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
96ac044
·
verified ·
1 Parent(s): eae143d

Copy SAE layer 19 from model repo

Browse files
Files changed (3) hide show
  1. layer_19_s0/checkpoint_full.pt +3 -0
  2. layer_19_s0/meta.json +129 -0
  3. layer_19_s0/sae.pt +3 -0
layer_19_s0/checkpoint_full.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cfc18e6d1286c0287a0c9c0f9286ceb8402fba8ea658ff13946a3ff3737f6792
3
+ size 1813939659
layer_19_s0/meta.json ADDED
@@ -0,0 +1,129 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "layer": 19,
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": 2501,
10
+ "lr": 0.0002,
11
+ "lambda_l0": 0.003644558588430284,
12
+ "tier": "mid",
13
+ "preflight": {
14
+ "activation_norm_probe": 1.6312718391418457,
15
+ "activation_norm_ref": 0.7526426911354065,
16
+ "initial_l0_probe": 71.94287109375,
17
+ "initial_lr_multiplier": 1.6349459923828127,
18
+ "early_pulse_multiplier": 1.11510859375,
19
+ "layer_lr_multiplier": 1.18
20
+ },
21
+ "scheduler_mode": "BASELINE",
22
+ "scheduler_phase": "DESCENT",
23
+ "scheduler_phase_summary": {
24
+ "phase": "DESCENT",
25
+ "phase_step": 2500,
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": 81952768,
33
+ "early_stopped": true,
34
+ "path": "rolling",
35
+ "best_ev": 0.8227164403743537,
36
+ "best_ev_step": 1750,
37
+ "final_metrics": {
38
+ "recon_loss": 0.522655,
39
+ "mean_l0": 51.44,
40
+ "dead_pct": 0.21,
41
+ "resampled": 0,
42
+ "ev": 0.8119032838346341,
43
+ "nonlinear_err": null,
44
+ "linear_err": null,
45
+ "qo_gap": 0.5327,
46
+ "activation_norm_probe": 1.6312718391418457,
47
+ "activation_norm_ref": 0.7526426911354065,
48
+ "initial_l0_probe": 71.94287109375,
49
+ "initial_lr_multiplier": 1.6349459923828127,
50
+ "early_pulse_multiplier": 1.11510859375,
51
+ "layer_lr_multiplier": 1.18
52
+ },
53
+ "training_curve": {
54
+ "recon_loss": [
55
+ 0.721347,
56
+ 0.628954,
57
+ 0.555032,
58
+ 0.520621,
59
+ 0.502434,
60
+ 0.486892,
61
+ 0.491858,
62
+ 0.482237,
63
+ 0.494107,
64
+ 0.522655
65
+ ],
66
+ "mean_l0": [
67
+ 28.44,
68
+ 30.61,
69
+ 31.57,
70
+ 39.94,
71
+ 50.94,
72
+ 54.26,
73
+ 52.77,
74
+ 48.76,
75
+ 50.95,
76
+ 51.44
77
+ ],
78
+ "dead_pct": [
79
+ 0.0,
80
+ 0.0,
81
+ 0.0,
82
+ 0.0,
83
+ 0.0,
84
+ 0.0,
85
+ 0.0,
86
+ 0.0,
87
+ 0.05,
88
+ 0.21
89
+ ],
90
+ "resampled": [
91
+ 0,
92
+ 0,
93
+ 0,
94
+ 0,
95
+ 0,
96
+ 0,
97
+ 0,
98
+ 0,
99
+ 0,
100
+ 0
101
+ ],
102
+ "ev": [
103
+ 0.7328418014900924,
104
+ 0.7704570282980081,
105
+ 0.7949702285200095,
106
+ 0.8088181758726142,
107
+ 0.8189262090278928,
108
+ 0.8227022556830943,
109
+ 0.8199513179202506,
110
+ 0.8217849642516819,
111
+ 0.8194079872371236,
112
+ 0.8119032838346341
113
+ ],
114
+ "nonlinear_err": [],
115
+ "linear_err": [],
116
+ "qo_gap": [
117
+ 0.706,
118
+ 0.6554,
119
+ 0.6226,
120
+ 0.5904,
121
+ 0.5595,
122
+ 0.542,
123
+ 0.5391,
124
+ 0.544,
125
+ 0.5498,
126
+ 0.5327
127
+ ]
128
+ }
129
+ }
layer_19_s0/sae.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a9a2b13302db22a25ce65e4973b7fc44c92484b06b1a17d17453041ad646a8c
3
+ size 604381549