astirex commited on
Commit
b0b3518
·
verified ·
1 Parent(s): 4d3deb2

Upload 8 files

Browse files
intervenable_model/config.json ADDED
@@ -0,0 +1,54 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "intervention_constant_sources": [
3
+ true,
4
+ true
5
+ ],
6
+ "intervention_dimensions": [
7
+ 3072,
8
+ 3072
9
+ ],
10
+ "intervention_types": [
11
+ "<class 'pyreft.interventions.GatedLoreftIntervention_Linear'>",
12
+ "<class 'pyreft.interventions.GatedLoreftIntervention_Linear'>"
13
+ ],
14
+ "mode": "parallel",
15
+ "representations": [
16
+ [
17
+ 22,
18
+ "block_output",
19
+ "pos",
20
+ 1,
21
+ 4,
22
+ null,
23
+ null,
24
+ null,
25
+ null,
26
+ null,
27
+ null,
28
+ null,
29
+ null,
30
+ null
31
+ ],
32
+ [
33
+ 24,
34
+ "block_output",
35
+ "pos",
36
+ 1,
37
+ 4,
38
+ null,
39
+ null,
40
+ null,
41
+ null,
42
+ null,
43
+ null,
44
+ null,
45
+ null,
46
+ null
47
+ ]
48
+ ],
49
+ "sorted_keys": [
50
+ "layer.22.comp.block_output.unit.pos.nunit.1#0",
51
+ "layer.24.comp.block_output.unit.pos.nunit.1#0"
52
+ ],
53
+ "transformers_version": "4.45.2"
54
+ }
intervenable_model/intkey_layer.22.comp.block_output.unit.pos.nunit.1#0.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d0acd9690863e5ce15eb3ff18b6730a99a5a6758b23aa2cc4300d15d9ae7496b
3
+ size 82721
intervenable_model/intkey_layer.24.comp.block_output.unit.pos.nunit.1#0.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b2c3b27973dc642900401f2edf0a3d38690e4e4a1f91f06f7e061c94b5947813
3
+ size 82721
intervenable_model/pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:049c26b844b79121ddd8379f7f69194e63f6fbf6aa007eeac0c66f17eebb8893
3
+ size 888
optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c1e1e3a602527b2607cdcabe51786782f779d1603eaa34bfaca517eadef9792
3
+ size 328448
rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3ceda789da21d0a82a08e7169a02f149733976358427f19a0644fca78fcf4fed
3
+ size 14244
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c64f2a890c26441da209d4d2660be5225e61c96fb16c8e91396996e345ad9e60
3
+ size 1064
trainer_state.json ADDED
@@ -0,0 +1,1633 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 40.0,
5
+ "eval_steps": 500,
6
+ "global_step": 1000,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.36,
13
+ "forget_cf_outputs.loss": -2.683056592941284,
14
+ "forget_loss": 2.683056592941284,
15
+ "gated_loss": 0.50390625,
16
+ "retain_loss": 0.2806420624256134,
17
+ "step": 9,
18
+ "warm_up_unlearning_weight": 1
19
+ },
20
+ {
21
+ "epoch": 0.4,
22
+ "grad_norm": 12.48087215423584,
23
+ "learning_rate": 0.00396,
24
+ "loss": 10.0439,
25
+ "step": 10
26
+ },
27
+ {
28
+ "epoch": 0.76,
29
+ "forget_cf_outputs.loss": -2.3281075954437256,
30
+ "forget_loss": 2.3281075954437256,
31
+ "gated_loss": 0.447265625,
32
+ "retain_loss": 0.30973613262176514,
33
+ "step": 19,
34
+ "warm_up_unlearning_weight": 1
35
+ },
36
+ {
37
+ "epoch": 0.8,
38
+ "grad_norm": 14.701934814453125,
39
+ "learning_rate": 0.00392,
40
+ "loss": 6.7906,
41
+ "step": 20
42
+ },
43
+ {
44
+ "epoch": 1.16,
45
+ "forget_cf_outputs.loss": -1.9086629152297974,
46
+ "forget_loss": 1.9086629152297974,
47
+ "gated_loss": 0.12890625,
48
+ "retain_loss": 0.25592905282974243,
49
+ "step": 29,
50
+ "warm_up_unlearning_weight": 1
51
+ },
52
+ {
53
+ "epoch": 1.2,
54
+ "grad_norm": 6.575860500335693,
55
+ "learning_rate": 0.0038799999999999998,
56
+ "loss": 5.0776,
57
+ "step": 30
58
+ },
59
+ {
60
+ "epoch": 1.56,
61
+ "forget_cf_outputs.loss": -1.8418117761611938,
62
+ "forget_loss": 1.8418117761611938,
63
+ "gated_loss": 0.26953125,
64
+ "retain_loss": 0.2785710096359253,
65
+ "step": 39,
66
+ "warm_up_unlearning_weight": 1
67
+ },
68
+ {
69
+ "epoch": 1.6,
70
+ "grad_norm": 16.85633087158203,
71
+ "learning_rate": 0.00384,
72
+ "loss": 4.3983,
73
+ "step": 40
74
+ },
75
+ {
76
+ "epoch": 1.96,
77
+ "forget_cf_outputs.loss": -1.5592725276947021,
78
+ "forget_loss": 1.5592725276947021,
79
+ "gated_loss": 0.09228515625,
80
+ "retain_loss": 0.2128174901008606,
81
+ "step": 49,
82
+ "warm_up_unlearning_weight": 1
83
+ },
84
+ {
85
+ "epoch": 2.0,
86
+ "grad_norm": 6.894440650939941,
87
+ "learning_rate": 0.0038,
88
+ "loss": 3.8443,
89
+ "step": 50
90
+ },
91
+ {
92
+ "epoch": 2.36,
93
+ "forget_cf_outputs.loss": -1.8075147867202759,
94
+ "forget_loss": 1.8075147867202759,
95
+ "gated_loss": 0.2099609375,
96
+ "retain_loss": 0.20431871712207794,
97
+ "step": 59,
98
+ "warm_up_unlearning_weight": 1
99
+ },
100
+ {
101
+ "epoch": 2.4,
102
+ "grad_norm": 12.495519638061523,
103
+ "learning_rate": 0.00376,
104
+ "loss": 3.1611,
105
+ "step": 60
106
+ },
107
+ {
108
+ "epoch": 2.76,
109
+ "forget_cf_outputs.loss": -1.648992896080017,
110
+ "forget_loss": 1.648992896080017,
111
+ "gated_loss": 0.08447265625,
112
+ "retain_loss": 0.23505711555480957,
113
+ "step": 69,
114
+ "warm_up_unlearning_weight": 1
115
+ },
116
+ {
117
+ "epoch": 2.8,
118
+ "grad_norm": 4.081427574157715,
119
+ "learning_rate": 0.00372,
120
+ "loss": 3.1274,
121
+ "step": 70
122
+ },
123
+ {
124
+ "epoch": 3.16,
125
+ "forget_cf_outputs.loss": -1.7901757955551147,
126
+ "forget_loss": 1.7901757955551147,
127
+ "gated_loss": 0.07861328125,
128
+ "retain_loss": 0.19908858835697174,
129
+ "step": 79,
130
+ "warm_up_unlearning_weight": 1
131
+ },
132
+ {
133
+ "epoch": 3.2,
134
+ "grad_norm": 5.019917011260986,
135
+ "learning_rate": 0.00368,
136
+ "loss": 2.9064,
137
+ "step": 80
138
+ },
139
+ {
140
+ "epoch": 3.56,
141
+ "forget_cf_outputs.loss": -1.6832540035247803,
142
+ "forget_loss": 1.6832540035247803,
143
+ "gated_loss": 0.1572265625,
144
+ "retain_loss": 0.31473198533058167,
145
+ "step": 89,
146
