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/clip-vit-b-16_layer10/consdl/x16/l0_128/code_metrics_step_final.json +136 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_128/dict_metrics_step_final.json +550 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_32/code_metrics_step_final.json +136 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_32/dict_metrics_step_final.json +550 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_32/seed1/config.json +24 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_32/seed1/lock.lock +0 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_32/seed2/config.json +24 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_32/seed2/lock.lock +0 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/code_metrics_step_final.json +136 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/dict_metrics_step_final.json +550 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed0/config.json +24 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed0/lock.lock +0 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed0/stats.imagenet.split_val.seed_no/metrics.json +9 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed1/config.json +24 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed1/lock.lock +0 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed1/stats.imagenet.split_val.seed_no/metrics.json +9 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed2/config.json +24 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed2/lock.lock +0 -0
- vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed2/stats.imagenet.split_val.seed_no/metrics.json +9 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed0/model.safetensors +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed1/model.safetensors +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/model.safetensors +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/active_count_hist_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/active_count_proportion_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/firing_rate_per_latent.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/max_act_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/mean_act.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/mean_act_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/sparsity.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/sparsity_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/model.safetensors +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/active_count_hist_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/active_count_proportion_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/firing_rate_per_latent.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/max_act_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/mean_act.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/mean_act_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/sparsity.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/sparsity_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/model.safetensors +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/active_count_hist_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/active_count_proportion_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/firing_rate_per_latent.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/max_act_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/mean_act.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/mean_act_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/sparsity.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/sparsity_per_token.pt +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed2/model.safetensors +3 -0
- vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed2/stats.imagenet.split_val.seed_no/active_count_hist_per_token.pt +3 -0
vision/clip-vit-b-16_layer10/consdl/x16/l0_128/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 591000,
|
| 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": 128.72902538071065,
|
| 20 |
+
"recon_r2": 0.9220997519580567,
|
| 21 |
+
"l1_per_token": 23.541516082042293
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 128.80150423011844,
|
| 25 |
+
"recon_r2": 0.9221203686554592,
|
| 26 |
+
"l1_per_token": 23.487404403787696
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 132.42130795262267,
|
| 30 |
+
"recon_r2": 0.9219594368798243,
|
| 31 |
+
"l1_per_token": 23.645934702859176
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.7430276669914712,
|
| 38 |
+
"mcc_abs": 0.7430278851720695,
|
| 39 |
+
"act_iou": 0.40716211389714535,
|
| 40 |
+
"act_iou_n_pairs": 12288,
|
| 41 |
+
"input_r2": 0.9975589442432786,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.783873331240773,
|
| 44 |
+
"0.7": 0.8219621896310739,
|
| 45 |
+
"0.8": 0.8787930926146232,
|
| 46 |
+
"0.9": 0.9437717748372098
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.783544910252085,
|
| 50 |
+
"0.7": 0.8219253067300413,
|
| 51 |
+
"0.8": 0.878816147123866,
|
| 52 |
+
"0.9": 0.9437713028438632
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.7421405569987973,
|
| 58 |
+
"mcc_abs": 0.7421408650387833,
|
| 59 |
+
"act_iou": 0.4059004880648401,
|
| 60 |
+
"act_iou_n_pairs": 12288,
|
| 61 |
+
"input_r2": 0.9975688213436134,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.7837816721563289,
|
| 64 |
+
"0.7": 0.8224386730273939,
|
| 65 |
+
"0.8": 0.8797112369836597,
|
| 66 |
+
"0.9": 0.9424956475706369
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.7841632511532908,
|
| 70 |
+
"0.7": 0.8222750504268559,
|
| 71 |
+
"0.8": 0.8795836400764604,
|
| 72 |
+
"0.9": 0.9424741794207664
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.742672418135363,
|
| 78 |
+
"mcc_abs": 0.7426736437509343,
|
| 79 |
+
"act_iou": 0.40557864488317247,
|
| 80 |
+
"act_iou_n_pairs": 12288,
|
| 81 |
+
"input_r2": 0.9975757320751364,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.7835981894948533,
|
| 84 |
+
"0.7": 0.8220574977571931,
|
| 85 |
+
"0.8": 0.877922075038517,
|
| 86 |
+
"0.9": 0.9418117477199773
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.7833263136042731,
|
| 90 |
+
"0.7": 0.8221036612018097,
|
| 91 |
+
"0.8": 0.8780253470365059,
|
| 92 |
+
"0.9": 0.9417569135546872
|
| 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": 129.98394585448392,
|
| 106 |
+
"l0_std": 2.11112855932209,
|
| 107 |
+
"recon_r2_mean": 0.9220598524977802,
|
| 108 |
+
"recon_r2_std": 8.757130979965736e-05,
|
| 109 |
+
"l1_per_token_mean": 23.55828506289639,
|
| 110 |
+
"l1_per_token_std": 0.08058450825119519,
|
| 111 |
+
"input_r2_mean": 0.9975678325540095,
|
| 112 |
+
"input_r2_std": 8.437482046306926e-06,
|
| 113 |
+
"mcc_signed_mean": 0.7426135473752105,
|
| 114 |
+
"mcc_signed_std": 0.0004464754859744715,
|
| 115 |
+
"mcc_abs_mean": 0.7426141313205957,
|
| 116 |
+
"mcc_abs_std": 0.0004464946542091884,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.7837510642973183,
|
| 119 |
+
"0.7": 0.8221527868052204,
|
| 120 |
+
"0.8": 0.8788088015455999,
|
| 121 |
+
"0.9": 0.9426930567092747
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.8568514273393534,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.7836781583365496,
|
| 126 |
+
"0.7": 0.8221013394529022,
|
| 127 |
+
"0.8": 0.8788083780789441,
|
| 128 |
+
"0.9": 0.9426674652731056
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.8568138352853754,
|
| 131 |
+
"act_iou_mean": 0.40621374894838597,
|
| 132 |
+
"act_iou_std": 0.0008369246175271848,
|
| 133 |
+
"act_iou_n_pairs": 12288.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_128/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,550 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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": 129,
|
| 22 |
+
"l0_raw": 129.024,
|
| 23 |
+
"recon_r2": 0.9223,
|
| 24 |
+
"l1_per_token": 23.5664,
|
| 25 |
+
"stats_source": "stats",
|
| 26 |
+
"n_dead": 0,
|
| 27 |
+
"dead_rate": 0.0,
|
| 28 |
+
"alive_source": "stats",
|
| 29 |
+
"n_dense": 0,
|
| 30 |
+
"dense_rate": 0.0,
|
| 31 |
+
"dense_threshold": 0.1,
|
| 32 |
+
"n_dup_signed": {
|
| 33 |
+
"0.5": 9164,
|
| 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": 3557,
|
| 41 |
+
"0.6": 0,
|
| 42 |
+
"0.7": 0,
|
| 43 |
+
"0.8": 0,
|
| 44 |
+
"0.9": 0
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 10,
|
| 48 |
+
"0.6": 1,
|
| 49 |
+
"0.7": 1,
|
| 50 |
+
"0.8": 1,
|
| 51 |
+
"0.9": 1
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.5085816979408264,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 9171,
|
| 56 |
+
"0.6": 6,
|
| 57 |
+
"0.7": 0,
|
| 58 |
+
"0.8": 0,
|
| 59 |
+
"0.9": 0
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 3561,
|
| 63 |
+
"0.6": 3,
|
| 64 |
+
"0.7": 0,
|
| 65 |
+
"0.8": 0,
|
| 66 |
+
"0.9": 0
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 10,
|
| 70 |
+
"0.6": 2,
|
| 71 |
+
"0.7": 1,
|
| 72 |
+
"0.8": 1,
|
| 73 |
+
"0.9": 1
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.509075939655304
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 12288,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 129,
|
| 81 |
+
"l0_raw": 129.024,
|
| 82 |
+
"recon_r2": 0.9223,
|
| 83 |
+
"l1_per_token": 23.5124,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 0,
|
| 89 |
+
"dense_rate": 0.0,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 9257,
|
| 93 |
+
"0.6": 5,
|
| 94 |
+
"0.7": 0,
|
| 95 |
+
"0.8": 0,
|
| 96 |
+
"0.9": 0
|
| 97 |
+
},
|
| 98 |
+
"n_dup_clusters_signed": {
|
| 99 |
+
"0.5": 3570,
|
| 100 |
+
"0.6": 2,
|
| 101 |
+
"0.7": 0,
|
| 102 |
+
"0.8": 0,
|
| 103 |
+
"0.9": 0
