Qin Liu commited on
Commit
a0486eb
1 Parent(s): bbc1ad4

Model save

Browse files
Files changed (4) hide show
  1. README.md +62 -0
  2. all_results.json +9 -0
  3. train_results.json +9 -0
  4. trainer_state.json +1557 -0
README.md ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: llama3
3
+ library_name: peft
4
+ tags:
5
+ - trl
6
+ - dpo
7
+ - generated_from_trainer
8
+ base_model: Jackie999/llama3-sudo-5epochs-tofu_full_sft
9
+ model-index:
10
+ - name: llama3-sudo-dpo-instruct-20epochs-forget10-1sft-2lora
11
+ results: []
12
+ ---
13
+
14
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
15
+ should probably proofread and complete it, then remove this comment. -->
16
+
17
+ # llama3-sudo-dpo-instruct-20epochs-forget10-1sft-2lora
18
+
19
+ This model is a fine-tuned version of [Jackie999/llama3-sudo-5epochs-tofu_full_sft](https://huggingface.co/Jackie999/llama3-sudo-5epochs-tofu_full_sft) on an unknown dataset.
20
+
21
+ ## Model description
22
+
23
+ More information needed
24
+
25
+ ## Intended uses & limitations
26
+
27
+ More information needed
28
+
29
+ ## Training and evaluation data
30
+
31
+ More information needed
32
+
33
+ ## Training procedure
34
+
35
+ ### Training hyperparameters
36
+
37
+ The following hyperparameters were used during training:
38
+ - learning_rate: 5e-06
39
+ - train_batch_size: 2
40
+ - eval_batch_size: 8
41
+ - seed: 42
42
+ - distributed_type: multi-GPU
43
+ - num_devices: 2
44
+ - gradient_accumulation_steps: 4
45
+ - total_train_batch_size: 16
46
+ - total_eval_batch_size: 16
47
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
48
+ - lr_scheduler_type: cosine
49
+ - lr_scheduler_warmup_ratio: 0.1
50
+ - num_epochs: 20
51
+
52
+ ### Training results
53
+
54
+
55
+
56
+ ### Framework versions
57
+
58
+ - PEFT 0.12.0
59
+ - Transformers 4.44.0
60
+ - Pytorch 2.1.2
61
+ - Datasets 3.0.0
62
+ - Tokenizers 0.19.1
all_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 20.0,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.07008945742505603,
5
+ "train_runtime": 8233.3372,
6
+ "train_samples": 800,
7
+ "train_samples_per_second": 1.943,
8
+ "train_steps_per_second": 0.121
9
+ }
train_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 20.0,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.07008945742505603,
5
+ "train_runtime": 8233.3372,
6
+ "train_samples": 800,
7
+ "train_samples_per_second": 1.943,
8
+ "train_steps_per_second": 0.121
9
+ }
trainer_state.json ADDED
@@ -0,0 +1,1557 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 20.0,
5
+ "eval_steps": 1000,
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": 0.02,
13
+ "grad_norm": 0.35460313427884693,
14
+ "learning_rate": 5.0000000000000004e-08,
15
+ "logits/chosen": -1.8231227397918701,
16
+ "logits/rejected": -1.5560479164123535,
17
+ "logps/chosen": -85.02825927734375,
18
+ "logps/rejected": -71.86117553710938,
19
+ "loss": 0.6931,
20
+ "rewards/accuracies": 0.0,
21
+ "rewards/chosen": 0.0,
22
+ "rewards/margins": 0.0,
23
+ "rewards/rejected": 0.0,
24
+ "step": 1
25
+ },
26
+ {
27
+ "epoch": 0.2,
28
+ "grad_norm": 0.30411563112440204,
29
+ "learning_rate": 5.000000000000001e-07,
30
+ "logits/chosen": -1.7207438945770264,
31
+ "logits/rejected": -1.698093295097351,
32
+ "logps/chosen": -75.15734100341797,
33
+ "logps/rejected": -80.09143829345703,
34
+ "loss": 0.6931,
35
+ "rewards/accuracies": 0.5,
36
+ "rewards/chosen": 9.971462714020163e-05,
37
+ "rewards/margins": 0.0002858855586964637,
38
+ "rewards/rejected": -0.00018617100431583822,
39
+ "step": 10
40
+ },
41
+ {
42
+ "epoch": 0.4,
43
+ "grad_norm": 0.29583445812044185,
44
+ "learning_rate": 1.0000000000000002e-06,
45
+ "logits/chosen": -1.7815284729003906,
46
+ "logits/rejected": -1.7640860080718994,
47
+ "logps/chosen": -74.8746566772461,
48
+ "logps/rejected": -77.6731185913086,
49
+ "loss": 0.6927,
50
+ "rewards/accuracies": 0.6499999761581421,
51
+ "rewards/chosen": 0.00024838969693519175,
52
+ "rewards/margins": 0.0008968595648184419,
53
+ "rewards/rejected": -0.0006484698969870806,
54
+ "step": 20
55
+ },
56
+ {
57
+ "epoch": 0.6,
58
+ "grad_norm": 0.2966957448432536,
59
+ "learning_rate": 1.5e-06,
60
+ "logits/chosen": -1.739800214767456,
61
+ "logits/rejected": -1.7370967864990234,
62
+ "logps/chosen": -75.44456481933594,
63
+ "logps/rejected": -77.48068237304688,
64
+ "loss": 0.6917,
65
+ "rewards/accuracies": 0.875,
66
+ "rewards/chosen": 0.0010354876285418868,
67
+ "rewards/margins": 0.0030451309867203236,
68
+ "rewards/rejected": -0.002009643241763115,
69
+ "step": 30
70
+ },
71
+ {
72
+ "epoch": 0.8,
73
+ "grad_norm": 0.33102320941887775,
74
+ "learning_rate": 2.0000000000000003e-06,
75
+ "logits/chosen": -1.7791082859039307,
76
+ "logits/rejected": -1.733616590499878,
77
+ "logps/chosen": -76.32530212402344,
78
+ "logps/rejected": -75.00889587402344,
79
+ "loss": 0.689,
80
+ "rewards/accuracies": 1.0,
81
+ "rewards/chosen": 0.007160990033298731,
82
+ "rewards/margins": 0.008440766483545303,
83
+ "rewards/rejected": -0.0012797765666618943,
84
+ "step": 40
85
+ },
86
+ {
87
+ "epoch": 1.0,
88
+ "grad_norm": 0.5247728899006847,
89
+ "learning_rate": 2.5e-06,
90
+ "logits/chosen": -1.7112547159194946,
91
+ "logits/rejected": -1.7392151355743408,
92
+ "logps/chosen": -73.26392364501953,
93
+ "logps/rejected": -79.31649780273438,
94
+ "loss": 0.6832,
95
+ "rewards/accuracies": 1.0,
96
+ "rewards/chosen": 0.013615945354104042,
97
+ "rewards/margins": 0.019877230748534203,
98
+ "rewards/rejected": -0.006261286325752735,
99
+ "step": 50
100
+ },
101
+ {
102
+ "epoch": 1.2,
103
+ "grad_norm": 0.4135013743852781,
104
+ "learning_rate": 3e-06,
105
+ "logits/chosen": -1.732405424118042,
106
+ "logits/rejected": -1.7449699640274048,
107
+ "logps/chosen": -71.77066040039062,
108
+ "logps/rejected": -80.100341796875,
109
+ "loss": 0.6732,
110
+ "rewards/accuracies": 1.0,
111
+ "rewards/chosen": 0.024410251528024673,
112
+ "rewards/margins": 0.03932947292923927,
113
+ "rewards/rejected": -0.01491921953856945,
114
+ "step": 60
115
+ },
116
+ {
117
+ "epoch": 1.4,
118
+ "grad_norm": 0.392415113956775,
119
+ "learning_rate": 3.5e-06,
120
+ "logits/chosen": -1.716195821762085,
121
+ "logits/rejected": -1.731308937072754,
122
+ "logps/chosen": -72.06253051757812,
123
+ "logps/rejected": -80.39039611816406,
124
+ "loss": 0.6542,
125
+ "rewards/accuracies": 1.0,
126
+ "rewards/chosen": 0.04634103178977966,
127
+ "rewards/margins": 0.07862894982099533,
128
+ "rewards/rejected": -0.032287921756505966,
129
+ "step": 70
130
+ },
131
+ {
132
+ "epoch": 1.6,
133
+ "grad_norm": 0.5321176658546207,
134
+ "learning_rate": 4.000000000000001e-06,
135
+ "logits/chosen": -1.69076406955719,
136
+ "logits/rejected": -1.6509063243865967,
137
+ "logps/chosen": -68.82267761230469,
138
+ "logps/rejected": -88.67816162109375,
139
+ "loss": 0.605,
140
+ "rewards/accuracies": 1.0,
141
+ "rewards/chosen": 0.08192411065101624,