+ "warm_up_unlearning_weight": 1
147
+ },
148
+ {
149
+ "epoch": 3.6,
150
+ "grad_norm": 25.354310989379883,
151
+ "learning_rate": 0.00364,
152
+ "loss": 2.7559,
153
+ "step": 90
154
+ },
155
+ {
156
+ "epoch": 3.96,
157
+ "forget_cf_outputs.loss": -1.5529592037200928,
158
+ "forget_loss": 1.5529592037200928,
159
+ "gated_loss": 0.0859375,
160
+ "retain_loss": 0.21396659314632416,
161
+ "step": 99,
162
+ "warm_up_unlearning_weight": 1
163
+ },
164
+ {
165
+ "epoch": 4.0,
166
+ "grad_norm": 13.010091781616211,
167
+ "learning_rate": 0.0036000000000000003,
168
+ "loss": 4.6507,
169
+ "step": 100
170
+ },
171
+ {
172
+ "epoch": 4.36,
173
+ "forget_cf_outputs.loss": -1.4749836921691895,
174
+ "forget_loss": 1.4749836921691895,
175
+ "gated_loss": 0.0751953125,
176
+ "retain_loss": 0.3322639465332031,
177
+ "step": 109,
178
+ "warm_up_unlearning_weight": 1
179
+ },
180
+ {
181
+ "epoch": 4.4,
182
+ "grad_norm": 19.40593147277832,
183
+ "learning_rate": 0.0035600000000000002,
184
+ "loss": 2.4806,
185
+ "step": 110
186
+ },
187
+ {
188
+ "epoch": 4.76,
189
+ "forget_cf_outputs.loss": -1.586808443069458,
190
+ "forget_loss": 1.586808443069458,
191
+ "gated_loss": 0.044677734375,
192
+ "retain_loss": 0.20972619950771332,
193
+ "step": 119,
194
+ "warm_up_unlearning_weight": 1
195
+ },
196
+ {
197
+ "epoch": 4.8,
198
+ "grad_norm": 3.9136083126068115,
199
+ "learning_rate": 0.00352,
200
+ "loss": 2.3875,
201
+ "step": 120
202
+ },
203
+ {
204
+ "epoch": 5.16,
205
+ "forget_cf_outputs.loss": -1.3828651905059814,
206
+ "forget_loss": 1.3828651905059814,
207
+ "gated_loss": 0.0308837890625,
208
+ "retain_loss": 0.2637632191181183,
209
+ "step": 129,
210
+ "warm_up_unlearning_weight": 1
211
+ },
212
+ {
213
+ "epoch": 5.2,
214
+ "grad_norm": 2.676300525665283,
215
+ "learning_rate": 0.00348,
216
+ "loss": 2.3602,
217
+ "step": 130
218
+ },
219
+ {
220
+ "epoch": 5.5600000000000005,
221
+ "forget_cf_outputs.loss": -1.353875756263733,
222
+ "forget_loss": 1.353875756263733,
223
+ "gated_loss": 0.03076171875,
224
+ "retain_loss": 0.2234891653060913,
225
+ "step": 139,
226
+ "warm_up_unlearning_weight": 1
227
+ },
228
+ {
229
+ "epoch": 5.6,
230
+ "grad_norm": 2.6358110904693604,
231
+ "learning_rate": 0.00344,
232
+ "loss": 2.14,
233
+ "step": 140
234
+ },
235
+ {
236
+ "epoch": 5.96,
237
+ "forget_cf_outputs.loss": -1.4662984609603882,
238
+ "forget_loss": 1.4662984609603882,
239
+ "gated_loss": 0.044921875,
240
+ "retain_loss": 0.22434721887111664,
241
+ "step": 149,
242
+ "warm_up_unlearning_weight": 1
243
+ },
244
+ {
245
+ "epoch": 6.0,
246
+ "grad_norm": 5.245201587677002,
247
+ "learning_rate": 0.0034,
248
+ "loss": 3.8186,
249
+ "step": 150
250
+ },
251
+ {
252
+ "epoch": 6.36,
253
+ "forget_cf_outputs.loss": -1.3115544319152832,
254
+ "forget_loss": 1.3115544319152832,
255
+ "gated_loss": 0.0208740234375,
256
+ "retain_loss": 0.27212581038475037,
257
+ "step": 159,
258
+ "warm_up_unlearning_weight": 1
259
+ },
260
+ {
261
+ "epoch": 6.4,
262
+ "grad_norm": 2.41806697845459,
263
+ "learning_rate": 0.00336,
264
+ "loss": 2.0658,
265
+ "step": 160
266
+ },
267
+ {
268
+ "epoch": 6.76,
269
+ "forget_cf_outputs.loss": -1.389865517616272,
270
+ "forget_loss": 1.389865517616272,
271
+ "gated_loss": 0.0157470703125,
272
+ "retain_loss": 0.22109903395175934,
273
+ "step": 169,
274
+ "warm_up_unlearning_weight": 1
275
+ },
276
+ {
277
+ "epoch": 6.8,
278
+ "grad_norm": 1.6761163473129272,
279
+ "learning_rate": 0.00332,
280
+ "loss": 2.0436,
281
+ "step": 170
282
+ },
283
+ {
284
+ "epoch": 7.16,
285
+ "forget_cf_outputs.loss": -1.4114540815353394,
286
+ "forget_loss": 1.4114540815353394,
287
+ "gated_loss": 0.01953125,
288
+ "retain_loss": 0.28868579864501953,
289
+ "step": 179,
290
+ "warm_up_unlearning_weight": 1
291
+ },
292
+ {
293
+ "epoch": 7.2,
294
+ "grad_norm": 2.110581159591675,
295
+ "learning_rate": 0.00328,
296
+ "loss": 1.9892,
297
+ "step": 180
298
+ },
299
+ {
300
+ "epoch": 7.5600000000000005,
301
+ "forget_cf_outputs.loss": -1.5645467042922974,
302
+ "forget_loss": 1.5645467042922974,
303
+ "gated_loss": 0.0135498046875,
304
+ "retain_loss": 0.18681125342845917,
305
+ "step": 189,
306
+ "warm_up_unlearning_weight": 1
307
+ },
308
+ {
309
+ "epoch": 7.6,
310
+ "grad_norm": 2.8026249408721924,
311
+ "learning_rate": 0.0032400000000000003,
312
+ "loss": 1.9509,
313
+ "step": 190
314
+ },
315
+ {
316
+ "epoch": 7.96,
317
+ "forget_cf_outputs.loss": -1.368323802947998,
318
+ "forget_loss": 1.368323802947998,
319
+ "gated_loss": 0.0142822265625,
320
+ "retain_loss": 0.23697978258132935,
321
+ "step": 199,
322
+ "warm_up_unlearning_weight": 1
323
+ },
324
+ {
325
+ "epoch": 8.0,
326
+ "grad_norm": 1.9209412336349487,
327
+ "learning_rate": 0.0032,
328
+ "loss": 1.8948,
329
+ "step": 200
330
+ },
331
+ {
332
+ "epoch": 8.36,
333
+ "forget_cf_outputs.loss": -1.199570894241333,
334
+ "forget_loss": 1.199570894241333,
335
+ "gated_loss": 0.00933837890625,
336
+ "retain_loss": 0.25221553444862366,
337
+ "step": 209,
338
+ "warm_up_unlearning_weight": 1
339
+ },
340
+ {
341
+ "epoch": 8.4,
342
+ "grad_norm": 1.5988783836364746,
343
+ "learning_rate": 0.00316,
344
+ "loss": 1.8466,
345
+ "step": 210
346
+ },
347
+ {
348
+ "epoch": 8.76,
349
+ "forget_cf_outputs.loss": -1.2841148376464844,
350
+ "forget_loss": 1.2841148376464844,
351
+ "gated_loss": 0.009765625,
352
+ "retain_loss": 0.24307847023010254,
353
+ "step": 219,
354
+ "warm_up_unlearning_weight": 1
355
+ },
356
+ {
357
+ "epoch": 8.8,
358
+ "grad_norm": 1.7333821058273315,
359
+ "learning_rate": 0.0031200000000000004,
360
+ "loss": 1.817,
361
+ "step": 220
362
+ },
363
+ {
364
+ "epoch": 9.16,
365
