
+
"0.8": 0.30815972222222227,
|
| 472 |
+
"0.9": 0.2729220920138889
|
| 473 |
+
},
|
| 474 |
+
"matched_frac_signed_band": 0.3237779405381944,
|
| 475 |
+
"matched_frac_signed_band_std": 0.001151129687104982,
|
| 476 |
+
"babel_mu1_abs_mean": 0.21169581015904745,
|
| 477 |
+
"babel_mu1_abs_std": 0.0012372180932078533,
|
| 478 |
+
"n_dup_abs_per_threshold": {
|
| 479 |
+
"0.5": 317.6666666666667,
|
| 480 |
+
"0.6": 220.0,
|
| 481 |
+
"0.7": 177.33333333333334,
|
| 482 |
+
"0.8": 150.66666666666666,
|
| 483 |
+
"0.9": 113.33333333333333
|
| 484 |
+
},
|
| 485 |
+
"n_dup_abs_band": 165.33333333333334,
|
| 486 |
+
"n_dup_abs_band_std": 7.34137816307901,
|
| 487 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 488 |
+
"0.5": 157.66666666666666,
|
| 489 |
+
"0.6": 109.66666666666667,
|
| 490 |
+
"0.7": 88.66666666666667,
|
| 491 |
+
"0.8": 75.33333333333333,
|
| 492 |
+
"0.9": 56.666666666666664
|
| 493 |
+
},
|
| 494 |
+
"n_dup_clusters_abs_band": 82.58333333333334,
|
| 495 |
+
"n_dup_clusters_abs_band_std": 3.660031875999625,
|
| 496 |
+
"largest_dup_abs_per_threshold": {
|
| 497 |
+
"0.5": 3.0,
|
| 498 |
+
"0.6": 2.6666666666666665,
|
| 499 |
+
"0.7": 2.0,
|
| 500 |
+
"0.8": 2.0,
|
| 501 |
+
"0.9": 2.0
|
| 502 |
+
},
|
| 503 |
+
"largest_dup_abs_band": 2.1666666666666665,
|
| 504 |
+
"largest_dup_abs_band_std": 0.14433756729740643,
|
| 505 |
+
"dcs_abs_mean": 0.4838961859544118,
|
| 506 |
+
"dcs_abs_std": 0.00030121519107826396,
|
| 507 |
+
"dedup_dcs_abs_per_threshold": {
|
| 508 |
+
"0.5": 0.4768550795827953,
|
| 509 |
+
"0.6": 0.47890209485356333,
|
| 510 |
+
"0.7": 0.4799524419552957,
|
| 511 |
+
"0.8": 0.4804211299674029,
|
| 512 |
+
"0.9": 0.4812468818005679
|
| 513 |
+
},
|
| 514 |
+
"dedup_dcs_abs_band": 0.48013063714420745,
|
| 515 |
+
"dedup_dcs_abs_band_std": 0.0003312692371722912,
|
| 516 |
+
"matched_cos_abs_per_threshold": {
|
| 517 |
+
"0.5": 0.4831460614999135,
|
| 518 |
+
"0.6": 0.48324108123779297,
|
| 519 |
+
"0.7": 0.48344077666600543,
|
| 520 |
+
"0.8": 0.48338454961776733,
|
| 521 |
+
"0.9": 0.4834764500459035
|
| 522 |
+
},
|
| 523 |
+
"matched_cos_abs_band": 0.4833857143918673,
|
| 524 |
+
"matched_cos_abs_band_std": 0.0002661793715341792,
|
| 525 |
+
"unique_frac_abs_per_threshold": {
|
| 526 |
+
"0.5": 0.9869791666666666,
|
| 527 |
+
"0.6": 0.9910210503472222,
|
| 528 |
+
"0.7": 0.9927842881944443,
|
| 529 |
+
"0.8": 0.9938693576388888,
|
| 530 |
+
"0.9": 0.995388454861111
|
| 531 |
+
},
|
| 532 |
+
"unique_frac_abs_band": 0.9932657877604166,
|
| 533 |
+
"unique_frac_abs_band_std": 0.00014985066592058786,
|
| 534 |
+
"matched_frac_abs_per_threshold": {
|
| 535 |
+
"0.5": 0.41118706597222215,
|
| 536 |
+
"0.6": 0.3714463975694444,
|
| 537 |
+
"0.7": 0.3372667100694444,
|
| 538 |
+
"0.8": 0.3066134982638889,
|
| 539 |
+
"0.9": 0.2722439236111111
|
| 540 |
+
},
|
| 541 |
+
"matched_frac_abs_band": 0.3218926323784722,
|
| 542 |
+
"matched_frac_abs_band_std": 0.0010240118320408336,
|
| 543 |
+
"alive_source": "stats"
|
| 544 |
+
}
|
| 545 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed0/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 0,
|
| 8 |
+
"top_k": 128,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed1/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 1,
|
| 8 |
+
"top_k": 128,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 2,
|
| 8 |
+
"top_k": 128,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.0769,
|
| 3 |
+
"r2_score": 0.9792,
|
| 4 |
+
"l0_per_token_0": 0.0104,
|
| 5 |
+
"l0_per_token_0.001": 0.0104,
|
| 6 |
+
"l0_per_token_0.01": 0.0104,
|
| 7 |
+
"l0_per_token_0.05": 0.0104,
|
| 8 |
+
"l0_per_token_0.1": 0.0104,
|
| 9 |
+
"l0_per_token_0.5": 0.0092,
|
| 10 |
+
"l0_per_token_1.0": 0.004,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0086,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 162.2623
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 771000,
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"l0": 32.066225680933854,
|
| 20 |
+
"recon_r2": 0.9596012381869284,
|
| 21 |
+
"l1_per_token": 88.7389955585258
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 31.94191699092088,
|
| 25 |
+
"recon_r2": 0.9595370158017628,
|
| 26 |
+
"l1_per_token": 88.11377621483324
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 32.092194552529186,
|
| 30 |
+
"recon_r2": 0.9595572117962637,
|
| 31 |
+
"l1_per_token": 88.71793095857717
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.2922865607425847,
|
| 38 |
+
"mcc_abs": 0.2923325621407628,
|
| 39 |
+
"act_iou": 0.2351127617084844,
|
| 40 |
+
"act_iou_n_pairs": 12288,
|
| 41 |
+
"input_r2": 0.9903172700981708,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.9226499440546785,
|
| 44 |
+
"0.7": 0.9416875519279435,
|
| 45 |
+
"0.8": 0.9596929818949981,
|
| 46 |
+
"0.9": 0.9741389230680151
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.9214993761291581,
|
| 50 |
+
"0.7": 0.9409964885517443,
|
| 51 |
+
"0.8": 0.9596522308362436,
|
| 52 |
+
"0.9": 0.9741243489213381
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.29196434929774934,
|
| 58 |
+
"mcc_abs": 0.2920104163144563,
|
| 59 |
+
"act_iou": 0.23533934112303767,
|
| 60 |
+
"act_iou_n_pairs": 12288,
|
| 61 |
+
"input_r2": 0.9904504920827524,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.9215389279795602,
|
| 64 |
+
"0.7": 0.9450532849562321,
|
| 65 |
+
"0.8": 0.9596452199724,
|
| 66 |
+
"0.9": 0.9734739970298455
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.9206283357639286,
|
| 70 |
+
"0.7": 0.9441064225887867,
|
| 71 |
+
"0.8": 0.9596223385053464,
|
| 72 |
+
"0.9": 0.973458579677056
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.2907593770608383,
|
| 78 |
+
"mcc_abs": 0.2908048226752531,
|
| 79 |
+
"act_iou": 0.23369424238384995,
|
| 80 |
+
"act_iou_n_pairs": 12288,
|
| 81 |
+
"input_r2": 0.9901690858040058,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.9233118596020079,
|
| 84 |
+
"0.7": 0.9420088898771856,
|
| 85 |
+
"0.8": 0.9592346095041473,
|
| 86 |
+
"0.9": 0.9730423107957126
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.9228729544659469,
|
| 90 |
+
"0.7": 0.9419104174946153,
|
| 91 |
+
"0.8": 0.9591800548613753,
|
| 92 |
+
"0.9": 0.9730269035409246
|
| 93 |
+
}
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"summary": {
|
| 97 |
+
"n_seeds": 3,
|
| 98 |
+
"n_pairs": 3,
|
| 99 |
+
"report_band": [
|
| 100 |
+
0.6,
|
| 101 |
+
0.7,
|
| 102 |
+
0.8,
|
| 103 |
+
0.9
|
| 104 |
+
],
|
| 105 |
+
"l0_mean": 32.0334457414613,
|
| 106 |
+
"l0_std": 0.08032265996966502,
|
| 107 |
+
"recon_r2_mean": 0.9595651552616516,
|
| 108 |
+
"recon_r2_std": 3.283980314905781e-05,
|
| 109 |
+
"l1_per_token_mean": 88.52356757731206,
|
| 110 |
+
"l1_per_token_std": 0.35504598281693356,
|
| 111 |
+
"input_r2_mean": 0.990312282661643,
|
| 112 |
+
"input_r2_std": 0.00014076941898667015,
|
| 113 |
+
"mcc_signed_mean": 0.29167009570039076,
|
| 114 |
+
"mcc_signed_std": 0.0008049915428853503,
|
| 115 |
+
"mcc_abs_mean": 0.2917159337101574,
|
| 116 |
+
"mcc_abs_std": 0.000805317932120582,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.9225002438787487,
|
| 119 |
+
"0.7": 0.9429165755871204,
|
| 120 |
+
"0.8": 0.9595242704571817,
|
| 121 |
+
"0.9": 0.973551743631191
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.9496232083885604,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.9216668887863445,
|
| 126 |
+
"0.7": 0.9423377762117155,
|
| 127 |
+
"0.8": 0.9594848747343218,
|
| 128 |
+
"0.9": 0.9735366107131062
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.949256537611372,
|
| 131 |
+
"act_iou_mean": 0.234715448405124,
|
| 132 |
+
"act_iou_std": 0.0008916169924298353,
|
| 133 |
+
"act_iou_n_pairs": 12288.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,545 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 2,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"linkage": "complete",
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"d_sae": 12288,
|
| 20 |
+
"d_in": 768,
|
| 21 |
+
"l0": 32,
|
| 22 |
+
"l0_raw": 31.9488,
|
| 23 |
+
"recon_r2": 0.9596,
|
| 24 |
+
"l1_per_token": 88.74,
|
| 25 |
+
"stats_source": "stats",
|
| 26 |
+
"n_dead": 0,
|
| 27 |
+
"dead_rate": 0.0,
|
| 28 |
+
"alive_source": "stats",
|
| 29 |
+
"n_dense": 6,
|
| 30 |
+
"dense_rate": 0.00048828125,
|
| 31 |
+
"dense_threshold": 0.1,
|
| 32 |
+
"n_dup_signed": {
|
| 33 |
+
"0.5": 28,
|
| 34 |
+
"0.6": 6,
|
| 35 |
+
"0.7": 2,
|
| 36 |
+
"0.8": 0,
|
| 37 |
+
"0.9": 0
|
| 38 |
+
},
|
| 39 |
+
"n_dup_clusters_signed": {
|
| 40 |
+
"0.5": 14,
|
| 41 |
+
"0.6": 3,
|
| 42 |
+
"0.7": 1,
|
| 43 |
+
"0.8": 0,
|
| 44 |
+
"0.9": 0
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 2,
|
| 48 |
+
"0.6": 2,
|
| 49 |
+
"0.7": 2,
|
| 50 |
+
"0.8": 1,
|
| 51 |
+
"0.9": 1
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.21009226143360138,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 50,
|
| 56 |
+
"0.6": 18,
|
| 57 |
+
"0.7": 8,
|
| 58 |
+
"0.8": 4,
|
| 59 |
+
"0.9": 4
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 25,
|
| 63 |
+
"0.6": 9,
|
| 64 |
+
"0.7": 4,
|
| 65 |
+
"0.8": 2,
|
| 66 |
+
"0.9": 2
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 2,
|
| 70 |
+
"0.6": 2,
|
| 71 |
+
"0.7": 2,
|
| 72 |
+
"0.8": 2,
|
| 73 |
+
"0.9": 2
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.21652503311634064
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 12288,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 32,
|
| 81 |
+
"l0_raw": 31.9488,
|
| 82 |
+
"recon_r2": 0.9595,
|
| 83 |
+
"l1_per_token": 88.1271,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 9,
|
| 89 |
+
"dense_rate": 0.000732421875,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 32,
|
| 93 |
+
"0.6": 12,
|
| 94 |
+
"0.7": 4,
|
| 95 |
+
"0.8": 0,
|
| 96 |
+
"0.9": 0
|
| 97 |
+
},
|
| 98 |
+
"n_dup_clusters_signed": {
|
| 99 |
+
"0.5": 16,
|
| 100 |
+
"0.6": 6,
|
| 101 |
+
"0.7": 2,
|
| 102 |
+
"0.8": 0,
|
| 103 |
+
"0.9": 0
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 2,
|
| 107 |
+
"0.6": 2,
|
| 108 |
+
"0.7": 2,
|
| 109 |
+
"0.8": 1,
|
| 110 |
+
"0.9": 1
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.2101074904203415,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 52,
|
| 115 |
+
"0.6": 26,
|
| 116 |
+
"0.7": 10,
|
| 117 |
+
"0.8": 4,
|
| 118 |
+
"0.9": 4
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 26,
|
| 122 |
+
"0.6": 13,
|
| 123 |
+
"0.7": 5,
|
| 124 |
+
"0.8": 2,
|
| 125 |
+
"0.9": 2
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 2,
|
| 129 |
+
"0.6": 2,
|
| 130 |
+
"0.7": 2,
|
| 131 |
+
"0.8": 2,
|
| 132 |
+
"0.9": 2
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.21666556596755981
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 12288,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 32,
|
| 140 |
+
"l0_raw": 31.9488,
|
| 141 |
+
"recon_r2": 0.9595,
|
| 142 |
+
"l1_per_token": 88.7149,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 7,
|
| 148 |
+
"dense_rate": 0.0005696614583333334,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 22,
|
| 152 |
+
"0.6": 4,
|
| 153 |
+
"0.7": 2,
|
| 154 |
+
"0.8": 0,
|
| 155 |
+
"0.9": 0
|
| 156 |
+
},
|
| 157 |
+
"n_dup_clusters_signed": {
|
| 158 |
+
"0.5": 11,
|
| 159 |
+
"0.6": 2,
|
| 160 |
+
"0.7": 1,
|
| 161 |
+
"0.8": 0,
|
| 162 |
+
"0.9": 0
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 2,
|
| 166 |
+
"0.6": 2,
|
| 167 |
+
"0.7": 2,
|
| 168 |
+
"0.8": 1,
|
| 169 |
+
"0.9": 1
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.21042774617671967,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 50,
|
| 174 |
+
"0.6": 20,
|
| 175 |
+
"0.7": 10,
|
| 176 |
+
"0.8": 4,
|
| 177 |
+
"0.9": 4
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 25,
|
| 181 |
+
"0.6": 10,
|
| 182 |
+
"0.7": 5,
|
| 183 |
+
"0.8": 2,
|
| 184 |
+
"0.9": 2
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 2,
|
| 188 |
+
"0.6": 2,
|
| 189 |
+
"0.7": 2,
|
| 190 |
+
"0.8": 2,
|
| 191 |
+
"0.9": 2
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.21692664921283722
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.42227792739868164,
|
| 199 |
+
"dcs_abs": 0.42629995942115784,
|
| 200 |
+
"frac_sign_flipped": 0.238525390625,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"n_alive_a": 12288,
|
| 203 |
+
"n_alive_b": 12288,
|
| 204 |
+
"matched_cos_signed": {
|
| 205 |
+
"0.5": 0.42199206352233887,
|
| 206 |
+
"0.6": 0.4222817122936249,
|
| 207 |
+
"0.7": 0.4222303330898285,
|
| 208 |
+
"0.8": 0.42227792739868164,
|
| 209 |
+
"0.9": 0.42227792739868164
|
| 210 |
+
},
|
| 211 |
+
"unique_frac_signed": {
|
| 212 |
+
"0.5": 0.998779296875,
|
| 213 |
+
"0.6": 0.9996337890625,
|
| 214 |
+
"0.7": 0.9998779296875,
|
| 215 |
+
"0.8": 1.0,
|
| 216 |
+
"0.9": 1.0
|
| 217 |
+
},
|
| 218 |
+
"dedup_dcs_signed": {
|
| 219 |
+
"0.5": 0.42147693649167195,
|
| 220 |
+
"0.6": 0.4221270681118767,
|
| 221 |
+
"0.7": 0.42217879130112124,
|
| 222 |
+
"0.8": 0.42227792739868164,
|
| 223 |
+
"0.9": 0.42227792739868164
|
| 224 |
+
},
|
| 225 |
+
"matched_frac_signed": {
|
| 226 |
+
"0.5": 0.3116861979166667,
|
| 227 |
+
"0.6": 0.298583984375,
|
| 228 |
+
"0.7": 0.2832845052083333,
|
| 229 |
+
"0.8": 0.2654622395833333,
|
| 230 |
+
"0.9": 0.244384765625
|
| 231 |
+
},
|
| 232 |
+
"matched_cos_abs": {
|
| 233 |
+
"0.5": 0.4255364239215851,
|
| 234 |
+
"0.6": 0.4259316921234131,
|
| 235 |
+
"0.7": 0.4261135160923004,
|
| 236 |
+
"0.8": 0.4262067377567291,
|
| 237 |
+
"0.9": 0.4262067377567291
|
| 238 |
+
},
|
| 239 |
+
"unique_frac_abs": {
|
| 240 |
+
"0.5": 0.9979248046875,
|
| 241 |
+
"0.6": 0.9991048177083334,
|
| 242 |
+
"0.7": 0.9996337890625,
|
| 243 |
+
"0.8": 0.9998372395833334,
|
| 244 |
+
"0.9": 0.9998372395833334
|
| 245 |
+
},
|
| 246 |
+
"dedup_dcs_abs": {
|
| 247 |
+
"0.5": 0.424653352729365,
|
| 248 |
+
"0.6": 0.4255504056151646,
|
| 249 |
+
"0.7": 0.42595746866209083,
|
| 250 |
+
"0.8": 0.42613736817050574,
|
| 251 |
+
"0.9": 0.42613736817050574
|
| 252 |
+
},
|
| 253 |
+
"matched_frac_abs": {
|
| 254 |
+
"0.5": 0.3113606770833333,
|
| 255 |
+
"0.6": 0.2985026041666667,
|
| 256 |
+
"0.7": 0.2836100260416667,
|
| 257 |
+
"0.8": 0.265380859375,
|
| 258 |
+
"0.9": 0.24422200520833334
|
| 259 |
+
}
|
| 260 |
+
},
|
| 261 |
+
"seed0\u2194seed2": {
|
| 262 |
+
"dcs_signed": 0.4223461449146271,
|
| 263 |
+
"dcs_abs": 0.4263066351413727,
|
| 264 |
+
"frac_sign_flipped": 0.240234375,
|
| 265 |
+
"alive_source": "stats",
|
| 266 |
+
"n_alive_a": 12288,
|
| 267 |
+
"n_alive_b": 12288,
|
| 268 |
+
"matched_cos_signed": {
|
| 269 |
+
"0.5": 0.42206889390945435,
|
| 270 |
+
"0.6": 0.4223623275756836,
|
| 271 |
+
"0.7": 0.4223664700984955,
|
| 272 |
+
"0.8": 0.4223461449146271,
|
| 273 |
+
"0.9": 0.4223461449146271
|
| 274 |
+
},
|
| 275 |
+
"unique_frac_signed": {
|
| 276 |
+
"0.5": 0.9989827473958334,
|
| 277 |
+
"0.6": 0.9997965494791666,
|
| 278 |
+
"0.7": 0.9999186197916666,
|
| 279 |
+
"0.8": 1.0,
|
| 280 |
+
"0.9": 1.0
|
| 281 |
+
},
|
| 282 |
+
"dedup_dcs_signed": {
|
| 283 |
+
"0.5": 0.42163954322798725,
|
| 284 |
+
"0.6": 0.4222763977401579,
|
| 285 |
+
"0.7": 0.42233209782716585,
|
| 286 |
+
"0.8": 0.4223461449146271,
|
| 287 |
+
"0.9": 0.4223461449146271
|
| 288 |
+
},
|
| 289 |
+
"matched_frac_signed": {
|
| 290 |
+
"0.5": 0.311279296875,
|
| 291 |
+
"0.6": 0.2993977864583333,
|
| 292 |
+
"0.7": 0.2814127604166667,
|
| 293 |
+
"0.8": 0.2662760416666667,
|
| 294 |
+
"0.9": 0.244384765625
|
| 295 |
+
},
|
| 296 |
+
"matched_cos_abs": {
|
| 297 |
+
"0.5": 0.4255024492740631,
|
| 298 |
+
"0.6": 0.426001638174057,
|
| 299 |
+
"0.7": 0.42615342140197754,
|
| 300 |
+
"0.8": 0.42621341347694397,
|
| 301 |
+
"0.9": 0.42621341347694397
|
| 302 |
+
},
|
| 303 |
+
"unique_frac_abs": {
|
| 304 |
+
"0.5": 0.9979654947916666,
|
| 305 |
+
"0.6": 0.9992268880208334,
|
| 306 |
+
"0.7": 0.9996337890625,
|
| 307 |
+
"0.8": 0.9998372395833334,
|
| 308 |
+
"0.9": 0.9998372395833334
|
| 309 |
+
},
|
| 310 |
+
"dedup_dcs_abs": {
|
| 311 |
+
"0.5": 0.4246367623248564,
|
| 312 |
+
"0.6": 0.42567229120444006,
|
| 313 |
+
"0.7": 0.4259973593580071,
|
| 314 |
+
"0.8": 0.42614404280417756,
|
| 315 |
+
"0.9": 0.42614404280417756
|
| 316 |
+
},
|
| 317 |
+
"matched_frac_abs": {
|
| 318 |
+
"0.5": 0.3103841145833333,
|
| 319 |
+
"0.6": 0.2992350260416667,
|
| 320 |
+
"0.7": 0.28173828125,
|
| 321 |
+
"0.8": 0.26611328125,
|
| 322 |
+
"0.9": 0.24422200520833334
|
| 323 |
+
}
|
| 324 |
+
},
|
| 325 |
+
"seed1\u2194seed2": {
|
| 326 |
+
"dcs_signed": 0.4210472106933594,
|
| 327 |
+
"dcs_abs": 0.4248305559158325,
|
| 328 |
+
"frac_sign_flipped": 0.23323567708333334,
|
| 329 |
+
"alive_source": "stats",
|
| 330 |
+
"n_alive_a": 12288,
|
| 331 |
+
"n_alive_b": 12288,
|
| 332 |
+
"matched_cos_signed": {
|
| 333 |
+
"0.5": 0.4209243059158325,
|
| 334 |
+
"0.6": 0.42112892866134644,
|
| 335 |
+
"0.7": 0.42109277844429016,
|
| 336 |
+
"0.8": 0.4210472106933594,
|
| 337 |
+
"0.9": 0.4210472106933594
|
| 338 |
+
},
|
| 339 |
+
"unique_frac_signed": {
|
| 340 |
+
"0.5": 0.9989013671875,
|
| 341 |
+
"0.6": 0.9996744791666666,
|
| 342 |
+
"0.7": 0.9998779296875,
|
| 343 |
+
"0.8": 1.0,
|
| 344 |
+
"0.9": 1.0
|
| 345 |
+
},
|
| 346 |
+
"dedup_dcs_signed": {
|
| 347 |
+
"0.5": 0.4204618646617746,
|
| 348 |
+
"0.6": 0.4209918424215478,
|
| 349 |
+
"0.7": 0.421041375517234,
|
| 350 |
+
"0.8": 0.4210472106933594,
|
| 351 |
+
"0.9": 0.4210472106933594
|
| 352 |
+
},
|
| 353 |
+
"matched_frac_signed": {
|
| 354 |
+
"0.5": 0.3089192708333333,
|
| 355 |
+
"0.6": 0.2954915364583333,
|
| 356 |
+
"0.7": 0.2808430989583333,
|
| 357 |
+
"0.8": 0.26416015625,
|
| 358 |
+
"0.9": 0.243896484375
|
| 359 |
+
},
|
| 360 |
+
"matched_cos_abs": {
|
| 361 |
+
"0.5": 0.4241715669631958,
|
| 362 |
+
"0.6": 0.42455750703811646,
|
| 363 |
+
"0.7": 0.42469891905784607,
|
| 364 |
+
"0.8": 0.4247370958328247,
|
| 365 |
+
"0.9": 0.4247370958328247
|
| 366 |
+
},
|
| 367 |
+
"unique_frac_abs": {
|
| 368 |
+
"0.5": 0.9979248046875,
|
| 369 |
+
"0.6": 0.9990641276041666,
|
| 370 |
+
"0.7": 0.9995930989583334,
|
| 371 |
+
"0.8": 0.9998372395833334,
|
| 372 |
+
"0.9": 0.9998372395833334
|
| 373 |
+
},
|
| 374 |
+
"dedup_dcs_abs": {
|
| 375 |
+
"0.5": 0.423291328115738,
|
| 376 |
+
"0.6": 0.42416017538683565,
|
| 377 |
+
"0.7": 0.4245261086252867,
|
| 378 |
+
"0.8": 0.42466796544613317,
|
| 379 |
+
"0.9": 0.42466796544613317
|
| 380 |
+
},
|
| 381 |
+
"matched_frac_abs": {
|
| 382 |
+
"0.5": 0.3084309895833333,
|
| 383 |
+
"0.6": 0.2950846354166667,
|
| 384 |
+
"0.7": 0.2808430989583333,
|
| 385 |
+
"0.8": 0.2640787760416667,
|
| 386 |
+
"0.9": 0.24373372395833334
|
| 387 |
+
}
|
| 388 |
+
}
|
| 389 |
+
},
|
| 390 |
+
"summary": {
|
| 391 |
+
"n_seeds": 3,
|
| 392 |
+
"n_pairs": 3,
|
| 393 |
+
"report_band": [
|
| 394 |
+
0.6,
|
| 395 |
+
0.7,
|
| 396 |
+
0.8,
|
| 397 |
+
0.9
|
| 398 |
+
],
|
| 399 |
+
"n_dead_mean": 0,
|
| 400 |
+
"n_dead_std": 0.0,
|
| 401 |
+
"n_dense_mean": 7,
|
| 402 |
+
"n_dense_std": 1.5275252316519465,
|
| 403 |
+
"l0_mean": 32,
|
| 404 |
+
"l0_std": 4.351167857633658e-15,
|
| 405 |
+
"recon_r2_mean": 0.9595333333333333,
|
| 406 |
+
"recon_r2_std": 0.0,
|
| 407 |
+
"l1_per_token_mean": 88.52733333333333,
|
| 408 |
+
"l1_per_token_std": 0.34559687980844317,
|
| 409 |
+
"babel_mu1_signed_mean": 0.21020916601022085,
|
| 410 |
+
"babel_mu1_signed_std": 0.00018944906280916654,
|
| 411 |
+
"n_dup_signed_per_threshold": {
|
| 412 |
+
"0.5": 27.333333333333332,
|
| 413 |
+
"0.6": 7.333333333333333,
|
| 414 |
+
"0.7": 2.6666666666666665,
|
| 415 |
+
"0.8": 0.0,
|
| 416 |
+
"0.9": 0.0
|
| 417 |
+
},
|
| 418 |
+
"n_dup_signed_band": 2.5,
|
| 419 |
+
"n_dup_signed_band_std": 1.3228756555322954,
|
| 420 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 421 |
+
"0.5": 13.666666666666666,
|
| 422 |
+
"0.6": 3.6666666666666665,
|
| 423 |
+
"0.7": 1.3333333333333333,
|
| 424 |
+
"0.8": 0.0,
|
| 425 |
+
"0.9": 0.0
|
| 426 |
+
},
|
| 427 |
+
"n_dup_clusters_signed_band": 1.25,
|
| 428 |
+
"n_dup_clusters_signed_band_std": 0.6614378277661477,
|
| 429 |
+
"largest_dup_signed_per_threshold": {
|
| 430 |
+
"0.5": 2.0,
|
| 431 |
+
"0.6": 2.0,
|
| 432 |
+