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 10,
|
| 107 |
+
"0.6": 3,
|
| 108 |
+
"0.7": 1,
|
| 109 |
+
"0.8": 1,
|
| 110 |
+
"0.9": 1
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.5089403986930847,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 9250,
|
| 115 |
+
"0.6": 17,
|
| 116 |
+
"0.7": 0,
|
| 117 |
+
"0.8": 0,
|
| 118 |
+
"0.9": 0
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 3567,
|
| 122 |
+
"0.6": 8,
|
| 123 |
+
"0.7": 0,
|
| 124 |
+
"0.8": 0,
|
| 125 |
+
"0.9": 0
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 10,
|
| 129 |
+
"0.6": 3,
|
| 130 |
+
"0.7": 1,
|
| 131 |
+
"0.8": 1,
|
| 132 |
+
"0.9": 1
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.5094455480575562
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 12288,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 133,
|
| 140 |
+
"l0_raw": 132.7104,
|
| 141 |
+
"recon_r2": 0.9222,
|
| 142 |
+
"l1_per_token": 23.6716,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 3,
|
| 148 |
+
"dense_rate": 0.000244140625,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 9259,
|
| 152 |
+
"0.6": 7,
|
| 153 |
+
"0.7": 0,
|
| 154 |
+
"0.8": 0,
|
| 155 |
+
"0.9": 0
|
| 156 |
+
},
|
| 157 |
+
"n_dup_clusters_signed": {
|
| 158 |
+
"0.5": 3558,
|
| 159 |
+
"0.6": 3,
|
| 160 |
+
"0.7": 0,
|
| 161 |
+
"0.8": 0,
|
| 162 |
+
"0.9": 0
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 11,
|
| 166 |
+
"0.6": 3,
|
| 167 |
+
"0.7": 1,
|
| 168 |
+
"0.8": 1,
|
| 169 |
+
"0.9": 1
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.5092904567718506,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 9257,
|
| 174 |
+
"0.6": 19,
|
| 175 |
+
"0.7": 0,
|
| 176 |
+
"0.8": 0,
|
| 177 |
+
"0.9": 0
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 3556,
|
| 181 |
+
"0.6": 9,
|
| 182 |
+
"0.7": 0,
|
| 183 |
+
"0.8": 0,
|
| 184 |
+
"0.9": 0
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 11,
|
| 188 |
+
"0.6": 3,
|
| 189 |
+
"0.7": 1,
|
| 190 |
+
"0.8": 1,
|
| 191 |
+
"0.9": 1
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.5100248456001282
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.7802608013153076,
|
| 199 |
+
"dcs_abs": 0.7802907824516296,
|
| 200 |
+
"frac_sign_flipped": 0.0013020833333333333,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"frac_denominator": "d_sae",
|
| 203 |
+
"n_alive_a": 12288,
|
| 204 |
+
"n_alive_b": 12288,
|
| 205 |
+
"matched_cos_signed": {
|
| 206 |
+
"0.5": 0.8014480471611023,
|
| 207 |
+
"0.6": 0.7803129553794861,
|
| 208 |
+
"0.7": 0.7802608013153076,
|
| 209 |
+
"0.8": 0.7802608013153076,
|
| 210 |
+
"0.9": 0.7802608013153076
|
| 211 |
+
},
|
| 212 |
+
"unique_frac_signed": {
|
| 213 |
+
"0.5": 0.5404459635416666,
|
| 214 |
+
"0.6": 0.9998779296875,
|
| 215 |
+
"0.7": 1.0,
|
| 216 |
+
"0.8": 1.0,
|
| 217 |
+
"0.9": 1.0
|
| 218 |
+
},
|
| 219 |
+
"dedup_dcs_signed": {
|
| 220 |
+
"0.5": 0.433139362076569,
|
| 221 |
+
"0.6": 0.7802177023331751,
|
| 222 |
+
"0.7": 0.7802608013153076,
|
| 223 |
+
"0.8": 0.7802608013153076,
|
| 224 |
+
"0.9": 0.7802608013153076
|
| 225 |
+
},
|
| 226 |
+
"matched_frac_signed": {
|
| 227 |
+
"0.5": 0.507568359375,
|
| 228 |
+
"0.6": 0.8978678385416666,
|
| 229 |
+
"0.7": 0.7430013020833334,
|
| 230 |
+
"0.8": 0.4878743489583333,
|
| 231 |
+
"0.9": 0.22517903645833334
|
| 232 |
+
},
|
| 233 |
+
"matched_cos_abs": {
|
| 234 |
+
"0.5": 0.8011406660079956,
|
| 235 |
+
"0.6": 0.7804637551307678,
|
| 236 |
+
"0.7": 0.7802907824516296,
|
| 237 |
+
"0.8": 0.7802907824516296,
|
| 238 |
+
"0.9": 0.7802907824516296
|
| 239 |
+
},
|
| 240 |
+
"unique_frac_abs": {
|
| 241 |
+
"0.5": 0.5404866536458334,
|
| 242 |
+
"0.6": 0.99951171875,
|
| 243 |
+
"0.7": 1.0,
|
| 244 |
+
"0.8": 1.0,
|
| 245 |
+
"0.9": 1.0
|
| 246 |
+
},
|
| 247 |
+
"dedup_dcs_abs": {
|
| 248 |
+
"0.5": 0.4330058376702558,
|
| 249 |
+
"0.6": 0.7800826693128329,
|
| 250 |
+
"0.7": 0.7802907824516296,
|
| 251 |
+
"0.8": 0.7802907824516296,
|
| 252 |
+
"0.9": 0.7802907824516296
|
| 253 |
+
},
|
| 254 |
+
"matched_frac_abs": {
|
| 255 |
+
"0.5": 0.50732421875,
|
| 256 |
+
"0.6": 0.898193359375,
|
| 257 |
+
"0.7": 0.742919921875,
|
| 258 |
+
"0.8": 0.4881998697916667,
|
| 259 |
+
"0.9": 0.22517903645833334
|
| 260 |
+
}
|
| 261 |
+
},
|
| 262 |
+
"seed0\u2194seed2": {
|
| 263 |
+
"dcs_signed": 0.779227077960968,
|
| 264 |
+
"dcs_abs": 0.7792566418647766,
|
| 265 |
+
"frac_sign_flipped": 0.0020345052083333335,
|
| 266 |
+
"alive_source": "stats",
|
| 267 |
+
"frac_denominator": "d_sae",
|
| 268 |
+
"n_alive_a": 12288,
|
| 269 |
+
"n_alive_b": 12288,
|
| 270 |
+
"matched_cos_signed": {
|
| 271 |
+
"0.5": 0.8003079295158386,
|
| 272 |
+
"0.6": 0.7793368697166443,
|
| 273 |
+
"0.7": 0.7792271971702576,
|
| 274 |
+
"0.8": 0.7792271971702576,
|
| 275 |
+
"0.9": 0.7792271971702576
|
| 276 |
+
},
|
| 277 |
+
"unique_frac_signed": {
|
| 278 |
+
"0.5": 0.5398763020833334,
|
| 279 |
+
"0.6": 0.9998372395833334,
|
| 280 |
+
"0.7": 1.0,
|
| 281 |
+
"0.8": 1.0,
|
| 282 |
+
"0.9": 1.0
|
| 283 |
+
},
|
| 284 |
+
"dedup_dcs_signed": {
|
| 285 |
+
"0.5": 0.43206728551497997,
|
| 286 |
+
"0.6": 0.7792100245230056,
|
| 287 |
+
"0.7": 0.7792271971702576,
|
| 288 |
+
"0.8": 0.7792271971702576,
|
| 289 |
+
"0.9": 0.7792271971702576
|
| 290 |
+
},
|
| 291 |
+
"matched_frac_signed": {
|
| 292 |
+
"0.5": 0.5049641927083334,
|
| 293 |
+
"0.6": 0.8954264322916666,
|
| 294 |
+
"0.7": 0.7377115885416666,
|
| 295 |
+
"0.8": 0.4825846354166667,
|
| 296 |
+
"0.9": 0.22574869791666666
|
| 297 |
+
},
|
| 298 |
+
"matched_cos_abs": {
|
| 299 |
+
"0.5": 0.8002008199691772,
|
| 300 |
+
"0.6": 0.7794950604438782,
|
| 301 |
+
"0.7": 0.7792566418647766,
|
| 302 |
+
"0.8": 0.7792566418647766,
|
| 303 |
+
"0.9": 0.7792566418647766
|
| 304 |
+
},
|
| 305 |
+
"unique_frac_abs": {
|
| 306 |
+
"0.5": 0.5397542317708334,
|
| 307 |
+
"0.6": 0.9994710286458334,
|
| 308 |
+
"0.7": 1.0,
|
| 309 |
+
"0.8": 1.0,
|
| 310 |
+
"0.9": 1.0
|
| 311 |
+
},
|
| 312 |
+
"dedup_dcs_abs": {
|
| 313 |
+
"0.5": 0.4319117788448542,
|
| 314 |
+
"0.6": 0.779082729886189,
|
| 315 |
+
"0.7": 0.7792566418647766,
|
| 316 |
+
"0.8": 0.7792566418647766,
|
| 317 |
+
"0.9": 0.7792566418647766
|
| 318 |
+
},
|
| 319 |
+
"matched_frac_abs": {
|
| 320 |
+
"0.5": 0.50537109375,
|
| 321 |
+
"0.6": 0.8944498697916666,
|
| 322 |
+
"0.7": 0.7381998697916666,
|
| 323 |
+
"0.8": 0.4832356770833333,
|
| 324 |
+
"0.9": 0.225830078125
|
| 325 |
+
}
|
| 326 |
+
},
|
| 327 |
+
"seed1\u2194seed2": {
|
| 328 |
+
"dcs_signed": 0.7793281674385071,
|
| 329 |
+
"dcs_abs": 0.7793558239936829,
|
| 330 |
+
"frac_sign_flipped": 0.0013020833333333333,
|
| 331 |
+
"alive_source": "stats",
|
| 332 |
+
"frac_denominator": "d_sae",
|
| 333 |
+
"n_alive_a": 12288,
|
| 334 |
+
"n_alive_b": 12288,
|
| 335 |
+
"matched_cos_signed": {
|
| 336 |
+
"0.5": 0.7971920967102051,
|
| 337 |
+
"0.6": 0.7792015671730042,
|
| 338 |
+
"0.7": 0.7793281674385071,
|
| 339 |
+
"0.8": 0.7793281674385071,
|
| 340 |
+
"0.9": 0.7793281674385071
|
| 341 |
+
},
|
| 342 |
+
"unique_frac_signed": {
|
| 343 |
+
"0.5": 0.53662109375,
|
| 344 |
+
"0.6": 0.9997151692708334,
|
| 345 |
+
"0.7": 1.0,
|
| 346 |
+
"0.8": 1.0,
|
| 347 |
+
"0.9": 1.0
|
| 348 |
+
},
|
| 349 |
+
"dedup_dcs_signed": {
|
| 350 |
+
"0.5": 0.427790094865486,
|
| 351 |
+
"0.6": 0.7789796266224585,
|
| 352 |
+
"0.7": 0.7793281674385071,
|
| 353 |
+
"0.8": 0.7793281674385071,
|
| 354 |
+
"0.9": 0.7793281674385071
|
| 355 |
+
},
|
| 356 |
+
"matched_frac_signed": {
|
| 357 |
+
"0.5": 0.5033365885416666,
|
| 358 |
+
"0.6": 0.8948567708333334,
|
| 359 |
+
"0.7": 0.739990234375,
|
| 360 |
+
"0.8": 0.4881184895833333,
|
| 361 |
+
"0.9": 0.2236328125
|
| 362 |
+
},
|
| 363 |
+
"matched_cos_abs": {
|
| 364 |
+
"0.5": 0.7974773645401001,
|
| 365 |
+
"0.6": 0.7792859077453613,
|
| 366 |
+
"0.7": 0.7793558239936829,
|
| 367 |
+
"0.8": 0.7793558239936829,
|
| 368 |
+
"0.9": 0.7793558239936829
|
| 369 |
+
},
|
| 370 |
+
"unique_frac_abs": {
|
| 371 |
+
"0.5": 0.5367838541666666,
|
| 372 |
+
"0.6": 0.9992268880208334,
|
| 373 |
+
"0.7": 1.0,
|
| 374 |
+
"0.8": 1.0,
|
| 375 |
+
"0.9": 1.0
|
| 376 |
+
},
|
| 377 |
+
"dedup_dcs_abs": {
|
| 378 |
+
"0.5": 0.42807297334851074,
|
| 379 |
+
"0.6": 0.7786834324748877,
|
| 380 |
+
"0.7": 0.7793558239936829,
|
| 381 |
+
"0.8": 0.7793558239936829,
|
| 382 |
+
"0.9": 0.7793558239936829
|
| 383 |
+
},
|
| 384 |
+
"matched_frac_abs": {
|
| 385 |
+
"0.5": 0.5035807291666666,
|
| 386 |
+
"0.6": 0.89404296875,
|
| 387 |
+
"0.7": 0.7398274739583334,
|
| 388 |
+
"0.8": 0.4873046875,
|
| 389 |
+
"0.9": 0.22371419270833334
|
| 390 |
+
}
|
| 391 |
+
}
|
| 392 |
+
},
|
| 393 |
+
"summary": {
|
| 394 |
+
"n_seeds": 3,
|
| 395 |
+
"n_pairs": 3,
|
| 396 |
+
"report_band": [
|
| 397 |
+
0.6,
|
| 398 |
+
0.7,
|
| 399 |
+
0.8,
|
| 400 |
+
0.9
|
| 401 |
+
],
|
| 402 |
+
"n_dead_mean": 0,
|
| 403 |
+
"n_dead_std": 0.0,
|
| 404 |
+
"n_dense_mean": 1,
|
| 405 |
+
"n_dense_std": 1.7320508075688772,
|
| 406 |
+
"l0_mean": 130,
|
| 407 |
+
"l0_std": 2.1283440323406317,
|
| 408 |
+
"recon_r2_mean": 0.9222666666666667,
|
| 409 |
+
"recon_r2_std": 5.773502691895622e-05,
|
| 410 |
+
"l1_per_token_mean": 23.583466666666666,
|
| 411 |
+
"l1_per_token_std": 0.08096056653293303,
|
| 412 |
+
"babel_mu1_signed_mean": 0.5089375178019205,
|
| 413 |
+
"babel_mu1_signed_std": 0.00035438819786938686,
|
| 414 |
+
"n_dup_signed_per_threshold": {
|
| 415 |
+
"0.5": 9226.666666666666,
|
| 416 |
+
"0.6": 4.0,
|
| 417 |
+
"0.7": 0.0,
|
| 418 |
+
"0.8": 0.0,
|
| 419 |
+
"0.9": 0.0
|
| 420 |
+
},
|
| 421 |
+
"n_dup_signed_band": 1.0,
|
| 422 |
+
"n_dup_signed_band_std": 0.9013878188659973,