+ "forget_cf_outputs.loss": -1.0533504486083984,
366
+ "forget_loss": 1.0533504486083984,
367
+ "gated_loss": 0.005462646484375,
368
+ "retain_loss": 0.22837677597999573,
369
+ "step": 229,
370
+ "warm_up_unlearning_weight": 1
371
+ },
372
+ {
373
+ "epoch": 9.2,
374
+ "grad_norm": 1.5669026374816895,
375
+ "learning_rate": 0.0030800000000000003,
376
+ "loss": 1.7654,
377
+ "step": 230
378
+ },
379
+ {
380
+ "epoch": 9.56,
381
+ "forget_cf_outputs.loss": -1.2518203258514404,
382
+ "forget_loss": 1.2518203258514404,
383
+ "gated_loss": 0.0107421875,
384
+ "retain_loss": 0.27540236711502075,
385
+ "step": 239,
386
+ "warm_up_unlearning_weight": 1
387
+ },
388
+ {
389
+ "epoch": 9.6,
390
+ "grad_norm": 1.7850066423416138,
391
+ "learning_rate": 0.00304,
392
+ "loss": 1.7462,
393
+ "step": 240
394
+ },
395
+ {
396
+ "epoch": 9.96,
397
+ "forget_cf_outputs.loss": -1.2480342388153076,
398
+ "forget_loss": 1.2480342388153076,
399
+ "gated_loss": 0.0079345703125,
400
+ "retain_loss": 0.2164781242609024,
401
+ "step": 249,
402
+ "warm_up_unlearning_weight": 1
403
+ },
404
+ {
405
+ "epoch": 10.0,
406
+ "grad_norm": 1.8285338878631592,
407
+ "learning_rate": 0.003,
408
+ "loss": 1.8334,
409
+ "step": 250
410
+ },
411
+ {
412
+ "epoch": 10.36,
413
+ "forget_cf_outputs.loss": -1.14595365524292,
414
+ "forget_loss": 1.14595365524292,
415
+ "gated_loss": 0.006072998046875,
416
+ "retain_loss": 0.22983184456825256,
417
+ "step": 259,
418
+ "warm_up_unlearning_weight": 1
419
+ },
420
+ {
421
+ "epoch": 10.4,
422
+ "grad_norm": 1.8580724000930786,
423
+ "learning_rate": 0.00296,
424
+ "loss": 1.6553,
425
+ "step": 260
426
+ },
427
+ {
428
+ "epoch": 10.76,
429
+ "forget_cf_outputs.loss": -1.2968733310699463,
430
+ "forget_loss": 1.2968733310699463,
431
+ "gated_loss": 0.008544921875,
432
+ "retain_loss": 0.20906659960746765,
433
+ "step": 269,
434
+ "warm_up_unlearning_weight": 1
435
+ },
436
+ {
437
+ "epoch": 10.8,
438
+ "grad_norm": 1.6227186918258667,
439
+ "learning_rate": 0.00292,
440
+ "loss": 1.7339,
441
+ "step": 270
442
+ },
443
+ {
444
+ "epoch": 11.16,
445
+ "forget_cf_outputs.loss": -1.21259605884552,
446
+ "forget_loss": 1.21259605884552,
447
+ "gated_loss": 0.00750732421875,
448
+ "retain_loss": 0.23951691389083862,
449
+ "step": 279,
450
+ "warm_up_unlearning_weight": 1
451
+ },
452
+ {
453
+ "epoch": 11.2,
454
+ "grad_norm": 1.3491921424865723,
455
+ "learning_rate": 0.0028799999999999997,
456
+ "loss": 1.6642,
457
+ "step": 280
458
+ },
459
+ {
460
+ "epoch": 11.56,
461
+ "forget_cf_outputs.loss": -1.203460454940796,
462
+ "forget_loss": 1.203460454940796,
463
+ "gated_loss": 0.0087890625,
464
+ "retain_loss": 0.23031194508075714,
465
+ "step": 289,
466
+ "warm_up_unlearning_weight": 1
467
+ },
468
+ {
469
+ "epoch": 11.6,
470
+ "grad_norm": 2.027022123336792,
471
+ "learning_rate": 0.00284,
472
+ "loss": 1.6413,
473
+ "step": 290
474
+ },
475
+ {
476
+ "epoch": 11.96,
477
+ "forget_cf_outputs.loss": -1.1864365339279175,
478
+ "forget_loss": 1.1864365339279175,
479
+ "gated_loss": 0.01043701171875,
480
+ "retain_loss": 0.23127436637878418,
481
+ "step": 299,
482
+ "warm_up_unlearning_weight": 1
483
+ },
484
+ {
485
+ "epoch": 12.0,
486
+ "grad_norm": 1.8040319681167603,
487
+ "learning_rate": 0.0028,
488
+ "loss": 1.761,
489
+ "step": 300
490
+ },
491
+ {
492
+ "epoch": 12.36,
493
+ "forget_cf_outputs.loss": -1.227767825126648,
494
+ "forget_loss": 1.227767825126648,
495
+ "gated_loss": 0.00616455078125,
496
+ "retain_loss": 0.22675465047359467,
497
+ "step": 309,
498
+ "warm_up_unlearning_weight": 1
499
+ },
500
+ {
501
+ "epoch": 12.4,
502
+ "grad_norm": 2.1620850563049316,
503
+ "learning_rate": 0.00276,
504
+ "loss": 1.5947,
505
+ "step": 310
506
+ },
507
+ {
508
+ "epoch": 12.76,
509
+ "forget_cf_outputs.loss": -1.0549200773239136,
510
+ "forget_loss": 1.0549200773239136,
511
+ "gated_loss": 0.0054931640625,
512
+ "retain_loss": 0.225913867354393,
513
+ "step": 319,
514
+ "warm_up_unlearning_weight": 1
515
+ },
516
+ {
517
+ "epoch": 12.8,
518
+ "grad_norm": 1.422467827796936,
519
+ "learning_rate": 0.00272,
520
+ "loss": 1.6839,
521
+ "step": 320
522
+ },
523
+ {
524
+ "epoch": 13.16,
525
+ "forget_cf_outputs.loss": -0.966583251953125,
526
+ "forget_loss": 0.966583251953125,
527
+ "gated_loss": 0.0050048828125,
528
+ "retain_loss": 0.21935530006885529,
529
+ "step": 329,
530
+ "warm_up_unlearning_weight": 1
531
+ },
532
+ {
533
+ "epoch": 13.2,
534
+ "grad_norm": 1.3767800331115723,
535
+ "learning_rate": 0.00268,
536
+ "loss": 1.6052,
537
+ "step": 330
538
+ },
539
+ {
540
+ "epoch": 13.56,
541
+ "forget_cf_outputs.loss": -1.0747449398040771,
542
+ "forget_loss": 1.0747449398040771,
543
+ "gated_loss": 0.008544921875,
544
+ "retain_loss": 0.23962758481502533,
545
+ "step": 339,
546
+ "warm_up_unlearning_weight": 1
547
+ },
548
+ {
549
+ "epoch": 13.6,
550
+ "grad_norm": 2.152151584625244,
551
+ "learning_rate": 0.00264,
552
+ "loss": 1.6239,
553
+ "step": 340
554
+ },
555
+ {
556
+ "epoch": 13.96,
557
+ "forget_cf_outputs.loss": -0.9583653211593628,
558
+ "forget_loss": 0.9583653211593628,
559
+ "gated_loss": 0.006500244140625,
560
+ "retain_loss": 0.21241800487041473,
561
+ "step": 349,
562
+ "warm_up_unlearning_weight": 1
563
+ },
564
+ {
565
+ "epoch": 14.0,
566
+ "grad_norm": 2.0913257598876953,
567
+ "learning_rate": 0.0026000000000000003,
568
+ "loss": 1.566,
569
+ "step": 350
570
+ },
571
+ {
572
+ "epoch": 14.36,
573
+ "forget_cf_outputs.loss": -0.8475239276885986,
574
+ "forget_loss": 0.8475239276885986,
575
+ "gated_loss": 0.01165771484375,
576
+ "retain_loss": 0.22438839077949524,
577
+ "step": 359,
578
