DuoNeural commited on
Commit
e0f3340
·
verified ·
1 Parent(s): 00d06e7

Upload geometry_results.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. geometry_results.json +217 -0
geometry_results.json ADDED
@@ -0,0 +1,217 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "probe_layers": [
3
+ 1,
4
+ 2,
5
+ 3,
6
+ 4,
7
+ 5,
8
+ 6,
9
+ 7,
10
+ 8,
11
+ 9,
12
+ 10,
13
+ 11,
14
+ 12,
15
+ 13,
16
+ 14,
17
+ 15,
18
+ 16
19
+ ],
20
+ "angles_by_layer": {
21
+ "1": {
22
+ "refusal|harm_awareness": 11.95,
23
+ "refusal|self_identity": 7.64,
24
+ "refusal|ethics": 10.11,
25
+ "refusal|benign_general": 12.4,
26
+ "harm_awareness|self_identity": 11.71,
27
+ "harm_awareness|ethics": 10.99,
28
+ "harm_awareness|benign_general": 11.02,
29
+ "self_identity|ethics": 9.29,
30
+ "self_identity|benign_general": 11.52,
31
+ "ethics|benign_general": 8.8
32
+ },
33
+ "2": {
34
+ "refusal|harm_awareness": 9.98,
35
+ "refusal|self_identity": 7.87,
36
+ "refusal|ethics": 8.27,
37
+ "refusal|benign_general": 10.43,
38
+ "harm_awareness|self_identity": 10.47,
39
+ "harm_awareness|ethics": 8.5,
40
+ "harm_awareness|benign_general": 10.09,
41
+ "self_identity|ethics": 7.83,
42
+ "self_identity|benign_general": 10.58,
43
+ "ethics|benign_general": 8.5
44
+ },
45
+ "3": {
46
+ "refusal|harm_awareness": 8.98,
47
+ "refusal|self_identity": 7.72,
48
+ "refusal|ethics": 8.02,
49
+ "refusal|benign_general": 9.32,
50
+ "harm_awareness|self_identity": 9.71,
51
+ "harm_awareness|ethics": 7.64,
52
+ "harm_awareness|benign_general": 8.7,
53
+ "self_identity|ethics": 7.66,
54
+ "self_identity|benign_general": 10.02,
55
+ "ethics|benign_general": 7.4
56
+ },
57
+ "4": {
58
+ "refusal|harm_awareness": 10.2,
59
+ "refusal|self_identity": 7.16,
60
+ "refusal|ethics": 8.42,
61
+ "refusal|benign_general": 10.31,
62
+ "harm_awareness|self_identity": 11.74,
63
+ "harm_awareness|ethics": 7.33,
64
+ "harm_awareness|benign_general": 10.07,
65
+ "self_identity|ethics": 9.05,
66
+ "self_identity|benign_general": 11.05,
67
+ "ethics|benign_general": 8.06
68
+ },
69
+ "5": {
70
+ "refusal|harm_awareness": 12.16,
71
+ "refusal|self_identity": 8.76,
72
+ "refusal|ethics": 10.71,
73
+ "refusal|benign_general": 13.29,
74
+ "harm_awareness|self_identity": 13.7,
75
+ "harm_awareness|ethics": 9.13,
76
+ "harm_awareness|benign_general": 12.17,
77
+ "self_identity|ethics": 10.42,
78
+ "self_identity|benign_general": 12.21,
79
+ "ethics|benign_general": 10.2
80
+ },
81
+ "6": {
82
+ "refusal|harm_awareness": 16.51,
83
+ "refusal|self_identity": 11.74,
84
+ "refusal|ethics": 15.29,
85
+ "refusal|benign_general": 16.71,
86
+ "harm_awareness|self_identity": 17.22,
87
+ "harm_awareness|ethics": 11.68,
88
+ "harm_awareness|benign_general": 15.37,
89
+ "self_identity|ethics": 13.94,
90
+ "self_identity|benign_general": 14.24,
91
+ "ethics|benign_general": 14.39
92
+ },
93
+ "7": {
94
+ "refusal|harm_awareness": 17.31,
95
+ "refusal|self_identity": 13.79,
96
+ "refusal|ethics": 15.28,
97
+ "refusal|benign_general": 19.14,
98
+ "harm_awareness|self_identity": 19.33,
99
+ "harm_awareness|ethics": 13.4,
100
+ "harm_awareness|benign_general": 18.73,
101
+ "self_identity|ethics": 15.47,
102
+ "self_identity|benign_general": 17.76,
103
+ "ethics|benign_general": 18.56
104
+ },
105
+ "8": {
106
+ "refusal|harm_awareness": 17.77,
107
+ "refusal|self_identity": 14.92,
108
+ "refusal|ethics": 16.16,
109
+ "refusal|benign_general": 20.1,
110
+ "harm_awareness|self_identity": 20.28,
111
+ "harm_awareness|ethics": 13.59,
112
+ "harm_awareness|benign_general": 18.91,
113