|
| 423 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 424 |
+
"0.5": 3561.6666666666665,
|
| 425 |
+
"0.6": 1.6666666666666667,
|
| 426 |
+
"0.7": 0.0,
|
| 427 |
+
"0.8": 0.0,
|
| 428 |
+
"0.9": 0.0
|
| 429 |
+
},
|
| 430 |
+
"n_dup_clusters_signed_band": 0.4166666666666667,
|
| 431 |
+
"n_dup_clusters_signed_band_std": 0.3818813079129867,
|
| 432 |
+
"largest_dup_signed_per_threshold": {
|
| 433 |
+
"0.5": 10.333333333333334,
|
| 434 |
+
"0.6": 2.3333333333333335,
|
| 435 |
+
"0.7": 1.0,
|
| 436 |
+
"0.8": 1.0,
|
| 437 |
+
"0.9": 1.0
|
| 438 |
+
},
|
| 439 |
+
"largest_dup_signed_band": 1.3333333333333335,
|
| 440 |
+
"largest_dup_signed_band_std": 0.2886751345948129,
|
| 441 |
+
"dcs_signed_mean": 0.7796053489049276,
|
| 442 |
+
"dcs_signed_std": 0.0005698843455667465,
|
| 443 |
+
"dedup_dcs_signed_per_threshold": {
|
| 444 |
+
"0.5": 0.430998914152345,
|
| 445 |
+
"0.6": 0.7794691178262131,
|
| 446 |
+
"0.7": 0.7796053886413574,
|
| 447 |
+
"0.8": 0.7796053886413574,
|
| 448 |
+
"0.9": 0.7796053886413574
|
| 449 |
+
},
|
| 450 |
+
"dedup_dcs_signed_band": 0.7795713209375713,
|
| 451 |
+
"dedup_dcs_signed_band_std": 0.0005878462039795222,
|
| 452 |
+
"matched_cos_signed_per_threshold": {
|
| 453 |
+
"0.5": 0.7996493577957153,
|
| 454 |
+
"0.6": 0.7796171307563782,
|
| 455 |
+
"0.7": 0.7796053886413574,
|
| 456 |
+
"0.8": 0.7796053886413574,
|
| 457 |
+
"0.9": 0.7796053886413574
|
| 458 |
+
},
|
| 459 |
+
"matched_cos_signed_band": 0.7796083241701126,
|
| 460 |
+
"matched_cos_signed_band_std": 0.0005767341391707553,
|
| 461 |
+
"unique_frac_signed_per_threshold": {
|
| 462 |
+
"0.5": 0.5389811197916666,
|
| 463 |
+
"0.6": 0.9998101128472223,
|
| 464 |
+
"0.7": 1.0,
|
| 465 |
+
"0.8": 1.0,
|
| 466 |
+
"0.9": 1.0
|
| 467 |
+
},
|
| 468 |
+
"unique_frac_signed_band": 0.9999525282118056,
|
| 469 |
+
"unique_frac_signed_band_std": 2.117580158960649e-05,
|
| 470 |
+
"matched_frac_signed_per_threshold": {
|
| 471 |
+
"0.5": 0.5052897135416666,
|
| 472 |
+
"0.6": 0.8960503472222222,
|
| 473 |
+
"0.7": 0.740234375,
|
| 474 |
+
"0.8": 0.4861924913194444,
|
| 475 |
+
"0.9": 0.224853515625
|
| 476 |
+
},
|
| 477 |
+
"matched_frac_signed_band": 0.5868326822916667,
|
| 478 |
+
"matched_frac_signed_band_std": 0.0015644538107881742,
|
| 479 |
+
"babel_mu1_abs_mean": 0.5095154444376627,
|
| 480 |
+
"babel_mu1_abs_std": 0.0004782988093176601,
|
| 481 |
+
"n_dup_abs_per_threshold": {
|
| 482 |
+
"0.5": 9226.0,
|
| 483 |
+
"0.6": 14.0,
|
| 484 |
+
"0.7": 0.0,
|
| 485 |
+
"0.8": 0.0,
|
| 486 |
+
"0.9": 0.0
|
| 487 |
+
},
|
| 488 |
+
"n_dup_abs_band": 3.5,
|
| 489 |
+
"n_dup_abs_band_std": 1.75,
|
| 490 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 491 |
+
"0.5": 3561.3333333333335,
|
| 492 |
+
"0.6": 6.666666666666667,
|
| 493 |
+
"0.7": 0.0,
|
| 494 |
+
"0.8": 0.0,
|
| 495 |
+
"0.9": 0.0
|
| 496 |
+
},
|
| 497 |
+
"n_dup_clusters_abs_band": 1.6666666666666667,
|
| 498 |
+
"n_dup_clusters_abs_band_std": 0.8036375634160795,
|
| 499 |
+
"largest_dup_abs_per_threshold": {
|
| 500 |
+
"0.5": 10.333333333333334,
|
| 501 |
+
"0.6": 2.6666666666666665,
|
| 502 |
+
"0.7": 1.0,
|
| 503 |
+
"0.8": 1.0,
|
| 504 |
+
"0.9": 1.0
|
| 505 |
+
},
|
| 506 |
+
"largest_dup_abs_band": 1.4166666666666665,
|
| 507 |
+
"largest_dup_abs_band_std": 0.14433756729740646,
|
| 508 |
+
"dcs_abs_mean": 0.779634416103363,
|
| 509 |
+
"dcs_abs_std": 0.0005705890474117245,
|
| 510 |
+
"dedup_dcs_abs_per_threshold": {
|
| 511 |
+
"0.5": 0.43099686328787357,
|
| 512 |
+
"0.6": 0.7792829438913031,
|
| 513 |
+
"0.7": 0.779634416103363,
|
| 514 |
+
"0.8": 0.779634416103363,
|
| 515 |
+
"0.9": 0.779634416103363
|
| 516 |
+
},
|
| 517 |
+
"dedup_dcs_abs_band": 0.7795465480503481,
|
| 518 |
+
"dedup_dcs_abs_band_std": 0.0005996029942678243,
|
| 519 |
+
"matched_cos_abs_per_threshold": {
|
| 520 |
+
"0.5": 0.7996062835057577,
|
| 521 |
+
"0.6": 0.7797482411066691,
|
| 522 |
+
"0.7": 0.779634416103363,
|
| 523 |
+
"0.8": 0.779634416103363,
|
| 524 |
+
"0.9": 0.779634416103363
|
| 525 |
+
},
|
| 526 |
+
"matched_cos_abs_band": 0.7796628723541895,
|
| 527 |
+
"matched_cos_abs_band_std": 0.0005813407917426763,
|
| 528 |
+
"unique_frac_abs_per_threshold": {
|
| 529 |
+
"0.5": 0.5390082465277778,
|
| 530 |
+
"0.6": 0.9994032118055557,
|
| 531 |
+
"0.7": 1.0,
|
| 532 |
+
"0.8": 1.0,
|
| 533 |
+
"0.9": 1.0
|
| 534 |
+
},
|
| 535 |
+
"unique_frac_abs_band": 0.999850802951389,
|
| 536 |
+
"unique_frac_abs_band_std": 3.8512562023914604e-05,
|
| 537 |
+
"matched_frac_abs_per_threshold": {
|
| 538 |
+
"0.5": 0.5054253472222222,
|
| 539 |
+
"0.6": 0.8955620659722222,
|
| 540 |
+
"0.7": 0.7403157552083334,
|
| 541 |
+
"0.8": 0.4862467447916667,
|
| 542 |
+
"0.9": 0.22490776909722224
|
| 543 |
+
},
|
| 544 |
+
"matched_frac_abs_band": 0.5867580837673612,
|
| 545 |
+
"matched_frac_abs_band_std": 0.0016631172717973322,
|
| 546 |
+
"alive_source": "stats",
|
| 547 |
+
"frac_denominator": "d_sae",
|
| 548 |
+
"d_sae": 12288
|
| 549 |
+
}
|
| 550 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_32/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 591000,
|
| 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": 34.894641285956006,
|
| 20 |
+
"recon_r2": 0.9237343207971792,
|
| 21 |
+
"l1_per_token": 23.55283047659831
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 35.89668358714044,
|
| 25 |
+
"recon_r2": 0.9238369012458022,
|
| 26 |
+
"l1_per_token": 23.89672657904016
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 35.78377326565144,
|
| 30 |
+
"recon_r2": 0.9239658248707565,
|
| 31 |
+
"l1_per_token": 23.959925247312796
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.5567062745506811,
|
| 38 |
+
"mcc_abs": 0.5567185464138801,
|
| 39 |
+
"act_iou": 0.3247426022459192,
|
| 40 |
+
"act_iou_n_pairs": 12288,
|
| 41 |
+
"input_r2": 0.9906743851959309,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.778489662241864,
|
| 44 |
+
"0.7": 0.8225123972905186,
|
| 45 |
+
"0.8": 0.8732037535280439,
|
| 46 |
+
"0.9": 0.9275916783516185
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.7787878304135807,
|
| 50 |
+
"0.7": 0.8221566389722251,
|
| 51 |
+
"0.8": 0.8730938988052966,
|
| 52 |
+
"0.9": 0.9275916783516182
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.5544664921628762,
|
| 58 |
+
"mcc_abs": 0.5544787361047171,
|
| 59 |
+
"act_iou": 0.3233801592734783,
|
| 60 |
+
"act_iou_n_pairs": 12288,
|
| 61 |
+
"input_r2": 0.9905543783110445,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.7778879330074708,
|
| 64 |
+
"0.7": 0.8217823593727598,
|
| 65 |
+
"0.8": 0.8744645519569243,
|
| 66 |
+
"0.9": 0.9304022028404454
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.77806854701958,
|
| 70 |
+
"0.7": 0.8216213914759847,
|
| 71 |
+
"0.8": 0.8743387646207568,
|
| 72 |
+
"0.9": 0.9303023086328086
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.5551743134365573,
|
| 78 |
+
"mcc_abs": 0.5551847888872837,
|
| 79 |
+
"act_iou": 0.3235662055516379,
|
| 80 |
+
"act_iou_n_pairs": 12288,
|
| 81 |
+
"input_r2": 0.9906795412121475,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.7799041044015064,
|
| 84 |
+
"0.7": 0.8240849755846157,
|
| 85 |
+
"0.8": 0.875262053128961,
|
| 86 |
+
"0.9": 0.928516927519262
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.7797712208670907,
|
| 90 |
+
"0.7": 0.8240847539589291,
|
| 91 |
+
"0.8": 0.8753568254740804,
|
| 92 |
+
"0.9": 0.9285860937856002
|
| 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": 35.52503271291596,
|
| 106 |
+
"l0_std": 0.5488462430983233,
|
| 107 |
+
"recon_r2_mean": 0.9238456823045794,
|
| 108 |
+
"recon_r2_std": 0.00011600157009996355,
|
| 109 |
+
"l1_per_token_mean": 23.80316076765042,
|
| 110 |
+
"l1_per_token_std": 0.21908322364237542,
|
| 111 |
+
"input_r2_mean": 0.9906361015730409,
|
| 112 |
+
"input_r2_std": 7.082135828484657e-05,
|
| 113 |
+
"mcc_signed_mean": 0.5554490267167048,
|
| 114 |
+
"mcc_signed_std": 0.0011448828874156942,
|
| 115 |
+
"mcc_abs_mean": 0.5554606904686269,
|
| 116 |
+
"mcc_abs_std": 0.0011451108318407293,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.7787605665502805,
|
| 119 |
+
"0.7": 0.8227932440826313,
|
| 120 |
+
"0.8": 0.8743101195379763,
|
| 121 |
+
"0.9": 0.9288369362371086
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.8511752166019992,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.7788758661000837,
|
| 126 |
+
"0.7": 0.822620928135713,
|
| 127 |
+
"0.8": 0.8742631629667112,
|
| 128 |
+
"0.9": 0.928826693590009
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.8511466626981292,
|
| 131 |
+
"act_iou_mean": 0.32389632235701177,
|
| 132 |
+
"act_iou_std": 0.00073877976568198,
|
| 133 |
+
"act_iou_n_pairs": 12288.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_32/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,550 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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": 34,
|
| 22 |
+
"l0_raw": 34.4064,
|
| 23 |
+
"recon_r2": 0.9239,
|
| 24 |
+
"l1_per_token": 23.5798,
|
| 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": 2759,
|
| 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": 1242,
|
| 41 |
+
"0.6": 0,
|
| 42 |
+
"0.7": 0,
|
| 43 |
+
"0.8": 0,
|
| 44 |
+
"0.9": 0
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 6,
|
| 48 |
+
"0.6": 1,
|
| 49 |
+
"0.7": 1,
|
| 50 |
+
"0.8": 1,
|
| 51 |
+
"0.9": 1
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.3883293867111206,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 2761,
|
| 56 |
+
"0.6": 12,
|
| 57 |
+
"0.7": 2,
|
| 58 |
+
"0.8": 0,
|
| 59 |
+
"0.9": 0
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 1243,