"0.7": 2.0,
|
| 433 |
+
"0.8": 1.0,
|
| 434 |
+
"0.9": 1.0
|
| 435 |
+
},
|
| 436 |
+
"largest_dup_signed_band": 1.5,
|
| 437 |
+
"largest_dup_signed_band_std": 0.0,
|
| 438 |
+
"dcs_signed_mean": 0.42189042766888935,
|
| 439 |
+
"dcs_signed_std": 0.0007310434721936154,
|
| 440 |
+
"dedup_dcs_signed_per_threshold": {
|
| 441 |
+
"0.5": 0.42119278146047795,
|
| 442 |
+
"0.6": 0.42179843609119416,
|
| 443 |
+
"0.7": 0.4218507548818404,
|
| 444 |
+
"0.8": 0.42189042766888935,
|
| 445 |
+
"0.9": 0.42189042766888935
|
| 446 |
+
},
|
| 447 |
+
"dedup_dcs_signed_band": 0.4218575115777033,
|
| 448 |
+
"dedup_dcs_signed_band_std": 0.0007170954782629625,
|
| 449 |
+
"matched_cos_signed_per_threshold": {
|
| 450 |
+
"0.5": 0.42166175444920856,
|
| 451 |
+
"0.6": 0.42192432284355164,
|
| 452 |
+
"0.7": 0.4218965272108714,
|
| 453 |
+
"0.8": 0.42189042766888935,
|
| 454 |
+
"0.9": 0.42189042766888935
|
| 455 |
+
},
|
| 456 |
+
"matched_cos_signed_band": 0.4219004263480504,
|
| 457 |
+
"matched_cos_signed_band_std": 0.0007127169404301337,
|
| 458 |
+
"unique_frac_signed_per_threshold": {
|
| 459 |
+
"0.5": 0.9988878038194445,
|
| 460 |
+
"0.6": 0.9997016059027777,
|
| 461 |
+
"0.7": 0.9998914930555555,
|
| 462 |
+
"0.8": 1.0,
|
| 463 |
+
"0.9": 1.0
|
| 464 |
+
},
|
| 465 |
+
"unique_frac_signed_band": 0.9998982747395833,
|
| 466 |
+
"unique_frac_signed_band_std": 2.6913974111543305e-05,
|
| 467 |
+
"matched_frac_signed_per_threshold": {
|
| 468 |
+
"0.5": 0.3106282552083333,
|
| 469 |
+
"0.6": 0.29782443576388884,
|
| 470 |
+
"0.7": 0.2818467881944444,
|
| 471 |
+
"0.8": 0.2652994791666667,
|
| 472 |
+
"0.9": 0.24422200520833334
|
| 473 |
+
},
|
| 474 |
+
"matched_frac_signed_band": 0.2722981770833333,
|
| 475 |
+
"matched_frac_signed_band_std": 0.001039988437429382,
|
| 476 |
+
"babel_mu1_abs_mean": 0.21670574943224588,
|
| 477 |
+
"babel_mu1_abs_std": 0.00020380114172133775,
|
| 478 |
+
"n_dup_abs_per_threshold": {
|
| 479 |
+
"0.5": 50.666666666666664,
|
| 480 |
+
"0.6": 21.333333333333332,
|
| 481 |
+
"0.7": 9.333333333333334,
|
| 482 |
+
"0.8": 4.0,
|
| 483 |
+
"0.9": 4.0
|
| 484 |
+
},
|
| 485 |
+
"n_dup_abs_band": 9.666666666666666,
|
| 486 |
+
"n_dup_abs_band_std": 1.2583057392117918,
|
| 487 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 488 |
+
"0.5": 25.333333333333332,
|
| 489 |
+
"0.6": 10.666666666666666,
|
| 490 |
+
"0.7": 4.666666666666667,
|
| 491 |
+
"0.8": 2.0,
|
| 492 |
+
"0.9": 2.0
|
| 493 |
+
},
|
| 494 |
+
"n_dup_clusters_abs_band": 4.833333333333333,
|
| 495 |
+
"n_dup_clusters_abs_band_std": 0.6291528696058959,
|
| 496 |
+
"largest_dup_abs_per_threshold": {
|
| 497 |
+
"0.5": 2.0,
|
| 498 |
+
"0.6": 2.0,
|
| 499 |
+
"0.7": 2.0,
|
| 500 |
+
"0.8": 2.0,
|
| 501 |
+
"0.9": 2.0
|
| 502 |
+
},
|
| 503 |
+
"largest_dup_abs_band": 2.0,
|
| 504 |
+
"largest_dup_abs_band_std": 0.0,
|
| 505 |
+
"dcs_abs_mean": 0.42581238349278766,
|
| 506 |
+
"dcs_abs_std": 0.0008502941752489702,
|
| 507 |
+
"dedup_dcs_abs_per_threshold": {
|
| 508 |
+
"0.5": 0.42419381438998643,
|
| 509 |
+
"0.6": 0.4251276240688134,
|
| 510 |
+
"0.7": 0.42549364554846153,
|
| 511 |
+
"0.8": 0.42564979214027215,
|
| 512 |
+
"0.9": 0.42564979214027215
|
| 513 |
+
},
|
| 514 |
+
"dedup_dcs_abs_band": 0.4254802134744548,
|
| 515 |
+
"dedup_dcs_abs_band_std": 0.0008443639151978224,
|
| 516 |
+
"matched_cos_abs_per_threshold": {
|
| 517 |
+
"0.5": 0.4250701467196147,
|
| 518 |
+
"0.6": 0.42549694577852887,
|
| 519 |
+
"0.7": 0.4256552855173747,
|
| 520 |
+
"0.8": 0.42571908235549927,
|
| 521 |
+
"0.9": 0.42571908235549927
|
| 522 |
+
},
|
| 523 |
+
"matched_cos_abs_band": 0.42564759900172555,
|
| 524 |
+
"matched_cos_abs_band_std": 0.0008358083964656375,
|
| 525 |
+
"unique_frac_abs_per_threshold": {
|
| 526 |
+
"0.5": 0.9979383680555555,
|
| 527 |
+
"0.6": 0.9991319444444445,
|
| 528 |
+
"0.7": 0.9996202256944445,
|
| 529 |
+
"0.8": 0.9998372395833334,
|
| 530 |
+
"0.9": 0.9998372395833334
|
| 531 |
+
},
|
| 532 |
+
"unique_frac_abs_band": 0.999606662326389,
|
| 533 |
+
"unique_frac_abs_band_std": 2.5600295801038305e-05,
|
| 534 |
+
"matched_frac_abs_per_threshold": {
|
| 535 |
+
"0.5": 0.31005859375,
|
| 536 |
+
"0.6": 0.297607421875,
|
| 537 |
+
"0.7": 0.2820638020833333,
|
| 538 |
+
"0.8": 0.26519097222222227,
|
| 539 |
+
"0.9": 0.24405924479166666
|
| 540 |
+
},
|
| 541 |
+
"matched_frac_abs_band": 0.2722303602430556,
|
| 542 |
+
"matched_frac_abs_band_std": 0.0011229166394581532,
|
| 543 |
+
"alive_source": "stats"
|
| 544 |
+
}
|
| 545 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 0,
|
| 8 |
+
"top_k": 32,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1493,
|
| 3 |
+
"r2_score": 0.9596,
|
| 4 |
+
"l0_per_token_0": 0.0026,
|
| 5 |
+
"l0_per_token_0.001": 0.0026,
|
| 6 |
+
"l0_per_token_0.01": 0.0026,
|
| 7 |
+
"l0_per_token_0.05": 0.0026,
|
| 8 |
+
"l0_per_token_0.1": 0.0026,
|
| 9 |
+
"l0_per_token_0.5": 0.0026,
|
| 10 |
+
"l0_per_token_1.0": 0.0026,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0002,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 88.74
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 1,
|
| 8 |
+
"top_k": 32,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1495,
|
| 3 |
+
"r2_score": 0.9595,
|
| 4 |
+
"l0_per_token_0": 0.0026,
|
| 5 |
+
"l0_per_token_0.001": 0.0026,
|
| 6 |
+
"l0_per_token_0.01": 0.0026,
|
| 7 |
+
"l0_per_token_0.05": 0.0026,
|
| 8 |
+
"l0_per_token_0.1": 0.0026,
|
| 9 |
+
"l0_per_token_0.5": 0.0026,
|
| 10 |
+
"l0_per_token_1.0": 0.0026,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0002,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 88.1271
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed2/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 2,
|
| 8 |
+
"top_k": 32,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed2/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1495,
|
| 3 |
+
"r2_score": 0.9595,
|
| 4 |
+
"l0_per_token_0": 0.0026,
|
| 5 |
+
"l0_per_token_0.001": 0.0026,
|
| 6 |
+
"l0_per_token_0.01": 0.0026,
|
| 7 |
+
"l0_per_token_0.05": 0.0026,
|
| 8 |
+
"l0_per_token_0.1": 0.0026,
|
| 9 |
+
"l0_per_token_0.5": 0.0026,
|
| 10 |
+
"l0_per_token_1.0": 0.0026,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0002,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 88.7149
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_64/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 771000,
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"l0": 63.986028534370945,
|
| 20 |
+
"recon_r2": 0.9712022621359476,
|
| 21 |
+
"l1_per_token": 118.38319033642308
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 63.99791439688716,
|
| 25 |
+
"recon_r2": 0.9711364832301964,
|
| 26 |
+
"l1_per_token": 119.07314170474139
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 64.02235667963683,
|
| 30 |
+
"recon_r2": 0.9711494397424985,
|
| 31 |
+
"l1_per_token": 117.93598294451628
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.40906097915909045,
|
| 38 |
+
"mcc_abs": 0.4091015360633811,
|
| 39 |
+
"act_iou": 0.3005517784745941,
|
| 40 |
+
"act_iou_n_pairs": 12288,
|
| 41 |
+
"input_r2": 0.9914611032900283,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.9008698385798795,
|
| 44 |
+
"0.7": 0.9301295314312694,
|
| 45 |
+
"0.8": 0.9544456845458938,
|
| 46 |
+
"0.9": 0.9728371055364811
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.9003925905253838,
|
| 50 |
+
"0.7": 0.9300294284753231,
|
| 51 |
+
"0.8": 0.9544957676226858,
|
| 52 |
+
"0.9": 0.9726293801578932
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.4101303005094561,
|
| 58 |
+
"mcc_abs": 0.41016906972901346,
|
| 59 |
+
"act_iou": 0.3019254115590623,
|
| 60 |
+
"act_iou_n_pairs": 12288,
|
| 61 |
+
"input_r2": 0.9915667891424259,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.9014868083600571,
|
| 64 |
+
"0.7": 0.9298186919744063,
|
| 65 |
+
"0.8": 0.9545154652161274,
|
| 66 |
+
"0.9": 0.9723472703803464
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.9003765172295217,
|
| 70 |
+
"0.7": 0.9296826496488213,
|
| 71 |
+
"0.8": 0.9544772702458105,
|
| 72 |
+
"0.9": 0.972278323604367
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.40840945808578777,
|
| 78 |
+
"mcc_abs": 0.40845214918193085,
|
| 79 |
+
"act_iou": 0.300424878379523,
|
| 80 |
+
"act_iou_n_pairs": 12288,
|
| 81 |
+
"input_r2": 0.9914834507564594,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.9028514622141288,
|
| 84 |
+
"0.7": 0.9303472414239641,
|
| 85 |
+
"0.8": 0.9545406186983754,
|
| 86 |
+
"0.9": 0.971766142546762
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.9018616132976246,
|
| 90 |
+
"0.7": 0.9298828159903569,
|
| 91 |
+
"0.8": 0.9545960265304443,
|
| 92 |
+
"0.9": 0.9717768693306584
|
| 93 |
+
}
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"summary": {
|
| 97 |
+
"n_seeds": 3,
|
| 98 |
+
"n_pairs": 3,
|
| 99 |
+
"report_band": [
|
| 100 |
+
0.6,
|
| 101 |
+
0.7,
|
| 102 |
+
0.8,
|
| 103 |
+
0.9
|
| 104 |
+
],
|
| 105 |
+
"l0_mean": 64.00209987029831,
|
| 106 |
+
"l0_std": 0.01852220762682939,
|
| 107 |
+
"recon_r2_mean": 0.9711627283695474,
|
| 108 |
+
"recon_r2_std": 3.484475307837803e-05,
|
| 109 |
+
"l1_per_token_mean": 118.46410499522692,
|
| 110 |
+
"l1_per_token_std": 0.572881225033759,
|
| 111 |
+
"input_r2_mean": 0.9915037810629711,
|
| 112 |
+
"input_r2_std": 5.569888572851859e-05,
|
| 113 |
+
"mcc_signed_mean": 0.40920024591811144,
|
| 114 |
+
"mcc_signed_std": 0.0008688331740905866,
|
| 115 |
+
"mcc_abs_mean": 0.40924091832477516,
|
| 116 |
+
"mcc_abs_std": 0.0008669051864777931,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.9017360363846884,
|
| 119 |
+
"0.7": 0.9300984882765465,
|
| 120 |
+
"0.8": 0.9545005894867988,
|
| 121 |
+
"0.9": 0.9723168394878631
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.9396629884089741,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.9008769070175101,
|
| 126 |
+
"0.7": 0.9298649647048339,
|
| 127 |
+
"0.8": 0.9545230214663135,
|
| 128 |
+
"0.9": 0.9722281910309728
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.9393732710549075,
|
| 131 |
+
"act_iou_mean": 0.3009673561377264,
|
| 132 |
+
"act_iou_std": 0.0008321229183813511,
|
| 133 |
+
"act_iou_n_pairs": 12288.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_64/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,545 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 2,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"linkage": "complete",
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"d_sae": 12288,
|
| 20 |
+
"d_in": 768,
|
| 21 |
+
"l0": 64,
|
| 22 |
+
"l0_raw": 63.8976,
|
| 23 |
+
"recon_r2": 0.9712,
|
| 24 |
+
"l1_per_token": 118.3737,
|
| 25 |
+
"stats_source": "stats",
|
| 26 |
+
"n_dead": 0,
|
| 27 |
+
"dead_rate": 0.0,
|
| 28 |
+
"alive_source": "stats",
|
| 29 |
+
"n_dense": 31,
|
| 30 |
+
"dense_rate": 0.0025227864583333335,
|
| 31 |
+
"dense_threshold": 0.1,
|
| 32 |
+
"n_dup_signed": {
|
| 33 |
+
"0.5": 119,
|
| 34 |
+
"0.6": 69,
|
| 35 |
+
"0.7": 47,
|
| 36 |
+
"0.8": 26,
|
| 37 |
+
"0.9": 0
|
| 38 |
+
},
|
| 39 |
+
"n_dup_clusters_signed": {
|
| 40 |
+
"0.5": 58,
|
| 41 |
+
"0.6": 34,
|
| 42 |
+
"0.7": 23,
|
| 43 |
+
"0.8": 13,
|
| 44 |
+
"0.9": 0
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 3,
|
| 48 |
+
"0.6": 3,
|
| 49 |
+
"0.7": 3,
|
| 50 |
+
"0.8": 2,
|
| 51 |
+
"0.9": 1
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.19530417025089264,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 193,
|
| 56 |
+
"0.6": 123,
|
| 57 |
+
"0.7": 83,
|
| 58 |
+
"0.8": 46,
|
| 59 |
+
"0.9": 10
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 95,
|
| 63 |
+
"0.6": 61,
|
| 64 |
+
"0.7": 41,
|
| 65 |
+
"0.8": 23,
|
| 66 |
+
"0.9": 5
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 3,
|
| 70 |
+
"0.6": 3,
|
| 71 |
+
"0.7": 3,
|
| 72 |
+
"0.8": 2,
|
| 73 |
+
"0.9": 2
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.20532314479351044
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 12288,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 64,
|
| 81 |
+
"l0_raw": 63.8976,
|
| 82 |
+
"recon_r2": 0.9711,
|
| 83 |
+
"l1_per_token": 119.0517,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 31,
|
| 89 |
+
"dense_rate": 0.0025227864583333335,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 101,
|
| 93 |
+
"0.6": 57,
|
| 94 |
+
"0.7": 38,
|
| 95 |
+
"0.8": 16,
|
| 96 |
+
"0.9": 0
|
| 97 |
+
},
|
| 98 |
+
"n_dup_clusters_signed": {
|
| 99 |
+
"0.5": 50,
|
| 100 |
+
"0.6": 28,
|
| 101 |
+
"0.7": 19,
|
| 102 |
+
"0.8": 8,
|
| 103 |
+
"0.9": 0
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 3,
|
| 107 |
+
"0.6": 3,
|
| 108 |
+
"0.7": 2,
|
| 109 |
+
"0.8": 2,
|
| 110 |
+
"0.9": 1
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.1941591054201126,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 171,
|
| 115 |
+
"0.6": 109,
|
| 116 |
+
"0.7": 64,
|
| 117 |
+
"0.8": 32,
|
| 118 |
+
"0.9": 12
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 85,
|
| 122 |
+
"0.6": 54,
|
| 123 |
+
"0.7": 32,
|
| 124 |
+
"0.8": 16,
|
| 125 |
+
"0.9": 6
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 3,
|
| 129 |
+
"0.6": 3,
|
| 130 |
+
"0.7": 2,
|
| 131 |
+
"0.8": 2,
|
| 132 |
+
"0.9": 2
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.20381033420562744
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 12288,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 64,
|
| 140 |
+
"l0_raw": 63.8976,
|
| 141 |
+
"recon_r2": 0.9711,
|
| 142 |
+
"l1_per_token": 117.9227,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 32,
|
| 148 |
+
"dense_rate": 0.0026041666666666665,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 136,
|
| 152 |
+
"0.6": 83,
|
| 153 |
+
"0.7": 56,
|
| 154 |
+
"0.8": 28,
|
| 155 |
+
"0.9": 0
|
| 156 |
+
},
|
| 157 |
+
"n_dup_clusters_signed": {
|
| 158 |
+
"0.5": 66,
|
| 159 |
+
"0.6": 41,
|
| 160 |
+
"0.7": 28,
|
| 161 |
+
"0.8": 14,
|
| 162 |
+
"0.9": 0
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 3,
|
| 166 |
+
"0.6": 3,
|
| 167 |
+
"0.7": 2,
|
| 168 |
+
"0.8": 2,
|
| 169 |
+
"0.9": 1
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.19544319808483124,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 208,
|
| 174 |
+
"0.6": 133,
|
| 175 |
+
"0.7": 82,
|
| 176 |
+
"0.8": 46,
|
| 177 |
+
"0.9": 8
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 102,
|
| 181 |
+
"0.6": 66,
|
| 182 |
+
"0.7": 41,
|
| 183 |
+
"0.8": 23,
|
| 184 |
+
"0.9": 4
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 3,
|
| 188 |
+
"0.6": 3,
|
| 189 |
+
"0.7": 2,
|
| 190 |
+
"0.8": 2,
|
| 191 |
+
"0.9": 2
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.20510494709014893
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.48372605443000793,
|
| 199 |
+
"dcs_abs": 0.4857630729675293,
|
| 200 |
+
"frac_sign_flipped": 0.15234375,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"n_alive_a": 12288,
|
| 203 |
+
"n_alive_b": 12288,
|
| 204 |
+
"matched_cos_signed": {
|
| 205 |
+
"0.5": 0.4835529923439026,
|
| 206 |
+
"0.6": 0.483801007270813,
|
| 207 |
+
"0.7": 0.4838911294937134,
|
| 208 |
+
"0.8": 0.4838899075984955,
|
| 209 |
+
"0.9": 0.48372605443000793
|
| 210 |
+
},
|
| 211 |
+
"unique_frac_signed": {
|
| 212 |
+
"0.5": 0.9954427083333334,
|
| 213 |
+
"0.6": 0.9973958333333334,
|
| 214 |
+
"0.7": 0.9982503255208334,
|
| 215 |
+
"0.8": 0.9991455078125,
|
| 216 |
+
"0.9": 1.0
|
| 217 |
+
},
|
| 218 |
+
"dedup_dcs_signed": {
|
| 219 |
+
"0.5": 0.481349300321502,
|
| 220 |
+
"0.6": 0.4825411088143786,
|
| 221 |
+
"0.7": 0.48304447753374313,
|
| 222 |
+
"0.8": 0.48347642745284247,
|
| 223 |
+
"0.9": 0.48372605443000793
|
| 224 |
+
},
|
| 225 |
+
"matched_frac_signed": {
|
| 226 |
+
"0.5": 0.4327799479166667,
|
| 227 |
+
"0.6": 0.3964029947916667,
|
| 228 |
+
"0.7": 0.3577473958333333,
|
| 229 |
+
"0.8": 0.3209635416666667,
|
| 230 |
+
"0.9": 0.28173828125
|
| 231 |
+
},
|
| 232 |
+
"matched_cos_abs": {
|
| 233 |
+
"0.5": 0.48473066091537476,
|
| 234 |
+
"0.6": 0.4850594997406006,
|
| 235 |
+
"0.7": 0.48544371128082275,
|
| 236 |
+
"0.8": 0.48560550808906555,
|
| 237 |
+
"0.9": 0.4855424761772156
|
| 238 |
+
},
|
| 239 |
+
"unique_frac_abs": {
|
| 240 |
+
"0.5": 0.9925130208333334,
|
| 241 |
+
"0.6": 0.9952392578125,
|
| 242 |
+
"0.7": 0.9969889322916666,
|
| 243 |
+
"0.8": 0.9984130859375,
|
| 244 |
+
"0.9": 0.9995524088541666
|
| 245 |
+
},
|
| 246 |
+
"dedup_dcs_abs": {
|
| 247 |
+
"0.5": 0.4811014925556568,
|
| 248 |
+
"0.6": 0.48275025651673786,
|
| 249 |
+
"0.7": 0.4839820073975716,
|
| 250 |
+
"0.8": 0.48483489387945156,
|
| 251 |
+
"0.9": 0.4853251516639526
|
| 252 |
+
},
|
| 253 |
+
"matched_frac_abs": {
|
| 254 |
+
"0.5": 0.4305013020833333,
|
| 255 |
+
"0.6": 0.3951009114583333,
|
| 256 |
+
"0.7": 0.356689453125,
|
| 257 |
+
"0.8": 0.3203125,
|
| 258 |
+
"0.9": 0.281494140625
|
| 259 |
+
}
|
| 260 |
+
},
|
| 261 |
+
"seed0\u2194seed2": {
|
| 262 |
+
"dcs_signed": 0.4850541651248932,
|
| 263 |
+
"dcs_abs": 0.48714756965637207,
|
| 264 |
+
"frac_sign_flipped": 0.15323893229166666,
|
| 265 |
+
"alive_source": "stats",
|
| 266 |
+
"n_alive_a": 12288,
|
| 267 |
+
"n_alive_b": 12288,
|
| 268 |
+
"matched_cos_signed": {
|
| 269 |
+
"0.5": 0.48470693826675415,
|
| 270 |
+
"0.6": 0.4849589765071869,
|
| 271 |
+
"0.7": 0.48511987924575806,
|
| 272 |
+
"0.8": 0.48513081669807434,
|
| 273 |
+
"0.9": 0.4850541651248932
|
| 274 |
+
},
|
| 275 |
+
"unique_frac_signed": {
|
| 276 |
+
"0.5": 0.9946695963541666,
|
| 277 |
+
"0.6": 0.9968668619791666,
|
| 278 |
+
"0.7": 0.9978841145833334,
|
| 279 |
+
"0.8": 0.9989013671875,
|
| 280 |
+
"0.9": 1.0
|
| 281 |
+
},
|
| 282 |
+
"dedup_dcs_signed": {
|
| 283 |
+
"0.5": 0.4821232546358563,
|
| 284 |
+
"0.6": 0.4834395330993478,
|
| 285 |
+
"0.7": 0.48409342116792686,
|
| 286 |
+
"0.8": 0.4845978360644949,
|
| 287 |
+
"0.9": 0.4850541651248932
|
| 288 |
+
},
|
| 289 |
+
"matched_frac_signed": {
|
| 290 |
+
"0.5": 0.4310709635416667,
|
| 291 |
+
"0.6": 0.3951009114583333,
|
| 292 |
+
"0.7": 0.3594563802083333,
|
| 293 |
+
"0.8": 0.3224283854166667,
|
| 294 |
+
"0.9": 0.2850748697916667
|
| 295 |
+
},
|
| 296 |
+
"matched_cos_abs": {
|
| 297 |
+
"0.5": 0.48573005199432373,
|
| 298 |
+
"0.6": 0.48625609278678894,
|
| 299 |
+
"0.7": 0.4866196811199188,
|
| 300 |
+
"0.8": 0.4868212938308716,
|
| 301 |
+
"0.9": 0.4869636595249176
|
| 302 |
+
},
|
| 303 |
+
"unique_frac_abs": {
|
| 304 |
+
"0.5": 0.99169921875,
|
| 305 |
+
"0.6": 0.9947509765625,
|
| 306 |
+
"0.7": 0.9966227213541666,
|
| 307 |
+
"0.8": 0.9981282552083334,
|
| 308 |
+
"0.9": 0.9996337890625
|
| 309 |
+
},
|
| 310 |
+
"dedup_dcs_abs": {
|
| 311 |
+
"0.5": 0.4816981130861677,
|
| 312 |
+
"0.6": 0.4837037231591239,
|
| 313 |
+
"0.7": 0.4849762308622303,
|
| 314 |
+
"0.8": 0.4859100886096712,
|
| 315 |
+
"0.9": 0.48678532810663455
|
| 316 |
+
},
|
| 317 |
+
"matched_frac_abs": {
|
| 318 |
+
"0.5": 0.4285481770833333,
|
| 319 |
+
"0.6": 0.3941243489583333,
|
| 320 |
+
"0.7": 0.3582356770833333,
|
| 321 |
+
"0.8": 0.3216145833333333,
|
| 322 |
+
"0.9": 0.2847493489583333
|
| 323 |
+
}
|
| 324 |
+
},
|
| 325 |
+
"seed1\u2194seed2": {
|
| 326 |
+
"dcs_signed": 0.48296311497688293,
|
| 327 |
+
"dcs_abs": 0.4852086007595062,
|
| 328 |
+
"frac_sign_flipped": 0.15388997395833334,
|
| 329 |
+
"alive_source": "stats",
|
| 330 |
+
"n_alive_a": 12288,
|
| 331 |
+
"n_alive_b": 12288,
|
| 332 |
+
"matched_cos_signed": {
|
| 333 |
+
"0.5": 0.4829995036125183,
|
| 334 |
+
"0.6": 0.48306789994239807,
|
| 335 |
+
"0.7": 0.48308712244033813,
|
| 336 |
+
"0.8": 0.4831407964229584,
|
| 337 |
+
"0.9": 0.48296311497688293
|
| 338 |
+
},
|
| 339 |
+
"unique_frac_signed": {
|
| 340 |
+
"0.5": 0.9950764973958334,
|
| 341 |
+
"0.6": 0.9971110026041666,
|
| 342 |
+
"0.7": 0.9980875651041666,
|
| 343 |
+
"0.8": 0.9991048177083334,
|
| 344 |
+
"0.9": 1.0
|
| 345 |
+
},
|
| 346 |
+
"dedup_dcs_signed": {
|
| 347 |
+
"0.5": 0.48062145429867087,
|
| 348 |
+
"0.6": 0.4816723180374538,
|
| 349 |
+
"0.7": 0.4821632497696555,
|
| 350 |
+
"0.8": 0.48270829733761883,
|
| 351 |
+
"0.9": 0.48296311497688293
|
| 352 |
+
},
|
| 353 |
+
"matched_frac_signed": {
|
| 354 |