142
+ "rewards/margins": 0.19158878922462463,
143
+ "rewards/rejected": -0.1096646636724472,
144
+ "step": 80
145
+ },
146
+ {
147
+ "epoch": 1.8,
148
+ "grad_norm": 0.5696532480733786,
149
+ "learning_rate": 4.5e-06,
150
+ "logits/chosen": -1.7061433792114258,
151
+ "logits/rejected": -1.684195876121521,
152
+ "logps/chosen": -59.4539680480957,
153
+ "logps/rejected": -106.4749526977539,
154
+ "loss": 0.4742,
155
+ "rewards/accuracies": 1.0,
156
+ "rewards/chosen": 0.18270966410636902,
157
+ "rewards/margins": 0.5085535645484924,
158
+ "rewards/rejected": -0.325843870639801,
159
+ "step": 90
160
+ },
161
+ {
162
+ "epoch": 2.0,
163
+ "grad_norm": 0.3568428380556091,
164
+ "learning_rate": 5e-06,
165
+ "logits/chosen": -1.5610880851745605,
166
+ "logits/rejected": -1.4533908367156982,
167
+ "logps/chosen": -48.31792449951172,
168
+ "logps/rejected": -164.77566528320312,
169
+ "loss": 0.3175,
170
+ "rewards/accuracies": 1.0,
171
+ "rewards/chosen": 0.2694377899169922,
172
+ "rewards/margins": 1.1592490673065186,
173
+ "rewards/rejected": -0.8898113965988159,
174
+ "step": 100
175
+ },
176
+ {
177
+ "epoch": 2.2,
178
+ "grad_norm": 0.28794719016802656,
179
+ "learning_rate": 4.99847706754774e-06,
180
+ "logits/chosen": -1.5363928079605103,
181
+ "logits/rejected": -1.4212225675582886,
182
+ "logps/chosen": -42.558448791503906,
183
+ "logps/rejected": -187.63729858398438,
184
+ "loss": 0.2626,
185
+ "rewards/accuracies": 1.0,
186
+ "rewards/chosen": 0.39137277007102966,
187
+ "rewards/margins": 1.5502681732177734,
188
+ "rewards/rejected": -1.1588953733444214,
189
+ "step": 110
190
+ },
191
+ {
192
+ "epoch": 2.4,
193
+ "grad_norm": 0.15078063761313237,
194
+ "learning_rate": 4.993910125649561e-06,
195
+ "logits/chosen": -1.51887845993042,
196
+ "logits/rejected": -1.437011957168579,
197
+ "logps/chosen": -38.434608459472656,
198
+ "logps/rejected": -232.57766723632812,
199
+ "loss": 0.223,
200
+ "rewards/accuracies": 1.0,
201
+ "rewards/chosen": 0.42294931411743164,
202
+ "rewards/margins": 2.0381760597229004,
203
+ "rewards/rejected": -1.6152267456054688,
204
+ "step": 120
205
+ },
206
+ {
207
+ "epoch": 2.6,
208
+ "grad_norm": 0.4347861863265109,
209
+ "learning_rate": 4.986304738420684e-06,
210
+ "logits/chosen": -1.3788427114486694,
211
+ "logits/rejected": -1.37319016456604,
212
+ "logps/chosen": -33.37788391113281,
213
+ "logps/rejected": -288.50018310546875,
214
+ "loss": 0.1776,
215
+ "rewards/accuracies": 1.0,
216
+ "rewards/chosen": 0.44394317269325256,
217
+ "rewards/margins": 2.571049928665161,
218
+ "rewards/rejected": -2.1271069049835205,
219
+ "step": 130
220
+ },
221
+ {
222
+ "epoch": 2.8,
223
+ "grad_norm": 0.7461810630936248,
224
+ "learning_rate": 4.975670171853926e-06,
225
+ "logits/chosen": -1.190661072731018,
226
+ "logits/rejected": -1.1670948266983032,
227
+ "logps/chosen": -32.81340026855469,
228
+ "logps/rejected": -336.36962890625,
229
+ "loss": 0.1102,
230
+ "rewards/accuracies": 1.0,
231
+ "rewards/chosen": 0.4285716414451599,
232
+ "rewards/margins": 3.032412528991699,
233
+ "rewards/rejected": -2.6038408279418945,
234
+ "step": 140
235
+ },
236
+ {
237
+ "epoch": 3.0,
238
+ "grad_norm": 0.053630215551159044,
239
+ "learning_rate": 4.962019382530521e-06,
240
+ "logits/chosen": -1.1552760601043701,
241
+ "logits/rejected": -1.1048529148101807,
242
+ "logps/chosen": -60.1913948059082,
243
+ "logps/rejected": -567.5430908203125,
244
+ "loss": 0.0207,
245
+ "rewards/accuracies": 1.0,
246
+ "rewards/chosen": 0.1675727218389511,
247
+ "rewards/margins": 5.098362922668457,
248
+ "rewards/rejected": -4.930790901184082,
249
+ "step": 150
250
+ },
251
+ {
252
+ "epoch": 3.2,
253
+ "grad_norm": 0.031903287512855885,
254
+ "learning_rate": 4.9453690018345144e-06,
255
+ "logits/chosen": -1.0522866249084473,
256
+ "logits/rejected": -1.0280696153640747,
257
+ "logps/chosen": -102.0748062133789,
258
+ "logps/rejected": -891.6737060546875,
259
+ "loss": 0.0036,
260
+ "rewards/accuracies": 1.0,
261
+ "rewards/chosen": -0.2856892943382263,
262
+ "rewards/margins": 7.848362922668457,
263
+ "rewards/rejected": -8.134051322937012,
264
+ "step": 160
265
+ },
266
+ {
267
+ "epoch": 3.4,
268
+ "grad_norm": 0.0197914954559647,
269
+ "learning_rate": 4.925739315689991e-06,
270
+ "logits/chosen": -1.1597774028778076,
271
+ "logits/rejected": -1.1589906215667725,
272
+ "logps/chosen": -61.60524368286133,
273
+ "logps/rejected": -1039.020751953125,
274
+ "loss": 0.0022,
275
+ "rewards/accuracies": 1.0,
276
+ "rewards/chosen": 0.12755107879638672,
277
+ "rewards/margins": 9.746142387390137,
278
+ "rewards/rejected": -9.61859130859375,
279
+ "step": 170
280
+ },
281
+ {
282
+ "epoch": 3.6,
283
+ "grad_norm": 0.023011571746249113,
284
+ "learning_rate": 4.903154239845798e-06,
285
+ "logits/chosen": -1.2340686321258545,
286
+ "logits/rejected": -1.0457987785339355,
287
+ "logps/chosen": -59.74028778076172,
288
+ "logps/rejected": -1156.054931640625,
289
+ "loss": 0.0014,
290
+ "rewards/accuracies": 1.0,
291
+ "rewards/chosen": 0.25307127833366394,
292
+ "rewards/margins": 11.132636070251465,
293
+ "rewards/rejected": -10.879566192626953,
294
+ "step": 180
295
+ },
296
+ {
297
+ "epoch": 3.8,
298
+ "grad_norm": 0.010025707600426755,
299
+ "learning_rate": 4.8776412907378845e-06,
300
+ "logits/chosen": -1.0119726657867432,
301
+ "logits/rejected": -1.0299015045166016,
302
+ "logps/chosen": -51.19286346435547,
303
+ "logps/rejected": -1030.427490234375,
304
+ "loss": 0.0015,
305
+ "rewards/accuracies": 1.0,
306
+ "rewards/chosen": 0.22199800610542297,
307
+ "rewards/margins": 9.714365005493164,
308
+ "rewards/rejected": -9.492366790771484,
309
+ "step": 190
310
+ },
311
+ {
312
+ "epoch": 4.0,
313
+ "grad_norm": 0.009363825045997377,
314
+ "learning_rate": 4.849231551964771e-06,
315
+ "logits/chosen": -1.1452789306640625,
316
+ "logits/rejected": -1.0538662672042847,
317
+ "logps/chosen": -55.855430603027344,
318
+ "logps/rejected": -1116.4219970703125,
319
+ "loss": 0.0012,
320
+ "rewards/accuracies": 1.0,
321
+ "rewards/chosen": 0.2577177882194519,
322
+ "rewards/margins": 10.68994140625,
323
+ "rewards/rejected": -10.43222427368164,
324
+ "step": 200
325
+ },
326
+ {
327
+ "epoch": 4.2,
328
+ "grad_norm": 0.007917186035092368,
329
+ "learning_rate": 4.817959636416969e-06,
330
+ "logits/chosen": -0.979863166809082,
331
+ "logits/rejected": -0.954973042011261,
332
+ "logps/chosen": -51.2191276550293,
333
+ "logps/rejected": -1067.7679443359375,
334
+ "loss": 0.0012,
335
+ "rewards/accuracies": 1.0,
336
+ "rewards/chosen": 0.21226799488067627,
337
+ "rewards/margins": 10.104147911071777,
338
+ "rewards/rejected": -9.89188003540039,
339
+ "step": 210
340
+ },
341
+ {
342
+ "epoch": 4.4,
343
+ "grad_norm": 0.004467951143379201,
344
+ "learning_rate": 4.783863644106502e-06,
345
+ "logits/chosen": -1.0173983573913574,
346
+ "logits/rejected": -0.9798928499221802,
347