|
| 63 |
+
"0.6": 6,
|
| 64 |
+
"0.7": 1,
|
| 65 |
+
"0.8": 0,
|
| 66 |
+
"0.9": 0
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 6,
|
| 70 |
+
"0.6": 2,
|
| 71 |
+
"0.7": 2,
|
| 72 |
+
"0.8": 1,
|
| 73 |
+
"0.9": 1
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.389350026845932
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 12288,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 36,
|
| 81 |
+
"l0_raw": 35.6352,
|
| 82 |
+
"recon_r2": 0.924,
|
| 83 |
+
"l1_per_token": 23.9215,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 4,
|
| 89 |
+
"dense_rate": 0.0003255208333333333,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 2682,
|
| 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": 1207,
|
| 100 |
+
"0.6": 0,
|
| 101 |
+
"0.7": 0,
|
| 102 |
+
"0.8": 0,
|
| 103 |
+
"0.9": 0
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 6,
|
| 107 |
+
"0.6": 1,
|
| 108 |
+
"0.7": 1,
|
| 109 |
+
"0.8": 1,
|
| 110 |
+
"0.9": 1
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.38871124386787415,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 2688,
|
| 115 |
+
"0.6": 6,
|
| 116 |
+
"0.7": 4,
|
| 117 |
+
"0.8": 0,
|
| 118 |
+
"0.9": 0
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 1211,
|
| 122 |
+
"0.6": 3,
|
| 123 |
+
"0.7": 2,
|
| 124 |
+
"0.8": 0,
|
| 125 |
+
"0.9": 0
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 6,
|
| 129 |
+
"0.6": 2,
|
| 130 |
+
"0.7": 2,
|
| 131 |
+
"0.8": 1,
|
| 132 |
+
"0.9": 1
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.38972291350364685
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 12288,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 36,
|
| 140 |
+
"l0_raw": 35.6352,
|
| 141 |
+
"recon_r2": 0.9241,
|
| 142 |
+
"l1_per_token": 23.9877,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 5,
|
| 148 |
+
"dense_rate": 0.0004069010416666667,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 2596,
|
| 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": 1167,
|
| 159 |
+
"0.6": 0,
|
| 160 |
+
"0.7": 0,
|
| 161 |
+
"0.8": 0,
|
| 162 |
+
"0.9": 0
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 7,
|
| 166 |
+
"0.6": 1,
|
| 167 |
+
"0.7": 1,
|
| 168 |
+
"0.8": 1,
|
| 169 |
+
"0.9": 1
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.3845524787902832,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 2605,
|
| 174 |
+
"0.6": 12,
|
| 175 |
+
"0.7": 2,
|
| 176 |
+
"0.8": 0,
|
| 177 |
+
"0.9": 0
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 1170,
|
| 181 |
+
"0.6": 6,
|
| 182 |
+
"0.7": 1,
|
| 183 |
+
"0.8": 0,
|
| 184 |
+
"0.9": 0
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 7,
|
| 188 |
+
"0.6": 2,
|
| 189 |
+
"0.7": 2,
|
| 190 |
+
"0.8": 1,
|
| 191 |
+
"0.9": 1
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.3856949806213379
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.6458575129508972,
|
| 199 |
+
"dcs_abs": 0.6464142799377441,
|
| 200 |
+
"frac_sign_flipped": 0.041178385416666664,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"frac_denominator": "d_sae",
|
| 203 |
+
"n_alive_a": 12288,
|
| 204 |
+
"n_alive_b": 12288,
|
| 205 |
+
"matched_cos_signed": {
|
| 206 |
+
"0.5": 0.6300660371780396,
|
| 207 |
+
"0.6": 0.6458575129508972,
|
| 208 |
+
"0.7": 0.6458575129508972,
|
| 209 |
+
"0.8": 0.6458575129508972,
|
| 210 |
+
"0.9": 0.6458575129508972
|
| 211 |
+
},
|
| 212 |
+
"unique_frac_signed": {
|
| 213 |
+
"0.5": 0.8782552083333334,
|
| 214 |
+
"0.6": 1.0,
|
| 215 |
+
"0.7": 1.0,
|
| 216 |
+
"0.8": 1.0,
|
| 217 |
+
"0.9": 1.0
|
| 218 |
+
},
|
| 219 |
+
"dedup_dcs_signed": {
|
| 220 |
+
"0.5": 0.553358778745557,
|
| 221 |
+
"0.6": 0.6458575129508972,
|
| 222 |
+
"0.7": 0.6458575129508972,
|
| 223 |
+
"0.8": 0.6458575129508972,
|
| 224 |
+
"0.9": 0.6458575129508972
|
| 225 |
+
},
|
| 226 |
+
"matched_frac_signed": {
|
| 227 |
+
"0.5": 0.592041015625,
|
| 228 |
+
"0.6": 0.6446940104166666,
|
| 229 |
+
"0.7": 0.5508626302083334,
|
| 230 |
+
"0.8": 0.4207356770833333,
|
| 231 |
+
"0.9": 0.2589518229166667
|
| 232 |
+
},
|
| 233 |
+
"matched_cos_abs": {
|
| 234 |
+
"0.5": 0.6306042671203613,
|
| 235 |
+
"0.6": 0.6464684009552002,
|
| 236 |
+
"0.7": 0.646440327167511,
|
| 237 |
+
"0.8": 0.6464142799377441,
|
| 238 |
+
"0.9": 0.6464142799377441
|
| 239 |
+
},
|
| 240 |
+
"unique_frac_abs": {
|
| 241 |
+
"0.5": 0.8781331380208334,
|
| 242 |
+
"0.6": 0.9996337890625,
|
| 243 |
+
"0.7": 0.9998779296875,
|
| 244 |
+
"0.8": 1.0,
|
| 245 |
+
"0.9": 1.0
|
| 246 |
+
},
|
| 247 |
+
"dedup_dcs_abs": {
|
| 248 |
+
"0.5": 0.5537545039357308,
|
| 249 |
+
"0.6": 0.6462316571560223,
|
| 250 |
+
"0.7": 0.646361415994761,
|
| 251 |
+
"0.8": 0.6464142799377441,
|
| 252 |
+
"0.9": 0.6464142799377441
|
| 253 |
+
},
|
| 254 |
+
"matched_frac_abs": {
|
| 255 |
+
"0.5": 0.5918782552083334,
|
| 256 |
+
"0.6": 0.6444498697916666,
|
| 257 |
+
"0.7": 0.551513671875,
|
| 258 |
+
"0.8": 0.4205729166666667,
|
| 259 |
+
"0.9": 0.2589518229166667
|
| 260 |
+
}
|
| 261 |
+
},
|
| 262 |
+
"seed0\u2194seed2": {
|
| 263 |
+
"dcs_signed": 0.6431646943092346,
|
| 264 |
+
"dcs_abs": 0.6437967419624329,
|
| 265 |
+
"frac_sign_flipped": 0.045491536458333336,
|
| 266 |
+
"alive_source": "stats",
|
| 267 |
+
"frac_denominator": "d_sae",
|
| 268 |
+
"n_alive_a": 12288,
|
| 269 |
+
"n_alive_b": 12288,
|
| 270 |
+
"matched_cos_signed": {
|
| 271 |
+
"0.5": 0.627785325050354,
|
| 272 |
+
"0.6": 0.6431646943092346,
|
| 273 |
+
"0.7": 0.6431646943092346,
|
| 274 |
+
"0.8": 0.6431646943092346,
|
| 275 |
+
"0.9": 0.6431646943092346
|
| 276 |
+
},
|
| 277 |
+
"unique_frac_signed": {
|
| 278 |
+
"0.5": 0.880126953125,
|
| 279 |
+
"0.6": 1.0,
|
| 280 |
+
"0.7": 1.0,
|
| 281 |
+
"0.8": 1.0,
|
| 282 |
+
"0.9": 1.0
|
| 283 |
+
},
|
| 284 |
+
"dedup_dcs_signed": {
|
| 285 |
+
"0.5": 0.5525307853531558,
|
| 286 |
+
"0.6": 0.6431646943092346,
|
| 287 |
+
"0.7": 0.6431646943092346,
|
| 288 |
+
"0.8": 0.6431646943092346,
|
| 289 |
+
"0.9": 0.6431646943092346
|
| 290 |
+
},
|
| 291 |
+
"matched_frac_signed": {
|
| 292 |
+
"0.5": 0.592529296875,
|
| 293 |
+
"0.6": 0.6419270833333334,
|
| 294 |
+
"0.7": 0.546142578125,
|
| 295 |
+
"0.8": 0.4164225260416667,
|
| 296 |
+
"0.9": 0.2574055989583333
|
| 297 |
+
},
|
| 298 |
+
"matched_cos_abs": {
|
| 299 |
+
"0.5": 0.6283566951751709,
|
| 300 |
+
"0.6": 0.6437250971794128,
|
| 301 |
+
"0.7": 0.6437689065933228,
|
| 302 |
+
"0.8": 0.6437967419624329,
|
| 303 |
+
"0.9": 0.6437967419624329
|
| 304 |
+
},
|
| 305 |
+
"unique_frac_abs": {
|
| 306 |
+
"0.5": 0.8798421223958334,
|
| 307 |
+
"0.6": 0.99951171875,
|
| 308 |
+
"0.7": 0.9999186197916666,
|
| 309 |
+
"0.8": 1.0,
|
| 310 |
+
"0.9": 1.0
|
| 311 |
+
},
|
| 312 |
+
"dedup_dcs_abs": {
|
| 313 |
+
"0.5": 0.5528546883045541,
|
| 314 |
+
"0.6": 0.6434107782843057,
|
| 315 |
+
"0.7": 0.6437165165455856,
|
| 316 |
+
"0.8": 0.6437967419624329,
|
| 317 |
+
"0.9": 0.6437967419624329
|
| 318 |
+
},
|
| 319 |
+
"matched_frac_abs": {
|
| 320 |
+
"0.5": 0.5918782552083334,
|
| 321 |
+
"0.6": 0.64111328125,
|
| 322 |
+
"0.7": 0.5464680989583334,
|
| 323 |
+
"0.8": 0.4170735677083333,
|
| 324 |
+
"0.9": 0.257568359375
|
| 325 |
+
}
|
| 326 |
+
},
|
| 327 |
+
"seed1\u2194seed2": {
|
| 328 |
+
"dcs_signed": 0.6429621577262878,
|
| 329 |
+
"dcs_abs": 0.6435258388519287,
|
| 330 |
+
"frac_sign_flipped": 0.044677734375,
|
| 331 |
+
"alive_source": "stats",
|
| 332 |
+
"frac_denominator": "d_sae",
|
| 333 |
+
"n_alive_a": 12288,
|
| 334 |
+
"n_alive_b": 12288,
|
| 335 |
+
"matched_cos_signed": {
|
| 336 |
+
"0.5": 0.6270092725753784,
|
| 337 |
+
"0.6": 0.6429621577262878,
|
| 338 |
+
"0.7": 0.6429621577262878,
|
| 339 |
+
"0.8": 0.6429621577262878,
|
| 340 |
+
"0.9": 0.6429621577262878
|
| 341 |
+
},
|
| 342 |
+
"unique_frac_signed": {
|
| 343 |
+
"0.5": 0.8818359375,
|
| 344 |
+
"0.6": 1.0,
|
| 345 |
+
"0.7": 1.0,
|
| 346 |
+
"0.8": 1.0,
|
| 347 |
+
"0.9": 1.0
|
| 348 |
+
},
|
| 349 |
+
"dedup_dcs_signed": {
|
| 350 |
+
"0.5": 0.5529193097027019,
|
| 351 |
+
"0.6": 0.6429621577262878,
|
| 352 |
+
"0.7": 0.6429621577262878,
|
| 353 |
+
"0.8": 0.6429621577262878,
|
| 354 |
+
"0.9": 0.6429621577262878
|
| 355 |
+
},
|
| 356 |
+
"matched_frac_signed": {
|
| 357 |
+
"0.5": 0.591796875,
|
| 358 |
+
"0.6": 0.6395670572916666,
|
| 359 |
+
"0.7": 0.5450846354166666,
|
| 360 |
+
"0.8": 0.4197591145833333,
|
| 361 |
+
"0.9": 0.2607421875
|
| 362 |
+
},
|
| 363 |
+
"matched_cos_abs": {
|
| 364 |
+
"0.5": 0.6275225877761841,
|
| 365 |
+
"0.6": 0.6435590386390686,
|
| 366 |
+
"0.7": 0.6435487866401672,
|
| 367 |
+
"0.8": 0.6435258388519287,
|
| 368 |
+
"0.9": 0.6435258388519287
|
| 369 |
+
},
|
| 370 |
+
"unique_frac_abs": {
|
| 371 |
+
"0.5": 0.8815104166666666,
|
| 372 |
+
"0.6": 0.9996337890625,
|
| 373 |
+
"0.7": 0.9998779296875,
|
| 374 |
+
"0.8": 1.0,
|
| 375 |
+
"0.9": 1.0
|
| 376 |
+
},
|
| 377 |
+
"dedup_dcs_abs": {
|
| 378 |
+
"0.5": 0.5531676978183289,
|
| 379 |
+
"0.6": 0.643323360280192,
|
| 380 |
+
"0.7": 0.6434702284386731,
|
| 381 |
+
"0.8": 0.6435258388519287,
|
| 382 |
+
"0.9": 0.6435258388519287
|
| 383 |
+
},
|
| 384 |
+
"matched_frac_abs": {
|
| 385 |
+
"0.5": 0.5916341145833334,
|
| 386 |
+
"0.6": 0.639404296875,
|
| 387 |
+
"0.7": 0.544677734375,
|
| 388 |
+
"0.8": 0.41943359375,
|
| 389 |
+
"0.9": 0.2607421875
|
| 390 |
+
}
|
| 391 |
+
}
|
| 392 |
+
},
|
| 393 |
+
"summary": {
|
| 394 |
+
"n_seeds": 3,
|
| 395 |
+
"n_pairs": 3,
|
| 396 |
+
"report_band": [
|
| 397 |
+
0.6,
|
| 398 |
+
0.7,
|
| 399 |
+
0.8,
|
| 400 |
+
0.9
|
| 401 |
+
],
|
| 402 |
+
"n_dead_mean": 0,
|
| 403 |
+
"n_dead_std": 0.0,
|
| 404 |
+
"n_dense_mean": 5,
|
| 405 |
+
"n_dense_std": 1.0,