+
"0.5": 0.4312337239583333,
|
| 355 |
+
"0.6": 0.392822265625,
|
| 356 |
+
"0.7": 0.3575846354166667,
|
| 357 |
+
"0.8": 0.3207194010416667,
|
| 358 |
+
"0.9": 0.282958984375
|
| 359 |
+
},
|
| 360 |
+
"matched_cos_abs": {
|
| 361 |
+
"0.5": 0.48432889580726624,
|
| 362 |
+
"0.6": 0.4845554232597351,
|
| 363 |
+
"0.7": 0.4847815930843353,
|
| 364 |
+
"0.8": 0.48501279950141907,
|
| 365 |
+
"0.9": 0.4850381314754486
|
| 366 |
+
},
|
| 367 |
+
"unique_frac_abs": {
|
| 368 |
+
"0.5": 0.9921875,
|
| 369 |
+
"0.6": 0.9950358072916666,
|
| 370 |
+
"0.7": 0.9970296223958334,
|
| 371 |
+
"0.8": 0.9984130859375,
|
| 372 |
+
"0.9": 0.9995930989583334
|
| 373 |
+
},
|
| 374 |
+
"dedup_dcs_abs": {
|
| 375 |
+
"0.5": 0.48054507630877197,
|
| 376 |
+
"0.6": 0.4821499967608057,
|
| 377 |
+
"0.7": 0.4833416086973254,
|
| 378 |
+
"0.8": 0.4842431258693978,
|
| 379 |
+
"0.9": 0.48484076895450323
|
| 380 |
+
},
|
| 381 |
+
"matched_frac_abs": {
|
| 382 |
+
"0.5": 0.4296875,
|
| 383 |
+
"0.6": 0.3916015625,
|
| 384 |
+
"0.7": 0.3567708333333333,
|
| 385 |
+
"0.8": 0.320068359375,
|
| 386 |
+
"0.9": 0.28271484375
|
| 387 |
+
}
|
| 388 |
+
}
|
| 389 |
+
},
|
| 390 |
+
"summary": {
|
| 391 |
+
"n_seeds": 3,
|
| 392 |
+
"n_pairs": 3,
|
| 393 |
+
"report_band": [
|
| 394 |
+
0.6,
|
| 395 |
+
0.7,
|
| 396 |
+
0.8,
|
| 397 |
+
0.9
|
| 398 |
+
],
|
| 399 |
+
"n_dead_mean": 0,
|
| 400 |
+
"n_dead_std": 0.0,
|
| 401 |
+
"n_dense_mean": 31,
|
| 402 |
+
"n_dense_std": 0.5773502691896258,
|
| 403 |
+
"l0_mean": 64,
|
| 404 |
+
"l0_std": 8.702335715267317e-15,
|
| 405 |
+
"recon_r2_mean": 0.9711333333333333,
|
| 406 |
+
"recon_r2_std": 0.0,
|
| 407 |
+
"l1_per_token_mean": 118.44936666666666,
|
| 408 |
+
"l1_per_token_std": 0.5673503238740546,
|
| 409 |
+
"babel_mu1_signed_mean": 0.19496882458527884,
|
| 410 |
+
"babel_mu1_signed_std": 0.0007046744138114592,
|
| 411 |
+
"n_dup_signed_per_threshold": {
|
| 412 |
+
"0.5": 118.66666666666667,
|
| 413 |
+
"0.6": 69.66666666666667,
|
| 414 |
+
"0.7": 47.0,
|
| 415 |
+
"0.8": 23.333333333333332,
|
| 416 |
+
"0.9": 0.0
|
| 417 |
+
},
|
| 418 |
+
"n_dup_signed_band": 35.0,
|
| 419 |
+
"n_dup_signed_band_std": 7.013380069552769,
|
| 420 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 421 |
+
"0.5": 58.0,
|
| 422 |
+
"0.6": 34.333333333333336,
|
| 423 |
+
"0.7": 23.333333333333332,
|
| 424 |
+
"0.8": 11.666666666666666,
|
| 425 |
+
"0.9": 0.0
|
| 426 |
+
},
|
| 427 |
+
"n_dup_clusters_signed_band": 17.333333333333336,
|
| 428 |
+
"n_dup_clusters_signed_band_std": 3.5029749261639505,
|
| 429 |
+
"largest_dup_signed_per_threshold": {
|
| 430 |
+
"0.5": 3.0,
|
| 431 |
+
"0.6": 3.0,
|
| 432 |
+
"0.7": 2.3333333333333335,
|
| 433 |
+
"0.8": 2.0,
|
| 434 |
+
"0.9": 1.0
|
| 435 |
+
},
|
| 436 |
+
"largest_dup_signed_band": 2.0833333333333335,
|
| 437 |
+
"largest_dup_signed_band_std": 0.14433756729740643,
|
| 438 |
+
"dcs_signed_mean": 0.48391444484392804,
|
| 439 |
+
"dcs_signed_std": 0.0010581781000456216,
|
| 440 |
+
"dedup_dcs_signed_per_threshold": {
|
| 441 |
+
"0.5": 0.48136466975200976,
|
| 442 |
+
"0.6": 0.4825509866503934,
|
| 443 |
+
"0.7": 0.4831003828237752,
|
| 444 |
+
"0.8": 0.48359418695165207,
|
| 445 |
+
"0.9": 0.48391444484392804
|
| 446 |
+
},
|
| 447 |
+
"dedup_dcs_signed_band": 0.4832900003174372,
|
| 448 |
+
"dedup_dcs_signed_band_std": 0.0009631191823450227,
|
| 449 |
+
"matched_cos_signed_per_threshold": {
|
| 450 |
+
"0.5": 0.48375314474105835,
|
| 451 |
+
"0.6": 0.4839426279067993,
|
| 452 |
+
"0.7": 0.48403271039326984,
|
| 453 |
+
"0.8": 0.4840538402398427,
|
| 454 |
+
"0.9": 0.48391444484392804
|
| 455 |
+
},
|
| 456 |
+
"matched_cos_signed_band": 0.48398590584595996,
|
| 457 |
+
"matched_cos_signed_band_std": 0.0010100290779935623,
|
| 458 |
+
"unique_frac_signed_per_threshold": {
|
| 459 |
+
"0.5": 0.9950629340277778,
|
| 460 |
+
"0.6": 0.9971245659722222,
|
| 461 |
+
"0.7": 0.9980740017361112,
|
| 462 |
+
"0.8": 0.9990505642361112,
|
| 463 |
+
"0.9": 1.0
|
| 464 |
+
},
|
| 465 |
+
"unique_frac_signed_band": 0.9985622829861112,
|
| 466 |
+
"unique_frac_signed_band_std": 0.0001428989495558702,
|
| 467 |
+
"matched_frac_signed_per_threshold": {
|
| 468 |
+
"0.5": 0.43169487847222227,
|
| 469 |
+
"0.6": 0.394775390625,
|
| 470 |
+
"0.7": 0.3582628038194444,
|
| 471 |
+
"0.8": 0.3213704427083333,
|
| 472 |
+
"0.9": 0.28325737847222227
|
| 473 |
+
},
|
| 474 |
+
"matched_frac_signed_band": 0.33941650390625,
|
| 475 |
+
"matched_frac_signed_band_std": 0.001012358016326864,
|
| 476 |
+
"babel_mu1_abs_mean": 0.20474614202976227,
|
| 477 |
+
"babel_mu1_abs_std": 0.0008177437081519053,
|
| 478 |
+
"n_dup_abs_per_threshold": {
|
| 479 |
+
"0.5": 190.66666666666666,
|
| 480 |
+
"0.6": 121.66666666666667,
|
| 481 |
+
"0.7": 76.33333333333333,
|
| 482 |
+
"0.8": 41.333333333333336,
|
| 483 |
+
"0.9": 10.0
|
| 484 |
+
},
|
| 485 |
+
"n_dup_abs_band": 62.333333333333336,
|
| 486 |
+
"n_dup_abs_band_std": 7.054844671098955,
|
| 487 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 488 |
+
"0.5": 94.0,
|
| 489 |
+
"0.6": 60.333333333333336,
|
| 490 |
+
"0.7": 38.0,
|
| 491 |
+
"0.8": 20.666666666666668,
|
| 492 |
+
"0.9": 5.0
|
| 493 |
+
},
|
| 494 |
+
"n_dup_clusters_abs_band": 31.000000000000004,
|
| 495 |
+
"n_dup_clusters_abs_band_std": 3.5,
|
| 496 |
+
"largest_dup_abs_per_threshold": {
|
| 497 |
+
"0.5": 3.0,
|
| 498 |
+
"0.6": 3.0,
|
| 499 |
+
"0.7": 2.3333333333333335,
|
| 500 |
+
"0.8": 2.0,
|
| 501 |
+
"0.9": 2.0
|
| 502 |
+
},
|
| 503 |
+
"largest_dup_abs_band": 2.3333333333333335,
|
| 504 |
+
"largest_dup_abs_band_std": 0.14433756729740643,
|
| 505 |
+
"dcs_abs_mean": 0.4860397477944692,
|
| 506 |
+
"dcs_abs_std": 0.0009986550033169392,
|
| 507 |
+
"dedup_dcs_abs_per_threshold": {
|
| 508 |
+
"0.5": 0.4811148939835322,
|
| 509 |
+
"0.6": 0.4828679921455558,
|
| 510 |
+
"0.7": 0.4840999489857091,
|
| 511 |
+
"0.8": 0.48499603611950687,
|
| 512 |
+
"0.9": 0.4856504162416968
|
| 513 |
+
},
|
| 514 |
+
"dedup_dcs_abs_band": 0.48440359837311714,
|
| 515 |
+
"dedup_dcs_abs_band_std": 0.0008642414870952852,
|
| 516 |
+
"matched_cos_abs_per_threshold": {
|
| 517 |
+
"0.5": 0.4849298695723216,
|
| 518 |
+
"0.6": 0.4852903385957082,
|
| 519 |
+
"0.7": 0.4856149951616923,
|
| 520 |
+
"0.8": 0.4858132004737854,
|
| 521 |
+
"0.9": 0.4858480890591939
|
| 522 |
+
},
|
| 523 |
+
"matched_cos_abs_band": 0.48564165582259494,
|
| 524 |
+
"matched_cos_abs_band_std": 0.0009304514475728574,
|
| 525 |
+
"unique_frac_abs_per_threshold": {
|
| 526 |
+
"0.5": 0.9921332465277778,
|
| 527 |
+
"0.6": 0.9950086805555555,
|
| 528 |
+
"0.7": 0.9968804253472222,
|
| 529 |
+
"0.8": 0.9983181423611112,
|
| 530 |
+
"0.9": 0.9995930989583334
|
| 531 |
+
},
|
| 532 |
+
"unique_frac_abs_band": 0.9974500868055556,
|
| 533 |
+
"unique_frac_abs_band_std": 0.0001446980000362782,
|
| 534 |
+
"matched_frac_abs_per_threshold": {
|
| 535 |
+
"0.5": 0.4295789930555555,
|
| 536 |
+
"0.6": 0.39360894097222215,
|
| 537 |
+
"0.7": 0.35723198784722215,
|
| 538 |
+
"0.8": 0.3206651475694444,
|
| 539 |
+
"0.9": 0.2829861111111111
|
| 540 |
+
},
|
| 541 |
+
"matched_frac_abs_band": 0.338623046875,
|
| 542 |
+
"matched_frac_abs_band_std": 0.0009656937081954605,
|
| 543 |
+
"alive_source": "stats"
|
| 544 |
+
}
|
| 545 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed0/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 0,
|
| 8 |
+
"top_k": 64,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed0/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1065,
|
| 3 |
+
"r2_score": 0.9712,
|
| 4 |
+
"l0_per_token_0": 0.0052,
|
| 5 |
+
"l0_per_token_0.001": 0.0052,
|
| 6 |
+
"l0_per_token_0.01": 0.0052,
|
| 7 |
+
"l0_per_token_0.05": 0.0052,
|
| 8 |
+
"l0_per_token_0.1": 0.0052,
|
| 9 |
+
"l0_per_token_0.5": 0.0052,
|
| 10 |
+
"l0_per_token_1.0": 0.0033,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0007,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 118.3737
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed1/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 1,
|
| 8 |
+
"top_k": 64,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed1/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1067,
|
| 3 |
+
"r2_score": 0.9711,
|
| 4 |
+
"l0_per_token_0": 0.0052,
|
| 5 |
+
"l0_per_token_0.001": 0.0052,
|
| 6 |
+
"l0_per_token_0.01": 0.0052,
|
| 7 |
+
"l0_per_token_0.05": 0.0052,
|
| 8 |
+
"l0_per_token_0.1": 0.0052,
|
| 9 |
+
"l0_per_token_0.5": 0.0052,
|
| 10 |
+
"l0_per_token_1.0": 0.0033,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0007,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 119.0517
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed2/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 2,
|
| 8 |
+
"top_k": 64,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_64/seed2/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1067,
|
| 3 |
+
"r2_score": 0.9711,
|
| 4 |
+
"l0_per_token_0": 0.0052,
|
| 5 |
+
"l0_per_token_0.001": 0.0052,
|
| 6 |
+
"l0_per_token_0.01": 0.0052,
|
| 7 |
+
"l0_per_token_0.05": 0.0052,
|
| 8 |
+
"l0_per_token_0.1": 0.0052,
|
| 9 |
+
"l0_per_token_0.5": 0.0052,
|
| 10 |
+
"l0_per_token_1.0": 0.0033,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0005,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 117.9227
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_128/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 771000,
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"l0": 127.80829961089495,
|
| 20 |
+
"recon_r2": 0.9735455875373764,
|
| 21 |
+
"l1_per_token": 187.25399052523233
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 127.86043709468223,
|
| 25 |
+
"recon_r2": 0.9736414308587705,
|
| 26 |
+
"l1_per_token": 182.5364159883529
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 127.99200907911803,
|
| 30 |
+
"recon_r2": 0.9734413572045786,
|
| 31 |
+
"l1_per_token": 189.28349931992568
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.8310382538926251,
|
| 38 |
+
"mcc_abs": 0.8313759438421854,
|
| 39 |
+
"act_iou": 0.6113557366136791,
|
| 40 |
+
"act_iou_n_pairs": 3072,
|
| 41 |
+
"input_r2": 0.9867011138524244,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.9557461591451071,
|
| 44 |
+
"0.7": 0.9612340551719217,
|
| 45 |
+
"0.8": 0.9699012364200496,
|
| 46 |
+
"0.9": 0.9774658472091793
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.9536868251646441,
|
| 50 |
+
"0.7": 0.959800092633683,
|
| 51 |
+
"0.8": 0.9688045533610871,
|
| 52 |
+
"0.9": 0.9769401192774338
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.8322673416085912,
|
| 58 |
+
"mcc_abs": 0.8325973613865977,
|
| 59 |
+
"act_iou": 0.6088632552265963,
|
| 60 |
+
"act_iou_n_pairs": 3072,
|
| 61 |
+
"input_r2": 0.986246524453457,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.9535368650553465,
|
| 64 |
+
"0.7": 0.9603274002388307,
|
| 65 |
+
"0.8": 0.9684214582711727,
|
| 66 |
+
"0.9": 0.9773431019322654
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.9512741328552985,
|
| 70 |
+
"0.7": 0.9583123924881924,
|
| 71 |
+
"0.8": 0.9678896894621989,
|
| 72 |
+
"0.9": 0.9773043417895559
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.8269515515357918,
|
| 78 |
+
"mcc_abs": 0.8272907709735621,
|
| 79 |
+
"act_iou": 0.6065678940475335,
|
| 80 |
+
"act_iou_n_pairs": 3072,
|
| 81 |
+
"input_r2": 0.9864061925544492,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.9553876908202976,
|
| 84 |
+
"0.7": 0.9624656118274896,
|
| 85 |
+
"0.8": 0.9692501314616038,
|
| 86 |
+
"0.9": 0.9764376787161292
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.9536025106162943,
|
| 90 |
+
"0.7": 0.9612463099404988,
|
| 91 |
+
"0.8": 0.9686570765783925,
|
| 92 |
+
"0.9": 0.976089531212062
|
| 93 |
+
}
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"summary": {
|
| 97 |
+
"n_seeds": 3,
|
| 98 |
+
"n_pairs": 3,
|
| 99 |
+
"report_band": [
|
| 100 |
+
0.6,
|
| 101 |
+
0.7,
|
| 102 |
+
0.8,
|
| 103 |
+
0.9
|
| 104 |
+
],
|
| 105 |
+
"l0_mean": 127.88691526156508,
|
| 106 |
+
"l0_std": 0.09467371424956884,
|
| 107 |
+
"recon_r2_mean": 0.9735427918669085,
|
| 108 |
+
"recon_r2_std": 0.0001000661211672232,
|
| 109 |
+
"l1_per_token_mean": 186.35796861117032,
|
| 110 |
+
"l1_per_token_std": 3.4616361483634215,
|
| 111 |
+
"input_r2_mean": 0.9864512769534436,
|
| 112 |
+
"input_r2_std": 0.00023062379037712866,
|
| 113 |
+
"mcc_signed_mean": 0.8300857156790027,
|
| 114 |
+
"mcc_signed_std": 0.0027829665486358767,
|
| 115 |
+
"mcc_abs_mean": 0.830421358734115,
|
| 116 |
+
"mcc_abs_std": 0.002779100573808475,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.9548902383402504,
|
| 119 |
+
"0.7": 0.9613423557460806,
|
| 120 |
+
"0.8": 0.969190942050942,
|
| 121 |
+
"0.9": 0.9770822092858579
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.9656264363557827,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.9528544895454122,
|
| 126 |
+
"0.7": 0.9597862650207913,
|
| 127 |
+
"0.8": 0.9684504398005594,
|
| 128 |
+
"0.9": 0.9767779974263506
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.9644672979482783,
|
| 131 |
+
"act_iou_mean": 0.608928961962603,
|
| 132 |
+
"act_iou_std": 0.0023945974903762947,
|
| 133 |
+
"act_iou_n_pairs": 3072.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_128/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,545 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 2,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"linkage": "complete",
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"d_sae": 3072,
|
| 20 |
+
"d_in": 768,
|
| 21 |
+
"l0": 128,
|
| 22 |
+
"l0_raw": 127.7952,
|
| 23 |
+
"recon_r2": 0.9735,
|
| 24 |
+
"l1_per_token": 187.3075,
|
| 25 |
+
"stats_source": "stats",
|
| 26 |
+
"n_dead": 0,
|
| 27 |
+
"dead_rate": 0.0,
|
| 28 |
+
"alive_source": "stats",
|
| 29 |
+
"n_dense": 268,
|
| 30 |
+
"dense_rate": 0.08723958333333333,
|
| 31 |
+
"dense_threshold": 0.1,
|
| 32 |
+
"n_dup_signed": {
|
| 33 |
+
"0.5": 4,
|
| 34 |
+
"0.6": 2,
|
| 35 |
+
"0.7": 2,
|
| 36 |
+
"0.8": 2,
|
| 37 |
+
"0.9": 0
|
| 38 |
+
},
|
| 39 |
+
"n_dup_clusters_signed": {
|
| 40 |
+
"0.5": 2,
|
| 41 |
+
"0.6": 1,
|
| 42 |
+
"0.7": 1,
|
| 43 |
+
"0.8": 1,
|
| 44 |
+
"0.9": 0
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 2,
|
| 48 |
+
"0.6": 2,
|
| 49 |
+
"0.7": 2,
|
| 50 |
+
"0.8": 2,
|
| 51 |
+
"0.9": 1
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.16538286209106445,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 108,
|
| 56 |
+
"0.6": 96,
|
| 57 |
+
"0.7": 90,
|
| 58 |
+
"0.8": 90,
|
| 59 |
+
"0.9": 80
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 54,
|
| 63 |
+
"0.6": 48,
|
| 64 |
+
"0.7": 45,
|
| 65 |
+
"0.8": 45,
|
| 66 |
+
"0.9": 40
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 2,
|
| 70 |
+
"0.6": 2,
|
| 71 |
+
"0.7": 2,
|
| 72 |
+
"0.8": 2,
|
| 73 |
+
"0.9": 2
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.2064165621995926
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 3072,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 128,
|
| 81 |
+
"l0_raw": 127.7952,
|
| 82 |
+
"recon_r2": 0.9736,
|
| 83 |
+
"l1_per_token": 182.5727,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 265,
|
| 89 |
+
"dense_rate": 0.08626302083333333,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 0,
|
| 93 |
+
"0.6": 0,
|
| 94 |
+
"0.7": 0,
|
| 95 |
+
"0.8": 0,
|
| 96 |
+
"0.9": 0
|
| 97 |
+
},
|
| 98 |
+
"n_dup_clusters_signed": {
|
| 99 |
+
"0.5": 0,
|
| 100 |
+
"0.6": 0,
|
| 101 |
+
"0.7": 0,
|
| 102 |
+
"0.8": 0,
|
| 103 |
+
"0.9": 0
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 1,
|
| 107 |
+
"0.6": 1,
|
| 108 |
+
"0.7": 1,
|
| 109 |
+
"0.8": 1,
|
| 110 |
+
"0.9": 1
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.16470392048358917,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 114,
|
| 115 |
+
"0.6": 106,
|
| 116 |
+
"0.7": 104,
|
| 117 |
+
"0.8": 100,
|
| 118 |
+
"0.9": 90
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 57,
|
| 122 |
+
"0.6": 53,
|
| 123 |
+
"0.7": 52,
|
| 124 |
+
"0.8": 50,
|
| 125 |
+
"0.9": 45
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 2,
|
| 129 |
+
"0.6": 2,
|
| 130 |
+
"0.7": 2,
|
| 131 |
+
"0.8": 2,
|
| 132 |
+
"0.9": 2
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.21009886264801025
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 3072,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 128,
|
| 140 |
+
"l0_raw": 128.1024,
|
| 141 |
+
"recon_r2": 0.9734,
|
| 142 |
+
"l1_per_token": 189.3606,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 266,
|
| 148 |
+
"dense_rate": 0.08658854166666667,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 2,
|
| 152 |
+
"0.6": 2,
|
| 153 |
+
"0.7": 0,
|
| 154 |
+
"0.8": 0,
|
| 155 |
+
"0.9": 0
|
| 156 |
+
},
|
| 157 |
+
"n_dup_clusters_signed": {
|
| 158 |
+
"0.5": 1,
|
| 159 |
+
"0.6": 1,
|
| 160 |
+
"0.7": 0,
|
| 161 |
+
"0.8": 0,
|
| 162 |
+
"0.9": 0
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 2,
|
| 166 |
+
"0.6": 2,
|
| 167 |
+
"0.7": 1,
|
| 168 |
+
"0.8": 1,
|
| 169 |
+
"0.9": 1
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.1651996225118637,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 116,
|
| 174 |
+
"0.6": 100,
|
| 175 |
+
"0.7": 94,
|
| 176 |
+
"0.8": 92,
|
| 177 |
+
"0.9": 90
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 58,
|
| 181 |
+
"0.6": 50,
|
| 182 |
+
"0.7": 47,
|
| 183 |
+
"0.8": 46,
|
| 184 |
+
"0.9": 45
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 2,
|
| 188 |
+
"0.6": 2,
|
| 189 |
+
"0.7": 2,
|
| 190 |
+
"0.8": 2,
|
| 191 |
+
"0.9": 2
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.20794568955898285
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.8042056560516357,
|
| 199 |
+
"dcs_abs": 0.8056440353393555,
|
| 200 |
+
"frac_sign_flipped": 0.0302734375,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"n_alive_a": 3072,
|
| 203 |
+
"n_alive_b": 3072,
|
| 204 |
+
"matched_cos_signed": {
|
| 205 |
+
"0.5": 0.804673433303833,
|
| 206 |
+
"0.6": 0.804445207118988,
|
| 207 |
+
"0.7": 0.804445207118988,
|
| 208 |
+
"0.8": 0.804445207118988,
|
| 209 |
+
"0.9": 0.8042056560516357
|
| 210 |
+
},
|
| 211 |
+
"unique_frac_signed": {
|
| 212 |
+
"0.5": 0.9996744791666666,
|
| 213 |
+
"0.6": 0.9998372395833334,
|
| 214 |
+
"0.7": 0.9998372395833334,
|
| 215 |
+
"0.8": 0.9998372395833334,
|
| 216 |
+
"0.9": 1.0
|
| 217 |
+
},
|
| 218 |
+
"dedup_dcs_signed": {
|
| 219 |
+
"0.5": 0.8044114953372627,
|
| 220 |
+
"0.6": 0.8043142752818919,
|
| 221 |
+
"0.7": 0.8043142752818919,
|
| 222 |
+
"0.8": 0.8043142752818919,
|
| 223 |
+
"0.9": 0.8042056560516357
|
| 224 |
+
},
|
| 225 |
+
"matched_frac_signed": {
|
| 226 |
+
"0.5": 0.8193359375,
|
| 227 |
+
"0.6": 0.7945963541666666,
|
| 228 |
+
"0.7": 0.7763671875,
|
| 229 |
+
"0.8": 0.7408854166666666,
|
| 230 |
+
"0.9": 0.6868489583333334
|
| 231 |
+
},
|
| 232 |
+
"matched_cos_abs": {
|
| 233 |
+
"0.5": 0.813916027545929,
|
| 234 |
+
"0.6": 0.8139116168022156,
|
| 235 |
+
"0.7": 0.8137345910072327,
|
| 236 |
+
"0.8": 0.8133804202079773,
|
| 237 |
+
"0.9": 0.8124451637268066
|
| 238 |
+
},
|
| 239 |
+
"unique_frac_abs": {
|
| 240 |
+
"0.5": 0.98193359375,
|
| 241 |
+
"0.6": 0.9835611979166666,
|
| 242 |
+
"0.7": 0.9842122395833334,
|
| 243 |
+
"0.8": 0.9845377604166666,
|
| 244 |
+
"0.9": 0.9861653645833334
|
| 245 |
+
},
|
| 246 |
+
"dedup_dcs_abs": {
|
| 247 |
+
"0.5": 0.799211489938898,
|
| 248 |
+
"0.6": 0.800531884820278,
|
| 249 |
+
"0.7": 0.8008875442416563,
|
| 250 |
+
"0.8": 0.8008037372783292,
|
| 251 |
+
"0.9": 0.8012052810906123
|
| 252 |
+
},
|
| 253 |
+
"matched_frac_abs": {
|
| 254 |
+
"0.5": 0.81640625,
|
| 255 |
+
"0.6": 0.7939453125,
|
| 256 |
+
"0.7": 0.775390625,
|
| 257 |
+
"0.8": 0.7399088541666666,
|
| 258 |
+
"0.9": 0.6865234375
|
| 259 |
+
}
|
| 260 |
+
},
|
| 261 |
+
"seed0\u2194seed2": {
|
| 262 |
+
"dcs_signed": 0.7999176383018494,
|
| 263 |
+
"dcs_abs": 0.8014013767242432,
|
| 264 |
+
"frac_sign_flipped": 0.030598958333333332,
|
| 265 |
+
"alive_source": "stats",
|
| 266 |
+
"n_alive_a": 3072,
|
| 267 |
+
"n_alive_b": 3072,
|
| 268 |
+
"matched_cos_signed": {
|
| 269 |
+
"0.5": 0.8001644015312195,