+ "warm_up_unlearning_weight": 1
579
+ },
580
+ {
581
+ "epoch": 14.4,
582
+ "grad_norm": 1.8125321865081787,
583
+ "learning_rate": 0.00256,
584
+ "loss": 1.52,
585
+ "step": 360
586
+ },
587
+ {
588
+ "epoch": 14.76,
589
+ "forget_cf_outputs.loss": -1.2125965356826782,
590
+ "forget_loss": 1.2125965356826782,
591
+ "gated_loss": 0.004974365234375,
592
+ "retain_loss": 0.23040254414081573,
593
+ "step": 369,
594
+ "warm_up_unlearning_weight": 1
595
+ },
596
+ {
597
+ "epoch": 14.8,
598
+ "grad_norm": 1.811591386795044,
599
+ "learning_rate": 0.00252,
600
+ "loss": 1.5403,
601
+ "step": 370
602
+ },
603
+ {
604
+ "epoch": 15.16,
605
+ "forget_cf_outputs.loss": -0.9788862466812134,
606
+ "forget_loss": 0.9788862466812134,
607
+ "gated_loss": 0.006439208984375,
608
+ "retain_loss": 0.24656128883361816,
609
+ "step": 379,
610
+ "warm_up_unlearning_weight": 1
611
+ },
612
+ {
613
+ "epoch": 15.2,
614
+ "grad_norm": 1.5504097938537598,
615
+ "learning_rate": 0.00248,
616
+ "loss": 1.5363,
617
+ "step": 380
618
+ },
619
+ {
620
+ "epoch": 15.56,
621
+ "forget_cf_outputs.loss": -0.8573880195617676,
622
+ "forget_loss": 0.8573880195617676,
623
+ "gated_loss": 0.007080078125,
624
+ "retain_loss": 0.2557204067707062,
625
+ "step": 389,
626
+ "warm_up_unlearning_weight": 1
627
+ },
628
+ {
629
+ "epoch": 15.6,
630
+ "grad_norm": 1.5796666145324707,
631
+ "learning_rate": 0.00244,
632
+ "loss": 1.4844,
633
+ "step": 390
634
+ },
635
+ {
636
+ "epoch": 15.96,
637
+ "forget_cf_outputs.loss": -1.0210211277008057,
638
+ "forget_loss": 1.0210211277008057,
639
+ "gated_loss": 0.007080078125,
640
+ "retain_loss": 0.22587800025939941,
641
+ "step": 399,
642
+ "warm_up_unlearning_weight": 1
643
+ },
644
+ {
645
+ "epoch": 16.0,
646
+ "grad_norm": 1.902213215827942,
647
+ "learning_rate": 0.0024,
648
+ "loss": 1.5053,
649
+ "step": 400
650
+ },
651
+ {
652
+ "epoch": 16.36,
653
+ "forget_cf_outputs.loss": -0.7990767955780029,
654
+ "forget_loss": 0.7990767955780029,
655
+ "gated_loss": 0.0064697265625,
656
+ "retain_loss": 0.20731617510318756,
657
+ "step": 409,
658
+ "warm_up_unlearning_weight": 1
659
+ },
660
+ {
661
+ "epoch": 16.4,
662
+ "grad_norm": 1.7532883882522583,
663
+ "learning_rate": 0.00236,
664
+ "loss": 1.3702,
665
+ "step": 410
666
+ },
667
+ {
668
+ "epoch": 16.76,
669
+ "forget_cf_outputs.loss": -1.08427095413208,
670
+ "forget_loss": 1.08427095413208,
671
+ "gated_loss": 0.005950927734375,
672
+ "retain_loss": 0.23056094348430634,
673
+ "step": 419,
674
+ "warm_up_unlearning_weight": 1
675
+ },
676
+ {
677
+ "epoch": 16.8,
678
+ "grad_norm": 1.8573689460754395,
679
+ "learning_rate": 0.00232,
680
+ "loss": 1.4936,
681
+ "step": 420
682
+ },
683
+ {
684
+ "epoch": 17.16,
685
+ "forget_cf_outputs.loss": -0.8196872472763062,
686
+ "forget_loss": 0.8196872472763062,
687
+ "gated_loss": 0.006805419921875,
688
+ "retain_loss": 0.3019656836986542,
689
+ "step": 429,
690
+ "warm_up_unlearning_weight": 1
691
+ },
692
+ {
693
+ "epoch": 17.2,
694
+ "grad_norm": 1.6391781568527222,
695
+ "learning_rate": 0.00228,
696
+ "loss": 1.5101,
697
+ "step": 430
698
+ },
699
+ {
700
+ "epoch": 17.56,
701
+ "forget_cf_outputs.loss": -0.8601583242416382,
702
+ "forget_loss": 0.8601583242416382,
703
+ "gated_loss": 0.00677490234375,
704
+ "retain_loss": 0.30145329236984253,
705
+ "step": 439,
706
+ "warm_up_unlearning_weight": 1
707
+ },
708
+ {
709
+ "epoch": 17.6,
710
+ "grad_norm": 1.7495087385177612,
711
+ "learning_rate": 0.0022400000000000002,
712
+ "loss": 1.4067,
713
+ "step": 440
714
+ },
715
+ {
716
+ "epoch": 17.96,
717
+ "forget_cf_outputs.loss": -0.9261890053749084,
718
+ "forget_loss": 0.9261890053749084,
719
+ "gated_loss": 0.007110595703125,
720
+ "retain_loss": 0.2040254771709442,
721
+ "step": 449,
722
+ "warm_up_unlearning_weight": 1
723
+ },
724
+ {
725
+ "epoch": 18.0,
726
+ "grad_norm": 2.1609787940979004,
727
+ "learning_rate": 0.0022,
728
+ "loss": 1.3843,
729
+ "step": 450
730
+ },
731
+ {
732
+ "epoch": 18.36,
733
+ "forget_cf_outputs.loss": -0.7474625706672668,
734
+ "forget_loss": 0.7474625706672668,
735
+ "gated_loss": 0.006103515625,
736
+ "retain_loss": 0.20823714137077332,
737
+ "step": 459,
738
+ "warm_up_unlearning_weight": 1
739
+ },
740
+ {
741
+ "epoch": 18.4,
742
+ "grad_norm": 1.997226595878601,
743
+ "learning_rate": 0.00216,
744
+ "loss": 1.3081,
745
+ "step": 460
746
+ },
747
+ {
748
+ "epoch": 18.76,
749
+ "forget_cf_outputs.loss": -1.0060858726501465,
750
+ "forget_loss": 1.0060858726501465,
751
+ "gated_loss": 0.00579833984375,
752
+ "retain_loss": 0.23037730157375336,
753
+ "step": 469,
754
+ "warm_up_unlearning_weight": 1
755
+ },
756
+ {
757
+ "epoch": 18.8,
758
+ "grad_norm": 2.023531675338745,
759
+ "learning_rate": 0.0021200000000000004,
760
+ "loss": 1.4162,
761
+ "step": 470
762
+ },
763
+ {
764
+ "epoch": 19.16,
765
+ "forget_cf_outputs.loss": -0.7942442893981934,
766
+ "forget_loss": 0.7942442893981934,
767
+ "gated_loss": 0.00677490234375,
768
+ "retain_loss": 0.30291858315467834,
769
+ "step": 479,
770
+ "warm_up_unlearning_weight": 1
771
+ },
772
+ {
773
+ "epoch": 19.2,
774
+ "grad_norm": 2.0801267623901367,
775
+ "learning_rate": 0.0020800000000000003,
776
+ "loss": 1.4455,
777
+ "step": 480
778
+ },
779
+ {
780
+ "epoch": 19.56,
781
+ "forget_cf_outputs.loss": -0.7663432955741882,
782
+ "forget_loss": 0.7663432955741882,
783
+ "gated_loss": 0.00640869140625,
784
+ "retain_loss": 0.3002184331417084,
785
+ "step": 489,
786
+ "warm_up_unlearning_weight": 1
787
+ },
788
+ {
789
+ "epoch": 19.6,
790
+ "grad_norm": 1.6843624114990234,
791
+ "learning_rate": 0.00204,
792
+ "loss": 1.3348,
793
+ "step": 490
794
+ },
795
+ {