|
| 406 |
+
"l0_mean": 35,
|
| 407 |
+
"l0_std": 0.7094480107802119,
|
| 408 |
+
"recon_r2_mean": 0.924,
|
| 409 |
+
"recon_r2_std": 9.999999999998899e-05,
|
| 410 |
+
"l1_per_token_mean": 23.829666666666668,
|
| 411 |
+
"l1_per_token_std": 0.21890779641971148,
|
| 412 |
+
"babel_mu1_signed_mean": 0.38719770312309265,
|
| 413 |
+
"babel_mu1_signed_std": 0.002298774140755539,
|
| 414 |
+
"n_dup_signed_per_threshold": {
|
| 415 |
+
"0.5": 2679.0,
|
| 416 |
+
"0.6": 0.0,
|
| 417 |
+
"0.7": 0.0,
|
| 418 |
+
"0.8": 0.0,
|
| 419 |
+
"0.9": 0.0
|
| 420 |
+
},
|
| 421 |
+
"n_dup_signed_band": 0.0,
|
| 422 |
+
"n_dup_signed_band_std": 0.0,
|
| 423 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 424 |
+
"0.5": 1205.3333333333333,
|
| 425 |
+
"0.6": 0.0,
|
| 426 |
+
"0.7": 0.0,
|
| 427 |
+
"0.8": 0.0,
|
| 428 |
+
"0.9": 0.0
|
| 429 |
+
},
|
| 430 |
+
"n_dup_clusters_signed_band": 0.0,
|
| 431 |
+
"n_dup_clusters_signed_band_std": 0.0,
|
| 432 |
+
"largest_dup_signed_per_threshold": {
|
| 433 |
+
"0.5": 6.333333333333333,
|
| 434 |
+
"0.6": 1.0,
|
| 435 |
+
"0.7": 1.0,
|
| 436 |
+
"0.8": 1.0,
|
| 437 |
+
"0.9": 1.0
|
| 438 |
+
},
|
| 439 |
+
"largest_dup_signed_band": 1.0,
|
| 440 |
+
"largest_dup_signed_band_std": 0.0,
|
| 441 |
+
"dcs_signed_mean": 0.6439947883288065,
|
| 442 |
+
"dcs_signed_std": 0.001616342330747118,
|
| 443 |
+
"dedup_dcs_signed_per_threshold": {
|
| 444 |
+
"0.5": 0.5529362912671382,
|
| 445 |
+
"0.6": 0.6439947883288065,
|
| 446 |
+
"0.7": 0.6439947883288065,
|
| 447 |
+
"0.8": 0.6439947883288065,
|
| 448 |
+
"0.9": 0.6439947883288065
|
| 449 |
+
},
|
| 450 |
+
"dedup_dcs_signed_band": 0.6439947883288065,
|
| 451 |
+
"dedup_dcs_signed_band_std": 0.001616342330747118,
|
| 452 |
+
"matched_cos_signed_per_threshold": {
|
| 453 |
+
"0.5": 0.628286878267924,
|
| 454 |
+
"0.6": 0.6439947883288065,
|
| 455 |
+
"0.7": 0.6439947883288065,
|
| 456 |
+
"0.8": 0.6439947883288065,
|
| 457 |
+
"0.9": 0.6439947883288065
|
| 458 |
+
},
|
| 459 |
+
"matched_cos_signed_band": 0.6439947883288065,
|
| 460 |
+
"matched_cos_signed_band_std": 0.001616342330747118,
|
| 461 |
+
"unique_frac_signed_per_threshold": {
|
| 462 |
+
"0.5": 0.8800726996527778,
|
| 463 |
+
"0.6": 1.0,
|
| 464 |
+
"0.7": 1.0,
|
| 465 |
+
"0.8": 1.0,
|
| 466 |
+
"0.9": 1.0
|
| 467 |
+
},
|
| 468 |
+
"unique_frac_signed_band": 1.0,
|
| 469 |
+
"unique_frac_signed_band_std": 0.0,
|
| 470 |
+
"matched_frac_signed_per_threshold": {
|
| 471 |
+
"0.5": 0.5921223958333334,
|
| 472 |
+
"0.6": 0.6420627170138888,
|
| 473 |
+
"0.7": 0.54736328125,
|
| 474 |
+
"0.8": 0.4189724392361111,
|
| 475 |
+
"0.9": 0.259033203125
|
| 476 |
+
},
|
| 477 |
+
"matched_frac_signed_band": 0.46685791015625,
|
| 478 |
+
"matched_frac_signed_band_std": 0.0017397101502630082,
|
| 479 |
+
"babel_mu1_abs_mean": 0.38825597365697223,
|
| 480 |
+
"babel_mu1_abs_std": 0.0022257077776072047,
|
| 481 |
+
"n_dup_abs_per_threshold": {
|
| 482 |
+
"0.5": 2684.6666666666665,
|
| 483 |
+
"0.6": 10.0,
|
| 484 |
+
"0.7": 2.6666666666666665,
|
| 485 |
+
"0.8": 0.0,
|
| 486 |
+
"0.9": 0.0
|
| 487 |
+
},
|
| 488 |
+
"n_dup_abs_band": 3.1666666666666665,
|
| 489 |
+
"n_dup_abs_band_std": 0.5773502691896258,
|
| 490 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 491 |
+
"0.5": 1208.0,
|
| 492 |
+
"0.6": 5.0,
|
| 493 |
+
"0.7": 1.3333333333333333,
|
| 494 |
+
"0.8": 0.0,
|
| 495 |
+
"0.9": 0.0
|
| 496 |
+
},
|
| 497 |
+
"n_dup_clusters_abs_band": 1.5833333333333333,
|
| 498 |
+
"n_dup_clusters_abs_band_std": 0.2886751345948129,
|
| 499 |
+
"largest_dup_abs_per_threshold": {
|
| 500 |
+
"0.5": 6.333333333333333,
|
| 501 |
+
"0.6": 2.0,
|
| 502 |
+
"0.7": 2.0,
|
| 503 |
+
"0.8": 1.0,
|
| 504 |
+
"0.9": 1.0
|
| 505 |
+
},
|
| 506 |
+
"largest_dup_abs_band": 1.5,
|
| 507 |
+
"largest_dup_abs_band_std": 0.0,
|
| 508 |
+
"dcs_abs_mean": 0.6445789535840353,
|
| 509 |
+
"dcs_abs_std": 0.0015952003768442441,
|
| 510 |
+
"dedup_dcs_abs_per_threshold": {
|
| 511 |
+
"0.5": 0.5532589633528712,
|
| 512 |
+
"0.6": 0.64432193190684,
|
| 513 |
+
"0.7": 0.6445160536596732,
|
| 514 |
+
"0.8": 0.6445789535840353,
|
| 515 |
+
"0.9": 0.6445789535840353
|
| 516 |
+
},
|
| 517 |
+
"dedup_dcs_abs_band": 0.644498973183646,
|
| 518 |
+
"dedup_dcs_abs_band_std": 0.0016114404391061239,
|
| 519 |
+
"matched_cos_abs_per_threshold": {
|
| 520 |
+
"0.5": 0.6288278500239054,
|
| 521 |
+
"0.6": 0.6445841789245605,
|
| 522 |
+
"0.7": 0.6445860068003336,
|
| 523 |
+
"0.8": 0.6445789535840353,
|
| 524 |
+
"0.9": 0.6445789535840353
|
| 525 |
+
},
|
| 526 |
+
"matched_cos_abs_band": 0.6445820232232412,
|
| 527 |
+
"matched_cos_abs_band_std": 0.0016083263425873502,
|
| 528 |
+
"unique_frac_abs_per_threshold": {
|
| 529 |
+
"0.5": 0.8798285590277778,
|
| 530 |
+
"0.6": 0.9995930989583334,
|
| 531 |
+
"0.7": 0.9998914930555555,
|
| 532 |
+
"0.8": 1.0,
|
| 533 |
+
"0.9": 1.0
|
| 534 |
+
},
|
| 535 |
+
"unique_frac_abs_band": 0.9998711480034722,
|
| 536 |
+
"unique_frac_abs_band_std": 1.1746221297010822e-05,
|
| 537 |
+
"matched_frac_abs_per_threshold": {
|
| 538 |
+
"0.5": 0.591796875,
|
| 539 |
+
"0.6": 0.6416558159722222,
|
| 540 |
+
"0.7": 0.5475531684027778,
|
| 541 |
+
"0.8": 0.4190266927083333,
|
| 542 |
+
"0.9": 0.25908745659722227
|
| 543 |
+
},
|
| 544 |
+
"matched_frac_abs_band": 0.4668307834201389,
|
| 545 |
+
"matched_frac_abs_band_std": 0.00178600512638869,
|
| 546 |
+
"alive_source": "stats",
|
| 547 |
+
"frac_denominator": "d_sae",
|
| 548 |
+
"d_sae": 12288
|
| 549 |
+
}
|
| 550 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_32/seed1/config.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "fista",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 1,
|
| 8 |
+
"fista_t_iters": 50,
|
| 9 |
+
"fista_step": 0.0,
|
| 10 |
+
"fista_lam": 1.0,
|
| 11 |
+
"fista_nonneg": true,
|
| 12 |
+
"fista_power_iters": 32,
|
| 13 |
+
"fista_r2_tol": 0.0001,
|
| 14 |
+
"fista_check_every": 5,
|
| 15 |
+
"fista_use_bf16": false,
|
| 16 |
+
"fista_step_cache_every": 0,
|
| 17 |
+
"fista_compile": false,
|
| 18 |
+
"fista_topk_k": 0,
|
| 19 |
+
"refit": true,
|
| 20 |
+
"refit_iters": 50,
|
| 21 |
+
"refit_thresh": 1e-08,
|
| 22 |
+
"class_type": "FistaSAE",
|
| 23 |
+
"module_name": "consdl.models.fista"
|
| 24 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_32/seed1/lock.lock
ADDED
|
File without changes
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_32/seed2/config.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "fista",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 2,
|
| 8 |
+
"fista_t_iters": 50,
|
| 9 |
+
"fista_step": 0.0,
|
| 10 |
+
"fista_lam": 1.0,
|
| 11 |
+
"fista_nonneg": true,
|
| 12 |
+
"fista_power_iters": 32,
|
| 13 |
+
"fista_r2_tol": 0.0001,
|
| 14 |
+
"fista_check_every": 5,
|
| 15 |
+
"fista_use_bf16": false,
|
| 16 |
+
"fista_step_cache_every": 0,
|
| 17 |
+
"fista_compile": false,
|
| 18 |
+
"fista_topk_k": 0,
|
| 19 |
+
"refit": true,
|
| 20 |
+
"refit_iters": 50,
|
| 21 |
+
"refit_thresh": 1e-08,
|
| 22 |
+
"class_type": "FistaSAE",
|
| 23 |
+
"module_name": "consdl.models.fista"
|
| 24 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_32/seed2/lock.lock
ADDED
|
File without changes
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/code_metrics_step_final.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": 1,
|
| 3 |
+
"step": "final",
|
| 4 |
+
"n_tokens": 591000,
|
| 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": 62.45392893401015,
|
| 20 |
+
"recon_r2": 0.8902434129997655,
|
| 21 |
+
"l1_per_token": 16.152315646511116
|
| 22 |
+
},
|
| 23 |
+
"seed1": {
|
| 24 |
+
"l0": 64.08272758037225,
|
| 25 |
+
"recon_r2": 0.8897518435535895,
|
| 26 |
+
"l1_per_token": 16.09002509331421
|
| 27 |
+
},
|
| 28 |
+
"seed2": {
|
| 29 |
+
"l0": 64.26186802030458,
|
| 30 |
+
"recon_r2": 0.8898026373585812,
|
| 31 |
+
"l1_per_token": 16.178846104722584
|
| 32 |
+
}
|
| 33 |
+
},
|
| 34 |
+
"per_pair": {
|
| 35 |
+
"seed0\u2194seed1": {
|
| 36 |
+
"alive_source": "stats",
|
| 37 |
+
"mcc_signed": 0.6298145045031466,
|
| 38 |
+
"mcc_abs": 0.6298204493369133,
|
| 39 |
+
"act_iou": 0.35266287635441973,
|
| 40 |
+
"act_iou_n_pairs": 12288,
|
| 41 |
+
"input_r2": 0.9982799951166693,
|
| 42 |
+
"matched_mcc_signed": {
|
| 43 |
+
"0.6": 0.7689133979807746,
|
| 44 |
+
"0.7": 0.7997098112646116,
|
| 45 |
+
"0.8": 0.854694436318848,
|
| 46 |
+
"0.9": 0.9289660800661758
|
| 47 |
+
},
|
| 48 |
+
"matched_mcc_abs": {
|
| 49 |
+
"0.6": 0.7688255881319279,
|
| 50 |
+
"0.7": 0.7995988953524815,
|
| 51 |
+
"0.8": 0.8546826835382902,
|
| 52 |
+
"0.9": 0.9289660800661755
|
| 53 |
+
}
|
| 54 |
+
},
|
| 55 |
+
"seed0\u2194seed2": {
|
| 56 |
+
"alive_source": "stats",
|
| 57 |
+
"mcc_signed": 0.6300010637338659,
|
| 58 |
+
"mcc_abs": 0.6300072895711294,
|
| 59 |
+
"act_iou": 0.3528902506876659,
|
| 60 |
+
"act_iou_n_pairs": 12288,
|
| 61 |
+
"input_r2": 0.9982792422275155,
|
| 62 |
+
"matched_mcc_signed": {
|
| 63 |
+
"0.6": 0.7668736101602807,
|
| 64 |
+
"0.7": 0.7993908913874955,
|
| 65 |
+
"0.8": 0.8584231043906829,
|
| 66 |
+
"0.9": 0.930440042912585
|
| 67 |
+
},
|
| 68 |
+
"matched_mcc_abs": {
|
| 69 |
+
"0.6": 0.7668781573882703,
|
| 70 |
+
"0.7": 0.7994439402404528,
|
| 71 |
+
"0.8": 0.8582495935260688,
|
| 72 |
+
"0.9": 0.930468816192188
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"seed1\u2194seed2": {
|
| 76 |
+
"alive_source": "stats",
|
| 77 |
+
"mcc_signed": 0.6295303053134002,
|
| 78 |
+
"mcc_abs": 0.6295360471526074,
|
| 79 |
+
"act_iou": 0.3524498054689431,
|
| 80 |
+
"act_iou_n_pairs": 12288,
|
| 81 |
+
"input_r2": 0.9982967993734801,
|
| 82 |
+
"matched_mcc_signed": {
|
| 83 |
+