|
| 270 |
+
"0.6": 0.800041913986206,
|
| 271 |
+
"0.7": 0.8001572489738464,
|
| 272 |
+
"0.8": 0.8001572489738464,
|
| 273 |
+
"0.9": 0.7999176383018494
|
| 274 |
+
},
|
| 275 |
+
"unique_frac_signed": {
|
| 276 |
+
"0.5": 0.99951171875,
|
| 277 |
+
"0.6": 0.9996744791666666,
|
| 278 |
+
"0.7": 0.9998372395833334,
|
| 279 |
+
"0.8": 0.9998372395833334,
|
| 280 |
+
"0.9": 1.0
|
| 281 |
+
},
|
| 282 |
+
"dedup_dcs_signed": {
|
| 283 |
+
"0.5": 0.7997736962570343,
|
| 284 |
+
"0.6": 0.7997814836756636,
|
| 285 |
+
"0.7": 0.8000270150466047,
|
| 286 |
+
"0.8": 0.8000270150466047,
|
| 287 |
+
"0.9": 0.7999176383018494
|
| 288 |
+
},
|
| 289 |
+
"matched_frac_signed": {
|
| 290 |
+
"0.5": 0.8186848958333334,
|
| 291 |
+
"0.6": 0.7972005208333334,
|
| 292 |
+
"0.7": 0.7724609375,
|
| 293 |
+
"0.8": 0.736328125,
|
| 294 |
+
"0.9": 0.6702473958333334
|
| 295 |
+
},
|
| 296 |
+
"matched_cos_abs": {
|
| 297 |
+
"0.5": 0.8097802996635437,
|
| 298 |
+
"0.6": 0.8090988993644714,
|
| 299 |
+
"0.7": 0.808817446231842,
|
| 300 |
+
"0.8": 0.8085996508598328,
|
| 301 |
+
"0.9": 0.8084743022918701
|
| 302 |
+
},
|
| 303 |
+
"unique_frac_abs": {
|
| 304 |
+
"0.5": 0.9817708333333334,
|
| 305 |
+
"0.6": 0.9840494791666666,
|
| 306 |
+
"0.7": 0.9850260416666666,
|
| 307 |
+
"0.8": 0.9851888020833334,
|
| 308 |
+
"0.9": 0.9861653645833334
|
| 309 |
+
},
|
| 310 |
+
"dedup_dcs_abs": {
|
| 311 |
+
"0.5": 0.7950186796175938,
|
| 312 |
+
"0.6": 0.7961933505139314,
|
| 313 |
+
"0.7": 0.7967062474926934,
|
| 314 |
+
"0.8": 0.7966233213956002,
|
| 315 |
+
"0.9": 0.7972893550759181
|
| 316 |
+
},
|
| 317 |
+
"matched_frac_abs": {
|
| 318 |
+
"0.5": 0.81640625,
|
| 319 |
+
"0.6": 0.7958984375,
|
| 320 |
+
"0.7": 0.771484375,
|
| 321 |
+
"0.8": 0.7353515625,
|
| 322 |
+
"0.9": 0.669921875
|
| 323 |
+
}
|
| 324 |
+
},
|
| 325 |
+
"seed1\u2194seed2": {
|
| 326 |
+
"dcs_signed": 0.8003015518188477,
|
| 327 |
+
"dcs_abs": 0.8013460636138916,
|
| 328 |
+
"frac_sign_flipped": 0.034830729166666664,
|
| 329 |
+
"alive_source": "stats",
|
| 330 |
+
"n_alive_a": 3072,
|
| 331 |
+
"n_alive_b": 3072,
|
| 332 |
+
"matched_cos_signed": {
|
| 333 |
+
"0.5": 0.8005486130714417,
|
| 334 |
+
"0.6": 0.8005486130714417,
|
| 335 |
+
"0.7": 0.8003015518188477,
|
| 336 |
+
"0.8": 0.8003015518188477,
|
| 337 |
+
"0.9": 0.8003015518188477
|
| 338 |
+
},
|
| 339 |
+
"unique_frac_signed": {
|
| 340 |
+
"0.5": 0.9998372395833334,
|
| 341 |
+
"0.6": 0.9998372395833334,
|
| 342 |
+
"0.7": 1.0,
|
| 343 |
+
"0.8": 1.0,
|
| 344 |
+
"0.9": 1.0
|
| 345 |
+
},
|
| 346 |
+
"dedup_dcs_signed": {
|
| 347 |
+
"0.5": 0.8004183154456163,
|
| 348 |
+
"0.6": 0.8004183154456163,
|
| 349 |
+
"0.7": 0.8003015518188477,
|
| 350 |
+
"0.8": 0.8003015518188477,
|
| 351 |
+
"0.9": 0.8003015518188477
|
| 352 |
+
},
|
| 353 |
+
"matched_frac_signed": {
|
| 354 |
+
"0.5": 0.8154296875,
|
| 355 |
+
"0.6": 0.7903645833333334,
|
| 356 |
+
"0.7": 0.7659505208333334,
|
| 357 |
+
"0.8": 0.73828125,
|
| 358 |
+
"0.9": 0.6783854166666666
|
| 359 |
+
},
|
| 360 |
+
"matched_cos_abs": {
|
| 361 |
+
"0.5": 0.8090705275535583,
|
| 362 |
+
"0.6": 0.8089057207107544,
|
| 363 |
+
"0.7": 0.808976948261261,
|
| 364 |
+
"0.8": 0.8088078498840332,
|
| 365 |
+
"0.9": 0.8078117370605469
|
| 366 |
+
},
|
| 367 |
+
"unique_frac_abs": {
|
| 368 |
+
"0.5": 0.9812825520833334,
|
| 369 |
+
"0.6": 0.9832356770833334,
|
| 370 |
+
"0.7": 0.98388671875,
|
| 371 |
+
"0.8": 0.984375,
|
| 372 |
+
"0.9": 0.9853515625
|
| 373 |
+
},
|
| 374 |
+
"dedup_dcs_abs": {
|
| 375 |
+
"0.5": 0.7939267920931646,
|
| 376 |
+
"0.6": 0.7953449639996204,
|
| 377 |
+
"0.7": 0.7959416751691606,
|
| 378 |
+
"0.8": 0.7961702272295952,
|
| 379 |
+
"0.9": 0.795978557318449
|
| 380 |
+
},
|
| 381 |
+
"matched_frac_abs": {
|
| 382 |
+
"0.5": 0.8125,
|
| 383 |
+
"0.6": 0.7887369791666666,
|
| 384 |
+
"0.7": 0.7643229166666666,
|
| 385 |
+
"0.8": 0.7373046875,
|
| 386 |
+
"0.9": 0.677734375
|
| 387 |
+
}
|
| 388 |
+
}
|
| 389 |
+
},
|
| 390 |
+
"summary": {
|
| 391 |
+
"n_seeds": 3,
|
| 392 |
+
"n_pairs": 3,
|
| 393 |
+
"report_band": [
|
| 394 |
+
0.6,
|
| 395 |
+
0.7,
|
| 396 |
+
0.8,
|
| 397 |
+
0.9
|
| 398 |
+
],
|
| 399 |
+
"n_dead_mean": 0,
|
| 400 |
+
"n_dead_std": 0.0,
|
| 401 |
+
"n_dense_mean": 266,
|
| 402 |
+
"n_dense_std": 1.5275252316519465,
|
| 403 |
+
"l0_mean": 128,
|
| 404 |
+
"l0_std": 0.17736200269504993,
|
| 405 |
+
"recon_r2_mean": 0.9735,
|
| 406 |
+
"recon_r2_std": 9.999999999998899e-05,
|
| 407 |
+
"l1_per_token_mean": 186.4136,
|
| 408 |
+
"l1_per_token_std": 3.4815936470721836,
|
| 409 |
+
"babel_mu1_signed_mean": 0.16509546836217245,
|
| 410 |
+
"babel_mu1_signed_std": 0.0003512498993077609,
|
| 411 |
+
"n_dup_signed_per_threshold": {
|
| 412 |
+
"0.5": 2.0,
|
| 413 |
+
"0.6": 1.3333333333333333,
|
| 414 |
+
"0.7": 0.6666666666666666,
|
| 415 |
+
"0.8": 0.6666666666666666,
|
| 416 |
+
"0.9": 0.0
|
| 417 |
+
},
|
| 418 |
+
"n_dup_signed_band": 0.6666666666666666,
|
| 419 |
+
"n_dup_signed_band_std": 0.7637626158259733,
|
| 420 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 421 |
+
"0.5": 1.0,
|
| 422 |
+
"0.6": 0.6666666666666666,
|
| 423 |
+
"0.7": 0.3333333333333333,
|
| 424 |
+
"0.8": 0.3333333333333333,
|
| 425 |
+
"0.9": 0.0
|
| 426 |
+
},
|
| 427 |
+
"n_dup_clusters_signed_band": 0.3333333333333333,
|
| 428 |
+
"n_dup_clusters_signed_band_std": 0.38188130791298663,
|
| 429 |
+
"largest_dup_signed_per_threshold": {
|
| 430 |
+
"0.5": 1.6666666666666667,
|
| 431 |
+
"0.6": 1.6666666666666667,
|
| 432 |
+
"0.7": 1.3333333333333333,
|
| 433 |
+
"0.8": 1.3333333333333333,
|
| 434 |
+
"0.9": 1.0
|
| 435 |
+
},
|
| 436 |
+
"largest_dup_signed_band": 1.3333333333333333,
|
| 437 |
+
"largest_dup_signed_band_std": 0.38188130791298663,
|
| 438 |
+
"dcs_signed_mean": 0.801474948724111,
|
| 439 |
+
"dcs_signed_std": 0.0023726397279363364,
|
| 440 |
+
"dedup_dcs_signed_per_threshold": {
|
| 441 |
+
"0.5": 0.8015345023466378,
|
| 442 |
+
"0.6": 0.801504691467724,
|
| 443 |
+
"0.7": 0.8015476140491148,
|
| 444 |
+
"0.8": 0.8015476140491148,
|
| 445 |
+
"0.9": 0.801474948724111
|
| 446 |
+
},
|
| 447 |
+
"dedup_dcs_signed_band": 0.8015187170725162,
|
| 448 |
+
"dedup_dcs_signed_band_std": 0.0024055245209314044,
|
| 449 |
+
"matched_cos_signed_per_threshold": {
|
| 450 |
+
"0.5": 0.801795482635498,
|
| 451 |
+
"0.6": 0.8016785780588785,
|
| 452 |
+
"0.7": 0.801634669303894,
|
| 453 |
+
"0.8": 0.801634669303894,
|
| 454 |
+
"0.9": 0.801474948724111
|
| 455 |
+
},
|
| 456 |
+
"matched_cos_signed_band": 0.8016057163476944,
|
| 457 |
+
"matched_cos_signed_band_std": 0.002411715588716453,
|
| 458 |
+
"unique_frac_signed_per_threshold": {
|
| 459 |
+
"0.5": 0.9996744791666666,
|
| 460 |
+
"0.6": 0.9997829861111112,
|
| 461 |
+
"0.7": 0.9998914930555557,
|
| 462 |
+
"0.8": 0.9998914930555557,
|
| 463 |
+
"0.9": 1.0
|
| 464 |
+
},
|
| 465 |
+
"unique_frac_signed_band": 0.9998914930555557,
|
| 466 |
+
"unique_frac_signed_band_std": 6.215516079313338e-05,
|
| 467 |
+
"matched_frac_signed_per_threshold": {
|
| 468 |
+
"0.5": 0.8178168402777778,
|
| 469 |
+
"0.6": 0.7940538194444445,
|
| 470 |
+
"0.7": 0.7715928819444445,
|
| 471 |
+
"0.8": 0.7384982638888888,
|
| 472 |
+
"0.9": 0.6784939236111112
|
| 473 |
+
},
|
| 474 |
+
"matched_frac_signed_band": 0.7456597222222222,
|
| 475 |
+
"matched_frac_signed_band_std": 0.003500610439613744,
|
| 476 |
+
"babel_mu1_abs_mean": 0.20815370480219522,
|
| 477 |
+
"babel_mu1_abs_std": 0.001849942405633419,
|
| 478 |
+
"n_dup_abs_per_threshold": {
|
| 479 |
+
"0.5": 112.66666666666667,
|
| 480 |
+
"0.6": 100.66666666666667,
|
| 481 |
+
"0.7": 96.0,
|
| 482 |
+
"0.8": 94.0,
|
| 483 |
+
"0.9": 86.66666666666667
|
| 484 |
+
},
|
| 485 |
+
"n_dup_abs_band": 94.33333333333334,
|
| 486 |
+
"n_dup_abs_band_std": 5.507570547286102,
|
| 487 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 488 |
+
"0.5": 56.333333333333336,
|
| 489 |
+
"0.6": 50.333333333333336,
|
| 490 |
+
"0.7": 48.0,
|
| 491 |
+
"0.8": 47.0,
|
| 492 |
+
"0.9": 43.333333333333336
|
| 493 |
+
},
|
| 494 |
+
"n_dup_clusters_abs_band": 47.16666666666667,
|
| 495 |
+
"n_dup_clusters_abs_band_std": 2.753785273643051,
|
| 496 |
+
"largest_dup_abs_per_threshold": {
|
| 497 |
+
"0.5": 2.0,
|
| 498 |
+
"0.6": 2.0,
|
| 499 |
+
"0.7": 2.0,
|
| 500 |
+
"0.8": 2.0,
|
| 501 |
+
"0.9": 2.0
|
| 502 |
+
},
|
| 503 |
+
"largest_dup_abs_band": 2.0,
|
| 504 |
+
"largest_dup_abs_band_std": 0.0,
|
| 505 |
+
"dcs_abs_mean": 0.8027971585591634,
|
| 506 |
+
"dcs_abs_std": 0.002465622727880669,
|
| 507 |
+
"dedup_dcs_abs_per_threshold": {
|
| 508 |
+
"0.5": 0.7960523205498854,
|
| 509 |
+
"0.6": 0.7973567331112766,
|
| 510 |
+
"0.7": 0.7978451556345035,
|
| 511 |
+
"0.8": 0.7978657619678415,
|
| 512 |
+
"0.9": 0.7981577311616599
|
| 513 |
+
},
|
| 514 |
+
"dedup_dcs_abs_band": 0.7978063454688205,
|
| 515 |
+
"dedup_dcs_abs_band_std": 0.002675547689044904,
|
| 516 |
+
"matched_cos_abs_per_threshold": {
|
| 517 |
+
"0.5": 0.8109222849210104,
|
| 518 |
+
"0.6": 0.8106387456258138,
|
| 519 |
+
"0.7": 0.8105096618334452,
|
| 520 |
+
"0.8": 0.8102626403172811,
|
| 521 |
+
"0.9": 0.8095770676930746
|
| 522 |
+
},
|
| 523 |
+
"matched_cos_abs_band": 0.8102470288674037,
|
| 524 |
+
"matched_cos_abs_band_std": 0.0027034835912255356,
|
| 525 |
+
"unique_frac_abs_per_threshold": {
|
| 526 |
+
"0.5": 0.981662326388889,
|
| 527 |
+
"0.6": 0.9836154513888888,
|
| 528 |
+
"0.7": 0.984375,
|
| 529 |
+
"0.8": 0.9847005208333334,
|
| 530 |
+
"0.9": 0.9858940972222223
|
| 531 |
+
},
|
| 532 |
+
"unique_frac_abs_band": 0.9846462673611112,
|
| 533 |
+
"unique_frac_abs_band_std": 0.00044820723854865576,
|
| 534 |
+
"matched_frac_abs_per_threshold": {
|
| 535 |
+
"0.5": 0.8151041666666666,
|
| 536 |
+
"0.6": 0.7928602430555555,
|
| 537 |
+
"0.7": 0.7703993055555555,
|
| 538 |
+
"0.8": 0.7375217013888888,
|
| 539 |
+
"0.9": 0.6780598958333334
|
| 540 |
+
},
|
| 541 |
+
"matched_frac_abs_band": 0.7447102864583334,
|
| 542 |
+
"matched_frac_abs_band_std": 0.0037088310107766466,
|
| 543 |
+
"alive_source": "stats"
|
| 544 |
+
}
|
| 545 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed0/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 4,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 0,
|
| 8 |
+
"top_k": 128,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed0/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.0978,
|
| 3 |
+
"r2_score": 0.9735,
|
| 4 |
+
"l0_per_token_0": 0.0416,
|
| 5 |
+
"l0_per_token_0.001": 0.0416,
|
| 6 |
+
"l0_per_token_0.01": 0.0416,
|
| 7 |
+
"l0_per_token_0.05": 0.0416,
|
| 8 |
+
"l0_per_token_0.1": 0.0416,
|
| 9 |
+
"l0_per_token_0.5": 0.0407,
|
| 10 |
+
"l0_per_token_1.0": 0.0195,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0544,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 187.3075
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed1/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 4,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 1,
|
| 8 |
+
"top_k": 128,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed1/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.0974,
|
| 3 |
+
"r2_score": 0.9736,
|
| 4 |
+
"l0_per_token_0": 0.0416,
|
| 5 |
+
"l0_per_token_0.001": 0.0416,
|
| 6 |
+
"l0_per_token_0.01": 0.0416,
|
| 7 |
+
"l0_per_token_0.05": 0.0416,
|
| 8 |
+
"l0_per_token_0.1": 0.0416,
|
| 9 |
+
"l0_per_token_0.5": 0.0408,
|
| 10 |
+
"l0_per_token_1.0": 0.0195,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0521,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 182.5727
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed2/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 4,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 2,
|
| 8 |
+
"top_k": 128,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_128/seed2/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.0981,
|
| 3 |
+
"r2_score": 0.9734,
|
| 4 |
+
"l0_per_token_0": 0.0417,
|
| 5 |
+
"l0_per_token_0.001": 0.0417,
|
| 6 |
+
"l0_per_token_0.01": 0.0417,
|
| 7 |
+
"l0_per_token_0.05": 0.0417,
|
| 8 |
+
"l0_per_token_0.1": 0.0417,
|
| 9 |
+
"l0_per_token_0.5": 0.041,
|
| 10 |
+
"l0_per_token_1.0": 0.0197,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0544,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 189.3606
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_32/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 771000,
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"l0": 32.05078728923476,
|
| 20 |
+
"recon_r2": 0.9551329095079198,
|
| 21 |
+
"l1_per_token": 92.1583183642247
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 32.02948119325551,
|
| 25 |
+
"recon_r2": 0.9551697858025693,
|
| 26 |
+
"l1_per_token": 91.46734735581909
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 32.08818677042802,
|
| 30 |
+
"recon_r2": 0.9549875118596333,
|
| 31 |
+
"l1_per_token": 92.66057466775217
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.8092696332763701,
|
| 38 |
+
"mcc_abs": 0.809320736819735,
|
| 39 |
+
"act_iou": 0.6617149797421935,
|
| 40 |
+
"act_iou_n_pairs": 3072,
|
| 41 |
+
"input_r2": 0.9897753725043655,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.955324509817214,
|
| 44 |
+
"0.7": 0.9604592017666529,
|
| 45 |
+
"0.8": 0.9669165534081017,
|
| 46 |
+
"0.9": 0.9737470935613771
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.9554403111684986,
|
| 50 |
+
"0.7": 0.9603981082461872,
|
| 51 |
+
"0.8": 0.9669412955504316,
|
| 52 |
+
"0.9": 0.9737298452919518
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.8072503136971415,
|
| 58 |
+
"mcc_abs": 0.8073184298384352,
|
| 59 |
+
"act_iou": 0.6611586751094007,
|
| 60 |
+
"act_iou_n_pairs": 3072,
|
| 61 |
+
"input_r2": 0.9898314363046794,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.9512325816239874,
|
| 64 |
+
"0.7": 0.9593405843203472,
|
| 65 |
+
"0.8": 0.9657912633121669,
|
| 66 |
+
"0.9": 0.9734224952514393
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.950712095629045,
|
| 70 |
+
"0.7": 0.9592688156511522,
|
| 71 |
+
"0.8": 0.965658891535763,
|
| 72 |
+
"0.9": 0.9734034792912942
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.8070326176458901,
|
| 78 |
+
"mcc_abs": 0.8071133985380677,
|
| 79 |
+
"act_iou": 0.660548653429746,
|
| 80 |
+
"act_iou_n_pairs": 3072,
|
| 81 |
+
"input_r2": 0.9895853266321959,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.9536620503077697,
|
| 84 |
+
"0.7": 0.9612059647835028,
|
| 85 |
+
"0.8": 0.9679499861215163,
|
| 86 |
+
"0.9": 0.9739898945330548
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.9529595934941167,
|
| 90 |
+
"0.7": 0.9611494487697142,
|
| 91 |
+
"0.8": 0.9679294580261005,
|
| 92 |
+
"0.9": 0.9739736974175337
|
| 93 |
+
}
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"summary": {
|
| 97 |
+
"n_seeds": 3,
|
| 98 |
+
"n_pairs": 3,
|
| 99 |
+
"report_band": [
|
| 100 |
+
0.6,
|
| 101 |
+
0.7,
|
| 102 |
+
0.8,
|
| 103 |
+
0.9
|
| 104 |
+
],
|
| 105 |
+
"l0_mean": 32.05615175097277,
|
| 106 |
+
"l0_std": 0.029718164228179585,
|
| 107 |
+
"recon_r2_mean": 0.9550967357233741,
|
| 108 |
+
"recon_r2_std": 9.6370921883266e-05,
|
| 109 |
+
"l1_per_token_mean": 92.09541346259864,
|
| 110 |
+
"l1_per_token_std": 0.5990956722195556,
|
| 111 |
+
"input_r2_mean": 0.9897307118137469,
|
| 112 |
+
"input_r2_std": 0.0001289900216581967,
|
| 113 |
+
"mcc_signed_mean": 0.8078508548731339,
|
| 114 |
+
"mcc_signed_std": 0.0012335100367649142,
|
| 115 |
+
"mcc_abs_mean": 0.8079175217320792,
|
| 116 |
+
"mcc_abs_std": 0.0012195363443543187,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.9534063805829903,
|
| 119 |
+
"0.7": 0.9603352502901678,
|
| 120 |
+
"0.8": 0.966885934280595,
|
| 121 |
+
"0.9": 0.9737198277819571
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.9635868482339276,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.9530373334305535,
|
| 126 |
+
"0.7": 0.9602721242223512,
|
| 127 |
+
"0.8": 0.9668432150374318,
|
| 128 |
+
"0.9": 0.9737023406669266
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.9634637533393158,
|
| 131 |
+
"act_iou_mean": 0.6611407694271134,
|
| 132 |
+
"act_iou_std": 0.0005833692885904101,
|
| 133 |
+
"act_iou_n_pairs": 3072.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_32/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,545 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 2,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"linkage": "complete",
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"d_sae": 3072,
|
| 20 |
+
"d_in": 768,
|
| 21 |
+
"l0": 32,
|
| 22 |
+
"l0_raw": 31.9488,
|
| 23 |
+
"recon_r2": 0.9551,
|
| 24 |
+
"l1_per_token": 92.1742,
|
| 25 |
+
"stats_source": "stats",
|
| 26 |
+
"n_dead": 0,
|
| 27 |
+
"dead_rate": 0.0,
|
| 28 |
+
"alive_source": "stats",
|
| 29 |
+
"n_dense": 16,
|
| 30 |
+
"dense_rate": 0.005208333333333333,
|
| 31 |
+
"dense_threshold": 0.1,
|
| 32 |
+
"n_dup_signed": {
|
| 33 |
+
"0.5": 8,
|
| 34 |
+
"0.6": 2,
|
| 35 |
+
"0.7": 0,
|
| 36 |
+
"0.8": 0,
|
| 37 |
+
"0.9": 0
|
| 38 |
+
},
|
| 39 |
+
"n_dup_clusters_signed": {
|
| 40 |
+
"0.5": 4,
|
| 41 |
+
"0.6": 1,
|
| 42 |
+
"0.7": 0,
|
| 43 |
+
"0.8": 0,
|
| 44 |
+
"0.9": 0
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 2,
|
| 48 |
+
"0.6": 2,
|
| 49 |
+
"0.7": 1,
|
| 50 |
+
"0.8": 1,
|
| 51 |
+
"0.9": 1
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.19082637131214142,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 24,
|
| 56 |
+
"0.6": 12,
|
| 57 |
+
"0.7": 8,
|
| 58 |
+
"0.8": 8,
|
| 59 |
+
"0.9": 6
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 12,
|
| 63 |
+
"0.6": 6,
|
| 64 |
+
"0.7": 4,
|
| 65 |
+
"0.8": 4,
|
| 66 |
+
"0.9": 3
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 2,
|
| 70 |
+
"0.6": 2,
|
| 71 |
+
"0.7": 2,
|
| 72 |
+
"0.8": 2,
|
| 73 |
+
"0.9": 2
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.19878263771533966
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 3072,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 32,
|
| 81 |
+
"l0_raw": 31.9488,
|
| 82 |
+
"recon_r2": 0.9551,
|
| 83 |
+
"l1_per_token": 91.4717,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 16,
|
| 89 |
+
"dense_rate": 0.005208333333333333,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 6,
|
| 93 |
+
"0.6": 2,
|
| 94 |
+
"0.7": 0,
|
| 95 |
+
"0.8": 0,
|
| 96 |
+
"0.9": 0
|
| 97 |
+
},
|
| 98 |
+
"n_dup_clusters_signed": {
|
| 99 |
+
"0.5": 3,
|
| 100 |
+
"0.6": 1,
|
| 101 |
+
"0.7": 0,
|
| 102 |
+
"0.8": 0,
|
| 103 |
+
"0.9": 0
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 2,
|
| 107 |
+
"0.6": 2,
|
| 108 |
+
"0.7": 1,
|
| 109 |
+
"0.8": 1,
|
| 110 |
+
"0.9": 1
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.19185525178909302,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 20,
|
| 115 |
+
"0.6": 12,
|
| 116 |
+
"0.7": 10,
|
| 117 |
+
"0.8": 6,
|
| 118 |
+
"0.9": 6
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 10,
|
| 122 |
+
"0.6": 6,
|
| 123 |
+
"0.7": 5,
|
| 124 |
+
"0.8": 3,
|
| 125 |
+
"0.9": 3
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 2,
|
| 129 |
+
"0.6": 2,
|
| 130 |
+
"0.7": 2,
|
| 131 |
+
"0.8": 2,
|
| 132 |
+
"0.9": 2
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.19969023764133453
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 3072,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 32,
|
| 140 |
+
"l0_raw": 31.9488,
|
| 141 |
+
"recon_r2": 0.955,
|
| 142 |
+
"l1_per_token": 92.6854,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 14,
|
| 148 |
+
"dense_rate": 0.004557291666666667,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 4,
|
| 152 |
+
"0.6": 0,
|
| 153 |
+
"0.7": 0,
|
| 154 |
+
"0.8": 0,
|
| 155 |
+
"0.9": 0
|
| 156 |
+
},
|
| 157 |
+
"n_dup_clusters_signed": {
|
| 158 |
+
"0.5": 2,
|
| 159 |
+
"0.6": 0,
|
| 160 |
+
"0.7": 0,
|
| 161 |
+
"0.8": 0,
|
| 162 |
+
"0.9": 0
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 2,
|
| 166 |
+
"0.6": 1,
|
| 167 |
+
"0.7": 1,
|
| 168 |
+
"0.8": 1,
|
| 169 |
+
"0.9": 1
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.19006387889385223,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 12,
|
| 174 |