+ "logps/chosen": -52.295379638671875,
348
+ "logps/rejected": -1050.406494140625,
349
+ "loss": 0.001,
350
+ "rewards/accuracies": 1.0,
351
+ "rewards/chosen": 0.21525339782238007,
352
+ "rewards/margins": 9.93811321258545,
353
+ "rewards/rejected": -9.722859382629395,
354
+ "step": 220
355
+ },
356
+ {
357
+ "epoch": 4.6,
358
+ "grad_norm": 0.0036885023764270065,
359
+ "learning_rate": 4.746985115747918e-06,
360
+ "logits/chosen": -1.1039237976074219,
361
+ "logits/rejected": -0.9902089238166809,
362
+ "logps/chosen": -56.54585647583008,
363
+ "logps/rejected": -1109.8258056640625,
364
+ "loss": 0.0008,
365
+ "rewards/accuracies": 1.0,
366
+ "rewards/chosen": 0.2429007589817047,
367
+ "rewards/margins": 10.61030101776123,
368
+ "rewards/rejected": -10.367398262023926,
369
+ "step": 230
370
+ },
371
+ {
372
+ "epoch": 4.8,
373
+ "grad_norm": 0.0065247264850717725,
374
+ "learning_rate": 4.707368982147318e-06,
375
+ "logits/chosen": -1.1152589321136475,
376
+ "logits/rejected": -1.0077005624771118,
377
+ "logps/chosen": -54.775856018066406,
378
+ "logps/rejected": -1131.518310546875,
379
+ "loss": 0.0008,
380
+ "rewards/accuracies": 1.0,
381
+ "rewards/chosen": 0.22144417464733124,
382
+ "rewards/margins": 10.786369323730469,
383
+ "rewards/rejected": -10.564926147460938,
384
+ "step": 240
385
+ },
386
+ {
387
+ "epoch": 5.0,
388
+ "grad_norm": 0.0032674373668132675,
389
+ "learning_rate": 4.665063509461098e-06,
390
+ "logits/chosen": -1.0255931615829468,
391
+ "logits/rejected": -0.9475139379501343,
392
+ "logps/chosen": -56.231163024902344,
393
+ "logps/rejected": -1138.18115234375,
394
+ "loss": 0.0008,
395
+ "rewards/accuracies": 1.0,
396
+ "rewards/chosen": 0.22440902888774872,
397
+ "rewards/margins": 10.85453987121582,
398
+ "rewards/rejected": -10.630131721496582,
399
+ "step": 250
400
+ },
401
+ {
402
+ "epoch": 5.2,
403
+ "grad_norm": 0.004115527916331285,
404
+ "learning_rate": 4.620120240391065e-06,
405
+ "logits/chosen": -1.0400116443634033,
406
+ "logits/rejected": -0.9953556060791016,
407
+ "logps/chosen": -54.63758087158203,
408
+ "logps/rejected": -1113.4554443359375,
409
+ "loss": 0.0007,
410
+ "rewards/accuracies": 1.0,
411
+ "rewards/chosen": 0.21181125938892365,
412
+ "rewards/margins": 10.56176471710205,
413
+ "rewards/rejected": -10.34995174407959,
414
+ "step": 260
415
+ },
416
+ {
417
+ "epoch": 5.4,
418
+ "grad_norm": 0.0030662391394572284,
419
+ "learning_rate": 4.572593931387604e-06,
420
+ "logits/chosen": -1.100754976272583,
421
+ "logits/rejected": -0.9190059900283813,
422
+ "logps/chosen": -55.8875846862793,
423
+ "logps/rejected": -1182.8133544921875,
424
+ "loss": 0.0007,
425
+ "rewards/accuracies": 1.0,
426
+ "rewards/chosen": 0.2545715868473053,
427
+ "rewards/margins": 11.359747886657715,
428
+ "rewards/rejected": -11.105175018310547,
429
+ "step": 270
430
+ },
431
+ {
432
+ "epoch": 5.6,
433
+ "grad_norm": 0.004246284055742857,
434
+ "learning_rate": 4.522542485937369e-06,
435
+ "logits/chosen": -1.0873005390167236,
436
+ "logits/rejected": -0.91912841796875,
437
+ "logps/chosen": -57.15370559692383,
438
+ "logps/rejected": -1198.202392578125,
439
+ "loss": 0.0007,
440
+ "rewards/accuracies": 1.0,
441
+ "rewards/chosen": 0.24588795006275177,
442
+ "rewards/margins": 11.502182960510254,
443
+ "rewards/rejected": -11.256293296813965,
444
+ "step": 280
445
+ },
446
+ {
447
+ "epoch": 5.8,
448
+ "grad_norm": 0.004139329336955742,
449
+ "learning_rate": 4.470026884016805e-06,
450
+ "logits/chosen": -0.9481426477432251,
451
+ "logits/rejected": -0.9623401761054993,
452
+ "logps/chosen": -52.13285446166992,
453
+ "logps/rejected": -1074.572509765625,
454
+ "loss": 0.0007,
455
+ "rewards/accuracies": 1.0,
456
+ "rewards/chosen": 0.20849105715751648,
457
+ "rewards/margins": 10.169190406799316,
458
+ "rewards/rejected": -9.960699081420898,
459
+ "step": 290
460
+ },
461
+ {
462
+ "epoch": 6.0,
463
+ "grad_norm": 0.003034900790871541,
464
+ "learning_rate": 4.415111107797445e-06,
465
+ "logits/chosen": -1.0314205884933472,
466
+ "logits/rejected": -1.0098388195037842,
467
+ "logps/chosen": -54.42866134643555,
468
+ "logps/rejected": -1097.184326171875,
469
+ "loss": 0.0006,
470
+ "rewards/accuracies": 1.0,
471
+ "rewards/chosen": 0.20703566074371338,
472
+ "rewards/margins": 10.398859024047852,
473
+ "rewards/rejected": -10.19182300567627,
474
+ "step": 300
475
+ },
476
+ {
477
+ "epoch": 6.2,
478
+ "grad_norm": 0.0038510374493456907,
479
+ "learning_rate": 4.357862063693486e-06,
480
+ "logits/chosen": -1.026374101638794,
481
+ "logits/rejected": -0.9435133934020996,
482
+ "logps/chosen": -53.9544792175293,
483
+ "logps/rejected": -1142.4066162109375,
484
+ "loss": 0.0006,
485
+ "rewards/accuracies": 1.0,
486
+ "rewards/chosen": 0.21439966559410095,
487
+ "rewards/margins": 10.869270324707031,
488
+ "rewards/rejected": -10.654870986938477,
489
+ "step": 310
490
+ },
491
+ {
492
+ "epoch": 6.4,
493
+ "grad_norm": 0.0035014469049547427,
494
+ "learning_rate": 4.2983495008466285e-06,
495
+ "logits/chosen": -1.105359673500061,
496
+ "logits/rejected": -0.9964410066604614,
497
+ "logps/chosen": -57.14154815673828,
498
+ "logps/rejected": -1162.7977294921875,
499
+ "loss": 0.0006,
500
+ "rewards/accuracies": 1.0,
501
+ "rewards/chosen": 0.21488821506500244,
502
+ "rewards/margins": 11.113906860351562,
503
+ "rewards/rejected": -10.899017333984375,
504
+ "step": 320
505
+ },
506
+ {
507
+ "epoch": 6.6,
508
+ "grad_norm": 0.002634385062559098,
509
+ "learning_rate": 4.236645926147493e-06,
510
+ "logits/chosen": -1.0640833377838135,
511
+ "logits/rejected": -0.9242901802062988,
512
+ "logps/chosen": -56.49944305419922,
513
+ "logps/rejected": -1212.7708740234375,
514
+ "loss": 0.0005,
515
+ "rewards/accuracies": 1.0,
516
+ "rewards/chosen": 0.23322686553001404,
517
+ "rewards/margins": 11.629671096801758,
518
+ "rewards/rejected": -11.396444320678711,
519
+ "step": 330
520
+ },
521
+ {
522
+ "epoch": 6.8,
523
+ "grad_norm": 0.0030996294667543192,
524
+ "learning_rate": 4.172826515897146e-06,
525
+ "logits/chosen": -1.082220435142517,
526
+ "logits/rejected": -0.9698995351791382,
527
+ "logps/chosen": -59.28369903564453,
528
+ "logps/rejected": -1214.587646484375,
529
+ "loss": 0.0005,
530
+ "rewards/accuracies": 1.0,
531
+ "rewards/chosen": 0.19337359070777893,
532
+ "rewards/margins": 11.616683959960938,
533
+ "rewards/rejected": -11.423309326171875,
534
+ "step": 340
535
+ },
536
+ {
537
+ "epoch": 7.0,
538
+ "grad_norm": 0.0015310049705482352,
539
+ "learning_rate": 4.106969024216348e-06,
540
+ "logits/chosen": -0.9850069284439087,
541
+ "logits/rejected": -0.9987241625785828,
542
+ "logps/chosen": -54.036949157714844,
543
+ "logps/rejected": -1109.0908203125,
544
+ "loss": 0.0006,
545
+ "rewards/accuracies": 1.0,
546
+ "rewards/chosen": 0.19889585673809052,
547
+ "rewards/margins": 10.489348411560059,
548
+ "rewards/rejected": -10.29045295715332,
549
+ "step": 350
550