796
+ "epoch": 19.96,
797
+ "forget_cf_outputs.loss": -0.8520928025245667,
798
+ "forget_loss": 0.8520928025245667,
799
+ "gated_loss": 0.006591796875,
800
+ "retain_loss": 0.20439262688159943,
801
+ "step": 499,
802
+ "warm_up_unlearning_weight": 1
803
+ },
804
+ {
805
+ "epoch": 20.0,
806
+ "grad_norm": 2.177591323852539,
807
+ "learning_rate": 0.002,
808
+ "loss": 1.3115,
809
+ "step": 500
810
+ },
811
+ {
812
+ "epoch": 20.36,
813
+ "forget_cf_outputs.loss": -0.7113033533096313,
814
+ "forget_loss": 0.7113033533096313,
815
+ "gated_loss": 0.0062255859375,
816
+ "retain_loss": 0.20779718458652496,
817
+ "step": 509,
818
+ "warm_up_unlearning_weight": 1
819
+ },
820
+ {
821
+ "epoch": 20.4,
822
+ "grad_norm": 2.2269105911254883,
823
+ "learning_rate": 0.00196,
824
+ "loss": 1.2523,
825
+ "step": 510
826
+ },
827
+ {
828
+ "epoch": 20.76,
829
+ "forget_cf_outputs.loss": -0.9132112860679626,
830
+ "forget_loss": 0.9132112860679626,
831
+ "gated_loss": 0.005615234375,
832
+ "retain_loss": 0.23009441792964935,
833
+ "step": 519,
834
+ "warm_up_unlearning_weight": 1
835
+ },
836
+ {
837
+ "epoch": 20.8,
838
+ "grad_norm": 1.955623745918274,
839
+ "learning_rate": 0.00192,
840
+ "loss": 1.3478,
841
+ "step": 520
842
+ },
843
+ {
844
+ "epoch": 21.16,
845
+ "forget_cf_outputs.loss": -0.7705625891685486,
846
+ "forget_loss": 0.7705625891685486,
847
+ "gated_loss": 0.00677490234375,
848
+ "retain_loss": 0.30292266607284546,
849
+ "step": 529,
850
+ "warm_up_unlearning_weight": 1
851
+ },
852
+ {
853
+ "epoch": 21.2,
854
+ "grad_norm": 2.0390868186950684,
855
+ "learning_rate": 0.00188,
856
+ "loss": 1.3883,
857
+ "step": 530
858
+ },
859
+ {
860
+ "epoch": 21.56,
861
+ "forget_cf_outputs.loss": -0.7298972606658936,
862
+ "forget_loss": 0.7298972606658936,
863
+ "gated_loss": 0.006195068359375,
864
+ "retain_loss": 0.3011211156845093,
865
+ "step": 539,
866
+ "warm_up_unlearning_weight": 1
867
+ },
868
+ {
869
+ "epoch": 21.6,
870
+ "grad_norm": 1.9718871116638184,
871
+ "learning_rate": 0.00184,
872
+ "loss": 1.2773,
873
+ "step": 540
874
+ },
875
+ {
876
+ "epoch": 21.96,
877
+ "forget_cf_outputs.loss": -0.7989807724952698,
878
+ "forget_loss": 0.7989807724952698,
879
+ "gated_loss": 0.00640869140625,
880
+ "retain_loss": 0.20371052622795105,
881
+ "step": 549,
882
+ "warm_up_unlearning_weight": 1
883
+ },
884
+ {
885
+ "epoch": 22.0,
886
+ "grad_norm": 2.304124116897583,
887
+ "learning_rate": 0.0018000000000000002,
888
+ "loss": 1.2614,
889
+ "step": 550
890
+ },
891
+ {
892
+ "epoch": 22.36,
893
+ "forget_cf_outputs.loss": -0.6765010952949524,
894
+ "forget_loss": 0.6765010952949524,
895
+ "gated_loss": 0.006134033203125,
896
+ "retain_loss": 0.20700086653232574,
897
+ "step": 559,
898
+ "warm_up_unlearning_weight": 1
899
+ },
900
+ {
901
+ "epoch": 22.4,
902
+ "grad_norm": 2.3407692909240723,
903
+ "learning_rate": 0.00176,
904
+ "loss": 1.2052,
905
+ "step": 560
906
+ },
907
+ {
908
+ "epoch": 22.76,
909
+ "forget_cf_outputs.loss": -0.8482251167297363,
910
+ "forget_loss": 0.8482251167297363,
911
+ "gated_loss": 0.00555419921875,
912
+ "retain_loss": 0.23026637732982635,
913
+ "step": 569,
914
+ "warm_up_unlearning_weight": 1
915
+ },
916
+ {
917
+ "epoch": 22.8,
918
+ "grad_norm": 1.9227235317230225,
919
+ "learning_rate": 0.00172,
920
+ "loss": 1.2995,
921
+ "step": 570
922
+ },
923
+ {
924
+ "epoch": 23.16,
925
+ "forget_cf_outputs.loss": -0.7452784776687622,
926
+ "forget_loss": 0.7452784776687622,
927
+ "gated_loss": 0.006744384765625,
928
+ "retain_loss": 0.30324649810791016,
929
+ "step": 579,
930
+ "warm_up_unlearning_weight": 1
931
+ },
932
+ {
933
+ "epoch": 23.2,
934
+ "grad_norm": 2.1659557819366455,
935
+ "learning_rate": 0.00168,
936
+ "loss": 1.3367,
937
+ "step": 580
938
+ },
939
+ {
940
+ "epoch": 23.56,
941
+ "forget_cf_outputs.loss": -0.6973183155059814,
942
+ "forget_loss": 0.6973183155059814,
943
+ "gated_loss": 0.006011962890625,
944
+ "retain_loss": 0.30067041516304016,
945
+ "step": 589,
946
+ "warm_up_unlearning_weight": 1
947
+ },
948
+ {
949
+ "epoch": 23.6,
950
+ "grad_norm": 2.140836238861084,
951
+ "learning_rate": 0.00164,
952
+ "loss": 1.2336,
953
+ "step": 590
954
+ },
955
+ {
956
+ "epoch": 23.96,
957
+ "forget_cf_outputs.loss": -0.7499862313270569,
958
+ "forget_loss": 0.7499862313270569,
959
+ "gated_loss": 0.006317138671875,
960
+ "retain_loss": 0.20415130257606506,
961
+ "step": 599,
962
+ "warm_up_unlearning_weight": 1
963
+ },
964
+ {
965
+ "epoch": 24.0,
966
+ "grad_norm": 2.5653772354125977,
967
+ "learning_rate": 0.0016,
968
+ "loss": 1.2199,
969
+ "step": 600
970
+ },
971
+ {
972
+ "epoch": 24.36,
973
+ "forget_cf_outputs.loss": -0.6281754374504089,
974
+ "forget_loss": 0.6281754374504089,
975
+ "gated_loss": 0.006103515625,
976
+ "retain_loss": 0.2073870748281479,
977
+ "step": 609,
978
+ "warm_up_unlearning_weight": 1
979
+ },
980
+ {
981
+ "epoch": 24.4,
982
+ "grad_norm": 2.2012555599212646,
983
+ "learning_rate": 0.0015600000000000002,
984
+ "loss": 1.1692,
985
+ "step": 610
986
+ },
987
+ {
988
+ "epoch": 24.76,
989
+ "forget_cf_outputs.loss": -0.7929825782775879,
990
+ "forget_loss": 0.7929825782775879,
991
+ "gated_loss": 0.005584716796875,
992
+ "retain_loss": 0.23124848306179047,
993
+ "step": 619,
994
+ "warm_up_unlearning_weight": 1
995
+ },
996
+ {
997
+ "epoch": 24.8,
998
+ "grad_norm": 1.9918410778045654,
999
+ "learning_rate": 0.00152,
1000
+ "loss": 1.2547,
1001
+ "step": 620
1002
+ },
1003
+ {
1004
+ "epoch": 25.16,
1005
+ "forget_cf_outputs.loss": -0.7136563062667847,
1006
+ "forget_loss": 0.7136563062667847,
1007
+ "gated_loss": 0.00677490234375,
1008
+ "retain_loss": 0.3019442558288574,