+
"0.6": 6,
|
| 175 |
+
"0.7": 4,
|
| 176 |
+
"0.8": 4,
|
| 177 |
+
"0.9": 4
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 6,
|
| 181 |
+
"0.6": 3,
|
| 182 |
+
"0.7": 2,
|
| 183 |
+
"0.8": 2,
|
| 184 |
+
"0.9": 2
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 2,
|
| 188 |
+
"0.6": 2,
|
| 189 |
+
"0.7": 2,
|
| 190 |
+
"0.8": 2,
|
| 191 |
+
"0.9": 2
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.19755180180072784
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.828850269317627,
|
| 199 |
+
"dcs_abs": 0.830113410949707,
|
| 200 |
+
"frac_sign_flipped": 0.040364583333333336,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"n_alive_a": 3072,
|
| 203 |
+
"n_alive_b": 3072,
|
| 204 |
+
"matched_cos_signed": {
|
| 205 |
+
"0.5": 0.8289244771003723,
|
| 206 |
+
"0.6": 0.8288153409957886,
|
| 207 |
+
"0.7": 0.828850269317627,
|
| 208 |
+
"0.8": 0.828850269317627,
|
| 209 |
+
"0.9": 0.828850269317627
|
| 210 |
+
},
|
| 211 |
+
"unique_frac_signed": {
|
| 212 |
+
"0.5": 0.9988606770833334,
|
| 213 |
+
"0.6": 0.9996744791666666,
|
| 214 |
+
"0.7": 1.0,
|
| 215 |
+
"0.8": 1.0,
|
| 216 |
+
"0.9": 1.0
|
| 217 |
+
},
|
| 218 |
+
"dedup_dcs_signed": {
|
| 219 |
+
"0.5": 0.827980064447426,
|
| 220 |
+
"0.6": 0.8285455443353081,
|
| 221 |
+
"0.7": 0.828850269317627,
|
| 222 |
+
"0.8": 0.828850269317627,
|
| 223 |
+
"0.9": 0.828850269317627
|
| 224 |
+
},
|
| 225 |
+
"matched_frac_signed": {
|
| 226 |
+
"0.5": 0.8326822916666666,
|
| 227 |
+
"0.6": 0.8209635416666666,
|
| 228 |
+
"0.7": 0.8089192708333334,
|
| 229 |
+
"0.8": 0.7848307291666666,
|
| 230 |
+
"0.9": 0.748046875
|
| 231 |
+
},
|
| 232 |
+
"matched_cos_abs": {
|
| 233 |
+
"0.5": 0.8299980163574219,
|
| 234 |
+
"0.6": 0.8299298286437988,
|
| 235 |
+
"0.7": 0.8300195336341858,
|
| 236 |
+
"0.8": 0.8299959897994995,
|
| 237 |
+
"0.9": 0.8300079703330994
|
| 238 |
+
},
|
| 239 |
+
"unique_frac_abs": {
|
| 240 |
+
"0.5": 0.9964192708333334,
|
| 241 |
+
"0.6": 0.998046875,
|
| 242 |
+
"0.7": 0.99853515625,
|
| 243 |
+
"0.8": 0.9988606770833334,
|
| 244 |
+
"0.9": 0.9990234375
|
| 245 |
+
},
|
| 246 |
+
"dedup_dcs_abs": {
|
| 247 |
+
"0.5": 0.8270260182519754,
|
| 248 |
+
"0.6": 0.8283088719472289,
|
| 249 |
+
"0.7": 0.8288036847079638,
|
| 250 |
+
"0.8": 0.8290503563475795,
|
| 251 |
+
"0.9": 0.829197415674571
|
| 252 |
+
},
|
| 253 |
+
"matched_frac_abs": {
|
| 254 |
+
"0.5": 0.8297526041666666,
|
| 255 |
+
"0.6": 0.8199869791666666,
|
| 256 |
+
"0.7": 0.8082682291666666,
|
| 257 |
+
"0.8": 0.7838541666666666,
|
| 258 |
+
"0.9": 0.7473958333333334
|
| 259 |
+
}
|
| 260 |
+
},
|
| 261 |
+
"seed0\u2194seed2": {
|
| 262 |
+
"dcs_signed": 0.828951358795166,
|
| 263 |
+
"dcs_abs": 0.8302128911018372,
|
| 264 |
+
"frac_sign_flipped": 0.036783854166666664,
|
| 265 |
+
"alive_source": "stats",
|
| 266 |
+
"n_alive_a": 3072,
|
| 267 |
+
"n_alive_b": 3072,
|
| 268 |
+
"matched_cos_signed": {
|
| 269 |
+
"0.5": 0.8293708562850952,
|
| 270 |
+
"0.6": 0.8291928172111511,
|
| 271 |
+
"0.7": 0.828951358795166,
|
| 272 |
+
"0.8": 0.828951358795166,
|
| 273 |
+
"0.9": 0.828951358795166
|
| 274 |
+
},
|
| 275 |
+
"unique_frac_signed": {
|
| 276 |
+
"0.5": 0.9990234375,
|
| 277 |
+
"0.6": 0.9998372395833334,
|
| 278 |
+
"0.7": 1.0,
|
| 279 |
+
"0.8": 1.0,
|
| 280 |
+
"0.9": 1.0
|
| 281 |
+
},
|
| 282 |
+
"dedup_dcs_signed": {
|
| 283 |
+
"0.5": 0.8285609238082543,
|
| 284 |
+
"0.6": 0.8290578574427249,
|
| 285 |
+
"0.7": 0.828951358795166,
|
| 286 |
+
"0.8": 0.828951358795166,
|
| 287 |
+
"0.9": 0.828951358795166
|
| 288 |
+
},
|
| 289 |
+
"matched_frac_signed": {
|
| 290 |
+
"0.5": 0.8382161458333334,
|
| 291 |
+
"0.6": 0.8216145833333334,
|
| 292 |
+
"0.7": 0.802734375,
|
| 293 |
+
"0.8": 0.7819010416666666,
|
| 294 |
+
"0.9": 0.7415364583333334
|
| 295 |
+
},
|
| 296 |
+
"matched_cos_abs": {
|
| 297 |
+
"0.5": 0.830889880657196,
|
| 298 |
+
"0.6": 0.8306258320808411,
|
| 299 |
+
"0.7": 0.8302557468414307,
|
| 300 |
+
"0.8": 0.8302557468414307,
|
| 301 |
+
"0.9": 0.8302267789840698
|
| 302 |
+
},
|
| 303 |
+
"unique_frac_abs": {
|
| 304 |
+
"0.5": 0.9970703125,
|
| 305 |
+
"0.6": 0.99853515625,
|
| 306 |
+
"0.7": 0.9990234375,
|
| 307 |
+
"0.8": 0.9990234375,
|
| 308 |
+
"0.9": 0.9991861979166666
|
| 309 |
+
},
|
| 310 |
+
"dedup_dcs_abs": {
|
| 311 |
+
"0.5": 0.8284556329599582,
|
| 312 |
+
"0.6": 0.8294090950221289,
|
| 313 |
+
"0.7": 0.8294449502136558,
|
| 314 |
+
"0.8": 0.8294449502136558,
|
| 315 |
+
"0.9": 0.8295511387016934
|
| 316 |
+
},
|
| 317 |
+
"matched_frac_abs": {
|
| 318 |
+
"0.5": 0.8372395833333334,
|
| 319 |
+
"0.6": 0.8209635416666666,
|
| 320 |
+
"0.7": 0.8024088541666666,
|
| 321 |
+
"0.8": 0.7815755208333334,
|
| 322 |
+
"0.9": 0.7408854166666666
|
| 323 |
+
}
|
| 324 |
+
},
|
| 325 |
+
"seed1\u2194seed2": {
|
| 326 |
+
"dcs_signed": 0.8261902332305908,
|
| 327 |
+
"dcs_abs": 0.826953113079071,
|
| 328 |
+
"frac_sign_flipped": 0.033528645833333336,
|
| 329 |
+
"alive_source": "stats",
|
| 330 |
+
"n_alive_a": 3072,
|
| 331 |
+
"n_alive_b": 3072,
|
| 332 |
+
"matched_cos_signed": {
|
| 333 |
+
"0.5": 0.8262224793434143,
|
| 334 |
+
"0.6": 0.8264197707176208,
|
| 335 |
+
"0.7": 0.8261902332305908,
|
| 336 |
+
"0.8": 0.8261902332305908,
|
| 337 |
+
"0.9": 0.8261902332305908
|
| 338 |
+
},
|
| 339 |
+
"unique_frac_signed": {
|
| 340 |
+
"0.5": 0.9991861979166666,
|
| 341 |
+
"0.6": 0.9998372395833334,
|
| 342 |
+
"0.7": 1.0,
|
| 343 |
+
"0.8": 1.0,
|
| 344 |
+
"0.9": 1.0
|
| 345 |
+
},
|
| 346 |
+
"dedup_dcs_signed": {
|
| 347 |
+
"0.5": 0.8255500977684278,
|
| 348 |
+
"0.6": 0.8262852622913973,
|
| 349 |
+
"0.7": 0.8261902332305908,
|
| 350 |
+
"0.8": 0.8261902332305908,
|
| 351 |
+
"0.9": 0.8261902332305908
|
| 352 |
+
},
|
| 353 |
+
"matched_frac_signed": {
|
| 354 |
+
"0.5": 0.833984375,
|
| 355 |
+
"0.6": 0.8193359375,
|
| 356 |
+
"0.7": 0.8011067708333334,
|
| 357 |
+
"0.8": 0.7783203125,
|
| 358 |
+
"0.9": 0.7434895833333334
|
| 359 |
+
},
|
| 360 |
+
"matched_cos_abs": {
|
| 361 |
+
"0.5": 0.8272656202316284,
|
| 362 |
+
"0.6": 0.8273926377296448,
|
| 363 |
+
"0.7": 0.8272384405136108,
|
| 364 |
+
"0.8": 0.8270357251167297,
|
| 365 |
+
"0.9": 0.8270357251167297
|
| 366 |
+
},
|
| 367 |
+
"unique_frac_abs": {
|
| 368 |
+
"0.5": 0.9973958333333334,
|
| 369 |
+
"0.6": 0.99853515625,
|
| 370 |
+
"0.7": 0.9988606770833334,
|
| 371 |
+
"0.8": 0.9991861979166666,
|
| 372 |
+
"0.9": 0.9991861979166666
|
| 373 |
+
},
|
| 374 |
+
"dedup_dcs_abs": {
|
| 375 |
+
"0.5": 0.825111282678942,
|
| 376 |
+
"0.6": 0.8261806367954705,
|
| 377 |
+
"0.7": 0.8262959488007862,
|
| 378 |
+
"0.8": 0.8263626817206386,
|
| 379 |
+
"0.9": 0.8263626817206386
|
| 380 |
+
},
|
| 381 |
+
"matched_frac_abs": {
|
| 382 |
+
"0.5": 0.83203125,
|
| 383 |
+
"0.6": 0.8196614583333334,
|
| 384 |
+
"0.7": 0.8004557291666666,
|
| 385 |
+
"0.8": 0.7776692708333334,
|
| 386 |
+
"0.9": 0.7428385416666666
|
| 387 |
+
}
|
| 388 |
+
}
|
| 389 |
+
},
|
| 390 |
+
"summary": {
|
| 391 |
+
"n_seeds": 3,
|
| 392 |
+
"n_pairs": 3,
|
| 393 |
+
"report_band": [
|
| 394 |
+
0.6,
|
| 395 |
+
0.7,
|
| 396 |
+
0.8,
|
| 397 |
+
0.9
|
| 398 |
+
],
|
| 399 |
+
"n_dead_mean": 0,
|
| 400 |
+
"n_dead_std": 0.0,
|
| 401 |
+
"n_dense_mean": 15,
|
| 402 |
+
"n_dense_std": 1.1547005383792517,
|
| 403 |
+
"l0_mean": 32,
|
| 404 |
+
"l0_std": 4.351167857633658e-15,
|
| 405 |
+
"recon_r2_mean": 0.9550666666666666,
|
| 406 |
+
"recon_r2_std": 5.773502691895622e-05,
|
| 407 |
+
"l1_per_token_mean": 92.11043333333333,
|
| 408 |
+
"l1_per_token_std": 0.6085769220074007,
|
| 409 |
+
"babel_mu1_signed_mean": 0.19091516733169556,
|
| 410 |
+
"babel_mu1_signed_std": 0.0008989815138629554,
|
| 411 |
+
"n_dup_signed_per_threshold": {
|
| 412 |
+
"0.5": 6.0,
|
| 413 |
+
"0.6": 1.3333333333333333,
|
| 414 |
+
"0.7": 0.0,
|
| 415 |
+
"0.8": 0.0,
|
| 416 |
+
"0.9": 0.0
|
| 417 |
+
},
|
| 418 |
+
"n_dup_signed_band": 0.3333333333333333,
|
| 419 |
+
"n_dup_signed_band_std": 0.2886751345948129,
|
| 420 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 421 |
+
"0.5": 3.0,
|
| 422 |
+
"0.6": 0.6666666666666666,
|
| 423 |
+
"0.7": 0.0,
|
| 424 |
+
"0.8": 0.0,
|
| 425 |
+
"0.9": 0.0
|
| 426 |
+
},
|
| 427 |
+
"n_dup_clusters_signed_band": 0.16666666666666666,
|
| 428 |
+
"n_dup_clusters_signed_band_std": 0.14433756729740646,
|
| 429 |
+
"largest_dup_signed_per_threshold": {
|
| 430 |
+
"0.5": 2.0,
|
| 431 |
+
"0.6": 1.6666666666666667,
|
| 432 |
+
"0.7": 1.0,
|
| 433 |
+
"0.8": 1.0,
|
| 434 |
+
"0.9": 1.0
|
| 435 |
+
},
|
| 436 |
+
"largest_dup_signed_band": 1.1666666666666667,
|
| 437 |
+
"largest_dup_signed_band_std": 0.14433756729740646,
|
| 438 |
+
"dcs_signed_mean": 0.8279972871144613,
|
| 439 |
+
"dcs_signed_std": 0.001565770601022295,
|
| 440 |
+
"dedup_dcs_signed_per_threshold": {
|
| 441 |
+
"0.5": 0.8273636953413694,
|
| 442 |
+
"0.6": 0.8279628880231434,
|
| 443 |
+
"0.7": 0.8279972871144613,
|
| 444 |
+
"0.8": 0.8279972871144613,
|
| 445 |
+
"0.9": 0.8279972871144613
|
| 446 |
+
},
|
| 447 |
+
"dedup_dcs_signed_band": 0.8279886873416319,
|
| 448 |
+
"dedup_dcs_signed_band_std": 0.00154031003469465,
|
| 449 |
+
"matched_cos_signed_per_threshold": {
|
| 450 |
+
"0.5": 0.8281726042429606,
|
| 451 |
+
"0.6": 0.8281426429748535,
|
| 452 |
+
"0.7": 0.8279972871144613,
|
| 453 |
+
"0.8": 0.8279972871144613,
|
| 454 |
+
"0.9": 0.8279972871144613
|
| 455 |
+
},
|
| 456 |
+
"matched_cos_signed_band": 0.8280336260795594,
|
| 457 |
+
"matched_cos_signed_band_std": 0.0015490676366562578,
|
| 458 |
+
"unique_frac_signed_per_threshold": {
|
| 459 |
+
"0.5": 0.9990234375,
|
| 460 |
+
"0.6": 0.9997829861111112,
|
| 461 |
+
"0.7": 1.0,
|
| 462 |
+
"0.8": 1.0,
|
| 463 |
+
"0.9": 1.0
|
| 464 |
+
},
|
| 465 |
+
"unique_frac_signed_band": 0.9999457465277778,
|
| 466 |
+
"unique_frac_signed_band_std": 2.3492442594021645e-05,
|
| 467 |
+
"matched_frac_signed_per_threshold": {
|
| 468 |
+
"0.5": 0.8349609375,
|
| 469 |
+
"0.6": 0.8206380208333334,
|
| 470 |
+
"0.7": 0.8042534722222223,
|
| 471 |
+
"0.8": 0.7816840277777777,
|
| 472 |
+
"0.9": 0.744357638888889
|
| 473 |
+
},
|
| 474 |
+
"matched_frac_signed_band": 0.7877332899305556,
|
| 475 |
+
"matched_frac_signed_band_std": 0.0026524621887845864,
|
| 476 |
+
"babel_mu1_abs_mean": 0.1986748923858007,
|
| 477 |
+
"babel_mu1_abs_std": 0.001073281767815846,
|
| 478 |
+
"n_dup_abs_per_threshold": {
|
| 479 |
+
"0.5": 18.666666666666668,
|
| 480 |
+
"0.6": 10.0,
|
| 481 |
+
"0.7": 7.333333333333333,
|
| 482 |
+
"0.8": 6.0,
|
| 483 |
+
"0.9": 5.333333333333333
|
| 484 |
+
},
|
| 485 |
+
"n_dup_abs_band": 7.166666666666666,
|
| 486 |
+
"n_dup_abs_band_std": 2.309401076758503,
|
| 487 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 488 |
+
"0.5": 9.333333333333334,
|
| 489 |
+
"0.6": 5.0,
|
| 490 |
+
"0.7": 3.6666666666666665,
|
| 491 |
+
"0.8": 3.0,
|
| 492 |
+
"0.9": 2.6666666666666665
|
| 493 |
+
},
|
| 494 |
+
"n_dup_clusters_abs_band": 3.583333333333333,
|
| 495 |
+
"n_dup_clusters_abs_band_std": 1.1547005383792515,
|
| 496 |
+
"largest_dup_abs_per_threshold": {
|
| 497 |
+
"0.5": 2.0,
|
| 498 |
+
"0.6": 2.0,
|
| 499 |
+
"0.7": 2.0,
|
| 500 |
+
"0.8": 2.0,
|
| 501 |
+
"0.9": 2.0
|
| 502 |
+
},
|
| 503 |
+
"largest_dup_abs_band": 2.0,
|
| 504 |
+
"largest_dup_abs_band_std": 0.0,
|
| 505 |
+
"dcs_abs_mean": 0.8290931383768717,
|
| 506 |
+
"dcs_abs_std": 0.0018539836249511993,
|
| 507 |
+
"dedup_dcs_abs_per_threshold": {
|
| 508 |
+
"0.5": 0.8268643112969585,
|
| 509 |
+
"0.6": 0.8279662012549428,
|
| 510 |
+
"0.7": 0.8281815279074687,
|
| 511 |
+
"0.8": 0.828285996093958,
|
| 512 |
+
"0.9": 0.8283704120323009
|
| 513 |
+
},
|
| 514 |
+
"dedup_dcs_abs_band": 0.8282010343221676,
|
| 515 |
+
"dedup_dcs_abs_band_std": 0.0016750882901845469,
|
| 516 |
+
"matched_cos_abs_per_threshold": {
|
| 517 |
+
"0.5": 0.8293845057487488,
|
| 518 |
+
"0.6": 0.8293160994847616,
|
| 519 |
+
"0.7": 0.8291712403297424,
|
| 520 |
+
"0.8": 0.8290958205858866,
|
| 521 |
+
"0.9": 0.8290901581446329
|
| 522 |
+
},
|
| 523 |
+
"matched_cos_abs_band": 0.8291683296362559,
|
| 524 |
+
"matched_cos_abs_band_std": 0.0017347135468773211,
|
| 525 |
+
"unique_frac_abs_per_threshold": {
|
| 526 |
+
"0.5": 0.9969618055555557,
|
| 527 |
+
"0.6": 0.9983723958333334,
|
| 528 |
+
"0.7": 0.9988064236111112,
|
| 529 |
+
"0.8": 0.9990234375,
|
| 530 |
+
"0.9": 0.9991319444444443
|
| 531 |
+
},
|
| 532 |
+
"unique_frac_abs_band": 0.9988335503472222,
|
| 533 |
+
"unique_frac_abs_band_std": 0.00018793954075178857,
|
| 534 |
+
"matched_frac_abs_per_threshold": {
|
| 535 |
+
"0.5": 0.8330078125,
|
| 536 |
+
"0.6": 0.8202039930555555,
|
| 537 |
+
"0.7": 0.8037109375,
|
| 538 |
+
"0.8": 0.7810329861111112,
|
| 539 |
+
"0.9": 0.7437065972222222
|
| 540 |
+
},
|
| 541 |
+
"matched_frac_abs_band": 0.7871636284722222,
|
| 542 |
+
"matched_frac_abs_band_std": 0.002437786668691543,
|
| 543 |
+
"alive_source": "stats"
|
| 544 |
+
}
|
| 545 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed0/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 4,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 0,
|
| 8 |
+
"top_k": 32,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed0/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1658,
|
| 3 |
+
"r2_score": 0.9551,
|
| 4 |
+
"l0_per_token_0": 0.0104,
|
| 5 |
+
"l0_per_token_0.001": 0.0104,
|
| 6 |
+
"l0_per_token_0.01": 0.0104,
|
| 7 |
+
"l0_per_token_0.05": 0.0104,
|
| 8 |
+
"l0_per_token_0.1": 0.0104,
|
| 9 |
+
"l0_per_token_0.5": 0.0104,
|
| 10 |
+
"l0_per_token_1.0": 0.0104,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.001,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 92.1742
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed1/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 4,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 1,
|
| 8 |
+
"top_k": 32,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed1/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1656,
|
| 3 |
+
"r2_score": 0.9551,
|
| 4 |
+
"l0_per_token_0": 0.0104,
|
| 5 |
+
"l0_per_token_0.001": 0.0104,
|
| 6 |
+
"l0_per_token_0.01": 0.0104,
|
| 7 |
+
"l0_per_token_0.05": 0.0104,
|
| 8 |
+
"l0_per_token_0.1": 0.0104,
|
| 9 |
+
"l0_per_token_0.5": 0.0104,
|
| 10 |
+
"l0_per_token_1.0": 0.0104,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.001,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 91.4717
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed2/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 4,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 2,
|
| 8 |
+
"top_k": 32,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_32/seed2/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1662,
|
| 3 |
+
"r2_score": 0.955,
|
| 4 |
+
"l0_per_token_0": 0.0104,
|
| 5 |
+
"l0_per_token_0.001": 0.0104,
|
| 6 |
+
"l0_per_token_0.01": 0.0104,
|
| 7 |
+
"l0_per_token_0.05": 0.0104,
|
| 8 |
+
"l0_per_token_0.1": 0.0104,
|
| 9 |
+
"l0_per_token_0.5": 0.0104,
|
| 10 |
+
"l0_per_token_1.0": 0.0104,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.001,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 92.6854
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_64/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 771000,
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"l0": 64.04089364461738,
|
| 20 |
+
"recon_r2": 0.965752672449887,
|
| 21 |
+
"l1_per_token": 125.83079287655183
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 63.992158236057065,
|
| 25 |
+
"recon_r2": 0.9657185498781529,
|
| 26 |
+
"l1_per_token": 125.96973285575599
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 64.02423216601815,
|
| 30 |
+
"recon_r2": 0.9656546594517439,
|
| 31 |
+
"l1_per_token": 126.65735384191404
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.8702998050927376,
|
| 38 |
+
"mcc_abs": 0.8704329626964453,
|
| 39 |
+
"act_iou": 0.6951362133245468,
|
| 40 |
+
"act_iou_n_pairs": 3072,
|
| 41 |
+
"input_r2": 0.9902146681688715,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.9568676209550181,
|
| 44 |
+
"0.7": 0.964607984510467,
|
| 45 |
+
"0.8": 0.9698950072666856,
|
| 46 |
+
"0.9": 0.9756163905924504
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.9559989005176703,
|
| 50 |
+
"0.7": 0.963736197776778,
|
| 51 |
+
"0.8": 0.9695182644096563,
|
| 52 |
+
"0.9": 0.9756171293346912
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.8711676598713242,
|
| 58 |
+
"mcc_abs": 0.8712361245683505,
|
| 59 |
+
"act_iou": 0.6929568737629443,
|
| 60 |
+
"act_iou_n_pairs": 3072,
|
| 61 |
+
"input_r2": 0.9900454873721835,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.9561251875052224,
|
| 64 |
+
"0.7": 0.965563551223705,
|
| 65 |
+
"0.8": 0.9727549611432701,
|
| 66 |
+
"0.9": 0.978281732100719
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.9554516087780373,
|
| 70 |
+
"0.7": 0.9646630862662121,
|
| 71 |
+
"0.8": 0.9722371210087354,
|
| 72 |
+
"0.9": 0.9779264744898296
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.8686435406564256,
|
| 78 |
+
"mcc_abs": 0.8687428035790106,
|
| 79 |
+
"act_iou": 0.6891907852858336,
|
| 80 |
+
"act_iou_n_pairs": 3072,
|
| 81 |
+
"input_r2": 0.9898687111270811,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.9550353941525525,
|
| 84 |
+
"0.7": 0.9634882674794839,
|
| 85 |
+
"0.8": 0.9707775262499349,
|
| 86 |
+
"0.9": 0.9781402920964573
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.9546873870015037,
|
| 90 |
+
"0.7": 0.9627200121662747,
|
| 91 |
+
"0.8": 0.9704425067353233,
|
| 92 |
+
"0.9": 0.9781314988573983
|
| 93 |
+
}
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"summary": {
|
| 97 |
+
"n_seeds": 3,
|
| 98 |
+
"n_pairs": 3,
|
| 99 |
+
"report_band": [
|
| 100 |
+
0.6,
|
| 101 |
+
0.7,
|
| 102 |
+
0.8,
|
| 103 |
+
0.9
|
| 104 |
+
],
|
| 105 |
+
"l0_mean": 64.01909468223087,
|
| 106 |
+
"l0_std": 0.02477055341799523,
|
| 107 |
+
"recon_r2_mean": 0.9657086272599279,
|
| 108 |
+
"recon_r2_std": 4.975420299412952e-05,
|
| 109 |
+
"l1_per_token_mean": 126.15262652474063,
|
| 110 |
+
"l1_per_token_std": 0.4425927337644904,
|
| 111 |
+
"input_r2_mean": 0.9900429555560454,
|
| 112 |
+
"input_r2_std": 0.00017299241677256022,
|
| 113 |
+
"mcc_signed_mean": 0.8700370018734959,
|
| 114 |
+
"mcc_signed_std": 0.001282417093533916,
|
| 115 |
+
"mcc_abs_mean": 0.8701372969479354,
|
| 116 |
+
"mcc_abs_std": 0.0012726845897952175,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.956009400870931,
|
| 119 |
+
"0.7": 0.9645532677378853,
|
| 120 |
+
"0.8": 0.9711424982199635,
|
| 121 |
+
"0.9": 0.977346138263209
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.9672628262729972,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.9553792987657371,
|
| 126 |
+
"0.7": 0.9637064320697549,
|
| 127 |
+
"0.8": 0.9707326307179049,
|
| 128 |
+
"0.9": 0.9772250342273064
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.9667608489451759,
|
| 131 |
+
"act_iou_mean": 0.6924279574577749,
|
| 132 |
+
"act_iou_std": 0.003007797031763437,
|
| 133 |
+
"act_iou_n_pairs": 3072.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_64/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,545 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 2,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"linkage": "complete",
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"d_sae": 3072,
|
| 20 |
+
"d_in": 768,
|
| 21 |
+
"l0": 64,
|
| 22 |
+
"l0_raw": 64.20479999999999,
|
| 23 |
+
"recon_r2": 0.9657,
|
| 24 |
+
"l1_per_token": 125.8463,
|
| 25 |
+
"stats_source": "stats",
|
| 26 |
+
"n_dead": 0,
|
| 27 |
+
"dead_rate": 0.0,
|
| 28 |
+
"alive_source": "stats",
|
| 29 |
+
"n_dense": 54,
|
| 30 |
+
"dense_rate": 0.017578125,
|
| 31 |
+
"dense_threshold": 0.1,
|
| 32 |
+
"n_dup_signed": {
|
| 33 |
+
"0.5": 0,
|
| 34 |
+
"0.6": 0,
|
| 35 |
+
"0.7": 0,
|
| 36 |
+
"0.8": 0,
|
| 37 |
+
"0.9": 0
|
| 38 |
+
},
|
| 39 |
+
"n_dup_clusters_signed": {
|
| 40 |
+
"0.5": 0,
|
| 41 |
+
"0.6": 0,
|
| 42 |
+
"0.7": 0,
|
| 43 |
+
"0.8": 0,
|
| 44 |
+
"0.9": 0
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 1,
|
| 48 |
+
"0.6": 1,
|
| 49 |
+
"0.7": 1,
|
| 50 |
+
"0.8": 1,
|
| 51 |
+
"0.9": 1