"0.6": 0.7674345259288563,
|
| 84 |
+
"0.7": 0.7987904908705277,
|
| 85 |
+
"0.8": 0.8566658740698354,
|
| 86 |
+
"0.9": 0.9321740696122842
|
| 87 |
+
},
|
| 88 |
+
"matched_mcc_abs": {
|
| 89 |
+
"0.6": 0.7673784943199115,
|
| 90 |
+
"0.7": 0.7985323714824797,
|
| 91 |
+
"0.8": 0.8567602880723721,
|
| 92 |
+
"0.9": 0.9321832619828021
|
| 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": 63.599508178228994,
|
| 106 |
+
"l0_std": 0.996135873208844,
|
| 107 |
+
"recon_r2_mean": 0.8899326313039788,
|
| 108 |
+
"recon_r2_std": 0.00027034043257244427,
|
| 109 |
+
"l1_per_token_mean": 16.1403956148493,
|
| 110 |
+
"l1_per_token_std": 0.04559449948201637,
|
| 111 |
+
"input_r2_mean": 0.998285345572555,
|
| 112 |
+
"input_r2_std": 9.926423184827114e-06,
|
| 113 |
+
"mcc_signed_mean": 0.6297819578501375,
|
| 114 |
+
"mcc_signed_std": 0.00023706082779830192,
|
| 115 |
+
"mcc_abs_mean": 0.6297879286868834,
|
| 116 |
+
"mcc_abs_std": 0.00023729843817943885,
|
| 117 |
+
"matched_mcc_signed_per_threshold": {
|
| 118 |
+
"0.6": 0.7677405113566372,
|
| 119 |
+
"0.7": 0.7992970645075449,
|
| 120 |
+
"0.8": 0.8565944715931222,
|
| 121 |
+
"0.9": 0.9305267308636817
|
| 122 |
+
},
|
| 123 |
+
"matched_mcc_signed_band": 0.8385396945802465,
|
| 124 |
+
"matched_mcc_abs_per_threshold": {
|
| 125 |
+
"0.6": 0.7676940799467032,
|
| 126 |
+
"0.7": 0.7991917356918047,
|
| 127 |
+
"0.8": 0.8565641883789104,
|
| 128 |
+
"0.9": 0.9305393860803886
|
| 129 |
+
},
|
| 130 |
+
"matched_mcc_abs_band": 0.8384973475244517,
|
| 131 |
+
"act_iou_mean": 0.3526676441703429,
|
| 132 |
+
"act_iou_std": 0.00022026131463660605,
|
| 133 |
+
"act_iou_n_pairs": 12288.0,
|
| 134 |
+
"alive_source": "stats"
|
| 135 |
+
}
|
| 136 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/dict_metrics_step_final.json
ADDED
|
@@ -0,0 +1,550 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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": 63,
|
| 22 |
+
"l0_raw": 62.668800000000005,
|
| 23 |
+
"recon_r2": 0.9009,
|
| 24 |
+
"l1_per_token": 16.1758,
|
| 25 |
+
"stats_source": "stats",
|
| 26 |
+
"n_dead": 0,
|
| 27 |
+
"dead_rate": 0.0,
|
| 28 |
+
"alive_source": "stats",
|
| 29 |
+
"n_dense": 0,
|
| 30 |
+
"dense_rate": 0.0,
|
| 31 |
+
"dense_threshold": 0.1,
|
| 32 |
+
"n_dup_signed": {
|
| 33 |
+
"0.5": 9122,
|
| 34 |
+
"0.6": 4,
|
| 35 |
+
"0.7": 0,
|
| 36 |
+
"0.8": 0,
|
| 37 |
+
"0.9": 0
|
| 38 |
+
},
|
| 39 |
+
"n_dup_clusters_signed": {
|
| 40 |
+
"0.5": 3559,
|
| 41 |
+
"0.6": 2,
|
| 42 |
+
"0.7": 0,
|
| 43 |
+
"0.8": 0,
|
| 44 |
+
"0.9": 0
|
| 45 |
+
},
|
| 46 |
+
"largest_dup_signed": {
|
| 47 |
+
"0.5": 12,
|
| 48 |
+
"0.6": 2,
|
| 49 |
+
"0.7": 1,
|
| 50 |
+
"0.8": 1,
|
| 51 |
+
"0.9": 1
|
| 52 |
+
},
|
| 53 |
+
"babel_mu1_signed": 0.4621741771697998,
|
| 54 |
+
"n_dup_abs": {
|
| 55 |
+
"0.5": 9120,
|
| 56 |
+
"0.6": 20,
|
| 57 |
+
"0.7": 2,
|
| 58 |
+
"0.8": 0,
|
| 59 |
+
"0.9": 0
|
| 60 |
+
},
|
| 61 |
+
"n_dup_clusters_abs": {
|
| 62 |
+
"0.5": 3559,
|
| 63 |
+
"0.6": 10,
|
| 64 |
+
"0.7": 1,
|
| 65 |
+
"0.8": 0,
|
| 66 |
+
"0.9": 0
|
| 67 |
+
},
|
| 68 |
+
"largest_dup_abs": {
|
| 69 |
+
"0.5": 10,
|
| 70 |
+
"0.6": 2,
|
| 71 |
+
"0.7": 2,
|
| 72 |
+
"0.8": 1,
|
| 73 |
+
"0.9": 1
|
| 74 |
+
},
|
| 75 |
+
"babel_mu1_abs": 0.46327534317970276
|
| 76 |
+
},
|
| 77 |
+
"seed1": {
|
| 78 |
+
"d_sae": 12288,
|
| 79 |
+
"d_in": 768,
|
| 80 |
+
"l0": 64,
|
| 81 |
+
"l0_raw": 63.8976,
|
| 82 |
+
"recon_r2": 0.9004,
|
| 83 |
+
"l1_per_token": 16.1139,
|
| 84 |
+
"stats_source": "stats",
|
| 85 |
+
"n_dead": 0,
|
| 86 |
+
"dead_rate": 0.0,
|
| 87 |
+
"alive_source": "stats",
|
| 88 |
+
"n_dense": 0,
|
| 89 |
+
"dense_rate": 0.0,
|
| 90 |
+
"dense_threshold": 0.1,
|
| 91 |
+
"n_dup_signed": {
|
| 92 |
+
"0.5": 9206,
|
| 93 |
+
"0.6": 9,
|
| 94 |
+
"0.7": 0,
|
| 95 |
+
"0.8": 0,
|
| 96 |
+
"0.9": 0
|
| 97 |
+
},
|
| 98 |
+
"n_dup_clusters_signed": {
|
| 99 |
+
"0.5": 3599,
|
| 100 |
+
"0.6": 4,
|
| 101 |
+
"0.7": 0,
|
| 102 |
+
"0.8": 0,
|
| 103 |
+
"0.9": 0
|
| 104 |
+
},
|
| 105 |
+
"largest_dup_signed": {
|
| 106 |
+
"0.5": 13,
|
| 107 |
+
"0.6": 3,
|
| 108 |
+
"0.7": 1,
|
| 109 |
+
"0.8": 1,
|
| 110 |
+
"0.9": 1
|
| 111 |
+
},
|
| 112 |
+
"babel_mu1_signed": 0.46308183670043945,
|
| 113 |
+
"n_dup_abs": {
|
| 114 |
+
"0.5": 9207,
|
| 115 |
+
"0.6": 27,
|
| 116 |
+
"0.7": 2,
|
| 117 |
+
"0.8": 0,
|
| 118 |
+
"0.9": 0
|
| 119 |
+
},
|
| 120 |
+
"n_dup_clusters_abs": {
|
| 121 |
+
"0.5": 3601,
|
| 122 |
+
"0.6": 13,
|
| 123 |
+
"0.7": 1,
|
| 124 |
+
"0.8": 0,
|
| 125 |
+
"0.9": 0
|
| 126 |
+
},
|
| 127 |
+
"largest_dup_abs": {
|
| 128 |
+
"0.5": 15,
|
| 129 |
+
"0.6": 3,
|
| 130 |
+
"0.7": 2,
|
| 131 |
+
"0.8": 1,
|
| 132 |
+
"0.9": 1
|
| 133 |
+
},
|
| 134 |
+
"babel_mu1_abs": 0.4641289710998535
|
| 135 |
+
},
|
| 136 |
+
"seed2": {
|
| 137 |
+
"d_sae": 12288,
|
| 138 |
+
"d_in": 768,
|
| 139 |
+
"l0": 64,
|
| 140 |
+
"l0_raw": 63.8976,
|
| 141 |
+
"recon_r2": 0.8901,
|
| 142 |
+
"l1_per_token": 16.202,
|
| 143 |
+
"stats_source": "stats",
|
| 144 |
+
"n_dead": 0,
|
| 145 |
+
"dead_rate": 0.0,
|
| 146 |
+
"alive_source": "stats",
|
| 147 |
+
"n_dense": 0,
|
| 148 |
+
"dense_rate": 0.0,
|
| 149 |
+
"dense_threshold": 0.1,
|
| 150 |
+
"n_dup_signed": {
|
| 151 |
+
"0.5": 9270,
|
| 152 |
+
"0.6": 9,
|
| 153 |
+
"0.7": 0,
|
| 154 |
+
"0.8": 0,
|
| 155 |
+
"0.9": 0
|
| 156 |
+
},
|
| 157 |
+
"n_dup_clusters_signed": {
|
| 158 |
+
"0.5": 3638,
|
| 159 |
+
"0.6": 4,
|
| 160 |
+
"0.7": 0,
|
| 161 |
+
"0.8": 0,
|
| 162 |
+
"0.9": 0
|
| 163 |
+
},
|
| 164 |
+
"largest_dup_signed": {
|
| 165 |
+
"0.5": 11,
|
| 166 |
+
"0.6": 3,
|
| 167 |
+
"0.7": 1,
|
| 168 |
+
"0.8": 1,
|
| 169 |
+
"0.9": 1
|
| 170 |
+
},
|
| 171 |
+
"babel_mu1_signed": 0.4634845554828644,
|
| 172 |
+
"n_dup_abs": {
|
| 173 |
+
"0.5": 9263,
|
| 174 |
+
"0.6": 21,
|
| 175 |
+
"0.7": 0,
|
| 176 |
+
"0.8": 0,
|
| 177 |
+
"0.9": 0
|
| 178 |
+
},
|
| 179 |
+
"n_dup_clusters_abs": {
|
| 180 |
+
"0.5": 3640,
|
| 181 |
+
"0.6": 10,
|
| 182 |
+
"0.7": 0,
|
| 183 |
+
"0.8": 0,
|
| 184 |
+
"0.9": 0
|
| 185 |
+
},
|
| 186 |
+
"largest_dup_abs": {
|
| 187 |
+
"0.5": 11,
|
| 188 |
+
"0.6": 3,
|
| 189 |
+
"0.7": 1,
|
| 190 |
+
"0.8": 1,
|
| 191 |
+
"0.9": 1
|
| 192 |
+
},
|
| 193 |
+
"babel_mu1_abs": 0.4645220935344696
|
| 194 |
+
}
|
| 195 |
+
},
|
| 196 |
+
"per_pair": {
|
| 197 |
+
"seed0\u2194seed1": {
|
| 198 |
+
"dcs_signed": 0.6726120114326477,
|
| 199 |
+
"dcs_abs": 0.6732103228569031,
|
| 200 |
+
"frac_sign_flipped": 0.05029296875,
|
| 201 |
+
"alive_source": "stats",
|
| 202 |
+
"frac_denominator": "d_sae",
|
| 203 |
+
"n_alive_a": 12288,
|
| 204 |
+
"n_alive_b": 12288,
|
| 205 |
+
"matched_cos_signed": {
|
| 206 |
+
"0.5": 0.6060582995414734,
|
| 207 |
+
"0.6": 0.6726253628730774,
|
| 208 |
+
"0.7": 0.6726120114326477,
|
| 209 |
+
"0.8": 0.6726120114326477,
|
| 210 |
+
"0.9": 0.6726120114326477
|
| 211 |
+
},
|
| 212 |
+
"unique_frac_signed": {
|
| 213 |
+
"0.5": 0.5454915364583334,
|
| 214 |
+
"0.6": 0.9997151692708334,
|
| 215 |
+
"0.7": 1.0,
|
| 216 |
+
"0.8": 1.0,
|
| 217 |
+
"0.9": 1.0
|
| 218 |
+
},
|
| 219 |
+
"dedup_dcs_signed": {
|
| 220 |
+
"0.5": 0.33059967300020315,
|
| 221 |
+
"0.6": 0.6724337785005143,
|
| 222 |
+
"0.7": 0.6726120114326477,
|
| 223 |
+
"0.8": 0.6726120114326477,
|
| 224 |
+
"0.9": 0.6726120114326477
|
| 225 |
+
},
|
| 226 |
+
"matched_frac_signed": {
|
| 227 |
+
"0.5": 0.3648274739583333,
|
| 228 |
+
"0.6": 0.764404296875,
|
| 229 |
+
"0.7": 0.6441243489583334,
|
| 230 |
+
"0.8": 0.3984375,
|
| 231 |
+
"0.9": 0.14029947916666666
|
| 232 |
+
},
|
| 233 |
+
"matched_cos_abs": {
|
| 234 |
+
"0.5": 0.6073630452156067,
|
| 235 |
+
"0.6": 0.6732158660888672,
|
| 236 |
+
"0.7": 0.6732053756713867,
|
| 237 |
+
"0.8": 0.6732103228569031,
|
| 238 |
+
"0.9": 0.6732103228569031
|
| 239 |
+
},
|
| 240 |
+
"unique_frac_abs": {
|
| 241 |
+
"0.5": 0.5456136067708334,
|
| 242 |
+
"0.6": 0.9990234375,
|
| 243 |
+
"0.7": 0.9999186197916666,
|
| 244 |
+
"0.8": 1.0,
|
| 245 |
+
"0.9": 1.0
|
| 246 |
+
},
|
| 247 |
+
"dedup_dcs_abs": {
|
| 248 |
+
"0.5": 0.33138554171940393,
|
| 249 |
+
"0.6": 0.6725584287196398,
|
| 250 |
+
"0.7": 0.6731505900776634,
|
| 251 |
+
"0.8": 0.6732103228569031,
|
| 252 |
+
"0.9": 0.6732103228569031
|
| 253 |
+
},
|
| 254 |
+
"matched_frac_abs": {
|
| 255 |
+
"0.5": 0.3648274739583333,
|
| 256 |
+
"0.6": 0.7637532552083334,
|
| 257 |
+
"0.7": 0.6444498697916666,
|
| 258 |
+
"0.8": 0.3985188802083333,
|
| 259 |
+
"0.9": 0.14029947916666666
|
| 260 |
+
}
|
| 261 |
+
},
|
| 262 |
+
"seed0\u2194seed2": {
|
| 263 |
+
"dcs_signed": 0.6733407378196716,
|
| 264 |
+
"dcs_abs": 0.6739850640296936,
|
| 265 |
+
"frac_sign_flipped": 0.052490234375,
|
| 266 |
+
"alive_source": "stats",
|
| 267 |
+
"frac_denominator": "d_sae",
|
| 268 |
+
"n_alive_a": 12288,
|
| 269 |
+
"n_alive_b": 12288,
|
| 270 |
+
"matched_cos_signed": {
|
| 271 |
+
"0.5": 0.6074126362800598,
|
| 272 |
+
"0.6": 0.6733916401863098,
|
| 273 |
+
"0.7": 0.6733407378196716,
|
| 274 |
+
"0.8": 0.6733407378196716,
|
| 275 |
+
"0.9": 0.6733407378196716
|
| 276 |
+
},
|
| 277 |
+
"unique_frac_signed": {
|
| 278 |
+
"0.5": 0.5444742838541666,
|
| 279 |
+
"0.6": 0.9997151692708334,
|
| 280 |
+
"0.7": 1.0,
|
| 281 |
+
"0.8": 1.0,
|
| 282 |
+
"0.9": 1.0
|
| 283 |
+
},
|
| 284 |
+
"dedup_dcs_signed": {
|
| 285 |
+
"0.5": 0.33072056014255696,
|
| 286 |
+
"0.6": 0.6731998375544208,
|
| 287 |
+
"0.7": 0.6733407378196716,
|
| 288 |
+
"0.8": 0.6733407378196716,
|
| 289 |
+
"0.9": 0.6733407378196716
|
| 290 |
+
},
|
| 291 |
+
"matched_frac_signed": {
|
| 292 |
+
"0.5": 0.365478515625,
|
| 293 |
+
"0.6": 0.7666015625,
|
| 294 |
+
"0.7": 0.6426595052083334,