1009
+ "step": 629,
1010
+ "warm_up_unlearning_weight": 1
1011
+ },
1012
+ {
1013
+ "epoch": 25.2,
1014
+ "grad_norm": 2.17232608795166,
1015
+ "learning_rate": 0.00148,
1016
+ "loss": 1.2906,
1017
+ "step": 630
1018
+ },
1019
+ {
1020
+ "epoch": 25.56,
1021
+ "forget_cf_outputs.loss": -0.6733591556549072,
1022
+ "forget_loss": 0.6733591556549072,
1023
+ "gated_loss": 0.00604248046875,
1024
+ "retain_loss": 0.3010113835334778,
1025
+ "step": 639,
1026
+ "warm_up_unlearning_weight": 1
1027
+ },
1028
+ {
1029
+ "epoch": 25.6,
1030
+ "grad_norm": 2.1968352794647217,
1031
+ "learning_rate": 0.0014399999999999999,
1032
+ "loss": 1.198,
1033
+ "step": 640
1034
+ },
1035
+ {
1036
+ "epoch": 25.96,
1037
+ "forget_cf_outputs.loss": -0.6898148655891418,
1038
+ "forget_loss": 0.6898148655891418,
1039
+ "gated_loss": 0.00628662109375,
1040
+ "retain_loss": 0.20451340079307556,
1041
+ "step": 649,
1042
+ "warm_up_unlearning_weight": 1
1043
+ },
1044
+ {
1045
+ "epoch": 26.0,
1046
+ "grad_norm": 2.708749294281006,
1047
+ "learning_rate": 0.0014,
1048
+ "loss": 1.1784,
1049
+ "step": 650
1050
+ },
1051
+ {
1052
+ "epoch": 26.36,
1053
+ "forget_cf_outputs.loss": -0.5750948786735535,
1054
+ "forget_loss": 0.5750948786735535,
1055
+ "gated_loss": 0.006011962890625,
1056
+ "retain_loss": 0.20745757222175598,
1057
+ "step": 659,
1058
+ "warm_up_unlearning_weight": 1
1059
+ },
1060
+ {
1061
+ "epoch": 26.4,
1062
+ "grad_norm": 2.0222957134246826,
1063
+ "learning_rate": 0.00136,
1064
+ "loss": 1.1326,
1065
+ "step": 660
1066
+ },
1067
+ {
1068
+ "epoch": 26.76,
1069
+ "forget_cf_outputs.loss": -0.7313442826271057,
1070
+ "forget_loss": 0.7313442826271057,
1071
+ "gated_loss": 0.005523681640625,
1072
+ "retain_loss": 0.2309209704399109,
1073
+ "step": 669,
1074
+ "warm_up_unlearning_weight": 1
1075
+ },
1076
+ {
1077
+ "epoch": 26.8,
1078
+ "grad_norm": 1.8242149353027344,
1079
+ "learning_rate": 0.00132,
1080
+ "loss": 1.2022,
1081
+ "step": 670
1082
+ },
1083
+ {
1084
+ "epoch": 27.16,
1085
+ "forget_cf_outputs.loss": -0.6695391535758972,
1086
+ "forget_loss": 0.6695391535758972,
1087
+ "gated_loss": 0.006744384765625,
1088
+ "retain_loss": 0.30264630913734436,
1089
+ "step": 679,
1090
+ "warm_up_unlearning_weight": 1
1091
+ },
1092
+ {
1093
+ "epoch": 27.2,
1094
+ "grad_norm": 2.184037208557129,
1095
+ "learning_rate": 0.00128,
1096
+ "loss": 1.24,
1097
+ "step": 680
1098
+ },
1099
+ {
1100
+ "epoch": 27.56,
1101
+ "forget_cf_outputs.loss": -0.6346314549446106,
1102
+ "forget_loss": 0.6346314549446106,
1103
+ "gated_loss": 0.005950927734375,
1104
+ "retain_loss": 0.3012670874595642,
1105
+ "step": 689,
1106
+ "warm_up_unlearning_weight": 1
1107
+ },
1108
+ {
1109
+ "epoch": 27.6,
1110
+ "grad_norm": 2.071834087371826,
1111
+ "learning_rate": 0.00124,
1112
+ "loss": 1.1582,
1113
+ "step": 690
1114
+ },
1115
+ {
1116
+ "epoch": 27.96,
1117
+ "forget_cf_outputs.loss": -0.6438873410224915,
1118
+ "forget_loss": 0.6438873410224915,
1119
+ "gated_loss": 0.00634765625,
1120
+ "retain_loss": 0.20389188826084137,
1121
+ "step": 699,
1122
+ "warm_up_unlearning_weight": 1
1123
+ },
1124
+ {
1125
+ "epoch": 28.0,
1126
+ "grad_norm": 2.627547264099121,
1127
+ "learning_rate": 0.0012,
1128
+ "loss": 1.1426,
1129
+ "step": 700
1130
+ },
1131
+ {
1132
+ "epoch": 28.36,
1133
+ "forget_cf_outputs.loss": -0.5409132242202759,
1134
+ "forget_loss": 0.5409132242202759,
1135
+ "gated_loss": 0.00604248046875,
1136
+ "retain_loss": 0.20821160078048706,
1137
+ "step": 709,
1138
+ "warm_up_unlearning_weight": 1
1139
+ },
1140
+ {
1141
+ "epoch": 28.4,
1142
+ "grad_norm": 1.8756951093673706,
1143
+ "learning_rate": 0.00116,
1144
+ "loss": 1.0961,
1145
+ "step": 710
1146
+ },
1147
+ {
1148
+ "epoch": 28.76,
1149
+ "forget_cf_outputs.loss": -0.6844155788421631,
1150
+ "forget_loss": 0.6844155788421631,
1151
+ "gated_loss": 0.005523681640625,
1152
+ "retain_loss": 0.23081088066101074,
1153
+ "step": 719,
1154
+ "warm_up_unlearning_weight": 1
1155
+ },
1156
+ {
1157
+ "epoch": 28.8,
1158
+ "grad_norm": 1.882070541381836,
1159
+ "learning_rate": 0.0011200000000000001,
1160
+ "loss": 1.1636,
1161
+ "step": 720
1162
+ },
1163
+ {
1164
+ "epoch": 29.16,
1165
+ "forget_cf_outputs.loss": -0.6462154388427734,
1166
+ "forget_loss": 0.6462154388427734,
1167
+ "gated_loss": 0.0068359375,
1168
+ "retain_loss": 0.30321410298347473,
1169
+ "step": 729,
1170
+ "warm_up_unlearning_weight": 1
1171
+ },
1172
+ {
1173
+ "epoch": 29.2,
1174
+ "grad_norm": 2.2409791946411133,
1175
+ "learning_rate": 0.00108,
1176
+ "loss": 1.1954,
1177
+ "step": 730
1178
+ },
1179
+ {
1180
+ "epoch": 29.56,
1181
+ "forget_cf_outputs.loss": -0.5779778957366943,
1182
+ "forget_loss": 0.5779778957366943,
1183
+ "gated_loss": 0.005828857421875,
1184
+ "retain_loss": 0.301031231880188,
1185
+ "step": 739,
1186
+ "warm_up_unlearning_weight": 1
1187
+ },
1188
+ {
1189
+ "epoch": 29.6,
1190
+ "grad_norm": 1.9507259130477905,
1191
+ "learning_rate": 0.0010400000000000001,
1192
+ "loss": 1.1284,
1193
+ "step": 740
1194
+ },
1195
+ {
1196
+ "epoch": 29.96,
1197
+ "forget_cf_outputs.loss": -0.595399022102356,
1198
+ "forget_loss": 0.595399022102356,
1199
+ "gated_loss": 0.00628662109375,
1200
+ "retain_loss": 0.20331645011901855,
1201
+ "step": 749,
1202
+ "warm_up_unlearning_weight": 1
1203
+ },
1204
+ {
1205
+ "epoch": 30.0,
1206
+ "grad_norm": 2.5469954013824463,
1207
+ "learning_rate": 0.001,
1208
+ "loss": 1.1074,
1209
+ "step": 750
1210
+ },
1211
+ {
1212
+ "epoch": 30.36,
1213
+ "forget_cf_outputs.loss": -0.5063520669937134,
1214
+ "forget_loss": 0.5063520669937134,
1215
+ "gated_loss": 0.0059814453125,
1216