+ },
551
+ {
552
+ "epoch": 7.2,
553
+ "grad_norm": 0.001899366605942392,
554
+ "learning_rate": 4.039153688314146e-06,
555
+ "logits/chosen": -1.0366054773330688,
556
+ "logits/rejected": -0.9512063264846802,
557
+ "logps/chosen": -56.377410888671875,
558
+ "logps/rejected": -1194.2122802734375,
559
+ "loss": 0.0005,
560
+ "rewards/accuracies": 1.0,
561
+ "rewards/chosen": 0.2178737223148346,
562
+ "rewards/margins": 11.424546241760254,
563
+ "rewards/rejected": -11.206673622131348,
564
+ "step": 360
565
+ },
566
+ {
567
+ "epoch": 7.4,
568
+ "grad_norm": 0.003085448477067617,
569
+ "learning_rate": 3.969463130731183e-06,
570
+ "logits/chosen": -1.058009147644043,
571
+ "logits/rejected": -0.9505330324172974,
572
+ "logps/chosen": -55.610870361328125,
573
+ "logps/rejected": -1222.853515625,
574
+ "loss": 0.0005,
575
+ "rewards/accuracies": 1.0,
576
+ "rewards/chosen": 0.2119036167860031,
577
+ "rewards/margins": 11.69975471496582,
578
+ "rewards/rejected": -11.4878511428833,
579
+ "step": 370
580
+ },
581
+ {
582
+ "epoch": 7.6,
583
+ "grad_norm": 0.002631100351440524,
584
+ "learning_rate": 3.897982258676867e-06,
585
+ "logits/chosen": -0.9967459440231323,
586
+ "logits/rejected": -1.021655797958374,
587
+ "logps/chosen": -52.216705322265625,
588
+ "logps/rejected": -1100.6519775390625,
589
+ "loss": 0.0005,
590
+ "rewards/accuracies": 1.0,
591
+ "rewards/chosen": 0.19106148183345795,
592
+ "rewards/margins": 10.369633674621582,
593
+ "rewards/rejected": -10.178572654724121,
594
+ "step": 380
595
+ },
596
+ {
597
+ "epoch": 7.8,
598
+ "grad_norm": 0.0031329070413845662,
599
+ "learning_rate": 3.824798160583012e-06,
600
+ "logits/chosen": -1.0442421436309814,
601
+ "logits/rejected": -0.9009174108505249,
602
+ "logps/chosen": -56.89820098876953,
603
+ "logps/rejected": -1184.45263671875,
604
+ "loss": 0.0005,
605
+ "rewards/accuracies": 1.0,
606
+ "rewards/chosen": 0.22130858898162842,
607
+ "rewards/margins": 11.319818496704102,
608
+ "rewards/rejected": -11.098509788513184,
609
+ "step": 390
610
+ },
611
+ {
612
+ "epoch": 8.0,
613
+ "grad_norm": 0.003423988945465294,
614
+ "learning_rate": 3.7500000000000005e-06,
615
+ "logits/chosen": -1.035902500152588,
616
+ "logits/rejected": -0.9636209607124329,
617
+ "logps/chosen": -56.99961471557617,
618
+ "logps/rejected": -1206.747802734375,
619
+ "loss": 0.0005,
620
+ "rewards/accuracies": 1.0,
621
+ "rewards/chosen": 0.18911413848400116,
622
+ "rewards/margins": 11.49039363861084,
623
+ "rewards/rejected": -11.301280975341797,
624
+ "step": 400
625
+ },
626
+ {
627
+ "epoch": 8.2,
628
+ "grad_norm": 0.003864669112701001,
629
+ "learning_rate": 3.6736789069647273e-06,
630
+ "logits/chosen": -0.9635273218154907,
631
+ "logits/rejected": -0.8587255477905273,
632
+ "logps/chosen": -58.7963981628418,
633
+ "logps/rejected": -1161.4803466796875,
634
+ "loss": 0.0005,
635
+ "rewards/accuracies": 1.0,
636
+ "rewards/chosen": 0.19340120255947113,
637
+ "rewards/margins": 11.068787574768066,
638
+ "rewards/rejected": -10.875386238098145,
639
+ "step": 410
640
+ },
641
+ {
642
+ "epoch": 8.4,
643
+ "grad_norm": 0.003245040701864265,
644
+ "learning_rate": 3.595927866972694e-06,
645
+ "logits/chosen": -1.0136067867279053,
646
+ "logits/rejected": -0.9350364804267883,
647
+ "logps/chosen": -55.97736358642578,
648
+ "logps/rejected": -1197.21484375,
649
+ "loss": 0.0004,
650
+ "rewards/accuracies": 1.0,
651
+ "rewards/chosen": 0.20746412873268127,
652
+ "rewards/margins": 11.428886413574219,
653
+ "rewards/rejected": -11.221421241760254,
654
+ "step": 420
655
+ },
656
+ {
657
+ "epoch": 8.6,
658
+ "grad_norm": 0.0023560411348610777,
659
+ "learning_rate": 3.516841607689501e-06,
660
+ "logits/chosen": -0.9947836995124817,
661
+ "logits/rejected": -0.9126423597335815,
662
+ "logps/chosen": -57.561622619628906,
663
+ "logps/rejected": -1165.994384765625,
664
+ "loss": 0.0004,
665
+ "rewards/accuracies": 1.0,
666
+ "rewards/chosen": 0.18999549746513367,
667
+ "rewards/margins": 11.074243545532227,
668
+ "rewards/rejected": -10.884247779846191,
669
+ "step": 430
670
+ },
671
+ {
672
+ "epoch": 8.8,
673
+ "grad_norm": 0.0025971993823853895,
674
+ "learning_rate": 3.436516483539781e-06,
675
+ "logits/chosen": -1.0256311893463135,
676
+ "logits/rejected": -0.8973512649536133,
677
+ "logps/chosen": -59.285438537597656,
678
+ "logps/rejected": -1248.3521728515625,
679
+ "loss": 0.0004,
680
+ "rewards/accuracies": 1.0,
681
+ "rewards/chosen": 0.18996688723564148,
682
+ "rewards/margins": 11.936816215515137,
683
+ "rewards/rejected": -11.746848106384277,
684
+ "step": 440
685
+ },
686
+ {
687
+ "epoch": 9.0,
688
+ "grad_norm": 0.0032315637570330567,
689
+ "learning_rate": 3.3550503583141726e-06,
690
+ "logits/chosen": -1.0244309902191162,
691
+ "logits/rejected": -0.8780086636543274,
692
+ "logps/chosen": -57.84431076049805,
693
+ "logps/rejected": -1234.871337890625,
694
+ "loss": 0.0004,
695
+ "rewards/accuracies": 1.0,
696
+ "rewards/chosen": 0.1998531073331833,
697
+ "rewards/margins": 11.807432174682617,
698
+ "rewards/rejected": -11.607580184936523,
699
+ "step": 450
700
+ },
701
+ {
702
+ "epoch": 9.2,
703
+ "grad_norm": 0.0036226784518706475,
704
+ "learning_rate": 3.272542485937369e-06,
705
+ "logits/chosen": -0.9375013113021851,
706
+ "logits/rejected": -0.8737777471542358,
707
+ "logps/chosen": -54.31487274169922,
708
+ "logps/rejected": -1172.191162109375,
709
+ "loss": 0.0004,
710
+ "rewards/accuracies": 1.0,
711
+ "rewards/chosen": 0.19032403826713562,
712
+ "rewards/margins": 11.127653121948242,
713
+ "rewards/rejected": -10.937328338623047,
714
+ "step": 460
715
+ },
716
+ {
717
+ "epoch": 9.4,
718
+ "grad_norm": 0.002997930586361418,
719
+ "learning_rate": 3.189093389542498e-06,
720
+ "logits/chosen": -1.0432255268096924,
721
+ "logits/rejected": -0.9277008175849915,
722
+ "logps/chosen": -58.5209846496582,
723
+ "logps/rejected": -1200.554931640625,
724
+ "loss": 0.0004,
725
+ "rewards/accuracies": 1.0,
726
+ "rewards/chosen": 0.19790109992027283,
727
+ "rewards/margins": 11.468005180358887,
728
+ "rewards/rejected": -11.270103454589844,
729
+ "step": 470
730
+ },
731
+ {
732
+ "epoch": 9.6,
733
+ "grad_norm": 0.0022499622323506746,
734
+ "learning_rate": 3.1048047389991693e-06,
735
+ "logits/chosen": -1.0687272548675537,
736
+ "logits/rejected": -0.9239826202392578,
737
+ "logps/chosen": -60.52696990966797,
738
+ "logps/rejected": -1241.221435546875,
739
+ "loss": 0.0004,
740
+ "rewards/accuracies": 1.0,
741
+ "rewards/chosen": 0.18703293800354004,
742
+ "rewards/margins": 11.873010635375977,
743
+ "rewards/rejected": -11.685976028442383,
744
+ "step": 480
745
+ },
746
+ {
747
+ "epoch": 9.8,
748
+ "grad_norm": 0.0019412808634179527,
749
+ "learning_rate": 3.019779227044398e-06,
750
+ "logits/chosen": -0.9789573550224304,
751
+ "logits/rejected": -0.7663965225219727,
752
+ "logps/chosen": -61.904319763183594,
753