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.1727079153060913,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 50,
|
| 56 |
+
"0.6": 38,
|
| 57 |
+
"0.7": 30,
|
| 58 |
+
"0.8": 20,
|
| 59 |
+
"0.9": 18
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 25,
|
| 63 |
+
"0.6": 19,
|
| 64 |
+
"0.7": 15,
|
| 65 |
+
"0.8": 10,
|
| 66 |
+
"0.9": 9
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 2,
|
| 70 |
+
"0.6": 2,
|
| 71 |
+
"0.7": 2,
|
| 72 |
+
"0.8": 2,
|
| 73 |
+
"0.9": 2
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.19299983978271484
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 3072,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 64,
|
| 81 |
+
"l0_raw": 63.8976,
|
| 82 |
+
"recon_r2": 0.9657,
|
| 83 |
+
"l1_per_token": 125.9709,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 59,
|
| 89 |
+
"dense_rate": 0.019205729166666668,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 2,
|
| 93 |
+
"0.6": 2,
|
| 94 |
+
"0.7": 2,
|
| 95 |
+
"0.8": 0,
|
| 96 |
+
"0.9": 0
|
| 97 |
+
},
|
| 98 |
+
"n_dup_clusters_signed": {
|
| 99 |
+
"0.5": 1,
|
| 100 |
+
"0.6": 1,
|
| 101 |
+
"0.7": 1,
|
| 102 |
+
"0.8": 0,
|
| 103 |
+
"0.9": 0
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 2,
|
| 107 |
+
"0.6": 2,
|
| 108 |
+
"0.7": 2,
|
| 109 |
+
"0.8": 1,
|
| 110 |
+
"0.9": 1
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.17320948839187622,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 50,
|
| 115 |
+
"0.6": 36,
|
| 116 |
+
"0.7": 36,
|
| 117 |
+
"0.8": 30,
|
| 118 |
+
"0.9": 20
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 25,
|
| 122 |
+
"0.6": 18,
|
| 123 |
+
"0.7": 18,
|
| 124 |
+
"0.8": 15,
|
| 125 |
+
"0.9": 10
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 2,
|
| 129 |
+
"0.6": 2,
|
| 130 |
+
"0.7": 2,
|
| 131 |
+
"0.8": 2,
|
| 132 |
+
"0.9": 2
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.19439299404621124
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 3072,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 64,
|
| 140 |
+
"l0_raw": 64.20479999999999,
|
| 141 |
+
"recon_r2": 0.9656,
|
| 142 |
+
"l1_per_token": 126.6698,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 62,
|
| 148 |
+
"dense_rate": 0.020182291666666668,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 0,
|
| 152 |
+
"0.6": 0,
|
| 153 |
+
"0.7": 0,
|
| 154 |
+
"0.8": 0,
|
| 155 |
+
"0.9": 0
|
| 156 |
+
},
|
| 157 |
+
"n_dup_clusters_signed": {
|
| 158 |
+
"0.5": 0,
|
| 159 |
+
"0.6": 0,
|
| 160 |
+
"0.7": 0,
|
| 161 |
+
"0.8": 0,
|
| 162 |
+
"0.9": 0
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 1,
|
| 166 |
+
"0.6": 1,
|
| 167 |
+
"0.7": 1,
|
| 168 |
+
"0.8": 1,
|
| 169 |
+
"0.9": 1
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.1734081655740738,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 52,
|
| 174 |
+
"0.6": 40,
|
| 175 |
+
"0.7": 38,
|
| 176 |
+
"0.8": 28,
|
| 177 |
+
"0.9": 24
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 26,
|
| 181 |
+
"0.6": 20,
|
| 182 |
+
"0.7": 19,
|
| 183 |
+
"0.8": 14,
|
| 184 |
+
"0.9": 12
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 2,
|
| 188 |
+
"0.6": 2,
|
| 189 |
+
"0.7": 2,
|
| 190 |
+
"0.8": 2,
|
| 191 |
+
"0.9": 2
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.19495642185211182
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.8765752911567688,
|
| 199 |
+
"dcs_abs": 0.8779502511024475,
|
| 200 |
+
"frac_sign_flipped": 0.019205729166666668,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"n_alive_a": 3072,
|
| 203 |
+
"n_alive_b": 3072,
|
| 204 |
+
"matched_cos_signed": {
|
| 205 |
+
"0.5": 0.876835286617279,
|
| 206 |
+
"0.6": 0.876835286617279,
|
| 207 |
+
"0.7": 0.876835286617279,
|
| 208 |
+
"0.8": 0.8765752911567688,
|
| 209 |
+
"0.9": 0.8765752911567688
|
| 210 |
+
},
|
| 211 |
+
"unique_frac_signed": {
|
| 212 |
+
"0.5": 0.9998372395833334,
|
| 213 |
+
"0.6": 0.9998372395833334,
|
| 214 |
+
"0.7": 0.9998372395833334,
|
| 215 |
+
"0.8": 1.0,
|
| 216 |
+
"0.9": 1.0
|
| 217 |
+
},
|
| 218 |
+
"dedup_dcs_signed": {
|
| 219 |
+
"0.5": 0.8766925725406812,
|
| 220 |
+
"0.6": 0.8766925725406812,
|
| 221 |
+
"0.7": 0.8766925725406812,
|
| 222 |
+
"0.8": 0.8765752911567688,
|
| 223 |
+
"0.9": 0.8765752911567688
|
| 224 |
+
},
|
| 225 |
+
"matched_frac_signed": {
|
| 226 |
+
"0.5": 0.8909505208333334,
|
| 227 |
+
"0.6": 0.8733723958333334,
|
| 228 |
+
"0.7": 0.8522135416666666,
|
| 229 |
+
"0.8": 0.8307291666666666,
|
| 230 |
+
"0.9": 0.796875
|
| 231 |
+
},
|
| 232 |
+
"matched_cos_abs": {
|
| 233 |
+
"0.5": 0.8790980577468872,
|
| 234 |
+
"0.6": 0.8791306614875793,
|
| 235 |
+
"0.7": 0.8793507814407349,
|
| 236 |
+
"0.8": 0.8791425824165344,
|
| 237 |
+
"0.9": 0.878437340259552
|
| 238 |
+
},
|
| 239 |
+
"unique_frac_abs": {
|
| 240 |
+
"0.5": 0.9918619791666666,
|
| 241 |
+
"0.6": 0.9939778645833334,
|
| 242 |
+
"0.7": 0.99462890625,
|
| 243 |
+
"0.8": 0.9959309895833334,
|
| 244 |
+
"0.9": 0.9969075520833334
|
| 245 |
+
},
|
| 246 |
+
"dedup_dcs_abs": {
|
| 247 |
+
"0.5": 0.8719439394384001,
|
| 248 |
+
"0.6": 0.8738364175951574,
|
| 249 |
+
"0.7": 0.8746277059544809,
|
| 250 |
+
"0.8": 0.8755653420909464,
|
| 251 |
+
"0.9": 0.8757208185367442
|
| 252 |
+
},
|
| 253 |
+
"matched_frac_abs": {
|
| 254 |
+
"0.5": 0.8863932291666666,
|
| 255 |
+
"0.6": 0.87109375,
|
| 256 |
+
"0.7": 0.8505859375,
|
| 257 |
+
"0.8": 0.8287760416666666,
|
| 258 |
+
"0.9": 0.7955729166666666
|
| 259 |
+
}
|
| 260 |
+
},
|
| 261 |
+
"seed0\u2194seed2": {
|
| 262 |
+
"dcs_signed": 0.8746981024742126,
|
| 263 |
+
"dcs_abs": 0.8758165240287781,
|
| 264 |
+
"frac_sign_flipped": 0.017903645833333332,
|
| 265 |
+
"alive_source": "stats",
|
| 266 |
+
"n_alive_a": 3072,
|
| 267 |
+
"n_alive_b": 3072,
|
| 268 |
+
"matched_cos_signed": {
|
| 269 |
+
"0.5": 0.8746981024742126,
|
| 270 |
+
"0.6": 0.8746981024742126,
|
| 271 |
+
"0.7": 0.8746981024742126,
|
| 272 |
+
"0.8": 0.8746981024742126,
|
| 273 |
+
"0.9": 0.8746981024742126
|
| 274 |
+
},
|
| 275 |
+
"unique_frac_signed": {
|
| 276 |
+
"0.5": 1.0,
|
| 277 |
+
"0.6": 1.0,
|
| 278 |
+
"0.7": 1.0,
|
| 279 |
+
"0.8": 1.0,
|
| 280 |
+
"0.9": 1.0
|
| 281 |
+
},
|
| 282 |
+
"dedup_dcs_signed": {
|
| 283 |
+
"0.5": 0.8746981024742126,
|
| 284 |
+
"0.6": 0.8746981024742126,
|
| 285 |
+
"0.7": 0.8746981024742126,
|
| 286 |
+
"0.8": 0.8746981024742126,
|
| 287 |
+
"0.9": 0.8746981024742126
|
| 288 |
+
},
|
| 289 |
+
"matched_frac_signed": {
|
| 290 |
+
"0.5": 0.8942057291666666,
|
| 291 |
+
"0.6": 0.8753255208333334,
|
| 292 |
+
"0.7": 0.8512369791666666,
|
| 293 |
+
"0.8": 0.8232421875,
|
| 294 |
+
"0.9": 0.7884114583333334
|
| 295 |
+
},
|
| 296 |
+
"matched_cos_abs": {
|
| 297 |
+
"0.5": 0.876774787902832,
|
| 298 |
+
"0.6": 0.8765795826911926,
|
| 299 |
+
"0.7": 0.877038836479187,
|
| 300 |
+
"0.8": 0.8766720294952393,
|
| 301 |
+
"0.9": 0.8766370415687561
|
| 302 |
+
},
|
| 303 |
+
"unique_frac_abs": {
|
| 304 |
+
"0.5": 0.99169921875,
|
| 305 |
+
"0.6": 0.99365234375,
|
| 306 |
+
"0.7": 0.9944661458333334,
|
| 307 |
+
"0.8": 0.99609375,
|
| 308 |
+
"0.9": 0.99658203125
|
| 309 |
+
},
|
| 310 |
+
"dedup_dcs_abs": {
|
| 311 |
+
"0.5": 0.8694968721829355,
|
| 312 |
+
"0.6": 0.8710153568245005,
|
| 313 |
+
"0.7": 0.8721854314596083,
|
| 314 |
+
"0.8": 0.8732475293800235,
|
| 315 |
+
"0.9": 0.8736407235555816
|
| 316 |
+
},
|
| 317 |
+
"matched_frac_abs": {
|
| 318 |
+
"0.5": 0.8883463541666666,
|
| 319 |
+
"0.6": 0.87109375,
|
| 320 |
+
"0.7": 0.8483072916666666,
|
| 321 |
+
"0.8": 0.822265625,
|
| 322 |
+
"0.9": 0.7874348958333334
|
| 323 |
+
}
|
| 324 |
+
},
|
| 325 |
+
"seed1\u2194seed2": {
|
| 326 |
+
"dcs_signed": 0.8737357258796692,
|
| 327 |
+
"dcs_abs": 0.8745160102844238,
|
| 328 |
+
"frac_sign_flipped": 0.01953125,
|
| 329 |
+
"alive_source": "stats",
|
| 330 |
+
"n_alive_a": 3072,
|
| 331 |
+
"n_alive_b": 3072,
|
| 332 |
+
"matched_cos_signed": {
|
| 333 |
+
"0.5": 0.8739960193634033,
|
| 334 |
+
"0.6": 0.8739960193634033,
|
| 335 |
+
"0.7": 0.8739960193634033,
|
| 336 |
+
"0.8": 0.8737358450889587,
|
| 337 |
+
"0.9": 0.8737358450889587
|
| 338 |
+
},
|
| 339 |
+
"unique_frac_signed": {
|
| 340 |
+
"0.5": 0.9998372395833334,
|
| 341 |
+
"0.6": 0.9998372395833334,
|
| 342 |
+
"0.7": 0.9998372395833334,
|
| 343 |
+
"0.8": 1.0,
|
| 344 |
+
"0.9": 1.0
|
| 345 |
+
},
|
| 346 |
+
"dedup_dcs_signed": {
|
| 347 |
+
"0.5": 0.8738537674071267,
|
| 348 |
+
"0.6": 0.8738537674071267,
|
| 349 |
+
"0.7": 0.8738537674071267,
|
| 350 |
+
"0.8": 0.8737358450889587,
|
| 351 |
+
"0.9": 0.8737358450889587
|
| 352 |
+
},
|
| 353 |
+
"matched_frac_signed": {
|
| 354 |
+
"0.5": 0.8909505208333334,
|
| 355 |
+
"0.6": 0.873046875,
|
| 356 |
+
"0.7": 0.8509114583333334,
|
| 357 |
+
"0.8": 0.8229166666666666,
|
| 358 |
+
"0.9": 0.7789713541666666
|
| 359 |
+
},
|
| 360 |
+
"matched_cos_abs": {
|
| 361 |
+
"0.5": 0.8760044574737549,
|
| 362 |
+
"0.6": 0.8758426308631897,
|
| 363 |
+
"0.7": 0.8757514357566833,
|
| 364 |
+
"0.8": 0.8751859068870544,
|
| 365 |
+
"0.9": 0.8751087188720703
|
| 366 |
+
},
|
| 367 |
+
"unique_frac_abs": {
|
| 368 |
+
"0.5": 0.99169921875,
|
| 369 |
+
"0.6": 0.9938151041666666,
|
| 370 |
+
"0.7": 0.9939778645833334,
|
| 371 |
+
"0.8": 0.9952799479166666,
|
| 372 |
+
"0.9": 0.9964192708333334
|
| 373 |
+
},
|
| 374 |
+
"dedup_dcs_abs": {
|
| 375 |
+
"0.5": 0.8687329360982403,
|
| 376 |
+
"0.6": 0.8704256354249083,
|
| 377 |
+
"0.7": 0.8704775420192163,
|
| 378 |
+
"0.8": 0.8710549838239482,
|
| 379 |
+
"0.9": 0.8719751915584009
|
| 380 |
+
},
|
| 381 |
+
"matched_frac_abs": {
|
| 382 |
+
"0.5": 0.8857421875,
|
| 383 |
+
"0.6": 0.8697916666666666,
|
| 384 |
+
"0.7": 0.8489583333333334,
|
| 385 |
+
"0.8": 0.8209635416666666,
|
| 386 |
+
"0.9": 0.7770182291666666
|
| 387 |
+
}
|
| 388 |
+
}
|
| 389 |
+
},
|
| 390 |
+
"summary": {
|
| 391 |
+
"n_seeds": 3,
|
| 392 |
+
"n_pairs": 3,
|
| 393 |
+
"report_band": [
|
| 394 |
+
0.6,
|
| 395 |
+
0.7,
|
| 396 |
+
0.8,
|
| 397 |
+
0.9
|
| 398 |
+
],
|
| 399 |
+
"n_dead_mean": 0,
|
| 400 |
+
"n_dead_std": 0.0,
|
| 401 |
+
"n_dense_mean": 58,
|
| 402 |
+
"n_dense_std": 4.041451884327381,
|
| 403 |
+
"l0_mean": 64,
|
| 404 |
+
"l0_std": 0.17736200269504993,
|
| 405 |
+
"recon_r2_mean": 0.9656666666666668,
|
| 406 |
+
"recon_r2_std": 5.773502691895622e-05,
|
| 407 |
+
"l1_per_token_mean": 126.16233333333334,
|
| 408 |
+
"l1_per_token_std": 0.4436593325214025,
|
| 409 |
+
"babel_mu1_signed_mean": 0.17310852309068045,
|
| 410 |
+
"babel_mu1_signed_std": 0.00036087823914244366,
|
| 411 |
+
"n_dup_signed_per_threshold": {
|
| 412 |
+
"0.5": 0.6666666666666666,
|
| 413 |
+
"0.6": 0.6666666666666666,
|
| 414 |
+
"0.7": 0.6666666666666666,
|
| 415 |
+
"0.8": 0.0,
|
| 416 |
+
"0.9": 0.0
|
| 417 |
+
},
|
| 418 |
+
"n_dup_signed_band": 0.3333333333333333,
|
| 419 |
+
"n_dup_signed_band_std": 0.5773502691896258,
|
| 420 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 421 |
+
"0.5": 0.3333333333333333,
|
| 422 |
+
"0.6": 0.3333333333333333,
|
| 423 |
+
"0.7": 0.3333333333333333,
|
| 424 |
+
"0.8": 0.0,
|
| 425 |
+
"0.9": 0.0
|
| 426 |
+
},
|
| 427 |
+
"n_dup_clusters_signed_band": 0.16666666666666666,
|
| 428 |
+
"n_dup_clusters_signed_band_std": 0.2886751345948129,
|
| 429 |
+
"largest_dup_signed_per_threshold": {
|
| 430 |
+
"0.5": 1.3333333333333333,
|
| 431 |
+
"0.6": 1.3333333333333333,
|
| 432 |
+
"0.7": 1.3333333333333333,
|
| 433 |
+
"0.8": 1.0,
|
| 434 |
+
"0.9": 1.0
|
| 435 |
+
},
|
| 436 |
+
"largest_dup_signed_band": 1.1666666666666665,
|
| 437 |
+
"largest_dup_signed_band_std": 0.2886751345948129,
|
| 438 |
+
"dcs_signed_mean": 0.8750030398368835,
|
| 439 |
+
"dcs_signed_std": 0.0014441339401495823,
|
| 440 |
+
"dedup_dcs_signed_per_threshold": {
|
| 441 |
+
"0.5": 0.8750814808073403,
|
| 442 |
+
"0.6": 0.8750814808073403,
|
| 443 |
+
"0.7": 0.8750814808073403,
|
| 444 |
+
"0.8": 0.8750030795733134,
|
| 445 |
+
"0.9": 0.8750030795733134
|
| 446 |
+
},
|
| 447 |
+
"dedup_dcs_signed_band": 0.8750422801903268,
|
| 448 |
+
"dedup_dcs_signed_band_std": 0.0014505179314005848,
|
| 449 |
+
"matched_cos_signed_per_threshold": {
|
| 450 |
+
"0.5": 0.875176469484965,
|
| 451 |
+
"0.6": 0.875176469484965,
|
| 452 |
+
"0.7": 0.875176469484965,
|
| 453 |
+
"0.8": 0.8750030795733134,
|
| 454 |
+
"0.9": 0.8750030795733134
|
| 455 |
+
},
|
| 456 |
+
"matched_cos_signed_band": 0.8750897745291392,
|
| 457 |
+
"matched_cos_signed_band_std": 0.001459637561460386,
|
| 458 |
+
"unique_frac_signed_per_threshold": {
|
| 459 |
+
"0.5": 0.9998914930555557,
|
| 460 |
+
"0.6": 0.9998914930555557,
|
| 461 |
+
"0.7": 0.9998914930555557,
|
| 462 |
+
"0.8": 1.0,
|
| 463 |
+
"0.9": 1.0
|
| 464 |
+
},
|
| 465 |
+
"unique_frac_signed_band": 0.9999457465277779,
|
| 466 |
+
"unique_frac_signed_band_std": 4.698488518791509e-05,
|
| 467 |
+
"matched_frac_signed_per_threshold": {
|
| 468 |
+
"0.5": 0.8920355902777778,
|
| 469 |
+
"0.6": 0.8739149305555557,
|
| 470 |
+
"0.7": 0.8514539930555555,
|
| 471 |
+
"0.8": 0.8256293402777777,
|
| 472 |
+
"0.9": 0.7880859375
|
| 473 |
+
},
|
| 474 |
+
"matched_frac_signed_band": 0.8347710503472222,
|
| 475 |
+
"matched_frac_signed_band_std": 0.003423131847731629,
|
| 476 |
+
"babel_mu1_abs_mean": 0.19411641856034598,
|
| 477 |
+
"babel_mu1_abs_std": 0.0010071861040084125,
|
| 478 |
+
"n_dup_abs_per_threshold": {
|
| 479 |
+
"0.5": 50.666666666666664,
|
| 480 |
+
"0.6": 38.0,
|
| 481 |
+
"0.7": 34.666666666666664,
|
| 482 |
+
"0.8": 26.0,
|
| 483 |
+
"0.9": 20.666666666666668
|
| 484 |
+
},
|
| 485 |
+
"n_dup_abs_band": 29.833333333333332,
|
| 486 |
+
"n_dup_abs_band_std": 3.0550504633038935,
|
| 487 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 488 |
+
"0.5": 25.333333333333332,
|
| 489 |
+
"0.6": 19.0,
|
| 490 |
+
"0.7": 17.333333333333332,
|
| 491 |
+
"0.8": 13.0,
|
| 492 |
+
"0.9": 10.333333333333334
|
| 493 |
+
},
|
| 494 |
+
"n_dup_clusters_abs_band": 14.916666666666666,
|
| 495 |
+
"n_dup_clusters_abs_band_std": 1.5275252316519468,
|
| 496 |
+
"largest_dup_abs_per_threshold": {
|
| 497 |
+
"0.5": 2.0,
|
| 498 |
+
"0.6": 2.0,
|
| 499 |
+
"0.7": 2.0,
|
| 500 |
+
"0.8": 2.0,
|
| 501 |
+
"0.9": 2.0
|
| 502 |
+
},
|
| 503 |
+
"largest_dup_abs_band": 2.0,
|
| 504 |
+
"largest_dup_abs_band_std": 0.0,
|
| 505 |
+
"dcs_abs_mean": 0.8760942618052164,
|
| 506 |
+
"dcs_abs_std": 0.0017338847145616976,
|
| 507 |
+
"dedup_dcs_abs_per_threshold": {
|
| 508 |
+
"0.5": 0.8700579159065253,
|
| 509 |
+
"0.6": 0.8717591366148554,
|
| 510 |
+
"0.7": 0.8724302264777686,
|
| 511 |
+
"0.8": 0.8732892850983061,
|
| 512 |
+
"0.9": 0.8737789112169089
|
| 513 |
+
},
|
| 514 |
+
"dedup_dcs_abs_band": 0.8728143898519597,
|
| 515 |
+
"dedup_dcs_abs_band_std": 0.001993237087729517,
|
| 516 |
+
"matched_cos_abs_per_threshold": {
|
| 517 |
+
"0.5": 0.8772924343744913,
|
| 518 |
+
"0.6": 0.8771842916806539,
|
| 519 |
+
"0.7": 0.877380351225535,
|
| 520 |
+
"0.8": 0.8770001729329427,
|
| 521 |
+
"0.9": 0.8767277002334595
|
| 522 |
+
},
|
| 523 |
+
"matched_cos_abs_band": 0.8770731290181478,
|
| 524 |
+
"matched_cos_abs_band_std": 0.001796065808975983,
|
| 525 |
+
"unique_frac_abs_per_threshold": {
|
| 526 |
+
"0.5": 0.9917534722222222,
|
| 527 |
+
"0.6": 0.9938151041666666,
|
| 528 |
+
"0.7": 0.994357638888889,
|
| 529 |
+
"0.8": 0.9957682291666666,
|
| 530 |
+
"0.9": 0.9966362847222223
|
| 531 |
+
},
|
| 532 |
+
"unique_frac_abs_band": 0.9951443142361112,
|
| 533 |
+
"unique_frac_abs_band_std": 0.00024862064317256985,
|
| 534 |
+
"matched_frac_abs_per_threshold": {
|
| 535 |
+
"0.5": 0.8868272569444443,
|
| 536 |
+
"0.6": 0.8706597222222222,
|
| 537 |
+
"0.7": 0.8492838541666666,
|
| 538 |
+
"0.8": 0.824001736111111,
|
| 539 |
+
"0.9": 0.7866753472222222
|
| 540 |
+
},
|
| 541 |
+
"matched_frac_abs_band": 0.8326551649305556,
|
| 542 |
+
"matched_frac_abs_band_std": 0.003676848714161764,
|
| 543 |
+
"alive_source": "stats"
|
| 544 |
+
}
|
| 545 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed0/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 4,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 0,
|
| 8 |
+
"top_k": 64,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed0/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1266,
|
| 3 |
+
"r2_score": 0.9657,
|
| 4 |
+
"l0_per_token_0": 0.0209,
|
| 5 |
+
"l0_per_token_0.001": 0.0209,
|
| 6 |
+
"l0_per_token_0.01": 0.0209,
|
| 7 |
+
"l0_per_token_0.05": 0.0209,
|
| 8 |
+
"l0_per_token_0.1": 0.0209,
|
| 9 |
+
"l0_per_token_0.5": 0.0209,
|
| 10 |
+
"l0_per_token_1.0": 0.014,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0091,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 125.8463
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed1/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 4,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 1,
|
| 8 |
+
"top_k": 64,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed1/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1267,
|
| 3 |
+
"r2_score": 0.9657,
|
| 4 |
+
"l0_per_token_0": 0.0208,
|
| 5 |
+
"l0_per_token_0.001": 0.0208,
|
| 6 |
+
"l0_per_token_0.01": 0.0208,
|
| 7 |
+
"l0_per_token_0.05": 0.0208,
|
| 8 |
+
"l0_per_token_0.1": 0.0208,
|
| 9 |
+
"l0_per_token_0.5": 0.0208,
|
| 10 |
+
"l0_per_token_1.0": 0.014,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0088,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 125.9709
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed2/config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 4,
|
| 4 |
+
"architecture": "batchtopk",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 2,
|
| 8 |
+
"top_k": 64,
|
| 9 |
+
"top_k_inference": 0,
|
| 10 |
+
"class_type": "BatchTopKSAE",
|
| 11 |
+
"module_name": "consdl.models.batchtopk"
|
| 12 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x4/l0_64/seed2/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.1269,
|
| 3 |
+
"r2_score": 0.9656,
|
| 4 |
+
"l0_per_token_0": 0.0209,
|
| 5 |
+
"l0_per_token_0.001": 0.0209,
|
| 6 |
+
"l0_per_token_0.01": 0.0209,
|
| 7 |
+
"l0_per_token_0.05": 0.0209,
|
| 8 |
+
"l0_per_token_0.1": 0.0209,
|
| 9 |
+
"l0_per_token_0.5": 0.0209,
|
| 10 |
+
"l0_per_token_1.0": 0.014,
|
| 11 |
+
"dead_rate": 0.0,
|
| 12 |
+
"dense_rate_0.2": 0.0081,
|
| 13 |
+
"live_latent_frac": 1.0,
|
| 14 |
+
"l1_per_token": 126.6698
|
| 15 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x8/l0_128/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 771000,
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"l0": 128.17673540856032,
|
| 20 |
+
"recon_r2": 0.9781049622527158,
|
| 21 |
+
"l1_per_token": 171.14638242202204
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 128.05976134889752,
|
| 25 |
+
"recon_r2": 0.978073063471785,
|
| 26 |
+
"l1_per_token": 171.03828205499408
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 128.18109079118028,
|
| 30 |
+
"recon_r2": 0.97796500138674,
|
| 31 |
+
"l1_per_token": 175.10224707070805
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.6493513968951606,
|
| 38 |
+
"mcc_abs": 0.6494521136794168,
|
| 39 |
+
"act_iou": 0.4864558583688097,
|
| 40 |
+
"act_iou_n_pairs": 6144,
|
| 41 |
+
"input_r2": 0.9902896807373807,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.9413211694415936,
|
| 44 |
+
"0.7": 0.9590139860011857,
|
| 45 |
+
"0.8": 0.9721719678283405,
|
| 46 |
+
"0.9": 0.9814095140411624
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.9416036257984107,
|
| 50 |
+
"0.7": 0.9588685576638132,
|
| 51 |
+
"0.8": 0.972227168163407,
|
| 52 |
+
"0.9": 0.9813314464790195
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.6438854109809963,
|
| 58 |
+
"mcc_abs": 0.6440184649461962,
|
| 59 |
+
"act_iou": 0.48099965556596663,
|
| 60 |
+
"act_iou_n_pairs": 6144,
|
| 61 |
+
"input_r2": 0.990054414777374,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.9439928013136646,
|
| 64 |
+
"0.7": 0.9598661266606123,
|
| 65 |
+
"0.8": 0.9726184513374846,
|
| 66 |
+
"0.9": 0.9827579221623697
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.9433412485562724,
|
| 70 |
+
"0.7": 0.9595313799901276,
|
| 71 |
+
"0.8": 0.9722973816310976,
|
| 72 |
+
"0.9": 0.9826175831258779