+ "retain_loss": 0.20728009939193726,
1217
+ "step": 759,
1218
+ "warm_up_unlearning_weight": 1
1219
+ },
1220
+ {
1221
+ "epoch": 30.4,
1222
+ "grad_norm": 1.9551663398742676,
1223
+ "learning_rate": 0.00096,
1224
+ "loss": 1.0691,
1225
+ "step": 760
1226
+ },
1227
+ {
1228
+ "epoch": 30.76,
1229
+ "forget_cf_outputs.loss": -0.6612439751625061,
1230
+ "forget_loss": 0.6612439751625061,
1231
+ "gated_loss": 0.005523681640625,
1232
+ "retain_loss": 0.23078040778636932,
1233
+ "step": 769,
1234
+ "warm_up_unlearning_weight": 1
1235
+ },
1236
+ {
1237
+ "epoch": 30.8,
1238
+ "grad_norm": 2.112478494644165,
1239
+ "learning_rate": 0.00092,
1240
+ "loss": 1.1296,
1241
+ "step": 770
1242
+ },
1243
+ {
1244
+ "epoch": 31.16,
1245
+ "forget_cf_outputs.loss": -0.6047573685646057,
1246
+ "forget_loss": 0.6047573685646057,
1247
+ "gated_loss": 0.00677490234375,
1248
+ "retain_loss": 0.3026810586452484,
1249
+ "step": 779,
1250
+ "warm_up_unlearning_weight": 1
1251
+ },
1252
+ {
1253
+ "epoch": 31.2,
1254
+ "grad_norm": 2.141299247741699,
1255
+ "learning_rate": 0.00088,
1256
+ "loss": 1.1566,
1257
+ "step": 780
1258
+ },
1259
+ {
1260
+ "epoch": 31.56,
1261
+ "forget_cf_outputs.loss": -0.5377554297447205,
1262
+ "forget_loss": 0.5377554297447205,
1263
+ "gated_loss": 0.005828857421875,
1264
+ "retain_loss": 0.30064091086387634,
1265
+ "step": 789,
1266
+ "warm_up_unlearning_weight": 1
1267
+ },
1268
+ {
1269
+ "epoch": 31.6,
1270
+ "grad_norm": 1.9649981260299683,
1271
+ "learning_rate": 0.00084,
1272
+ "loss": 1.097,
1273
+ "step": 790
1274
+ },
1275
+ {
1276
+ "epoch": 31.96,
1277
+ "forget_cf_outputs.loss": -0.5527829527854919,
1278
+ "forget_loss": 0.5527829527854919,
1279
+ "gated_loss": 0.0062255859375,
1280
+ "retain_loss": 0.20365992188453674,
1281
+ "step": 799,
1282
+ "warm_up_unlearning_weight": 1
1283
+ },
1284
+ {
1285
+ "epoch": 32.0,
1286
+ "grad_norm": 2.628004312515259,
1287
+ "learning_rate": 0.0008,
1288
+ "loss": 1.0695,
1289
+ "step": 800
1290
+ },
1291
+ {
1292
+ "epoch": 32.36,
1293
+ "forget_cf_outputs.loss": -0.47989267110824585,
1294
+ "forget_loss": 0.47989267110824585,
1295
+ "gated_loss": 0.005950927734375,
1296
+ "retain_loss": 0.20727092027664185,
1297
+ "step": 809,
1298
+ "warm_up_unlearning_weight": 1
1299
+ },
1300
+ {
1301
+ "epoch": 32.4,
1302
+ "grad_norm": 1.9921866655349731,
1303
+ "learning_rate": 0.00076,
1304
+ "loss": 1.0319,
1305
+ "step": 810
1306
+ },
1307
+ {
1308
+ "epoch": 32.76,
1309
+ "forget_cf_outputs.loss": -0.6170799136161804,
1310
+ "forget_loss": 0.6170799136161804,
1311
+ "gated_loss": 0.00555419921875,
1312
+ "retain_loss": 0.23075489699840546,
1313
+ "step": 819,
1314
+ "warm_up_unlearning_weight": 1
1315
+ },
1316
+ {
1317
+ "epoch": 32.8,
1318
+ "grad_norm": 2.0169899463653564,
1319
+ "learning_rate": 0.0007199999999999999,
1320
+ "loss": 1.091,
1321
+ "step": 820
1322
+ },
1323
+ {
1324
+ "epoch": 33.16,
1325
+ "forget_cf_outputs.loss": -0.5823113322257996,
1326
+ "forget_loss": 0.5823113322257996,
1327
+ "gated_loss": 0.0068359375,
1328
+ "retain_loss": 0.30245938897132874,
1329
+ "step": 829,
1330
+ "warm_up_unlearning_weight": 1
1331
+ },
1332
+ {
1333
+ "epoch": 33.2,
1334
+ "grad_norm": 2.2369046211242676,
1335
+ "learning_rate": 0.00068,
1336
+ "loss": 1.126,
1337
+ "step": 830
1338
+ },
1339
+ {
1340
+ "epoch": 33.56,
1341
+ "forget_cf_outputs.loss": -0.5231561660766602,
1342
+ "forget_loss": 0.5231561660766602,
1343
+ "gated_loss": 0.005767822265625,
1344
+ "retain_loss": 0.3002181351184845,
1345
+ "step": 839,
1346
+ "warm_up_unlearning_weight": 1
1347
+ },
1348
+ {
1349
+ "epoch": 33.6,
1350
+ "grad_norm": 2.4003753662109375,
1351
+ "learning_rate": 0.00064,
1352
+ "loss": 1.0696,
1353
+ "step": 840
1354
+ },
1355
+ {
1356
+ "epoch": 33.96,
1357
+ "forget_cf_outputs.loss": -0.4980463981628418,
1358
+ "forget_loss": 0.4980463981628418,
1359
+ "gated_loss": 0.0062255859375,
1360
+ "retain_loss": 0.20359160006046295,
1361
+ "step": 849,
1362
+ "warm_up_unlearning_weight": 1
1363
+ },
1364
+ {
1365
+ "epoch": 34.0,
1366
+ "grad_norm": 2.4804441928863525,
1367
+ "learning_rate": 0.0006,
1368
+ "loss": 1.0311,
1369
+ "step": 850
1370
+ },
1371
+ {
1372
+ "epoch": 34.36,
1373
+ "forget_cf_outputs.loss": -0.4477725327014923,
1374
+ "forget_loss": 0.4477725327014923,
1375
+ "gated_loss": 0.00592041015625,
1376
+ "retain_loss": 0.20766329765319824,
1377
+ "step": 859,
1378
+ "warm_up_unlearning_weight": 1
1379
+ },
1380
+ {
1381
+ "epoch": 34.4,
1382
+ "grad_norm": 1.8642009496688843,
1383
+ "learning_rate": 0.0005600000000000001,
1384
+ "loss": 0.9964,
1385
+ "step": 860
1386
+ },
1387
+ {
1388
+ "epoch": 34.76,
1389
+ "forget_cf_outputs.loss": -0.5641895532608032,
1390
+ "forget_loss": 0.5641895532608032,
1391
+ "gated_loss": 0.00555419921875,
1392
+ "retain_loss": 0.23121777176856995,
1393
+ "step": 869,
1394
+ "warm_up_unlearning_weight": 1
1395
+ },
1396
+ {
1397
+ "epoch": 34.8,
1398
+ "grad_norm": 1.8299639225006104,
1399
+ "learning_rate": 0.0005200000000000001,
1400
+ "loss": 1.055,
1401
+ "step": 870
1402
+ },
1403
+ {
1404
+ "epoch": 35.16,
1405
+ "forget_cf_outputs.loss": -0.5485031604766846,
1406
+ "forget_loss": 0.5485031604766846,
1407
+ "gated_loss": 0.006805419921875,
1408
+ "retain_loss": 0.3044523596763611,
1409
+ "step": 879,
1410
+ "warm_up_unlearning_weight": 1
1411
+ },
1412
+ {
1413
+ "epoch": 35.2,
1414
+ "grad_norm": 2.0533902645111084,
1415
+ "learning_rate": 0.00048,
1416
+ "loss": 1.0938,
1417
+ "step": 880
1418
+ },
1419
+ {
1420
+ "epoch": 35.56,
1421
+ "forget_cf_outputs.loss": -0.5070799589157104,
1422
+ "forget_loss": 0.5070799589157104,
1423