+ "logps/rejected": -1237.324951171875,
754
+ "loss": 0.0003,
755
+ "rewards/accuracies": 1.0,
756
+ "rewards/chosen": 0.1967342346906662,
757
+ "rewards/margins": 11.862494468688965,
758
+ "rewards/rejected": -11.66576099395752,
759
+ "step": 490
760
+ },
761
+ {
762
+ "epoch": 10.0,
763
+ "grad_norm": 0.002323126626066593,
764
+ "learning_rate": 2.9341204441673267e-06,
765
+ "logits/chosen": -1.0079948902130127,
766
+ "logits/rejected": -0.9249541163444519,
767
+ "logps/chosen": -58.3544807434082,
768
+ "logps/rejected": -1210.3780517578125,
769
+ "loss": 0.0004,
770
+ "rewards/accuracies": 1.0,
771
+ "rewards/chosen": 0.16591690480709076,
772
+ "rewards/margins": 11.488636016845703,
773
+ "rewards/rejected": -11.32271957397461,
774
+ "step": 500
775
+ },
776
+ {
777
+ "epoch": 10.2,
778
+ "grad_norm": 0.0018387058134298486,
779
+ "learning_rate": 2.847932752400164e-06,
780
+ "logits/chosen": -0.9845121502876282,
781
+ "logits/rejected": -0.8383527994155884,
782
+ "logps/chosen": -59.52973556518555,
783
+ "logps/rejected": -1259.748291015625,
784
+ "loss": 0.0004,
785
+ "rewards/accuracies": 1.0,
786
+ "rewards/chosen": 0.17635658383369446,
787
+ "rewards/margins": 12.032411575317383,
788
+ "rewards/rejected": -11.856054306030273,
789
+ "step": 510
790
+ },
791
+ {
792
+ "epoch": 10.4,
793
+ "grad_norm": 0.0013643138700466504,
794
+ "learning_rate": 2.761321158169134e-06,
795
+ "logits/chosen": -1.0216516256332397,
796
+ "logits/rejected": -0.9077258110046387,
797
+ "logps/chosen": -59.731300354003906,
798
+ "logps/rejected": -1231.614501953125,
799
+ "loss": 0.0004,
800
+ "rewards/accuracies": 1.0,
801
+ "rewards/chosen": 0.1921447217464447,
802
+ "rewards/margins": 11.769120216369629,
803
+ "rewards/rejected": -11.576974868774414,
804
+ "step": 520
805
+ },
806
+ {
807
+ "epoch": 10.6,
808
+ "grad_norm": 0.002977922307910112,
809
+ "learning_rate": 2.6743911843603134e-06,
810
+ "logits/chosen": -1.0051695108413696,
811
+ "logits/rejected": -0.8601359128952026,
812
+ "logps/chosen": -62.12017822265625,
813
+ "logps/rejected": -1268.072509765625,
814
+ "loss": 0.0003,
815
+ "rewards/accuracies": 1.0,
816
+ "rewards/chosen": 0.1786160171031952,
817
+ "rewards/margins": 12.141373634338379,
818
+ "rewards/rejected": -11.962759017944336,
819
+ "step": 530
820
+ },
821
+ {
822
+ "epoch": 10.8,
823
+ "grad_norm": 0.0019745371274113147,
824
+ "learning_rate": 2.587248741756253e-06,
825
+ "logits/chosen": -1.0047599077224731,
826
+ "logits/rejected": -0.8642932176589966,
827
+ "logps/chosen": -61.087615966796875,
828
+ "logps/rejected": -1251.1107177734375,
829
+ "loss": 0.0003,
830
+ "rewards/accuracies": 1.0,
831
+ "rewards/chosen": 0.18852102756500244,
832
+ "rewards/margins": 11.956985473632812,
833
+ "rewards/rejected": -11.768463134765625,
834
+ "step": 540
835
+ },
836
+ {
837
+ "epoch": 11.0,
838
+ "grad_norm": 0.0018986393686723218,
839
+ "learning_rate": 2.5e-06,
840
+ "logits/chosen": -1.0309114456176758,
841
+ "logits/rejected": -0.9270389676094055,
842
+ "logps/chosen": -59.46849822998047,
843
+ "logps/rejected": -1210.13720703125,
844
+ "loss": 0.0003,
845
+ "rewards/accuracies": 1.0,
846
+ "rewards/chosen": 0.18577028810977936,
847
+ "rewards/margins": 11.525289535522461,
848
+ "rewards/rejected": -11.339518547058105,
849
+ "step": 550
850
+ },
851
+ {
852
+ "epoch": 11.2,
853
+ "grad_norm": 0.0016611092816266825,
854
+ "learning_rate": 2.4127512582437486e-06,
855
+ "logits/chosen": -0.9609254598617554,
856
+ "logits/rejected": -0.9043042063713074,
857
+ "logps/chosen": -54.84800338745117,
858
+ "logps/rejected": -1216.4305419921875,
859
+ "loss": 0.0003,
860
+ "rewards/accuracies": 1.0,
861
+ "rewards/chosen": 0.20413394272327423,
862
+ "rewards/margins": 11.576135635375977,
863
+ "rewards/rejected": -11.372001647949219,
864
+ "step": 560
865
+ },
866
+ {
867
+ "epoch": 11.4,
868
+ "grad_norm": 0.0019080916476901313,
869
+ "learning_rate": 2.325608815639687e-06,
870
+ "logits/chosen": -0.9941130876541138,
871
+ "logits/rejected": -0.8515572547912598,
872
+ "logps/chosen": -59.786705017089844,
873
+ "logps/rejected": -1280.828857421875,
874
+ "loss": 0.0003,
875
+ "rewards/accuracies": 1.0,
876
+ "rewards/chosen": 0.19204822182655334,
877
+ "rewards/margins": 12.262724876403809,
878
+ "rewards/rejected": -12.070676803588867,
879
+ "step": 570
880
+ },
881
+ {
882
+ "epoch": 11.6,
883
+ "grad_norm": 0.001417529626290414,
884
+ "learning_rate": 2.238678841830867e-06,
885
+ "logits/chosen": -1.0167783498764038,
886
+ "logits/rejected": -0.8408117294311523,
887
+ "logps/chosen": -61.850563049316406,
888
+ "logps/rejected": -1273.6541748046875,
889
+ "loss": 0.0003,
890
+ "rewards/accuracies": 1.0,
891
+ "rewards/chosen": 0.17794856429100037,
892
+ "rewards/margins": 12.197797775268555,
893
+ "rewards/rejected": -12.019848823547363,
894
+ "step": 580
895
+ },
896
+ {
897
+ "epoch": 11.8,
898
+ "grad_norm": 0.001985700625240649,
899
+ "learning_rate": 2.1520672475998374e-06,
900
+ "logits/chosen": -0.9488497972488403,
901
+ "logits/rejected": -0.8910242319107056,
902
+ "logps/chosen": -56.951904296875,
903
+ "logps/rejected": -1185.1470947265625,
904
+ "loss": 0.0003,
905
+ "rewards/accuracies": 1.0,
906
+ "rewards/chosen": 0.17158612608909607,
907
+ "rewards/margins": 11.239965438842773,
908
+ "rewards/rejected": -11.068379402160645,
909
+ "step": 590
910
+ },
911
+ {
912
+ "epoch": 12.0,
913
+ "grad_norm": 0.0014036347657331724,
914
+ "learning_rate": 2.0658795558326745e-06,
915
+ "logits/chosen": -0.9536435008049011,
916
+ "logits/rejected": -0.8297941088676453,
917
+ "logps/chosen": -57.86296463012695,
918
+ "logps/rejected": -1237.7889404296875,
919
+ "loss": 0.0003,
920
+ "rewards/accuracies": 1.0,
921
+ "rewards/chosen": 0.1768932044506073,
922
+ "rewards/margins": 11.793856620788574,
923
+ "rewards/rejected": -11.616963386535645,
924
+ "step": 600
925
+ },
926
+ {
927
+ "epoch": 12.2,
928
+ "grad_norm": 0.0014637474046611807,
929
+ "learning_rate": 1.9802207729556023e-06,
930
+ "logits/chosen": -0.9591706395149231,
931
+ "logits/rejected": -0.8812329173088074,
932
+ "logps/chosen": -56.78215789794922,
933
+ "logps/rejected": -1231.258056640625,
934
+ "loss": 0.0003,
935
+ "rewards/accuracies": 1.0,
936
+ "rewards/chosen": 0.1855459213256836,
937
+ "rewards/margins": 11.722236633300781,
938
+ "rewards/rejected": -11.536690711975098,
939
+ "step": 610
940
+ },
941
+ {
942
+ "epoch": 12.4,
943
+ "grad_norm": 0.0017805033618398401,
944
+ "learning_rate": 1.895195261000831e-06,
945
+ "logits/chosen": -1.019749641418457,
946
+ "logits/rejected": -0.9892827868461609,
947
+ "logps/chosen": -58.21343994140625,
948
+ "logps/rejected": -1209.8052978515625,
949
+ "loss": 0.0003,
950
+ "rewards/accuracies": 1.0,
951
+ "rewards/chosen": 0.16973453760147095,
952
+ "rewards/margins": 11.481805801391602,
953
+ "rewards/rejected": -11.312070846557617,
954
+ "step": 620
955
+ },
956
+ {
957
+ "epoch": 12.6,