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.6453431116944831,
|
| 78 |
+
"mcc_abs": 0.645485473413518,
|
| 79 |
+
"act_iou": 0.48406864569777275,
|
| 80 |
+
"act_iou_n_pairs": 6144,
|
| 81 |
+
"input_r2": 0.9902022484985527,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.9444904583034658,
|
| 84 |
+
"0.7": 0.962270797428444,
|
| 85 |
+
"0.8": 0.973184751817828,
|
| 86 |
+
"0.9": 0.9825403598322766
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.9447691282453197,
|
| 90 |
+
"0.7": 0.9623379291978087,
|
| 91 |
+
"0.8": 0.9729660769514571,
|
| 92 |
+
"0.9": 0.9824534492581076
|
| 93 |
+
}
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"summary": {
|
| 97 |
+
"n_seeds": 3,
|
| 98 |
+
"n_pairs": 3,
|
| 99 |
+
"report_band": [
|
| 100 |
+
0.6,
|
| 101 |
+
0.7,
|
| 102 |
+
0.8,
|
| 103 |
+
0.9
|
| 104 |
+
],
|
| 105 |
+
"l0_mean": 128.13919584954604,
|
| 106 |
+
"l0_std": 0.06882675540370213,
|
| 107 |
+
"recon_r2_mean": 0.978047675703747,
|
| 108 |
+
"recon_r2_std": 7.335301681348162e-05,
|
| 109 |
+
"l1_per_token_mean": 172.42897051590805,
|
| 110 |
+
"l1_per_token_std": 2.3157562644134995,
|
| 111 |
+
"input_r2_mean": 0.9901821146711024,
|
| 112 |
+
"input_r2_std": 0.00011891823342242886,
|
| 113 |
+
"mcc_signed_mean": 0.6461933065235467,
|
| 114 |
+
"mcc_signed_std": 0.002830437057944519,
|
| 115 |
+
"mcc_abs_mean": 0.6463186840130436,
|
| 116 |
+
"mcc_abs_std": 0.002811016642849859,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.9432681430195746,
|
| 119 |
+
"0.7": 0.9603836366967474,
|
| 120 |
+
"0.8": 0.9726583903278844,
|
| 121 |
+
"0.9": 0.9822359320119363
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.9646365255140357,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.9432380008666676,
|
| 126 |
+
"0.7": 0.9602459556172498,
|
| 127 |
+
"0.8": 0.9724968755819873,
|
| 128 |
+
"0.9": 0.9821341596210017
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.9645287479217266,
|
| 131 |
+
"act_iou_mean": 0.483841386544183,
|
| 132 |
+
"act_iou_std": 0.0027351914555668556,
|
| 133 |
+
"act_iou_n_pairs": 6144.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x8/l0_128/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,545 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 2,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"linkage": "complete",
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"d_sae": 6144,
|
| 20 |
+
"d_in": 768,
|
| 21 |
+
"l0": 128,
|
| 22 |
+
"l0_raw": 128.40959999999998,
|
| 23 |
+
"recon_r2": 0.9781,
|
| 24 |
+
"l1_per_token": 171.1687,
|
| 25 |
+
"stats_source": "stats",
|
| 26 |
+
"n_dead": 0,
|
| 27 |
+
"dead_rate": 0.0,
|
| 28 |
+
"alive_source": "stats",
|
| 29 |
+
"n_dense": 202,
|
| 30 |
+
"dense_rate": 0.032877604166666664,
|
| 31 |
+
"dense_threshold": 0.1,
|
| 32 |
+
"n_dup_signed": {
|
| 33 |
+
"0.5": 20,
|
| 34 |
+
"0.6": 14,
|
| 35 |
+
"0.7": 10,
|
| 36 |
+
"0.8": 10,
|
| 37 |
+
"0.9": 0
|
| 38 |
+
},
|
| 39 |
+
"n_dup_clusters_signed": {
|
| 40 |
+
"0.5": 10,
|
| 41 |
+
"0.6": 7,
|
| 42 |
+
"0.7": 5,
|
| 43 |
+
"0.8": 5,
|
| 44 |
+
"0.9": 0
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 2,
|
| 48 |
+
"0.6": 2,
|
| 49 |
+
"0.7": 2,
|
| 50 |
+
"0.8": 2,
|
| 51 |
+
"0.9": 1
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.18168111145496368,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 243,
|
| 56 |
+
"0.6": 225,
|
| 57 |
+
"0.7": 212,
|
| 58 |
+
"0.8": 198,
|
| 59 |
+
"0.9": 170
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 121,
|
| 63 |
+
"0.6": 112,
|
| 64 |
+
"0.7": 106,
|
| 65 |
+
"0.8": 99,
|
| 66 |
+
"0.9": 85
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 3,
|
| 70 |
+
"0.6": 3,
|
| 71 |
+
"0.7": 2,
|
| 72 |
+
"0.8": 2,
|
| 73 |
+
"0.9": 2
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.22454096376895905
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 6144,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 128,
|
| 81 |
+
"l0_raw": 127.7952,
|
| 82 |
+
"recon_r2": 0.978,
|
| 83 |
+
"l1_per_token": 171.0602,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 213,
|
| 89 |
+
"dense_rate": 0.03466796875,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 24,
|
| 93 |
+
"0.6": 14,
|
| 94 |
+
"0.7": 14,
|
| 95 |
+
"0.8": 10,
|
| 96 |
+
"0.9": 0
|
| 97 |
+
},
|
| 98 |
+
"n_dup_clusters_signed": {
|
| 99 |
+
"0.5": 12,
|
| 100 |
+
"0.6": 7,
|
| 101 |
+
"0.7": 7,
|
| 102 |
+
"0.8": 5,
|
| 103 |
+
"0.9": 0
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 2,
|
| 107 |
+
"0.6": 2,
|
| 108 |
+
"0.7": 2,
|
| 109 |
+
"0.8": 2,
|
| 110 |
+
"0.9": 1
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.18108181655406952,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 266,
|
| 115 |
+
"0.6": 240,
|
| 116 |
+
"0.7": 232,
|
| 117 |
+
"0.8": 216,
|
| 118 |
+
"0.9": 186
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 133,
|
| 122 |
+
"0.6": 120,
|
| 123 |
+
"0.7": 116,
|
| 124 |
+
"0.8": 108,
|
| 125 |
+
"0.9": 93
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 2,
|
| 129 |
+
"0.6": 2,
|
| 130 |
+
"0.7": 2,
|
| 131 |
+
"0.8": 2,
|
| 132 |
+
"0.9": 2
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.22488540410995483
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 6144,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 128,
|
| 140 |
+
"l0_raw": 128.40959999999998,
|
| 141 |
+
"recon_r2": 0.9779,
|
| 142 |
+
"l1_per_token": 175.1126,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 216,
|
| 148 |
+
"dense_rate": 0.03515625,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 22,
|
| 152 |
+
"0.6": 14,
|
| 153 |
+
"0.7": 14,
|
| 154 |
+
"0.8": 10,
|
| 155 |
+
"0.9": 0
|
| 156 |
+
},
|
| 157 |
+
"n_dup_clusters_signed": {
|
| 158 |
+
"0.5": 11,
|
| 159 |
+
"0.6": 7,
|
| 160 |
+
"0.7": 7,
|
| 161 |
+
"0.8": 5,
|
| 162 |
+
"0.9": 0
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 2,
|
| 166 |
+
"0.6": 2,
|
| 167 |
+
"0.7": 2,
|
| 168 |
+
"0.8": 2,
|
| 169 |
+
"0.9": 1
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.17938290536403656,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 272,
|
| 174 |
+
"0.6": 236,
|
| 175 |
+
"0.7": 226,
|
| 176 |
+
"0.8": 210,
|
| 177 |
+
"0.9": 186
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 136,
|
| 181 |
+
"0.6": 118,
|
| 182 |
+
"0.7": 113,
|
| 183 |
+
"0.8": 105,
|
| 184 |
+
"0.9": 93
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 2,
|
| 188 |
+
"0.6": 2,
|
| 189 |
+
"0.7": 2,
|
| 190 |
+
"0.8": 2,
|
| 191 |
+
"0.9": 2
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.22333462536334991
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.656007707118988,
|
| 199 |
+
"dcs_abs": 0.6574788689613342,
|
| 200 |
+
"frac_sign_flipped": 0.07047526041666667,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"n_alive_a": 6144,
|
| 203 |
+
"n_alive_b": 6144,
|
| 204 |
+
"matched_cos_signed": {
|
| 205 |
+
"0.5": 0.6567835211753845,
|
| 206 |
+
"0.6": 0.6564295887947083,
|
| 207 |
+
"0.7": 0.6564483642578125,
|
| 208 |
+
"0.8": 0.6562615036964417,
|
| 209 |
+
"0.9": 0.6560077667236328
|
| 210 |
+
},
|
| 211 |
+
"unique_frac_signed": {
|
| 212 |
+
"0.5": 0.9982096354166666,
|
| 213 |
+
"0.6": 0.9988606770833334,
|
| 214 |
+
"0.7": 0.9990234375,
|
| 215 |
+
"0.8": 0.9991861979166666,
|
| 216 |
+
"0.9": 1.0
|
| 217 |
+
},
|
| 218 |
+
"dedup_dcs_signed": {
|
| 219 |
+
"0.5": 0.6556076392201551,
|
| 220 |
+
"0.6": 0.6556817035210164,
|
| 221 |
+
"0.7": 0.655807301402092,
|
| 222 |
+
"0.8": 0.655727436717522,
|
| 223 |
+
"0.9": 0.6560077667236328
|
| 224 |
+
},
|
| 225 |
+
"matched_frac_signed": {
|
| 226 |
+
"0.5": 0.6359049479166666,
|
| 227 |
+
"0.6": 0.59619140625,
|
| 228 |
+
"0.7": 0.55712890625,
|
| 229 |
+
"0.8": 0.5193684895833334,
|
| 230 |
+
"0.9": 0.4728190104166667
|
| 231 |
+
},
|
| 232 |
+
"matched_cos_abs": {
|
| 233 |
+
"0.5": 0.6630678176879883,
|
| 234 |
+
"0.6": 0.662459671497345,
|
| 235 |
+
"0.7": 0.6623132824897766,
|
| 236 |
+
"0.8": 0.6618468761444092,
|
| 237 |
+
"0.9": 0.6612928509712219
|
| 238 |
+
},
|
| 239 |
+
"unique_frac_abs": {
|
| 240 |
+
"0.5": 0.979248046875,
|
| 241 |
+
"0.6": 0.9810384114583334,
|
| 242 |
+
"0.7": 0.98193359375,
|
| 243 |
+
"0.8": 0.983154296875,
|
| 244 |
+
"0.9": 0.9855143229166666
|
| 245 |
+
},
|
| 246 |
+
"dedup_dcs_abs": {
|
| 247 |
+
"0.5": 0.6493078654166311,
|
| 248 |
+
"0.6": 0.6498983837809647,
|
| 249 |
+
"0.7": 0.6503476616635453,
|
| 250 |
+
"0.8": 0.6506976001546718,
|
| 251 |
+
"0.9": 0.6517135762745359
|
| 252 |
+
},
|
| 253 |
+
"matched_frac_abs": {
|
| 254 |
+
"0.5": 0.6305338541666666,
|
| 255 |
+
"0.6": 0.5916341145833334,
|
| 256 |
+
"0.7": 0.5543619791666666,
|
| 257 |
+
"0.8": 0.5172526041666666,
|
| 258 |
+
"0.9": 0.4723307291666667
|
| 259 |
+
}
|
| 260 |
+
},
|
| 261 |
+
"seed0\u2194seed2": {
|
| 262 |
+
"dcs_signed": 0.6484585404396057,
|
| 263 |
+
"dcs_abs": 0.6501538753509521,
|
| 264 |
+
"frac_sign_flipped": 0.07958984375,
|
| 265 |
+
"alive_source": "stats",
|
| 266 |
+
"n_alive_a": 6144,
|
| 267 |
+
"n_alive_b": 6144,
|
| 268 |
+
"matched_cos_signed": {
|
| 269 |
+
"0.5": 0.649176836013794,
|
| 270 |
+
"0.6": 0.6489683985710144,
|
| 271 |
+
"0.7": 0.6489858627319336,
|
| 272 |
+
"0.8": 0.6488386988639832,
|
| 273 |
+
"0.9": 0.6484585404396057
|
| 274 |
+
},
|
| 275 |
+
"unique_frac_signed": {
|
| 276 |
+
"0.5": 0.998291015625,
|
| 277 |
+
"0.6": 0.9988606770833334,
|
| 278 |
+
"0.7": 0.9990234375,
|
| 279 |
+
"0.8": 0.9991861979166666,
|
| 280 |
+
"0.9": 1.0
|
| 281 |
+
},
|
| 282 |
+
"dedup_dcs_signed": {
|
| 283 |
+
"0.5": 0.6480674029444344,
|
| 284 |
+
"0.6": 0.64822901400233,
|
| 285 |
+
"0.7": 0.6483520874753594,
|
| 286 |
+
"0.8": 0.6483106725791004,
|
| 287 |
+
"0.9": 0.6484585404396057
|
| 288 |
+
},
|
| 289 |
+
"matched_frac_signed": {
|
| 290 |
+
"0.5": 0.6272786458333334,
|
| 291 |
+
"0.6": 0.5888671875,
|
| 292 |
+
"0.7": 0.5530598958333334,
|
| 293 |
+
"0.8": 0.5167643229166666,
|
| 294 |
+
"0.9": 0.4654947916666667
|
| 295 |
+
},
|
| 296 |
+
"matched_cos_abs": {
|
| 297 |
+
"0.5": 0.6562765836715698,
|
| 298 |
+
"0.6": 0.6555381417274475,
|
| 299 |
+
"0.7": 0.6552789807319641,
|
| 300 |
+
"0.8": 0.6549323797225952,
|
| 301 |
+
"0.9": 0.6543703079223633
|
| 302 |
+
},
|
| 303 |
+
"unique_frac_abs": {
|
| 304 |
+
"0.5": 0.97900390625,
|
| 305 |
+
"0.6": 0.981201171875,
|
| 306 |
+
"0.7": 0.982177734375,
|
| 307 |
+
"0.8": 0.9833984375,
|
| 308 |
+
"0.9": 0.9855143229166666
|
| 309 |
+
},
|
| 310 |
+
"dedup_dcs_abs": {
|
| 311 |
+
"0.5": 0.6424973389948718,
|
| 312 |
+
"0.6": 0.6432147928717313,
|
| 313 |
+
"0.7": 0.6436004246788798,
|
| 314 |
+
"0.8": 0.6440594788873568,
|
| 315 |
+
"0.9": 0.6448913109488785
|
| 316 |
+
},
|
| 317 |
+
"matched_frac_abs": {
|
| 318 |
+
"0.5": 0.6228841145833334,
|
| 319 |
+
"0.6": 0.5867513020833334,
|
| 320 |
+
"0.7": 0.55078125,
|
| 321 |
+
"0.8": 0.5154622395833334,
|
| 322 |
+
"0.9": 0.4650065104166667
|
| 323 |
+
}
|
| 324 |
+
},
|
| 325 |
+
"seed1\u2194seed2": {
|
| 326 |
+
"dcs_signed": 0.6490408778190613,
|
| 327 |
+
"dcs_abs": 0.6505702137947083,
|
| 328 |
+
"frac_sign_flipped": 0.0751953125,
|
| 329 |
+
"alive_source": "stats",
|
| 330 |
+
"n_alive_a": 6144,
|
| 331 |
+
"n_alive_b": 6144,
|
| 332 |
+
"matched_cos_signed": {
|
| 333 |
+
"0.5": 0.6494475603103638,
|
| 334 |
+
"0.6": 0.6491867303848267,
|
| 335 |
+
"0.7": 0.6491867303848267,
|
| 336 |
+
"0.8": 0.6490247845649719,
|
| 337 |
+
"0.9": 0.6490408778190613
|
| 338 |
+
},
|
| 339 |
+
"unique_frac_signed": {
|
| 340 |
+
"0.5": 0.9981282552083334,
|
| 341 |
+
"0.6": 0.9988606770833334,
|
| 342 |
+
"0.7": 0.9988606770833334,
|
| 343 |
+
"0.8": 0.9991861979166666,
|
| 344 |
+
"0.9": 1.0
|
| 345 |
+
},
|
| 346 |
+
"dedup_dcs_signed": {
|
| 347 |
+
"0.5": 0.6482319602218922,
|
| 348 |
+
"0.6": 0.6484470970657034,
|
| 349 |
+
"0.7": 0.6484470970657034,
|
| 350 |
+
"0.8": 0.6484966068431579,
|
| 351 |
+
"0.9": 0.6490408778190613
|
| 352 |
+
},
|
| 353 |
+
"matched_frac_signed": {
|
| 354 |
+
"0.5": 0.6297200520833334,
|
| 355 |
+
"0.6": 0.58935546875,
|
| 356 |
+
"0.7": 0.55126953125,
|
| 357 |
+
"0.8": 0.5172526041666666,
|
| 358 |
+
"0.9": 0.4698893229166667
|
| 359 |
+
},
|
| 360 |
+
"matched_cos_abs": {
|
| 361 |
+
"0.5": 0.6563108563423157,
|
| 362 |
+
"0.6": 0.6558352708816528,
|
| 363 |
+
"0.7": 0.6557228565216064,
|
| 364 |
+
"0.8": 0.6552143692970276,
|
| 365 |
+
"0.9": 0.6548299193382263
|
| 366 |
+
},
|
| 367 |
+
"unique_frac_abs": {
|
| 368 |
+
"0.5": 0.9781087239583334,
|
| 369 |
+
"0.6": 0.9806315104166666,
|
| 370 |
+
"0.7": 0.9813639322916666,
|
| 371 |
+
"0.8": 0.982666015625,
|
| 372 |
+
"0.9": 0.98486328125
|
| 373 |
+
},
|
| 374 |
+
"dedup_dcs_abs": {
|
| 375 |
+
"0.5": 0.6419433742169834,
|
| 376 |
+
"0.6": 0.6431327322691989,
|
| 377 |
+
"0.7": 0.643502760969568,
|
| 378 |
+
"0.8": 0.6438568936573574,
|
| 379 |
+
"0.9": 0.6449179430201184
|
| 380 |
+
},
|
| 381 |
+
"matched_frac_abs": {
|
| 382 |
+
"0.5": 0.6251627604166666,
|
| 383 |
+
"0.6": 0.5857747395833334,
|
| 384 |
+
"0.7": 0.548828125,
|
| 385 |
+
"0.8": 0.5157877604166666,
|
| 386 |
+
"0.9": 0.46923828125
|
| 387 |
+
}
|
| 388 |
+
}
|
| 389 |
+
},
|
| 390 |
+
"summary": {
|
| 391 |
+
"n_seeds": 3,
|
| 392 |
+
"n_pairs": 3,
|
| 393 |
+
"report_band": [
|
| 394 |
+
0.6,
|
| 395 |
+
0.7,
|
| 396 |
+
0.8,
|
| 397 |
+
0.9
|
| 398 |
+
],
|
| 399 |
+
"n_dead_mean": 0,
|
| 400 |
+
"n_dead_std": 0.0,
|
| 401 |
+
"n_dense_mean": 210,
|
| 402 |
+
"n_dense_std": 7.371114795831994,
|
| 403 |
+
"l0_mean": 128,
|
| 404 |
+
"l0_std": 0.35472400539009985,
|
| 405 |
+
"recon_r2_mean": 0.9780000000000001,
|
| 406 |
+
"recon_r2_std": 5.773502691895622e-05,
|
| 407 |
+
"l1_per_token_mean": 172.44716666666667,
|
| 408 |
+
"l1_per_token_std": 2.3084603404289528,
|
| 409 |
+
"babel_mu1_signed_mean": 0.18071527779102325,
|
| 410 |
+
"babel_mu1_signed_std": 0.0011921412700285345,
|
| 411 |
+
"n_dup_signed_per_threshold": {
|
| 412 |
+
"0.5": 22.0,
|
| 413 |
+
"0.6": 14.0,
|
| 414 |
+
"0.7": 12.666666666666666,
|
| 415 |
+
"0.8": 10.0,
|
| 416 |
+
"0.9": 0.0
|
| 417 |
+
},
|
| 418 |
+
"n_dup_signed_band": 9.166666666666666,
|
| 419 |
+
"n_dup_signed_band_std": 0.5773502691896257,
|
| 420 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 421 |
+
"0.5": 11.0,
|
| 422 |
+
"0.6": 7.0,
|
| 423 |
+
"0.7": 6.333333333333333,
|
| 424 |
+
"0.8": 5.0,
|
| 425 |
+
"0.9": 0.0
|
| 426 |
+
},
|
| 427 |
+
"n_dup_clusters_signed_band": 4.583333333333333,
|
| 428 |
+
"n_dup_clusters_signed_band_std": 0.28867513459481287,
|
| 429 |
+
"largest_dup_signed_per_threshold": {
|
| 430 |
+
"0.5": 2.0,
|
| 431 |
+
"0.6": 2.0,
|
| 432 |
+
"0.7": 2.0,
|
| 433 |
+
"0.8": 2.0,
|
| 434 |
+
"0.9": 1.0
|
| 435 |
+
},
|
| 436 |
+
"largest_dup_signed_band": 1.75,
|
| 437 |
+
"largest_dup_signed_band_std": 0.0,
|
| 438 |
+
"dcs_signed_mean": 0.6511690417925516,
|
| 439 |
+
"dcs_signed_std": 0.004200510779872004,
|
| 440 |
+
"dedup_dcs_signed_per_threshold": {
|
| 441 |
+
"0.5": 0.6506356674621606,
|
| 442 |
+
"0.6": 0.6507859381963499,
|
| 443 |
+
"0.7": 0.6508688286477183,
|
| 444 |
+
"0.8": 0.6508449053799267,
|
| 445 |
+
"0.9": 0.6511690616607666
|
| 446 |
+
},
|
| 447 |
+
"dedup_dcs_signed_band": 0.6509171834711904,
|
| 448 |
+
"dedup_dcs_signed_band_std": 0.0042360415910276715,
|
| 449 |
+
"matched_cos_signed_per_threshold": {
|
| 450 |
+
"0.5": 0.651802639166514,
|
| 451 |
+
"0.6": 0.6515282392501831,
|
| 452 |
+
"0.7": 0.6515403191248575,
|
| 453 |
+
"0.8": 0.6513749957084656,
|
| 454 |
+
"0.9": 0.6511690616607666
|
| 455 |
+
},
|
| 456 |
+
"matched_cos_signed_band": 0.6514031539360682,
|
| 457 |
+
"matched_cos_signed_band_std": 0.004231971217361345,
|
| 458 |
+
"unique_frac_signed_per_threshold": {
|
| 459 |
+
"0.5": 0.9982096354166666,
|
| 460 |
+
"0.6": 0.9988606770833334,
|
| 461 |
+
"0.7": 0.9989691840277778,
|
| 462 |
+
"0.8": 0.9991861979166666,
|
| 463 |
+
"0.9": 1.0
|
| 464 |
+
},
|
| 465 |
+
"unique_frac_signed_band": 0.9992540147569444,
|
| 466 |
+
"unique_frac_signed_band_std": 2.3492442593957545e-05,
|
| 467 |
+
"matched_frac_signed_per_threshold": {
|
| 468 |
+
"0.5": 0.6309678819444445,
|
| 469 |
+
"0.6": 0.5914713541666666,
|
| 470 |
+
"0.7": 0.5538194444444445,
|
| 471 |
+
"0.8": 0.5177951388888888,
|
| 472 |
+
"0.9": 0.4694010416666667
|
| 473 |
+
},
|
| 474 |
+
"matched_frac_signed_band": 0.5331217447916666,
|
| 475 |
+
"matched_frac_signed_band_std": 0.002854404281786419,
|
| 476 |
+
"babel_mu1_abs_mean": 0.22425366441408792,
|
| 477 |
+
"babel_mu1_abs_std": 0.0008143306267803023,
|
| 478 |
+
"n_dup_abs_per_threshold": {
|
| 479 |
+
"0.5": 260.3333333333333,
|
| 480 |
+
"0.6": 233.66666666666666,
|
| 481 |
+
"0.7": 223.33333333333334,
|
| 482 |
+
"0.8": 208.0,
|
| 483 |
+
"0.9": 180.66666666666666
|
| 484 |
+
},
|
| 485 |
+
"n_dup_abs_band": 211.41666666666666,
|
| 486 |
+
"n_dup_abs_band_std": 9.028888820521235,
|
| 487 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 488 |
+
"0.5": 130.0,
|
| 489 |
+
"0.6": 116.66666666666667,
|
| 490 |
+
"0.7": 111.66666666666667,
|
| 491 |
+
"0.8": 104.0,
|
| 492 |
+
"0.9": 90.33333333333333
|
| 493 |
+
},
|
| 494 |
+
"n_dup_clusters_abs_band": 105.66666666666667,
|
| 495 |
+
"n_dup_clusters_abs_band_std": 4.584848234492973,
|
| 496 |
+
"largest_dup_abs_per_threshold": {
|
| 497 |
+
"0.5": 2.3333333333333335,
|
| 498 |
+
"0.6": 2.3333333333333335,
|
| 499 |
+
"0.7": 2.0,
|
| 500 |
+
"0.8": 2.0,
|
| 501 |
+
"0.9": 2.0
|
| 502 |
+
},
|
| 503 |
+
"largest_dup_abs_band": 2.0833333333333335,
|
| 504 |
+
"largest_dup_abs_band_std": 0.14433756729740643,
|
| 505 |
+
"dcs_abs_mean": 0.6527343193689982,
|
| 506 |
+
"dcs_abs_std": 0.004114170335625304,
|
| 507 |
+
"dedup_dcs_abs_per_threshold": {
|
| 508 |
+
"0.5": 0.6445828595428288,
|
| 509 |
+
"0.6": 0.645415302973965,
|
| 510 |
+
"0.7": 0.6458169491039977,
|
| 511 |
+
"0.8": 0.6462046575664621,
|
| 512 |
+
"0.9": 0.6471742767478442
|
| 513 |
+
},
|
| 514 |
+
"dedup_dcs_abs_band": 0.6461527965980672,
|
| 515 |
+
"dedup_dcs_abs_band_std": 0.003907334242038059,
|
| 516 |
+
"matched_cos_abs_per_threshold": {
|
| 517 |
+
"0.5": 0.6585517525672913,
|
| 518 |
+
"0.6": 0.6579443613688151,
|
| 519 |
+
"0.7": 0.6577717065811157,
|
| 520 |
+
"0.8": 0.6573312083880106,
|
| 521 |
+
"0.9": 0.6568310260772705
|
| 522 |
+
},
|
| 523 |
+
"matched_cos_abs_band": 0.6574695756038029,
|
| 524 |
+
"matched_cos_abs_band_std": 0.003908953193257159,
|
| 525 |
+
"unique_frac_abs_per_threshold": {
|
| 526 |
+
"0.5": 0.9787868923611112,
|
| 527 |
+
"0.6": 0.98095703125,
|
| 528 |
+
"0.7": 0.9818250868055555,
|
| 529 |
+
"0.8": 0.9830729166666666,
|
| 530 |
+
"0.9": 0.9852973090277777
|
| 531 |
+
},
|
| 532 |
+
"unique_frac_abs_band": 0.9827880859374999,
|
| 533 |
+
"unique_frac_abs_band_std": 0.00036166155669418353,
|
| 534 |
+
"matched_frac_abs_per_threshold": {
|
| 535 |
+
"0.5": 0.6261935763888888,
|
| 536 |
+
"0.6": 0.5880533854166666,
|
| 537 |
+
"0.7": 0.5513237847222222,
|
| 538 |
+
"0.8": 0.5161675347222222,
|
| 539 |
+
"0.9": 0.4688585069444445
|
| 540 |
+
},
|
| 541 |
+
"matched_frac_abs_band": 0.531100802951389,
|
| 542 |
+
"matched_frac_abs_band_std": 0.0024282595977142053,
|
| 543 |
+
"alive_source": "stats"
|
| 544 |
+
}
|
| 545 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x8/l0_32/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 771000,
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"l0": 32.0865084306096,
|
| 20 |
+
"recon_r2": 0.958755501091873,
|
| 21 |
+
"l1_per_token": 89.91572485589657
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 32.06509857328145,
|
| 25 |
+
"recon_r2": 0.9586197211703241,
|
| 26 |
+
"l1_per_token": 90.83018726911867
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 32.10813878080415,
|
| 30 |
+
"recon_r2": 0.958593822482764,
|
| 31 |
+
"l1_per_token": 90.41584288713997
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.5262949178426933,
|
| 38 |
+
"mcc_abs": 0.526334355397068,
|
| 39 |
+
"act_iou": 0.43305336246654047,
|
| 40 |
+
"act_iou_n_pairs": 6144,
|
| 41 |
+
"input_r2": 0.9908650980605785,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.9402515128631861,
|
| 44 |
+
"0.7": 0.9527841160879091,
|
| 45 |
+