|
| 295 |
+
"0.8": 0.3937174479166667,
|
| 296 |
+
"0.9": 0.14306640625
|
| 297 |
+
},
|
| 298 |
+
"matched_cos_abs": {
|
| 299 |
+
"0.5": 0.6084281802177429,
|
| 300 |
+
"0.6": 0.673947811126709,
|
| 301 |
+
"0.7": 0.673995852470398,
|
| 302 |
+
"0.8": 0.6739850640296936,
|
| 303 |
+
"0.9": 0.6739850640296936
|
| 304 |
+
},
|
| 305 |
+
"unique_frac_abs": {
|
| 306 |
+
"0.5": 0.544921875,
|
| 307 |
+
"0.6": 0.9991455078125,
|
| 308 |
+
"0.7": 0.9999593098958334,
|
| 309 |
+
"0.8": 1.0,
|
| 310 |
+
"0.9": 1.0
|
| 311 |
+
},
|
| 312 |
+
"dedup_dcs_abs": {
|
| 313 |
+
"0.5": 0.3315458247670904,
|
| 314 |
+
"0.6": 0.6733719279873185,
|
| 315 |
+
"0.7": 0.673968427508953,
|
| 316 |
+
"0.8": 0.6739850640296936,
|
| 317 |
+
"0.9": 0.6739850640296936
|
| 318 |
+
},
|
| 319 |
+
"matched_frac_abs": {
|
| 320 |
+
"0.5": 0.3666178385416667,
|
| 321 |
+
"0.6": 0.7661946614583334,
|
| 322 |
+
"0.7": 0.6419270833333334,
|
| 323 |
+
"0.8": 0.39404296875,
|
| 324 |
+
"0.9": 0.14298502604166666
|
| 325 |
+
}
|
| 326 |
+
},
|
| 327 |
+
"seed1\u2194seed2": {
|
| 328 |
+
"dcs_signed": 0.6727239489555359,
|
| 329 |
+
"dcs_abs": 0.6733234524726868,
|
| 330 |
+
"frac_sign_flipped": 0.048909505208333336,
|
| 331 |
+
"alive_source": "stats",
|
| 332 |
+
"frac_denominator": "d_sae",
|
| 333 |
+
"n_alive_a": 12288,
|
| 334 |
+
"n_alive_b": 12288,
|
| 335 |
+
"matched_cos_signed": {
|
| 336 |
+
"0.5": 0.6065645217895508,
|
| 337 |
+
"0.6": 0.6726888418197632,
|
| 338 |
+
"0.7": 0.6727239489555359,
|
| 339 |
+
"0.8": 0.6727239489555359,
|
| 340 |
+
"0.9": 0.6727239489555359
|
| 341 |
+
},
|
| 342 |
+
"unique_frac_signed": {
|
| 343 |
+
"0.5": 0.5426839192708334,
|
| 344 |
+
"0.6": 0.9995930989583334,
|
| 345 |
+
"0.7": 1.0,
|
| 346 |
+
"0.8": 1.0,
|
| 347 |
+
"0.9": 1.0
|
| 348 |
+
},
|
| 349 |
+
"dedup_dcs_signed": {
|
| 350 |
+
"0.5": 0.3291728119753922,
|
| 351 |
+
"0.6": 0.6724151240293093,
|
| 352 |
+
"0.7": 0.6727239489555359,
|
| 353 |
+
"0.8": 0.6727239489555359,
|
| 354 |
+
"0.9": 0.6727239489555359
|
| 355 |
+
},
|
| 356 |
+
"matched_frac_signed": {
|
| 357 |
+
"0.5": 0.3658040364583333,
|
| 358 |
+
"0.6": 0.763427734375,
|
| 359 |
+
"0.7": 0.6404622395833334,
|
| 360 |
+
"0.8": 0.3956705729166667,
|
| 361 |
+
"0.9": 0.13932291666666666
|
| 362 |
+
},
|
| 363 |
+
"matched_cos_abs": {
|
| 364 |
+
"0.5": 0.6074586510658264,
|
| 365 |
+
"0.6": 0.6733308434486389,
|
| 366 |
+
"0.7": 0.6733475923538208,
|
| 367 |
+
"0.8": 0.6733234524726868,
|
| 368 |
+
"0.9": 0.6733234524726868
|
| 369 |
+
},
|
| 370 |
+
"unique_frac_abs": {
|
| 371 |
+
"0.5": 0.5430908203125,
|
| 372 |
+
"0.6": 0.9989827473958334,
|
| 373 |
+
"0.7": 0.9999593098958334,
|
| 374 |
+
"0.8": 1.0,
|
| 375 |
+
"0.9": 1.0
|
| 376 |
+
},
|
| 377 |
+
"dedup_dcs_abs": {
|
| 378 |
+
"0.5": 0.32990521711326437,
|
| 379 |
+
"0.6": 0.672645895894675,
|
| 380 |
+
"0.7": 0.6733201937701476,
|
| 381 |
+
"0.8": 0.6733234524726868,
|
| 382 |
+
"0.9": 0.6733234524726868
|
| 383 |
+
},
|
| 384 |
+
"matched_frac_abs": {
|
| 385 |
+
"0.5": 0.3663736979166667,
|
| 386 |
+
"0.6": 0.76318359375,
|
| 387 |
+
"0.7": 0.6410319010416666,
|
| 388 |
+
"0.8": 0.39501953125,
|
| 389 |
+
"0.9": 0.13924153645833334
|
| 390 |
+
}
|
| 391 |
+
}
|
| 392 |
+
},
|
| 393 |
+
"summary": {
|
| 394 |
+
"n_seeds": 3,
|
| 395 |
+
"n_pairs": 3,
|
| 396 |
+
"report_band": [
|
| 397 |
+
0.6,
|
| 398 |
+
0.7,
|
| 399 |
+
0.8,
|
| 400 |
+
0.9
|
| 401 |
+
],
|
| 402 |
+
"n_dead_mean": 0,
|
| 403 |
+
"n_dead_std": 0.0,
|
| 404 |
+
"n_dense_mean": 0,
|
| 405 |
+
"n_dense_std": 0.0,
|
| 406 |
+
"l0_mean": 64,
|
| 407 |
+
"l0_std": 0.7094480107802079,
|
| 408 |
+
"recon_r2_mean": 0.8971333333333332,
|
| 409 |
+
"recon_r2_std": 0.006096173663318112,
|
| 410 |
+
"l1_per_token_mean": 16.163899999999998,
|
| 411 |
+
"l1_per_token_std": 0.04523947391382887,
|
| 412 |
+
"babel_mu1_signed_mean": 0.46291352311770123,
|
| 413 |
+
"babel_mu1_signed_std": 0.0006712078124084428,
|
| 414 |
+
"n_dup_signed_per_threshold": {
|
| 415 |
+
"0.5": 9199.333333333334,
|
| 416 |
+
"0.6": 7.333333333333333,
|
| 417 |
+
"0.7": 0.0,
|
| 418 |
+
"0.8": 0.0,
|
| 419 |
+
"0.9": 0.0
|
| 420 |
+
},
|
| 421 |
+
"n_dup_signed_band": 1.8333333333333333,
|
| 422 |
+
"n_dup_signed_band_std": 0.7216878364870322,
|
| 423 |
+
"n_dup_clusters_signed_per_threshold": {
|
| 424 |
+
"0.5": 3598.6666666666665,
|
| 425 |
+
"0.6": 3.3333333333333335,
|
| 426 |
+
"0.7": 0.0,
|
| 427 |
+
"0.8": 0.0,
|
| 428 |
+
"0.9": 0.0
|
| 429 |
+
},
|
| 430 |
+
"n_dup_clusters_signed_band": 0.8333333333333334,
|
| 431 |
+
"n_dup_clusters_signed_band_std": 0.28867513459481287,
|
| 432 |
+
"largest_dup_signed_per_threshold": {
|
| 433 |
+
"0.5": 12.0,
|
| 434 |
+
"0.6": 2.6666666666666665,
|
| 435 |
+
"0.7": 1.0,
|
| 436 |
+
"0.8": 1.0,
|
| 437 |
+
"0.9": 1.0
|
| 438 |
+
},
|
| 439 |
+
"largest_dup_signed_band": 1.4166666666666665,
|
| 440 |
+
"largest_dup_signed_band_std": 0.14433756729740646,
|
| 441 |
+
"dcs_signed_mean": 0.6728922327359518,
|
| 442 |
+
"dcs_signed_std": 0.00039242847736820064,
|
| 443 |
+
"dedup_dcs_signed_per_threshold": {
|
| 444 |
+
"0.5": 0.33016434837271746,
|
| 445 |
+
"0.6": 0.6726829133614148,
|
| 446 |
+
"0.7": 0.6728922327359518,
|
| 447 |
+
"0.8": 0.6728922327359518,
|
| 448 |
+
"0.9": 0.6728922327359518
|
| 449 |
+
},
|
| 450 |
+
"dedup_dcs_signed_band": 0.6728399028923175,
|
| 451 |
+
"dedup_dcs_signed_band_std": 0.00040517417759863914,
|
| 452 |
+
"matched_cos_signed_per_threshold": {
|
| 453 |
+
"0.5": 0.6066784858703613,
|
| 454 |
+
"0.6": 0.6729019482930502,
|
| 455 |
+
"0.7": 0.6728922327359518,
|
| 456 |
+
"0.8": 0.6728922327359518,
|
| 457 |
+
"0.9": 0.6728922327359518
|
| 458 |
+
},
|
| 459 |
+
"matched_cos_signed_band": 0.6728946616252264,
|
| 460 |
+
"matched_cos_signed_band_std": 0.0004004565656654889,
|
| 461 |
+
"unique_frac_signed_per_threshold": {
|
| 462 |
+
"0.5": 0.5442165798611112,
|
| 463 |
+
"0.6": 0.9996744791666666,
|
| 464 |
+
"0.7": 1.0,
|
| 465 |
+
"0.8": 1.0,
|
| 466 |
+
"0.9": 1.0
|
| 467 |
+
},
|
| 468 |
+
"unique_frac_signed_band": 0.9999186197916666,
|
| 469 |
+
"unique_frac_signed_band_std": 1.7619331945484184e-05,
|
| 470 |
+
"matched_frac_signed_per_threshold": {
|
| 471 |
+
"0.5": 0.3653700086805555,
|
| 472 |
+
"0.6": 0.7648111979166666,
|
| 473 |
+
"0.7": 0.6424153645833334,
|
| 474 |
+
"0.8": 0.39594184027777785,
|
| 475 |
+
"0.9": 0.14089626736111108
|
| 476 |
+
},
|
| 477 |
+
"matched_frac_signed_band": 0.48601616753472227,
|
| 478 |
+
"matched_frac_signed_band_std": 0.0011320944467603541,
|
| 479 |
+
"babel_mu1_abs_mean": 0.463975469271342,
|
| 480 |
+
"babel_mu1_abs_std": 0.0006373921244376238,
|
| 481 |
+
"n_dup_abs_per_threshold": {
|
| 482 |
+
"0.5": 9196.666666666666,
|
| 483 |
+
"0.6": 22.666666666666668,
|
| 484 |
+
"0.7": 1.3333333333333333,
|
| 485 |
+
"0.8": 0.0,
|
| 486 |
+
"0.9": 0.0
|
| 487 |
+
},
|
| 488 |
+
"n_dup_abs_band": 6.0,
|
| 489 |
+
"n_dup_abs_band_std": 1.0897247358851685,
|
| 490 |
+
"n_dup_clusters_abs_per_threshold": {
|
| 491 |
+
"0.5": 3600.0,
|
| 492 |
+
"0.6": 11.0,
|
| 493 |
+
"0.7": 0.6666666666666666,
|
| 494 |
+
"0.8": 0.0,
|
| 495 |
+
"0.9": 0.0
|
| 496 |
+
},
|
| 497 |
+
"n_dup_clusters_abs_band": 2.9166666666666665,
|
| 498 |
+
"n_dup_clusters_abs_band_std": 0.5204164998665333,
|
| 499 |
+
"largest_dup_abs_per_threshold": {
|
| 500 |
+
"0.5": 12.0,
|
| 501 |
+
"0.6": 2.6666666666666665,
|
| 502 |
+
"0.7": 1.6666666666666667,
|
| 503 |
+
"0.8": 1.0,
|
| 504 |
+
"0.9": 1.0
|
| 505 |
+
},
|
| 506 |
+
"largest_dup_abs_band": 1.5833333333333333,
|
| 507 |
+
"largest_dup_abs_band_std": 0.14433756729740646,
|
| 508 |
+
"dcs_abs_mean": 0.6735062797864279,
|
| 509 |
+
"dcs_abs_std": 0.0004184797978296398,
|
| 510 |
+
"dedup_dcs_abs_per_threshold": {
|
| 511 |
+
"0.5": 0.3309455278665862,
|
| 512 |
+
"0.6": 0.6728587508672111,
|
| 513 |
+
"0.7": 0.6734797371189213,
|
| 514 |
+
"0.8": 0.6735062797864279,
|
| 515 |
+
"0.9": 0.6735062797864279
|
| 516 |
+
},
|
| 517 |
+
"dedup_dcs_abs_band": 0.673337761889747,
|
| 518 |
+
"dedup_dcs_abs_band_std": 0.0004285107980667797,
|
| 519 |
+
"matched_cos_abs_per_threshold": {
|
| 520 |
+
"0.5": 0.6077499588330587,
|
| 521 |
+
"0.6": 0.6734981735547384,
|
| 522 |
+
"0.7": 0.6735162734985352,
|
| 523 |
+
"0.8": 0.6735062797864279,
|
| 524 |
+
"0.9": 0.6735062797864279
|
| 525 |
+
},
|
| 526 |
+
"matched_cos_abs_band": 0.6735067516565324,
|
| 527 |
+
"matched_cos_abs_band_std": 0.0004129467338597957,
|
| 528 |
+
"unique_frac_abs_per_threshold": {
|
| 529 |
+
"0.5": 0.5445421006944445,
|
| 530 |
+
"0.6": 0.9990505642361112,
|
| 531 |
+
"0.7": 0.9999457465277778,
|
| 532 |
+
"0.8": 1.0,
|
| 533 |
+
"0.9": 1.0
|
| 534 |
+
},
|
| 535 |
+
"unique_frac_abs_band": 0.9997490776909722,
|
| 536 |
+
"unique_frac_abs_band_std": 2.3492442593989597e-05,
|
| 537 |
+
"matched_frac_abs_per_threshold": {
|
| 538 |
+
"0.5": 0.3659396701388889,
|
| 539 |
+
"0.6": 0.764377170138889,
|
| 540 |
+
"0.7": 0.6424696180555555,
|
| 541 |
+
"0.8": 0.3958604600694444,
|
| 542 |
+
"0.9": 0.14084201388888887
|
| 543 |
+
},
|
| 544 |
+
"matched_frac_abs_band": 0.4858873155381944,
|
| 545 |
+
"matched_frac_abs_band_std": 0.0011229166394581855,
|
| 546 |
+
"alive_source": "stats",
|
| 547 |
+
"frac_denominator": "d_sae",
|
| 548 |
+
"d_sae": 12288
|
| 549 |
+
}
|
| 550 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed0/config.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "fista",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 0,
|
| 8 |
+
"fista_t_iters": 30,
|
| 9 |
+
"fista_step": 0.0,
|
| 10 |
+
"fista_lam": 1.25,
|
| 11 |
+
"fista_nonneg": true,
|
| 12 |
+
"fista_power_iters": 32,