958
+ "grad_norm": 0.0011184561683295657,
959
+ "learning_rate": 1.8109066104575023e-06,
960
+ "logits/chosen": -0.9216413497924805,
961
+ "logits/rejected": -0.8319498300552368,
962
+ "logps/chosen": -58.66545486450195,
963
+ "logps/rejected": -1199.0224609375,
964
+ "loss": 0.0003,
965
+ "rewards/accuracies": 1.0,
966
+ "rewards/chosen": 0.17631840705871582,
967
+ "rewards/margins": 11.408339500427246,
968
+ "rewards/rejected": -11.232022285461426,
969
+ "step": 630
970
+ },
971
+ {
972
+ "epoch": 12.8,
973
+ "grad_norm": 0.0014408987665450045,
974
+ "learning_rate": 1.7274575140626318e-06,
975
+ "logits/chosen": -0.9979192614555359,
976
+ "logits/rejected": -0.7878575325012207,
977
+ "logps/chosen": -62.93827438354492,
978
+ "logps/rejected": -1300.874755859375,
979
+ "loss": 0.0003,
980
+ "rewards/accuracies": 1.0,
981
+ "rewards/chosen": 0.18857316672801971,
982
+ "rewards/margins": 12.50341510772705,
983
+ "rewards/rejected": -12.314840316772461,
984
+ "step": 640
985
+ },
986
+ {
987
+ "epoch": 13.0,
988
+ "grad_norm": 0.0020645792333772377,
989
+ "learning_rate": 1.6449496416858285e-06,
990
+ "logits/chosen": -0.9246803522109985,
991
+ "logits/rejected": -0.8380001783370972,
992
+ "logps/chosen": -57.9194221496582,
993
+ "logps/rejected": -1195.3837890625,
994
+ "loss": 0.0003,
995
+ "rewards/accuracies": 1.0,
996
+ "rewards/chosen": 0.16603608429431915,
997
+ "rewards/margins": 11.333303451538086,
998
+ "rewards/rejected": -11.167266845703125,
999
+ "step": 650
1000
+ },
1001
+ {
1002
+ "epoch": 13.2,
1003
+ "grad_norm": 0.0015139495857028652,
1004
+ "learning_rate": 1.56348351646022e-06,
1005
+ "logits/chosen": -1.0453752279281616,
1006
+ "logits/rejected": -0.8043217658996582,
1007
+ "logps/chosen": -63.8051872253418,
1008
+ "logps/rejected": -1392.04443359375,
1009
+ "loss": 0.0002,
1010
+ "rewards/accuracies": 1.0,
1011
+ "rewards/chosen": 0.19428420066833496,
1012
+ "rewards/margins": 13.428848266601562,
1013
+ "rewards/rejected": -13.234565734863281,
1014
+ "step": 660
1015
+ },
1016
+ {
1017
+ "epoch": 13.4,
1018
+ "grad_norm": 0.0013498029319408118,
1019
+ "learning_rate": 1.4831583923105e-06,
1020
+ "logits/chosen": -0.979832649230957,
1021
+ "logits/rejected": -0.8365589380264282,
1022
+ "logps/chosen": -60.9407844543457,
1023
+ "logps/rejected": -1260.807861328125,
1024
+ "loss": 0.0003,
1025
+ "rewards/accuracies": 1.0,
1026
+ "rewards/chosen": 0.16681666672229767,
1027
+ "rewards/margins": 12.033578872680664,
1028
+ "rewards/rejected": -11.866762161254883,
1029
+ "step": 670
1030
+ },
1031
+ {
1032
+ "epoch": 13.6,
1033
+ "grad_norm": 0.0018616946971343861,
1034
+ "learning_rate": 1.4040721330273063e-06,
1035
+ "logits/chosen": -0.9041559100151062,
1036
+ "logits/rejected": -0.8964747190475464,
1037
+ "logps/chosen": -54.83356857299805,
1038
+ "logps/rejected": -1143.5806884765625,
1039
+ "loss": 0.0003,
1040
+ "rewards/accuracies": 1.0,
1041
+ "rewards/chosen": 0.16955164074897766,
1042
+ "rewards/margins": 10.786935806274414,
1043
+ "rewards/rejected": -10.617385864257812,
1044
+ "step": 680
1045
+ },
1046
+ {
1047
+ "epoch": 13.8,
1048
+ "grad_norm": 0.0012150873329068216,
1049
+ "learning_rate": 1.3263210930352737e-06,
1050
+ "logits/chosen": -0.8522924184799194,
1051
+ "logits/rejected": -0.8536826968193054,
1052
+ "logps/chosen": -57.020355224609375,
1053
+ "logps/rejected": -1151.3582763671875,
1054
+ "loss": 0.0003,
1055
+ "rewards/accuracies": 1.0,
1056
+ "rewards/chosen": 0.16875942051410675,
1057
+ "rewards/margins": 10.881406784057617,
1058
+ "rewards/rejected": -10.712648391723633,
1059
+ "step": 690
1060
+ },
1061
+ {
1062
+ "epoch": 14.0,
1063
+ "grad_norm": 0.0015006535309411454,
1064
+ "learning_rate": 1.2500000000000007e-06,
1065
+ "logits/chosen": -0.9550180435180664,
1066
+ "logits/rejected": -0.8258699178695679,
1067
+ "logps/chosen": -58.025535583496094,
1068
+ "logps/rejected": -1249.10546875,
1069
+ "loss": 0.0003,
1070
+ "rewards/accuracies": 1.0,
1071
+ "rewards/chosen": 0.17841221392154694,
1072
+ "rewards/margins": 11.912469863891602,
1073
+ "rewards/rejected": -11.73405647277832,
1074
+ "step": 700
1075
+ },
1076
+ {
1077
+ "epoch": 14.2,
1078
+ "grad_norm": 0.0015715381830458688,
1079
+ "learning_rate": 1.1752018394169882e-06,
1080
+ "logits/chosen": -0.9631673693656921,
1081
+ "logits/rejected": -0.8420494794845581,
1082
+ "logps/chosen": -60.133888244628906,
1083
+ "logps/rejected": -1249.9853515625,
1084
+ "loss": 0.0002,
1085
+ "rewards/accuracies": 1.0,
1086
+ "rewards/chosen": 0.1643882542848587,
1087
+ "rewards/margins": 11.908964157104492,
1088
+ "rewards/rejected": -11.744573593139648,
1089
+ "step": 710
1090
+ },
1091
+ {
1092
+ "epoch": 14.4,
1093
+ "grad_norm": 0.0013660785683358696,
1094
+ "learning_rate": 1.1020177413231334e-06,
1095
+ "logits/chosen": -0.8709633946418762,
1096
+ "logits/rejected": -0.8834745287895203,
1097
+ "logps/chosen": -54.35590362548828,
1098
+ "logps/rejected": -1181.2337646484375,
1099
+ "loss": 0.0003,
1100
+ "rewards/accuracies": 1.0,
1101
+ "rewards/chosen": 0.17913100123405457,
1102
+ "rewards/margins": 11.173266410827637,
1103
+ "rewards/rejected": -10.994134902954102,
1104
+ "step": 720
1105
+ },
1106
+ {
1107
+ "epoch": 14.6,
1108
+ "grad_norm": 0.0016180177521581223,
1109
+ "learning_rate": 1.0305368692688175e-06,
1110
+ "logits/chosen": -0.9120365381240845,
1111
+ "logits/rejected": -0.7947243452072144,
1112
+ "logps/chosen": -59.50437545776367,
1113
+ "logps/rejected": -1239.4515380859375,
1114
+ "loss": 0.0003,
1115
+ "rewards/accuracies": 1.0,
1116
+ "rewards/chosen": 0.15415535867214203,
1117
+ "rewards/margins": 11.778593063354492,
1118
+ "rewards/rejected": -11.624438285827637,
1119
+ "step": 730
1120
+ },
1121
+ {
1122
+ "epoch": 14.8,
1123
+ "grad_norm": 0.0009131551082822374,
1124
+ "learning_rate": 9.608463116858544e-07,
1125
+ "logits/chosen": -0.9982080459594727,
1126
+ "logits/rejected": -0.8127807378768921,
1127
+ "logps/chosen": -62.322227478027344,
1128
+ "logps/rejected": -1302.121337890625,
1129
+ "loss": 0.0003,
1130
+ "rewards/accuracies": 1.0,
1131
+ "rewards/chosen": 0.17191007733345032,
1132
+ "rewards/margins": 12.468157768249512,
1133
+ "rewards/rejected": -12.296248435974121,
1134
+ "step": 740
1135
+ },
1136
+ {
1137
+ "epoch": 15.0,
1138
+ "grad_norm": 0.004586809943697903,
1139
+ "learning_rate": 8.930309757836517e-07,
1140
+ "logits/chosen": -0.9402343034744263,
1141
+ "logits/rejected": -0.8805861473083496,
1142
+ "logps/chosen": -56.05229568481445,
1143
+ "logps/rejected": -1228.699951171875,
1144
+ "loss": 0.0003,
1145
+ "rewards/accuracies": 1.0,
1146
+ "rewards/chosen": 0.19031982123851776,
1147
+ "rewards/margins": 11.688825607299805,
1148
+ "rewards/rejected": -11.498506546020508,
1149
+ "step": 750
1150
+ },
1151
+ {
1152
+ "epoch": 15.2,
1153
+ "grad_norm": 0.001855508098495543,
1154
+ "learning_rate": 8.271734841028553e-07,
1155