+ "gated_loss": 0.005828857421875,
1424
+ "retain_loss": 0.3008805215358734,
1425
+ "step": 889,
1426
+ "warm_up_unlearning_weight": 1
1427
+ },
1428
+ {
1429
+ "epoch": 35.6,
1430
+ "grad_norm": 2.342500925064087,
1431
+ "learning_rate": 0.00044,
1432
+ "loss": 1.0375,
1433
+ "step": 890
1434
+ },
1435
+ {
1436
+ "epoch": 35.96,
1437
+ "forget_cf_outputs.loss": -0.4470018446445465,
1438
+ "forget_loss": 0.4470018446445465,
1439
+ "gated_loss": 0.0062255859375,
1440
+ "retain_loss": 0.20359660685062408,
1441
+ "step": 899,
1442
+ "warm_up_unlearning_weight": 1
1443
+ },
1444
+ {
1445
+ "epoch": 36.0,
1446
+ "grad_norm": 2.555351734161377,
1447
+ "learning_rate": 0.0004,
1448
+ "loss": 0.9953,
1449
+ "step": 900
1450
+ },
1451
+ {
1452
+ "epoch": 36.36,
1453
+ "forget_cf_outputs.loss": -0.42907437682151794,
1454
+ "forget_loss": 0.42907437682151794,
1455
+ "gated_loss": 0.005950927734375,
1456
+ "retain_loss": 0.20733724534511566,
1457
+ "step": 909,
1458
+ "warm_up_unlearning_weight": 1
1459
+ },
1460
+ {
1461
+ "epoch": 36.4,
1462
+ "grad_norm": 1.745898962020874,
1463
+ "learning_rate": 0.00035999999999999997,
1464
+ "loss": 0.9637,
1465
+ "step": 910
1466
+ },
1467
+ {
1468
+ "epoch": 36.76,
1469
+ "forget_cf_outputs.loss": -0.5278509855270386,
1470
+ "forget_loss": 0.5278509855270386,
1471
+ "gated_loss": 0.00555419921875,
1472
+ "retain_loss": 0.23086762428283691,
1473
+ "step": 919,
1474
+ "warm_up_unlearning_weight": 1
1475
+ },
1476
+ {
1477
+ "epoch": 36.8,
1478
+ "grad_norm": 1.8470394611358643,
1479
+ "learning_rate": 0.00032,
1480
+ "loss": 1.0207,
1481
+ "step": 920
1482
+ },
1483
+ {
1484
+ "epoch": 37.16,
1485
+ "forget_cf_outputs.loss": -0.5195350646972656,
1486
+ "forget_loss": 0.5195350646972656,
1487
+ "gated_loss": 0.0068359375,
1488
+ "retain_loss": 0.3039552867412567,
1489
+ "step": 929,
1490
+ "warm_up_unlearning_weight": 1
1491
+ },
1492
+ {
1493
+ "epoch": 37.2,
1494
+ "grad_norm": 2.0206496715545654,
1495
+ "learning_rate": 0.00028000000000000003,
1496
+ "loss": 1.0645,
1497
+ "step": 930
1498
+ },
1499
+ {
1500
+ "epoch": 37.56,
1501
+ "forget_cf_outputs.loss": -0.4906051754951477,
1502
+ "forget_loss": 0.4906051754951477,
1503
+ "gated_loss": 0.005828857421875,
1504
+ "retain_loss": 0.3013906478881836,
1505
+ "step": 939,
1506
+ "warm_up_unlearning_weight": 1
1507
+ },
1508
+ {
1509
+ "epoch": 37.6,
1510
+ "grad_norm": 2.380035877227783,
1511
+ "learning_rate": 0.00024,
1512
+ "loss": 1.0092,
1513
+ "step": 940
1514
+ },
1515
+ {
1516
+ "epoch": 37.96,
1517
+ "forget_cf_outputs.loss": -0.41733187437057495,
1518
+ "forget_loss": 0.41733187437057495,
1519
+ "gated_loss": 0.0062255859375,
1520
+ "retain_loss": 0.20396247506141663,
1521
+ "step": 949,
1522
+ "warm_up_unlearning_weight": 1
1523
+ },
1524
+ {
1525
+ "epoch": 38.0,
1526
+ "grad_norm": 2.3801748752593994,
1527
+ "learning_rate": 0.0002,
1528
+ "loss": 0.9681,
1529
+ "step": 950
1530
+ },
1531
+ {
1532
+ "epoch": 38.36,
1533
+ "forget_cf_outputs.loss": -0.41866594552993774,
1534
+ "forget_loss": 0.41866594552993774,
1535
+ "gated_loss": 0.005950927734375,
1536
+ "retain_loss": 0.20748548209667206,
1537
+ "step": 959,
1538
+ "warm_up_unlearning_weight": 1
1539
+ },
1540
+ {
1541
+ "epoch": 38.4,
1542
+ "grad_norm": 1.8068156242370605,
1543
+ "learning_rate": 0.00016,
1544
+ "loss": 0.9408,
1545
+ "step": 960
1546
+ },
1547
+ {
1548
+ "epoch": 38.76,
1549
+ "forget_cf_outputs.loss": -0.503368079662323,
1550
+ "forget_loss": 0.503368079662323,
1551
+ "gated_loss": 0.00555419921875,
1552
+ "retain_loss": 0.2300793081521988,
1553
+ "step": 969,
1554
+ "warm_up_unlearning_weight": 1
1555
+ },
1556
+ {
1557
+ "epoch": 38.8,
1558
+ "grad_norm": 1.8037129640579224,
1559
+ "learning_rate": 0.00012,
1560
+ "loss": 0.9892,
1561
+ "step": 970
1562
+ },
1563
+ {
1564
+ "epoch": 39.16,
1565
+ "forget_cf_outputs.loss": -0.5070582628250122,
1566
+ "forget_loss": 0.5070582628250122,
1567
+ "gated_loss": 0.006805419921875,
1568
+ "retain_loss": 0.3026222288608551,
1569
+ "step": 979,
1570
+ "warm_up_unlearning_weight": 1
1571
+ },
1572
+ {
1573
+ "epoch": 39.2,
1574
+ "grad_norm": 2.055860757827759,
1575
+ "learning_rate": 8e-05,
1576
+ "loss": 1.0436,
1577
+ "step": 980
1578
+ },
1579
+ {
1580
+ "epoch": 39.56,
1581
+ "forget_cf_outputs.loss": -0.46062666177749634,
1582
+ "forget_loss": 0.46062666177749634,
1583
+ "gated_loss": 0.005828857421875,
1584
+ "retain_loss": 0.30048835277557373,
1585
+ "step": 989,
1586
+ "warm_up_unlearning_weight": 1
1587
+ },
1588
+ {
1589
+ "epoch": 39.6,
1590
+ "grad_norm": 2.1290202140808105,
1591
+ "learning_rate": 4e-05,
1592
+ "loss": 0.9838,
1593
+ "step": 990
1594
+ },
1595
+ {
1596
+ "epoch": 39.96,
1597
+ "forget_cf_outputs.loss": -0.39183497428894043,
1598
+ "forget_loss": 0.39183497428894043,
1599
+ "gated_loss": 0.0062255859375,
1600
+ "retain_loss": 0.20340043306350708,
1601
+ "step": 999,
1602
+ "warm_up_unlearning_weight": 1
1603
+ },
1604
+ {
1605
+ "epoch": 40.0,
1606
+ "grad_norm": 2.2874202728271484,
1607
+ "learning_rate": 0.0,
1608
+ "loss": 0.9411,
1609
+ "step": 1000
1610
+ }
1611
+ ],
1612
+ "logging_steps": 10,
1613
+ "max_steps": 1000,
1614
+ "num_input_tokens_seen": 0,
1615
+ "num_train_epochs": 40,
1616
+ "save_steps": 500,
1617
+ "stateful_callbacks": {
1618
+ "TrainerControl": {
1619
+ "args": {
1620
+ "should_epoch_stop": false,
1621
+ "should_evaluate": false,
1622
+ "should_log": false,
1623
+ "should_save": true,
1624
+ "should_training_stop": true
1625
+ },
1626
+ "attributes": {}
1627
+ }
1628
+ },
1629
+ "total_flos": 0.0,
1630
+ "train_batch_size": 16,
1631
+ "trial_name": null,
1632
+ "trial_params": null
1633
+ }