"0.8": 0.9645011304356255,
|
| 46 |
+
"0.9": 0.9750549384527898
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.9397122225999072,
|
| 50 |
+
"0.7": 0.9526168401761123,
|
| 51 |
+
"0.8": 0.9642998854265218,
|
| 52 |
+
"0.9": 0.9750244106801105
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.5270038318360919,
|
| 58 |
+
"mcc_abs": 0.5270382830494768,
|
| 59 |
+
"act_iou": 0.4334568885351235,
|
| 60 |
+
"act_iou_n_pairs": 6144,
|
| 61 |
+
"input_r2": 0.9909521334716191,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.9413386836216968,
|
| 64 |
+
"0.7": 0.952915418708591,
|
| 65 |
+
"0.8": 0.9637463272178458,
|
| 66 |
+
"0.9": 0.975637054417111
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.9413515697646627,
|
| 70 |
+
"0.7": 0.9527576533480522,
|
| 71 |
+
"0.8": 0.9637323097269777,
|
| 72 |
+
"0.9": 0.9756234891198798
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.5238004040460928,
|
| 78 |
+
"mcc_abs": 0.5238392893510958,
|
| 79 |
+
"act_iou": 0.4315482892044136,
|
| 80 |
+
"act_iou_n_pairs": 6144,
|
| 81 |
+
"input_r2": 0.9908737528802516,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.9430283513712964,
|
| 84 |
+
"0.7": 0.9555447071103701,
|
| 85 |
+
"0.8": 0.9660350354288983,
|
| 86 |
+
"0.9": 0.9752992295624631
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.9425256724951891,
|
| 90 |
+
"0.7": 0.9548439136272728,
|
| 91 |
+
"0.8": 0.9655781995010535,
|
| 92 |
+
"0.9": 0.9753303789133898
|
| 93 |
+
}
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"summary": {
|
| 97 |
+
"n_seeds": 3,
|
| 98 |
+
"n_pairs": 3,
|
| 99 |
+
"report_band": [
|
| 100 |
+
0.6,
|
| 101 |
+
0.7,
|
| 102 |
+
0.8,
|
| 103 |
+
0.9
|
| 104 |
+
],
|
| 105 |
+
"l0_mean": 32.08658192823174,
|
| 106 |
+
"l0_std": 0.02152019789231685,
|
| 107 |
+
"recon_r2_mean": 0.9586563482483204,
|
| 108 |
+
"recon_r2_std": 8.683979671073901e-05,
|
| 109 |
+
"l1_per_token_mean": 90.38725167071841,
|
| 110 |
+
"l1_per_token_std": 0.4579011569556732,
|
| 111 |
+
"input_r2_mean": 0.9908969948041498,
|
| 112 |
+
"input_r2_std": 4.794716846298545e-05,
|
| 113 |
+
"mcc_signed_mean": 0.5256997179082926,
|
| 114 |
+
"mcc_signed_std": 0.0016826124398884726,
|
| 115 |
+
"mcc_abs_mean": 0.5257373092658802,
|
| 116 |
+
"mcc_abs_std": 0.001680993228018776,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.9415395159520598,
|
| 119 |
+
"0.7": 0.9537480806356234,
|
| 120 |
+
"0.8": 0.9647608310274566,
|
| 121 |
+
"0.9": 0.9753304074774546
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.9588447087731485,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.9411964882865863,
|
| 126 |
+
"0.7": 0.9534061357171457,
|
| 127 |
+
"0.8": 0.9645367982181844,
|
| 128 |
+
"0.9": 0.97532609290446
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.9586163787815941,
|
| 131 |
+
"act_iou_mean": 0.4326861800686925,
|
| 132 |
+
"act_iou_std": 0.0010058852003402548,
|
| 133 |
+
"act_iou_n_pairs": 6144.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x8/l0_32/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,545 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 2,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"linkage": "complete",
|
| 5 |
+
"thresholds": [
|
| 6 |
+
0.5,
|
| 7 |
+
0.6,
|
| 8 |
+
0.7,
|
| 9 |
+
0.8,
|
| 10 |
+
0.9
|
| 11 |
+
],
|
| 12 |
+
"seeds": [
|
| 13 |
+
"seed0",
|
| 14 |
+
"seed1",
|
| 15 |
+
"seed2"
|
| 16 |
+
],
|
| 17 |
+
"per_seed": {
|
| 18 |
+
"seed0": {
|
| 19 |
+
"d_sae": 6144,
|
| 20 |
+
"d_in": 768,
|
| 21 |
+
"l0": 32,
|
| 22 |
+
"l0_raw": 31.9488,
|
| 23 |
+
"recon_r2": 0.9587,
|
| 24 |
+
"l1_per_token": 89.9078,
|
| 25 |
+
"stats_source": "stats",
|
| 26 |
+
"n_dead": 0,
|
| 27 |
+
"dead_rate": 0.0,
|
| 28 |
+
"alive_source": "stats",
|
| 29 |
+
"n_dense": 11,
|
| 30 |
+
"dense_rate": 0.0017903645833333333,
|
| 31 |
+
"dense_threshold": 0.1,
|
| 32 |
+
"n_dup_signed": {
|
| 33 |
+
"0.5": 12,
|
| 34 |
+
"0.6": 6,
|
| 35 |
+
"0.7": 2,
|
| 36 |
+
"0.8": 0,
|
| 37 |
+
"0.9": 0
|
| 38 |
+
},
|
| 39 |
+
"n_dup_clusters_signed": {
|
| 40 |
+
"0.5": 6,
|
| 41 |
+
"0.6": 3,
|
| 42 |
+
"0.7": 1,
|
| 43 |
+
"0.8": 0,
|
| 44 |
+
"0.9": 0
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 2,
|
| 48 |
+
"0.6": 2,
|
| 49 |
+
"0.7": 2,
|
| 50 |
+
"0.8": 1,
|
| 51 |
+
"0.9": 1
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.19783306121826172,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 38,
|
| 56 |
+
"0.6": 22,
|
| 57 |
+
"0.7": 10,
|
| 58 |
+
"0.8": 4,
|
| 59 |
+
"0.9": 4
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 19,
|
| 63 |
+
"0.6": 11,
|
| 64 |
+
"0.7": 5,
|
| 65 |
+
"0.8": 2,
|
| 66 |
+
"0.9": 2
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 2,
|
| 70 |
+
"0.6": 2,
|
| 71 |
+
"0.7": 2,
|
| 72 |
+
"0.8": 2,
|
| 73 |
+
"0.9": 2
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.20420557260513306
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 6144,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 32,
|
| 81 |
+
"l0_raw": 31.9488,
|
| 82 |
+
"recon_r2": 0.9586,
|
| 83 |
+
"l1_per_token": 90.8347,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 11,
|
| 89 |
+
"dense_rate": 0.0017903645833333333,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 16,
|
| 93 |
+
"0.6": 10,
|
| 94 |
+
"0.7": 8,
|
| 95 |
+
"0.8": 0,
|
| 96 |
+
"0.9": 0
|
| 97 |
+
},
|
| 98 |
+
"n_dup_clusters_signed": {
|
| 99 |
+
"0.5": 8,
|
| 100 |
+
"0.6": 5,
|
| 101 |
+
"0.7": 4,
|
| 102 |
+
"0.8": 0,
|
| 103 |
+
"0.9": 0
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 2,
|
| 107 |
+
"0.6": 2,
|
| 108 |
+
"0.7": 2,
|
| 109 |
+
"0.8": 1,
|
| 110 |
+
"0.9": 1
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.19802320003509521,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 38,
|
| 115 |
+
"0.6": 20,
|
| 116 |
+
"0.7": 14,
|
| 117 |
+
"0.8": 4,
|
| 118 |
+
"0.9": 4
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 19,
|
| 122 |
+
"0.6": 10,
|
| 123 |
+
"0.7": 7,
|
| 124 |
+
"0.8": 2,
|
| 125 |
+
"0.9": 2
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 2,
|
| 129 |
+
"0.6": 2,
|
| 130 |
+
"0.7": 2,
|
| 131 |
+
"0.8": 2,
|
| 132 |
+
"0.9": 2
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.20472045242786407
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 6144,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 32,
|
| 140 |
+
"l0_raw": 31.9488,
|
| 141 |
+
"recon_r2": 0.9586,
|
| 142 |
+
"l1_per_token": 90.4115,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 11,
|
| 148 |
+
"dense_rate": 0.0017903645833333333,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 22,
|
| 152 |
+
"0.6": 12,
|
| 153 |
+
"0.7": 8,
|
| 154 |
+
"0.8": 2,
|
| 155 |
+
"0.9": 0
|
| 156 |
+
},
|
| 157 |
+
"n_dup_clusters_signed": {
|
| 158 |
+
"0.5": 11,
|
| 159 |
+
"0.6": 6,
|
| 160 |
+
"0.7": 4,
|
| 161 |
+
"0.8": 1,
|
| 162 |
+
"0.9": 0
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 2,
|
| 166 |
+
"0.6": 2,
|
| 167 |
+
"0.7": 2,
|
| 168 |
+
"0.8": 2,
|
| 169 |
+
"0.9": 1
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.19730699062347412,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 52,
|
| 174 |
+
"0.6": 24,
|
| 175 |
+
"0.7": 16,
|
| 176 |
+
"0.8": 6,
|
| 177 |
+
"0.9": 4
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 26,
|
| 181 |
+
"0.6": 12,
|
| 182 |
+
"0.7": 8,
|
| 183 |
+
"0.8": 3,
|
| 184 |
+
"0.9": 2
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 2,
|
| 188 |
+
"0.6": 2,
|
| 189 |
+
"0.7": 2,
|
| 190 |
+
"0.8": 2,
|
| 191 |
+
"0.9": 2
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.20398755371570587
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.5966472625732422,
|
| 199 |
+
"dcs_abs": 0.598185122013092,
|
| 200 |
+
"frac_sign_flipped": 0.08268229166666667,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"n_alive_a": 6144,
|
| 203 |
+
"n_alive_b": 6144,
|
| 204 |
+
"matched_cos_signed": {
|
| 205 |
+
"0.5": 0.5965452790260315,
|
| 206 |
+
"0.6": 0.5967425107955933,
|
| 207 |
+
"0.7": 0.5967465043067932,
|
| 208 |
+
"0.8": 0.5966472625732422,
|
| 209 |
+
"0.9": 0.5966472625732422
|
| 210 |
+
},
|
| 211 |
+
"unique_frac_signed": {
|
| 212 |
+
"0.5": 0.9988606770833334,
|
| 213 |
+
"0.6": 0.9993489583333334,
|
| 214 |
+
"0.7": 0.9995930989583334,
|
| 215 |
+
"0.8": 1.0,
|
| 216 |
+
"0.9": 1.0
|
| 217 |
+
},
|
| 218 |
+
"dedup_dcs_signed": {
|
| 219 |
+
"0.5": 0.5958656213188078,
|
| 220 |
+
"0.6": 0.596354006556794,
|
| 221 |
+
"0.7": 0.5965036875325799,
|
| 222 |
+
"0.8": 0.5966472625732422,
|
| 223 |
+
"0.9": 0.5966472625732422
|
| 224 |
+
},
|
| 225 |
+
"matched_frac_signed": {
|
| 226 |
+
"0.5": 0.54833984375,
|
| 227 |
+
"0.6": 0.5369466145833334,
|
| 228 |
+
"0.7": 0.51904296875,
|
| 229 |
+
"0.8": 0.49609375,
|
| 230 |
+
"0.9": 0.4661458333333333
|
| 231 |
+
},
|
| 232 |
+
"matched_cos_abs": {
|
| 233 |
+
"0.5": 0.5974833369255066,
|
| 234 |
+
"0.6": 0.5978721380233765,
|
| 235 |
+
"0.7": 0.598053514957428,
|
| 236 |
+
"0.8": 0.598054826259613,
|
| 237 |
+
"0.9": 0.598054826259613
|
| 238 |
+
},
|
| 239 |
+
"unique_frac_abs": {
|
| 240 |
+
"0.5": 0.9969075520833334,
|
| 241 |
+
"0.6": 0.998291015625,
|
| 242 |
+
"0.7": 0.9990234375,
|
| 243 |
+
"0.8": 0.9996744791666666,
|
| 244 |
+
"0.9": 0.9996744791666666
|
| 245 |
+
},
|
| 246 |
+
"dedup_dcs_abs": {
|
| 247 |
+
"0.5": 0.5956356508249883,
|
| 248 |
+
"0.6": 0.5968503838812467,
|
| 249 |
+
"0.7": 0.5974694783217274,
|
| 250 |
+
"0.8": 0.59786014695419,
|
| 251 |
+
"0.9": 0.59786014695419
|
| 252 |
+
},
|
| 253 |
+
"matched_frac_abs": {
|
| 254 |
+
"0.5": 0.5470377604166666,
|
| 255 |
+
"0.6": 0.5361328125,
|
| 256 |
+
"0.7": 0.5185546875,
|
| 257 |
+
"0.8": 0.4959309895833333,
|
| 258 |
+
"0.9": 0.4658203125
|
| 259 |
+
}
|
| 260 |
+
},
|
| 261 |
+
"seed0\u2194seed2": {
|
| 262 |
+
"dcs_signed": 0.5975165367126465,
|
| 263 |
+
"dcs_abs": 0.5992233157157898,
|
| 264 |
+
"frac_sign_flipped": 0.08740234375,
|
| 265 |
+
"alive_source": "stats",
|
| 266 |
+
"n_alive_a": 6144,
|
| 267 |
+
"n_alive_b": 6144,
|
| 268 |
+
"matched_cos_signed": {
|
| 269 |
+
"0.5": 0.5976077914237976,
|
| 270 |
+
"0.6": 0.5976551175117493,
|
| 271 |
+
"0.7": 0.5976582169532776,
|
| 272 |
+
"0.8": 0.5975971221923828,
|
| 273 |
+
"0.9": 0.5975165367126465
|
| 274 |
+
},
|
| 275 |
+
"unique_frac_signed": {
|
| 276 |
+
"0.5": 0.9986165364583334,
|
| 277 |
+
"0.6": 0.999267578125,
|
| 278 |
+
"0.7": 0.9995930989583334,
|
| 279 |
+
"0.8": 0.9999186197916666,
|
| 280 |
+
"0.9": 1.0
|
| 281 |
+
},
|
| 282 |
+
"dedup_dcs_signed": {
|
| 283 |
+
"0.5": 0.5967810228321468,
|
| 284 |
+
"0.6": 0.597217381829978,
|
| 285 |
+
"0.7": 0.5974150292022387,
|
| 286 |
+
"0.8": 0.5975484896140794,
|
| 287 |
+
"0.9": 0.5975165367126465
|
| 288 |
+
},
|
| 289 |
+
"matched_frac_signed": {
|
| 290 |
+
"0.5": 0.55126953125,
|
| 291 |
+
"0.6": 0.537109375,
|
| 292 |
+
"0.7": 0.5213216145833334,
|
| 293 |
+
"0.8": 0.5017903645833334,
|
| 294 |
+
"0.9": 0.46533203125
|
| 295 |
+
},
|
| 296 |
+
"matched_cos_abs": {
|
| 297 |
+
"0.5": 0.5987836718559265,
|
| 298 |
+
"0.6": 0.598793089389801,
|
| 299 |
+
"0.7": 0.599134087562561,
|
| 300 |
+
"0.8": 0.5991737842559814,
|
| 301 |
+
"0.9": 0.5990933775901794
|
| 302 |
+
},
|
| 303 |
+
"unique_frac_abs": {
|
| 304 |
+
"0.5": 0.996337890625,
|
| 305 |
+
"0.6": 0.9981282552083334,
|
| 306 |
+
"0.7": 0.9989420572916666,
|
| 307 |
+
"0.8": 0.9995930989583334,
|
| 308 |
+
"0.9": 0.9996744791666666
|
| 309 |
+
},
|
| 310 |
+
"dedup_dcs_abs": {
|
| 311 |
+
"0.5": 0.596590860557626,
|
| 312 |
+
"0.6": 0.5976723015434497,
|
| 313 |
+
"0.7": 0.5985002380233102,
|
| 314 |
+
"0.8": 0.5989299798190284,
|
| 315 |
+
"0.9": 0.5988983602146618
|
| 316 |
+
},
|
| 317 |
+
"matched_frac_abs": {
|
| 318 |
+
"0.5": 0.5494791666666666,
|
| 319 |
+
"0.6": 0.5358072916666666,
|
| 320 |
+
"0.7": 0.5211588541666666,
|
| 321 |
+
"0.8": 0.5016276041666666,
|
| 322 |
+
"0.9": 0.4650065104166667
|
| 323 |
+
}
|
| 324 |
+
},
|
| 325 |
+
"seed1\u2194seed2": {
|
| 326 |
+
"dcs_signed": 0.5937657952308655,
|
| 327 |
+
"dcs_abs": 0.5953160524368286,
|
| 328 |
+
"frac_sign_flipped": 0.09326171875,
|
| 329 |
+
"alive_source": "stats",
|
| 330 |
+
"n_alive_a": 6144,
|
| 331 |
+
"n_alive_b": 6144,
|
| 332 |
+
"matched_cos_signed": {
|
| 333 |
+
"0.5": 0.5938093066215515,
|
| 334 |
+
"0.6": 0.5938435196876526,
|
| 335 |
+
"0.7": 0.5937299728393555,
|
| 336 |
+
"0.8": 0.593850314617157,
|
| 337 |
+
"0.9": 0.5937657952308655
|
| 338 |
+
},
|
| 339 |
+
"unique_frac_signed": {
|
| 340 |
+
"0.5": 0.9984537760416666,
|
| 341 |
+
"0.6": 0.9991048177083334,
|
| 342 |
+
"0.7": 0.9993489583333334,
|
| 343 |
+
"0.8": 0.9999186197916666,
|
| 344 |
+
"0.9": 1.0
|
| 345 |
+
},
|
| 346 |
+
"dedup_dcs_signed": {
|
| 347 |
+
"0.5": 0.5928911444449719,
|
| 348 |
+
"0.6": 0.5933119214848073,
|
| 349 |
+
"0.7": 0.5933434298882883,
|
| 350 |
+
"0.8": 0.5938019869548345,
|
| 351 |
+
"0.9": 0.5937657952308655
|
| 352 |
+
},
|
| 353 |
+
"matched_frac_signed": {
|
| 354 |
+
"0.5": 0.5465494791666666,
|
| 355 |
+
"0.6": 0.5320638020833334,
|
| 356 |
+
"0.7": 0.5128580729166666,
|
| 357 |
+
"0.8": 0.4933268229166667,
|
| 358 |
+
"0.9": 0.4666341145833333
|
| 359 |
+
},
|
| 360 |
+
"matched_cos_abs": {
|
| 361 |
+
"0.5": 0.5949174165725708,
|
| 362 |
+
"0.6": 0.595068633556366,
|
| 363 |
+
"0.7": 0.5951368808746338,
|
| 364 |
+
"0.8": 0.5952655673027039,
|
| 365 |
+
"0.9": 0.5951848030090332
|
| 366 |
+
},
|
| 367 |
+
"unique_frac_abs": {
|
| 368 |
+
"0.5": 0.996337890625,
|
| 369 |
+
"0.6": 0.9982096354166666,
|
| 370 |
+
"0.7": 0.998779296875,
|
| 371 |
+
"0.8": 0.9995930989583334,
|
| 372 |
+
"0.9": 0.9996744791666666
|
| 373 |
+
},
|
| 374 |
+
"dedup_dcs_abs": {
|
| 375 |
+
"0.5": 0.5927387639239896,
|
| 376 |
+
"0.6": 0.5940032437501941,
|
| 377 |
+
"0.7": 0.5944103954243474,
|
| 378 |
+
"0.8": 0.5950233531233001,
|
| 379 |
+
"0.9": 0.5949910579559703
|
| 380 |
+
},
|
| 381 |
+
"matched_frac_abs": {
|
| 382 |
+
"0.5": 0.5442708333333334,
|
| 383 |
+
"0.6": 0.5314127604166666,
|
| 384 |
+
"0.7": 0.5128580729166666,
|
| 385 |
+
"0.8": 0.4931640625,
|
| 386 |
+
"0.9": 0.4661458333333333
|
| 387 |
+
}
|
| 388 |
+
}
|
| 389 |
+
},
|
| 390 |
+
"summary": {
|
| 391 |
+
"n_seeds": 3,
|
| 392 |
+
"n_pairs": 3,
|
| 393 |
+
"report_band": [
|
| 394 |
+
0.6,
|
| 395 |
+
0.7,
|
| 396 |
+
0.8,
|
| 397 |
+
0.9
|
| 398 |
+
],
|
| 399 |
+
"n_dead_mean": 0,
|
| 400 |
+
"n_dead_std": 0.0,
|
| 401 |
+
"n_dense_mean": 11,
|
| 402 |
+
"n_dense_std": 0.0,
|
| 403 |
+
"l0_mean": 32,
|
| 404 |
+
"l0_std": 4.351167857633658e-15,
|
| 405 |
+
"recon_r2_mean": 0.9586333333333333,
|
| 406 |
+
"recon_r2_std": 5.773502691895622e-05,
|
| 407 |
+
"l1_per_token_mean": 90.38466666666666,
|
| 408 |
+
"l1_per_token_std": 0.463916166995716,
|
| 409 |
+
"babel_mu1_signed_mean": 0.1977210839589437,
|
| 410 |
+
"babel_mu1_signed_std": 0.0003710029114130265,
|
| 411 |
+
"n_dup_signed_per_threshold": {
|
| 412 |
+
"0.5": 16.666666666666668,
|
| 413 |
+
"0.6": 9.333333333333334,
|
| 414 |
+
"0.7": 6.0,
|
| 415 |
+
"0.8": 0.6666666666666666,
|
| 416 |
+
"0.9": 0.0
|
| 417 |
+
},
|
| 418 |
+
"n_dup_signed_band": 4.0,
|
| 419 |
+
"n_dup_signed_band_std": 1.8027756377319946,
|
| 420 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 421 |
+
"0.5": 8.333333333333334,
|
| 422 |
+
"0.6": 4.666666666666667,
|
| 423 |
+
"0.7": 3.0,
|
| 424 |
+
"0.8": 0.3333333333333333,
|
| 425 |
+
"0.9": 0.0
|
| 426 |
+
},
|
| 427 |
+
"n_dup_clusters_signed_band": 2.0,
|
| 428 |
+
"n_dup_clusters_signed_band_std": 0.9013878188659973,
|
| 429 |
+
"largest_dup_signed_per_threshold": {
|
| 430 |
+
"0.5": 2.0,
|
| 431 |
+
"0.6": 2.0,
|
| 432 |
+
"0.7": 2.0,
|
| 433 |
+
"0.8": 1.3333333333333333,
|
| 434 |
+
"0.9": 1.0
|
| 435 |
+
},
|
| 436 |
+
"largest_dup_signed_band": 1.5833333333333333,
|
| 437 |
+
"largest_dup_signed_band_std": 0.14433756729740646,
|
| 438 |
+
"dcs_signed_mean": 0.5959765315055847,
|
| 439 |
+
"dcs_signed_std": 0.0019632690950628475,
|
| 440 |
+
"dedup_dcs_signed_per_threshold": {
|
| 441 |
+
"0.5": 0.5951792628653089,
|
| 442 |
+
"0.6": 0.5956277699571931,
|
| 443 |
+
"0.7": 0.595754048874369,
|
| 444 |
+
"0.8": 0.5959992463807187,
|
| 445 |
+
"0.9": 0.5959765315055847
|
| 446 |
+
},
|
| 447 |
+
"dedup_dcs_signed_band": 0.5958393991794664,
|
| 448 |
+
"dedup_dcs_signed_band_std": 0.0020267115573688954,
|
| 449 |
+
"matched_cos_signed_per_threshold": {
|
| 450 |
+
"0.5": 0.5959874590237936,
|
| 451 |
+
"0.6": 0.5960803826649984,
|
| 452 |
+
"0.7": 0.5960448980331421,
|
| 453 |
+
"0.8": 0.5960315664609274,
|
| 454 |
+
"0.9": 0.5959765315055847
|
| 455 |
+
},
|
| 456 |
+
"matched_cos_signed_band": 0.5960333446661632,
|
| 457 |
+
"matched_cos_signed_band_std": 0.001989221556298344,
|
| 458 |
+
"unique_frac_signed_per_threshold": {
|
| 459 |
+
"0.5": 0.9986436631944445,
|
| 460 |
+
"0.6": 0.999240451388889,
|
| 461 |
+
"0.7": 0.99951171875,
|
| 462 |
+
"0.8": 0.9999457465277777,
|
| 463 |
+
"0.9": 1.0
|
| 464 |
+
},
|
| 465 |
+
"unique_frac_signed_band": 0.9996744791666666,
|
| 466 |
+
"unique_frac_signed_band_std": 7.335512848845405e-05,
|
| 467 |
+
"matched_frac_signed_per_threshold": {
|
| 468 |
+
"0.5": 0.5487196180555555,
|
| 469 |
+
"0.6": 0.535373263888889,
|
| 470 |
+
"0.7": 0.5177408854166666,
|
| 471 |
+
"0.8": 0.4970703125,
|
| 472 |
+
"0.9": 0.46603732638888884
|
| 473 |
+
},
|
| 474 |
+
"matched_frac_signed_band": 0.5040554470486112,
|
| 475 |
+
"matched_frac_signed_band_std": 0.002620118346848426,
|
| 476 |
+
"babel_mu1_abs_mean": 0.20430452624956766,
|
| 477 |
+
"babel_mu1_abs_std": 0.0003763362836360012,
|
| 478 |
+
"n_dup_abs_per_threshold": {
|
| 479 |
+
"0.5": 42.666666666666664,
|
| 480 |
+
"0.6": 22.0,
|
| 481 |
+
"0.7": 13.333333333333334,
|
| 482 |
+
"0.8": 4.666666666666667,
|
| 483 |
+
"0.9": 4.0
|
| 484 |
+
},
|
| 485 |
+
"n_dup_abs_band": 11.0,
|
| 486 |
+
"n_dup_abs_band_std": 1.3228756555322954,
|
| 487 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 488 |
+
"0.5": 21.333333333333332,
|
| 489 |
+
"0.6": 11.0,
|
| 490 |
+
"0.7": 6.666666666666667,
|
| 491 |
+
"0.8": 2.3333333333333335,
|
| 492 |
+
"0.9": 2.0
|
| 493 |
+
},
|
| 494 |
+
"n_dup_clusters_abs_band": 5.5,
|
| 495 |
+
"n_dup_clusters_abs_band_std": 0.6614378277661477,
|
| 496 |
+
"largest_dup_abs_per_threshold": {
|
| 497 |
+
"0.5": 2.0,
|
| 498 |
+
"0.6": 2.0,
|
| 499 |
+
"0.7": 2.0,
|
| 500 |
+
"0.8": 2.0,
|
| 501 |
+
"0.9": 2.0
|
| 502 |
+
},
|
| 503 |
+
"largest_dup_abs_band": 2.0,
|
| 504 |
+
"largest_dup_abs_band_std": 0.0,
|
| 505 |
+
"dcs_abs_mean": 0.5975748300552368,
|
| 506 |
+
"dcs_abs_std": 0.002023862344169359,
|
| 507 |
+
"dedup_dcs_abs_per_threshold": {
|
| 508 |
+
"0.5": 0.5949884251022013,
|
| 509 |
+
"0.6": 0.5961753097249635,
|
| 510 |
+
"0.7": 0.596793370589795,
|
| 511 |
+
"0.8": 0.5972711599655062,
|
| 512 |
+
"0.9": 0.5972498550416073
|
| 513 |
+
},
|
| 514 |
+
"dedup_dcs_abs_band": 0.596872423830468,
|
| 515 |
+
"dedup_dcs_abs_band_std": 0.0020234081883033993,
|
| 516 |
+
"matched_cos_abs_per_threshold": {
|
| 517 |
+
"0.5": 0.5970614751180013,
|
| 518 |
+
"0.6": 0.5972446203231812,
|
| 519 |
+
"0.7": 0.5974414944648743,
|
| 520 |
+
"0.8": 0.5974980592727661,
|
| 521 |
+
"0.9": 0.5974443356196085
|
| 522 |
+
},
|
| 523 |
+
"matched_cos_abs_band": 0.5974071274201075,
|
| 524 |
+
"matched_cos_abs_band_std": 0.0020109914874086162,
|
| 525 |
+
"unique_frac_abs_per_threshold": {
|
| 526 |
+
"0.5": 0.9965277777777778,
|
| 527 |
+
"0.6": 0.9982096354166666,
|
| 528 |
+
"0.7": 0.9989149305555555,
|
| 529 |
+
"0.8": 0.9996202256944445,
|
| 530 |
+
"0.9": 0.9996744791666666
|
| 531 |
+
},
|
| 532 |
+
"unique_frac_abs_band": 0.9991048177083333,
|
| 533 |
+
"unique_frac_abs_band_std": 5.3827948223128575e-05,
|
| 534 |
+
"matched_frac_abs_per_threshold": {
|
| 535 |
+
"0.5": 0.5469292534722222,
|
| 536 |
+
"0.6": 0.534450954861111,
|
| 537 |
+
"0.7": 0.5175238715277777,
|
| 538 |
+
"0.8": 0.4969075520833333,
|
| 539 |
+
"0.9": 0.4656575520833333
|
| 540 |
+
},
|
| 541 |
+
"matched_frac_abs_band": 0.5036349826388888,
|
| 542 |
+
"matched_frac_abs_band_std": 0.0025359871435582685,
|
| 543 |
+
"alive_source": "stats"
|
| 544 |
+
}
|
| 545 |
+
}
|