+ "logits/chosen": -0.9797531962394714,
1156
+ "logits/rejected": -0.8751746416091919,
1157
+ "logps/chosen": -59.685791015625,
1158
+ "logps/rejected": -1249.674560546875,
1159
+ "loss": 0.0003,
1160
+ "rewards/accuracies": 1.0,
1161
+ "rewards/chosen": 0.17554932832717896,
1162
+ "rewards/margins": 11.926473617553711,
1163
+ "rewards/rejected": -11.750923156738281,
1164
+ "step": 760
1165
+ },
1166
+ {
1167
+ "epoch": 15.4,
1168
+ "grad_norm": 0.0009559205861326566,
1169
+ "learning_rate": 7.633540738525066e-07,
1170
+ "logits/chosen": -0.9692495465278625,
1171
+ "logits/rejected": -0.8025411367416382,
1172
+ "logps/chosen": -60.701194763183594,
1173
+ "logps/rejected": -1309.8841552734375,
1174
+ "loss": 0.0002,
1175
+ "rewards/accuracies": 1.0,
1176
+ "rewards/chosen": 0.183763325214386,
1177
+ "rewards/margins": 12.551748275756836,
1178
+ "rewards/rejected": -12.367984771728516,
1179
+ "step": 770
1180
+ },
1181
+ {
1182
+ "epoch": 15.6,
1183
+ "grad_norm": 0.0016221612239979135,
1184
+ "learning_rate": 7.016504991533727e-07,
1185
+ "logits/chosen": -0.9435825347900391,
1186
+ "logits/rejected": -0.873455822467804,
1187
+ "logps/chosen": -58.68913650512695,
1188
+ "logps/rejected": -1226.319091796875,
1189
+ "loss": 0.0003,
1190
+ "rewards/accuracies": 1.0,
1191
+ "rewards/chosen": 0.16507244110107422,
1192
+ "rewards/margins": 11.652719497680664,
1193
+ "rewards/rejected": -11.487646102905273,
1194
+ "step": 780
1195
+ },
1196
+ {
1197
+ "epoch": 15.8,
1198
+ "grad_norm": 0.0011565013550164168,
1199
+ "learning_rate": 6.421379363065142e-07,
1200
+ "logits/chosen": -1.0095231533050537,
1201
+ "logits/rejected": -0.7704787850379944,
1202
+ "logps/chosen": -65.56999206542969,
1203
+ "logps/rejected": -1337.3360595703125,
1204
+ "loss": 0.0003,
1205
+ "rewards/accuracies": 1.0,
1206
+ "rewards/chosen": 0.18380019068717957,
1207
+ "rewards/margins": 12.87743091583252,
1208
+ "rewards/rejected": -12.693631172180176,
1209
+ "step": 790
1210
+ },
1211
+ {
1212
+ "epoch": 16.0,
1213
+ "grad_norm": 0.0012537423983021187,
1214
+ "learning_rate": 5.848888922025553e-07,
1215
+ "logits/chosen": -0.8937209248542786,
1216
+ "logits/rejected": -0.8747299313545227,
1217
+ "logps/chosen": -54.911643981933594,
1218
+ "logps/rejected": -1144.4791259765625,
1219
+ "loss": 0.0003,
1220
+ "rewards/accuracies": 1.0,
1221
+ "rewards/chosen": 0.16021445393562317,
1222
+ "rewards/margins": 10.787330627441406,
1223
+ "rewards/rejected": -10.627116203308105,
1224
+ "step": 800
1225
+ },
1226
+ {
1227
+ "epoch": 16.2,
1228
+ "grad_norm": 0.0014373038754308242,
1229
+ "learning_rate": 5.299731159831953e-07,
1230
+ "logits/chosen": -0.9762552976608276,
1231
+ "logits/rejected": -0.8652079701423645,
1232
+ "logps/chosen": -58.197059631347656,
1233
+ "logps/rejected": -1232.221435546875,
1234
+ "loss": 0.0003,
1235
+ "rewards/accuracies": 1.0,
1236
+ "rewards/chosen": 0.1831381916999817,
1237
+ "rewards/margins": 11.740102767944336,
1238
+ "rewards/rejected": -11.556963920593262,
1239
+ "step": 810
1240
+ },
1241
+ {
1242
+ "epoch": 16.4,
1243
+ "grad_norm": 0.000862440552553043,
1244
+ "learning_rate": 4.774575140626317e-07,
1245
+ "logits/chosen": -0.9985950589179993,
1246
+ "logits/rejected": -0.8936964273452759,
1247
+ "logps/chosen": -58.0911979675293,
1248
+ "logps/rejected": -1282.3681640625,
1249
+ "loss": 0.0003,
1250
+ "rewards/accuracies": 1.0,
1251
+ "rewards/chosen": 0.19094052910804749,
1252
+ "rewards/margins": 12.265246391296387,
1253
+ "rewards/rejected": -12.074307441711426,
1254
+ "step": 820
1255
+ },
1256
+ {
1257
+ "epoch": 16.6,
1258
+ "grad_norm": 0.0011309096722098083,
1259
+ "learning_rate": 4.27406068612396e-07,
1260
+ "logits/chosen": -0.9642303586006165,
1261
+ "logits/rejected": -0.8438752293586731,
1262
+ "logps/chosen": -59.80333709716797,
1263
+ "logps/rejected": -1246.0048828125,
1264
+ "loss": 0.0002,
1265
+ "rewards/accuracies": 1.0,
1266
+ "rewards/chosen": 0.16420656442642212,
1267
+ "rewards/margins": 11.867765426635742,
1268
+ "rewards/rejected": -11.703559875488281,
1269
+ "step": 830
1270
+ },
1271
+ {
1272
+ "epoch": 16.8,
1273
+ "grad_norm": 0.001634196377460084,
1274
+ "learning_rate": 3.798797596089351e-07,
1275
+ "logits/chosen": -0.9439367055892944,
1276
+ "logits/rejected": -0.8413814306259155,
1277
+ "logps/chosen": -59.905982971191406,
1278
+ "logps/rejected": -1256.216796875,
1279
+ "loss": 0.0003,
1280
+ "rewards/accuracies": 1.0,
1281
+ "rewards/chosen": 0.16103532910346985,
1282
+ "rewards/margins": 11.955886840820312,
1283
+ "rewards/rejected": -11.794851303100586,
1284
+ "step": 840
1285
+ },
1286
+ {
1287
+ "epoch": 17.0,
1288
+ "grad_norm": 0.0011128513110797832,
1289
+ "learning_rate": 3.3493649053890325e-07,
1290
+ "logits/chosen": -0.9567734003067017,
1291
+ "logits/rejected": -0.7875978350639343,
1292
+ "logps/chosen": -61.99500274658203,
1293
+ "logps/rejected": -1306.9703369140625,
1294
+ "loss": 0.0002,
1295
+ "rewards/accuracies": 1.0,
1296
+ "rewards/chosen": 0.1775272786617279,
1297
+ "rewards/margins": 12.505702018737793,
1298
+ "rewards/rejected": -12.32817554473877,
1299
+ "step": 850
1300
+ },
1301
+ {
1302
+ "epoch": 17.2,
1303
+ "grad_norm": 0.0013520172966032026,
1304
+ "learning_rate": 2.9263101785268253e-07,
1305
+ "logits/chosen": -0.9488283395767212,
1306
+ "logits/rejected": -0.8742234110832214,
1307
+ "logps/chosen": -56.80482864379883,
1308
+ "logps/rejected": -1268.548095703125,
1309
+ "loss": 0.0003,
1310
+ "rewards/accuracies": 1.0,
1311
+ "rewards/chosen": 0.17435792088508606,
1312
+ "rewards/margins": 12.08137321472168,
1313
+ "rewards/rejected": -11.907015800476074,
1314
+ "step": 860
1315
+ },
1316
+ {
1317
+ "epoch": 17.4,
1318
+ "grad_norm": 0.0012511118706732029,
1319
+ "learning_rate": 2.53014884252083e-07,
1320
+ "logits/chosen": -1.0541223287582397,
1321
+ "logits/rejected": -0.81927490234375,
1322
+ "logps/chosen": -63.943763732910156,
1323
+ "logps/rejected": -1374.1676025390625,
1324
+ "loss": 0.0002,
1325
+ "rewards/accuracies": 1.0,
1326
+ "rewards/chosen": 0.18310335278511047,
1327
+ "rewards/margins": 13.212549209594727,
1328
+ "rewards/rejected": -13.029446601867676,
1329
+ "step": 870
1330
+ },
1331
+ {
1332
+ "epoch": 17.6,
1333
+ "grad_norm": 0.0018809600968994445,
1334
+ "learning_rate": 2.1613635589349756e-07,
1335
+ "logits/chosen": -0.8839080929756165,
1336
+ "logits/rejected": -0.7969019412994385,
1337
+ "logps/chosen": -56.40065383911133,
1338
+ "logps/rejected": -1226.865478515625,
1339
+ "loss": 0.0003,
1340
+ "rewards/accuracies": 1.0,
1341
+ "rewards/chosen": 0.17095676064491272,
1342
+ "rewards/margins": 11.626653671264648,
1343
+ "rewards/rejected": -11.455695152282715,
1344
+ "step": 880
1345
+ },
1346
+ {
1347
+ "epoch": 17.8,
1348
+ "grad_norm": 0.0016373893137363934,
1349
+ "learning_rate": 1.8204036358303173e-07,
1350
+ "logits/chosen": -0.9771726727485657,
1351