+ "self_identity|ethics": 17.59,
114
+ "self_identity|benign_general": 18.96,
115
+ "ethics|benign_general": 19.0
116
+ },
117
+ "9": {
118
+ "refusal|harm_awareness": 17.91,
119
+ "refusal|self_identity": 16.12,
120
+ "refusal|ethics": 17.62,
121
+ "refusal|benign_general": 20.87,
122
+ "harm_awareness|self_identity": 20.76,
123
+ "harm_awareness|ethics": 13.68,
124
+ "harm_awareness|benign_general": 19.61,
125
+ "self_identity|ethics": 18.35,
126
+ "self_identity|benign_general": 19.71,
127
+ "ethics|benign_general": 19.28
128
+ },
129
+ "10": {
130
+ "refusal|harm_awareness": 20.57,
131
+ "refusal|self_identity": 18.94,
132
+ "refusal|ethics": 20.18,
133
+ "refusal|benign_general": 24.21,
134
+ "harm_awareness|self_identity": 25.15,
135
+ "harm_awareness|ethics": 15.51,
136
+ "harm_awareness|benign_general": 23.66,
137
+ "self_identity|ethics": 22.46,
138
+ "self_identity|benign_general": 23.88,
139
+ "ethics|benign_general": 23.98
140
+ },
141
+ "11": {
142
+ "refusal|harm_awareness": 21.37,
143
+ "refusal|self_identity": 18.24,
144
+ "refusal|ethics": 20.84,
145
+ "refusal|benign_general": 26.9,
146
+ "harm_awareness|self_identity": 25.82,
147
+ "harm_awareness|ethics": 15.67,
148
+ "harm_awareness|benign_general": 24.99,
149
+ "self_identity|ethics": 22.71,
150
+ "self_identity|benign_general": 26.38,
151
+ "ethics|benign_general": 25.94
152
+ },
153
+ "12": {
154
+ "refusal|harm_awareness": 28.21,
155
+ "refusal|self_identity": 23.62,
156
+ "refusal|ethics": 27.58,
157
+ "refusal|benign_general": 39.04,
158
+ "harm_awareness|self_identity": 33.08,
159
+ "harm_awareness|ethics": 21.19,
160
+ "harm_awareness|benign_general": 33.57,
161
+ "self_identity|ethics": 29.22,
162
+ "self_identity|benign_general": 36.32,
163
+ "ethics|benign_general": 34.97
164
+ },
165
+ "13": {
166
+ "refusal|harm_awareness": 32.03,
167
+ "refusal|self_identity": 25.77,
168
+ "refusal|ethics": 30.54,
169
+ "refusal|benign_general": 45.4,
170
+ "harm_awareness|self_identity": 36.79,
171
+ "harm_awareness|ethics": 23.9,
172
+ "harm_awareness|benign_general": 38.46,
173
+ "self_identity|ethics": 31.89,
174
+ "self_identity|benign_general": 42.56,
175
+ "ethics|benign_general": 38.94
176
+ },
177
+ "14": {
178
+ "refusal|harm_awareness": 30.45,
179
+ "refusal|self_identity": 21.31,
180
+ "refusal|ethics": 27.94,
181
+ "refusal|benign_general": 42.28,
182
+ "harm_awareness|self_identity": 32.8,
183
+ "harm_awareness|ethics": 19.14,
184
+ "harm_awareness|benign_general": 32.46,
185
+ "self_identity|ethics": 27.59,
186
+ "self_identity|benign_general": 38.89,
187
+ "ethics|benign_general": 32.47
188
+ },
189
+ "15": {
190
+ "refusal|harm_awareness": 27.97,
191
+ "refusal|self_identity": 19.56,
192
+ "refusal|ethics": 25.59,
193
+ "refusal|benign_general": 39.02,
194
+ "harm_awareness|self_identity": 29.0,
195
+ "harm_awareness|ethics": 17.55,
196
+ "harm_awareness|benign_general": 28.57,
197
+ "self_identity|ethics": 23.6,
198
+ "self_identity|benign_general": 35.86,
199
+ "ethics|benign_general": 29.06
200
+ },
201
+ "16": {
202
+ "refusal|harm_awareness": 26.94,
203
+ "refusal|self_identity": 18.72,
204
+ "refusal|ethics": 24.8,
205
+ "refusal|benign_general": 37.79,
206
+ "harm_awareness|self_identity": 27.62,
207
+ "harm_awareness|ethics": 17.28,
208
+ "harm_awareness|benign_general": 27.79,
209
+ "self_identity|ethics": 21.95,
210
+ "self_identity|benign_general": 34.77,
211
+ "ethics|benign_general": 27.96
212
+ }
213
+ },
214
+ "model_name": "10pct",
215
+ "refusal_frac": 0.1,
216
+ "n_layers": 16
217
+ }