|
| 13 |
+
"fista_r2_tol": 0.0001,
|
| 14 |
+
"fista_check_every": 5,
|
| 15 |
+
"fista_use_bf16": false,
|
| 16 |
+
"fista_step_cache_every": 0,
|
| 17 |
+
"fista_compile": false,
|
| 18 |
+
"fista_topk_k": 0,
|
| 19 |
+
"refit": true,
|
| 20 |
+
"refit_iters": 10,
|
| 21 |
+
"refit_thresh": 1e-08,
|
| 22 |
+
"class_type": "FistaSAE",
|
| 23 |
+
"module_name": "consdl.models.fista"
|
| 24 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed0/lock.lock
ADDED
|
File without changes
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed0/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.0497,
|
| 3 |
+
"r2_score": 0.9009,
|
| 4 |
+
"l0_per_token_0": 0.0051,
|
| 5 |
+
"dead_rate": 0.0,
|
| 6 |
+
"dense_rate_0.2": 0.0,
|
| 7 |
+
"live_latent_frac": 0.9998,
|
| 8 |
+
"l1_per_token": 16.1758
|
| 9 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed1/config.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "fista",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 1,
|
| 8 |
+
"fista_t_iters": 30,
|
| 9 |
+
"fista_step": 0.0,
|
| 10 |
+
"fista_lam": 1.25,
|
| 11 |
+
"fista_nonneg": true,
|
| 12 |
+
"fista_power_iters": 32,
|
| 13 |
+
"fista_r2_tol": 0.0001,
|
| 14 |
+
"fista_check_every": 5,
|
| 15 |
+
"fista_use_bf16": false,
|
| 16 |
+
"fista_step_cache_every": 0,
|
| 17 |
+
"fista_compile": false,
|
| 18 |
+
"fista_topk_k": 0,
|
| 19 |
+
"refit": true,
|
| 20 |
+
"refit_iters": 10,
|
| 21 |
+
"refit_thresh": 1e-08,
|
| 22 |
+
"class_type": "FistaSAE",
|
| 23 |
+
"module_name": "consdl.models.fista"
|
| 24 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed1/lock.lock
ADDED
|
File without changes
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed1/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.0499,
|
| 3 |
+
"r2_score": 0.9004,
|
| 4 |
+
"l0_per_token_0": 0.0052,
|
| 5 |
+
"dead_rate": 0.0,
|
| 6 |
+
"dense_rate_0.2": 0.0,
|
| 7 |
+
"live_latent_frac": 0.9998,
|
| 8 |
+
"l1_per_token": 16.1139
|
| 9 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed2/config.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"d_in": 768,
|
| 3 |
+
"d_sae_factor": 16,
|
| 4 |
+
"architecture": "fista",
|
| 5 |
+
"activation_name": "relu",
|
| 6 |
+
"device": "cpu",
|
| 7 |
+
"seed": 2,
|
| 8 |
+
"fista_t_iters": 30,
|
| 9 |
+
"fista_step": 0.0,
|
| 10 |
+
"fista_lam": 1.25,
|
| 11 |
+
"fista_nonneg": true,
|
| 12 |
+
"fista_power_iters": 32,
|
| 13 |
+
"fista_r2_tol": 0.0001,
|
| 14 |
+
"fista_check_every": 5,
|
| 15 |
+
"fista_use_bf16": false,
|
| 16 |
+
"fista_step_cache_every": 0,
|
| 17 |
+
"fista_compile": false,
|
| 18 |
+
"fista_topk_k": 0,
|
| 19 |
+
"refit": true,
|
| 20 |
+
"refit_iters": 10,
|
| 21 |
+
"refit_thresh": 1e-08,
|
| 22 |
+
"class_type": "FistaSAE",
|
| 23 |
+
"module_name": "consdl.models.fista"
|
| 24 |
+
}
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed2/lock.lock
ADDED
|
File without changes
|
vision/clip-vit-b-16_layer10/consdl/x16/l0_64/seed2/stats.imagenet.split_val.seed_no/metrics.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"l2_loss": 0.0499,
|
| 3 |
+
"r2_score": 0.8901,
|
| 4 |
+
"l0_per_token_0": 0.0052,
|
| 5 |
+
"dead_rate": 0.0,
|
| 6 |
+
"dense_rate_0.2": 0.0,
|
| 7 |
+
"live_latent_frac": 1.0,
|
| 8 |
+
"l1_per_token": 16.202
|
| 9 |
+
}
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed0/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f1fd4ec7dbadb22ae3c0337cbf13f403962c03d09bcc399bf2ff8e30e6409aec
|
| 3 |
+
size 75550084
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed1/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:84f8686032f5a98559ac769008e89215b89052b7fd9e8aad646b4904999acdbe
|
| 3 |
+
size 75550084
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2b39be7990f333af6bd1d4fe33642e913b6052c044f600eb34cebfb8a657ff76
|
| 3 |
+
size 75550084
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/active_count_hist_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:87a712e16d63cfad50194bb7e185543d26765253643c62e0d34b6731a1cf1318
|
| 3 |
+
size 7424
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/active_count_proportion_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ef23f87d97c90967b72b4bf600952e3179a0d59ab4ba1815df6a9fcc0e418ff0
|
| 3 |
+
size 4382
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/firing_rate_per_latent.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:451b5e45de8b969a6a44227e964a6e73af72b140e88a2c54f62f078c091ba1b2
|
| 3 |
+
size 50471
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/max_act_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:40e95389d9a8f8345b59327f2cb90b7b1fb0864e426b49f300ff6cea54eb3878
|
| 3 |
+
size 50382
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/mean_act.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bc8deb849c6a6fae5268741a0a62125e181ca828d3094884ee11fb14fdcd6a8a
|
| 3 |
+
size 50337
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/mean_act_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f1ac37b64b3c79aa4d85881941fb2655aab77dffda21f0bee70de81e7d4a12d2
|
| 3 |
+
size 50387
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/sparsity.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ef1daf9cdb3eae1058803b0b0dc8b6b8253ec020f347ee46687371ee95722d3f
|
| 3 |
+
size 50337
|
vision/dinov2-b_layer10/batchtopk/x16/l0_128/seed2/stats.imagenet.split_val.seed_no/sparsity_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e4bdde42429dfccc8bee8f6ebbb8903dcab118501400c78f216c255ad0da7c07
|
| 3 |
+
size 50387
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5e7f9d9db7ef7b9e93c28a337cd64ddc1dc73915a73531281b39c6d3854e7eb2
|
| 3 |
+
size 75550084
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/active_count_hist_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b34498538e63b214013d313e3eba2d292415e2c9926980e3a6fe147c355a4e8b
|
| 3 |
+
size 3840
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/active_count_proportion_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:67c0ad4c7d1de7a005d35d1871e867ee10a1a354644fc27fa8a1412c3d118b12
|
| 3 |
+
size 2654
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/firing_rate_per_latent.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1913c031ab4ad76ab1d56e8ee85ae84749e13c1ac30518fae71a099b26477352
|
| 3 |
+
size 50471
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/max_act_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:913bb2a33a1a229cf616a20e3cf891c5280418be0bb4e2cdfc3b9b56cafb3793
|
| 3 |
+
size 50382
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/mean_act.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1503d19c1f44a278c259dd905f936aa6ba7f21f4098e010837714f2702220fe1
|
| 3 |
+
size 50337
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/mean_act_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9e982d0f26b96b10ed65302a07cc700af25b6247d0280972a04bfd6563690674
|
| 3 |
+
size 50387
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/sparsity.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ed4e1029b12af1a95e69f48239805e8904e7a841d4a8aa9afbf69aa73ae3032c
|
| 3 |
+
size 50337
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed0/stats.imagenet.split_val.seed_no/sparsity_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d14fb0d809862f32274b4532e12482d55d7cb3c900ba08b10900b757155696e4
|
| 3 |
+
size 50387
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:46191d06733f2537f77009efe3e8d1ce4d3130a7fdea04a86b931257c3c15936
|
| 3 |
+
size 75550084
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/active_count_hist_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e2bc388a3386f50e6f2b37815ec74582d8e593857267e1f5b408e17701f2a673
|
| 3 |
+
size 4032
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/active_count_proportion_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5958ffa7539e9245a69f98be0705bbefbe9902c61167665183540d446f2420de
|
| 3 |
+
size 2718
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/firing_rate_per_latent.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fa7cd0af40f189cd1bb8f22cb0262d6bda1b28458940c5060f94fc47b8131d50
|
| 3 |
+
size 50471
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/max_act_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e28b3fcf750a89d13d071eb4538461ed09c57b40ad2451978f3582bf3f08d59b
|
| 3 |
+
size 50382
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/mean_act.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d6697c46530c86582a051e9bc2dbd3c1601159ecde70491f4be72157f8ac6384
|
| 3 |
+
size 50337
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/mean_act_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:482639bbd75e98023c5ed72b94aefa2147d90734e6d3ad72b3869887fc9c36ad
|
| 3 |
+
size 50387
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/sparsity.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7bbc43b71a52b1d5c5d14a89304816de90008e8b87eea2224c0cca3ad84463d0
|
| 3 |
+
size 50337
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed1/stats.imagenet.split_val.seed_no/sparsity_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2021acad8e54376f881a38175255ef08fb34b46edd88d68d0ca5ae1107b62be1
|
| 3 |
+
size 50387
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed2/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9d6d550fb3ce7cb9627d27e4a6ac89eca5a3c45f4c9aba40f9edadb3d7835467
|
| 3 |
+
size 75550084
|
vision/dinov2-b_layer10/batchtopk/x16/l0_32/seed2/stats.imagenet.split_val.seed_no/active_count_hist_per_token.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5d2d82d8d63a3e1fd2a24950a7aab50d2bd6339de614ccc74f4feb27c3bfbb89
|
| 3 |
+
size 4032
|