Hugging Face's logo

Datasets:
juiceb0xc0de
/
gemma-4-e2b-it-SAE

Tasks:
Feature Extraction
Size:
10B<n<100B
Tags:
sparse-autoencoder
mechanistic-interpretability
sae
gemma
brain-atlas
jumprelu
License:
Dataset card Files
xet
 Community
1
juiceb0xc0de commited on
Commit
8e177fe
·
verified ·
1 Parent(s): b044528

Copy SAE layer 13 from model repo

Browse files
Files changed (3) hide show
  1. layer_13_s0/checkpoint_full.pt +3 -0
  2. layer_13_s0/meta.json +111 -0
  3. layer_13_s0/sae.pt +3 -0
layer_13_s0/checkpoint_full.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3798290e326a7a1b7680dfe0abce539059947dddf5eba8c5be47834f9cb9cedd
3
+ size 1813939595
layer_13_s0/meta.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "layer": 13,
3
+ "seed": 0,
4
+ "model_id": "google/gemma-4-E2B-it",
5
+ "d_in": 1536,
6
+ "n_features": 49152,
7
+ "k": 50,
8
+ "batch_tokens": 32768,
9
+ "n_steps": 1751,
10
+ "lr": 0.0002,
11
+ "lambda_l0": 0.0020103728068267667,
12
+ "tier": "mid",
13
+ "preflight": {
14
+ "activation_norm_probe": 1.0544369220733643,
15
+ "activation_norm_ref": 0.7526426911354065,
16
+ "initial_l0_probe": 71.7044677734375,
17
+ "initial_lr_multiplier": 1.2144578303298121,
18
+ "early_pulse_multiplier": 1.0347271484375,
19
+ "layer_lr_multiplier": 1.0
20
+ },
21
+ "scheduler_mode": "BASELINE",
22
+ "scheduler_phase": "DESCENT",
23
+ "scheduler_phase_summary": {
24
+ "phase": "DESCENT",
25
+ "phase_step": 1750,
26
+ "pin_entry_step": null,
27
+ "pinned_lambda": null,
28
+ "pin_ev_count": 0,
29
+ "pin_retry_count": 0
30
+ },
31
+ "scheduler_transitions": 0,
32
+ "total_tokens": 57376768,
33
+ "early_stopped": true,
34
+ "path": "rolling",
35
+ "best_ev": 0.7890679836549388,
36
+ "best_ev_step": 1750,
37
+ "final_metrics": {
38
+ "recon_loss": 0.230602,
39
+ "mean_l0": 51.44,
40
+ "dead_pct": 0.0,
41
+ "resampled": 0,
42
+ "ev": 0.7866328055011582,
43
+ "nonlinear_err": null,
44
+ "linear_err": null,
45
+ "qo_gap": 0.3378,
46
+ "activation_norm_probe": 1.0544369220733643,
47
+ "activation_norm_ref": 0.7526426911354065,
48
+ "initial_l0_probe": 71.7044677734375,
49
+ "initial_lr_multiplier": 1.2144578303298121,
50
+ "early_pulse_multiplier": 1.0347271484375,
51
+ "layer_lr_multiplier": 1.0
52
+ },
53
+ "training_curve": {
54
+ "recon_loss": [
55
+ 0.414945,
56
+ 0.324932,
57
+ 0.278025,
58
+ 0.257128,
59
+ 0.235118,
60
+ 0.236338,
61
+ 0.230602
62
+ ],
63
+ "mean_l0": [
64
+ 26.38,
65
+ 28.07,
66
+ 33.46,
67
+ 42.37,
68
+ 49.04,
69
+ 51.44,
70
+ 51.44
71
+ ],
72
+ "dead_pct": [
73
+ 0.0,
74
+ 0.0,
75
+ 0.0,
76
+ 0.0,
77
+ 0.0,
78
+ 0.0,
79
+ 0.0
80
+ ],
81
+ "resampled": [
82
+ 0,
83
+ 0,
84
+ 0,
85
+ 0,
86
+ 0,
87
+ 0,
88
+ 0
89
+ ],
90
+ "ev": [
91
+ 0.6252005866611364,
92
+ 0.7063686359262689,
93
+ 0.7477332426911638,
94
+ 0.7674794653305139,
95
+ 0.78776118338137,
96
+ 0.7866328055011582,
97
+ 0.7866328055011582
98
+ ],
99
+ "nonlinear_err": [],
100
+ "linear_err": [],
101
+ "qo_gap": [
102
+ 0.4889,
103
+ 0.4331,
104
+ 0.4015,
105
+ 0.3683,
106
+ 0.3485,
107
+ 0.3352,
108
+ 0.3378
109
+ ]
110
+ }
111
+ }
layer_13_s0/sae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ecb46ff6d524ac1a5578f4563a07659c94e687593db52a65d234031717ce8c9
3
+ size 604381549