+ "logits/rejected": -0.8529565930366516,
1352
+ "logps/chosen": -57.20879364013672,
1353
+ "logps/rejected": -1290.327880859375,
1354
+ "loss": 0.0002,
1355
+ "rewards/accuracies": 1.0,
1356
+ "rewards/chosen": 0.18934224545955658,
1357
+ "rewards/margins": 12.324792861938477,
1358
+ "rewards/rejected": -12.135449409484863,
1359
+ "step": 890
1360
+ },
1361
+ {
1362
+ "epoch": 18.0,
1363
+ "grad_norm": 0.000786600665947747,
1364
+ "learning_rate": 1.507684480352292e-07,
1365
+ "logits/chosen": -0.9094891548156738,
1366
+ "logits/rejected": -0.8147133588790894,
1367
+ "logps/chosen": -59.374755859375,
1368
+ "logps/rejected": -1234.203125,
1369
+ "loss": 0.0003,
1370
+ "rewards/accuracies": 1.0,
1371
+ "rewards/chosen": 0.1541765034198761,
1372
+ "rewards/margins": 11.714710235595703,
1373
+ "rewards/rejected": -11.560534477233887,
1374
+ "step": 900
1375
+ },
1376
+ {
1377
+ "epoch": 18.2,
1378
+ "grad_norm": 0.002709830744526596,
1379
+ "learning_rate": 1.223587092621162e-07,
1380
+ "logits/chosen": -1.00485098361969,
1381
+ "logits/rejected": -0.7558645606040955,
1382
+ "logps/chosen": -67.38685607910156,
1383
+ "logps/rejected": -1367.7635498046875,
1384
+ "loss": 0.0002,
1385
+ "rewards/accuracies": 1.0,
1386
+ "rewards/chosen": 0.17096880078315735,
1387
+ "rewards/margins": 13.181119918823242,
1388
+ "rewards/rejected": -13.010151863098145,
1389
+ "step": 910
1390
+ },
1391
+ {
1392
+ "epoch": 18.4,
1393
+ "grad_norm": 0.0010941683716015594,
1394
+ "learning_rate": 9.684576015420277e-08,
1395
+ "logits/chosen": -0.9139689207077026,
1396
+ "logits/rejected": -0.8125830888748169,
1397
+ "logps/chosen": -58.29936599731445,
1398
+ "logps/rejected": -1264.84033203125,
1399
+ "loss": 0.0003,
1400
+ "rewards/accuracies": 1.0,
1401
+ "rewards/chosen": 0.16245688498020172,
1402
+ "rewards/margins": 12.02514934539795,
1403
+ "rewards/rejected": -11.862692832946777,
1404
+ "step": 920
1405
+ },
1406
+ {
1407
+ "epoch": 18.6,
1408
+ "grad_norm": 0.001790071139550289,
1409
+ "learning_rate": 7.426068431000883e-08,
1410
+ "logits/chosen": -0.9928874969482422,
1411
+ "logits/rejected": -0.8945854306221008,
1412
+ "logps/chosen": -59.90386962890625,
1413
+ "logps/rejected": -1225.345458984375,
1414
+ "loss": 0.0002,
1415
+ "rewards/accuracies": 1.0,
1416
+ "rewards/chosen": 0.18468160927295685,
1417
+ "rewards/margins": 11.685681343078613,
1418
+ "rewards/rejected": -11.500999450683594,
1419
+ "step": 930
1420
+ },
1421
+ {
1422
+ "epoch": 18.8,
1423
+ "grad_norm": 0.001444924500021094,
1424
+ "learning_rate": 5.463099816548578e-08,
1425
+ "logits/chosen": -0.8352192640304565,
1426
+ "logits/rejected": -0.8455368876457214,
1427
+ "logps/chosen": -53.367652893066406,
1428
+ "logps/rejected": -1094.0107421875,
1429
+ "loss": 0.0003,
1430
+ "rewards/accuracies": 1.0,
1431
+ "rewards/chosen": 0.15318068861961365,
1432
+ "rewards/margins": 10.247284889221191,
1433
+ "rewards/rejected": -10.094103813171387,
1434
+ "step": 940
1435
+ },
1436
+ {
1437
+ "epoch": 19.0,
1438
+ "grad_norm": 0.0011127218090168816,
1439
+ "learning_rate": 3.798061746947995e-08,
1440
+ "logits/chosen": -0.9996277689933777,
1441
+ "logits/rejected": -0.7810768485069275,
1442
+ "logps/chosen": -62.9672737121582,
1443
+ "logps/rejected": -1361.350341796875,
1444
+ "loss": 0.0002,
1445
+ "rewards/accuracies": 1.0,
1446
+ "rewards/chosen": 0.1840919703245163,
1447
+ "rewards/margins": 13.088468551635742,
1448
+ "rewards/rejected": -12.904377937316895,
1449
+ "step": 950
1450
+ },
1451
+ {
1452
+ "epoch": 19.2,
1453
+ "grad_norm": 0.0011199039962919363,
1454
+ "learning_rate": 2.4329828146074096e-08,
1455
+ "logits/chosen": -0.9857956171035767,
1456
+ "logits/rejected": -0.8834479451179504,
1457
+ "logps/chosen": -61.652801513671875,
1458
+ "logps/rejected": -1269.288330078125,
1459
+ "loss": 0.0003,
1460
+ "rewards/accuracies": 1.0,
1461
+ "rewards/chosen": 0.14881816506385803,
1462
+ "rewards/margins": 12.092996597290039,
1463
+ "rewards/rejected": -11.944178581237793,
1464
+ "step": 960
1465
+ },
1466
+ {
1467
+ "epoch": 19.4,
1468
+ "grad_norm": 0.0013686833523596474,
1469
+ "learning_rate": 1.3695261579316776e-08,
1470
+ "logits/chosen": -0.9996697306632996,
1471
+ "logits/rejected": -0.8662905693054199,
1472
+ "logps/chosen": -60.64554977416992,
1473
+ "logps/rejected": -1260.70458984375,
1474
+ "loss": 0.0003,
1475
+ "rewards/accuracies": 1.0,
1476
+ "rewards/chosen": 0.18075986206531525,
1477
+ "rewards/margins": 12.048669815063477,
1478
+ "rewards/rejected": -11.867908477783203,
1479
+ "step": 970
1480
+ },
1481
+ {
1482
+ "epoch": 19.6,
1483
+ "grad_norm": 0.0017745592515844095,
1484
+ "learning_rate": 6.089874350439507e-09,
1485
+ "logits/chosen": -1.0439860820770264,
1486
+ "logits/rejected": -0.9185269474983215,
1487
+ "logps/chosen": -59.79936981201172,
1488
+ "logps/rejected": -1267.5032958984375,
1489
+ "loss": 0.0003,
1490
+ "rewards/accuracies": 1.0,
1491
+ "rewards/chosen": 0.18838214874267578,
1492
+ "rewards/margins": 12.108388900756836,
1493
+ "rewards/rejected": -11.92000675201416,
1494
+ "step": 980
1495
+ },
1496
+ {
1497
+ "epoch": 19.8,
1498
+ "grad_norm": 0.0013032690672856834,
1499
+ "learning_rate": 1.5229324522605949e-09,
1500
+ "logits/chosen": -0.9330822229385376,
1501
+ "logits/rejected": -0.8125349879264832,
1502
+ "logps/chosen": -58.721580505371094,
1503
+ "logps/rejected": -1287.365966796875,
1504
+ "loss": 0.0002,
1505
+ "rewards/accuracies": 1.0,
1506
+ "rewards/chosen": 0.17665794491767883,
1507
+ "rewards/margins": 12.307037353515625,
1508
+ "rewards/rejected": -12.130378723144531,
1509
+ "step": 990
1510
+ },
1511
+ {
1512
+ "epoch": 20.0,
1513
+ "grad_norm": 0.003965487525700407,
1514
+ "learning_rate": 0.0,
1515
+ "logits/chosen": -0.9679548144340515,
1516
+ "logits/rejected": -0.8124309778213501,
1517
+ "logps/chosen": -61.12044143676758,
1518
+ "logps/rejected": -1297.4541015625,
1519
+ "loss": 0.0002,
1520
+ "rewards/accuracies": 1.0,
1521
+ "rewards/chosen": 0.1760241687297821,
1522
+ "rewards/margins": 12.41258716583252,
1523
+ "rewards/rejected": -12.236563682556152,
1524
+ "step": 1000
1525
+ },
1526
+ {
1527
+ "epoch": 20.0,
1528
+ "step": 1000,
1529
+ "total_flos": 0.0,
1530
+ "train_loss": 0.07008945742505603,
1531
+ "train_runtime": 8233.3372,
1532
+ "train_samples_per_second": 1.943,
1533
+ "train_steps_per_second": 0.121
1534
+ }
1535
+ ],
1536
+ "logging_steps": 10,
1537
+ "max_steps": 1000,
1538
+ "num_input_tokens_seen": 0,
1539
+ "num_train_epochs": 20,
1540
+ "save_steps": 1000,
1541
+ "stateful_callbacks": {
1542
+ "TrainerControl": {
1543
+ "args": {
1544
+ "should_epoch_stop": false,
1545
+ "should_evaluate": false,
1546
+ "should_log": false,
1547
+ "should_save": true,
1548
+ "should_training_stop": true
1549
+ },
1550
+ "attributes": {}
1551
+ }
1552
+ },
1553
+ "total_flos": 0.0,
1554
+ "train_batch_size": 2,
1555
+ "trial_name": null,
1556
+ "trial_params": null
1557
+ }