Jan Majkutewicz commited on
Commit
9ccced0
·
verified ·
1 Parent(s): dfd8333

Model save

Browse files
Files changed (4) hide show
  1. README.md +60 -0
  2. all_results.json +9 -0
  3. train_results.json +9 -0
  4. trainer_state.json +1377 -0
README.md ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: alignment-handbook/zephyr-7b-sft-full
3
+ library_name: peft
4
+ license: apache-2.0
5
+ tags:
6
+ - trl
7
+ - dpo
8
+ - generated_from_trainer
9
+ model-index:
10
+ - name: zephyr-7b-oasst1
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
+ [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/jan-majkutewicz/huggingface/runs/uistaj1t)
18
+ # zephyr-7b-oasst1
19
+
20
+ This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) on an unknown dataset.
21
+
22
+ ## Model description
23
+
24
+ More information needed
25
+
26
+ ## Intended uses & limitations
27
+
28
+ More information needed
29
+
30
+ ## Training and evaluation data
31
+
32
+ More information needed
33
+
34
+ ## Training procedure
35
+
36
+ ### Training hyperparameters
37
+
38
+ The following hyperparameters were used during training:
39
+ - learning_rate: 5e-06
40
+ - train_batch_size: 8
41
+ - eval_batch_size: 8
42
+ - seed: 42
43
+ - gradient_accumulation_steps: 2
44
+ - total_train_batch_size: 16
45
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
46
+ - lr_scheduler_type: cosine
47
+ - lr_scheduler_warmup_ratio: 0.1
48
+ - num_epochs: 1
49
+
50
+ ### Training results
51
+
52
+
53
+
54
+ ### Framework versions
55
+
56
+ - PEFT 0.10.0
57
+ - Transformers 4.42.3
58
+ - Pytorch 2.2.0
59
+ - Datasets 2.20.0
60
+ - Tokenizers 0.19.1
all_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 0.9994353472614342,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.6616038349388683,
5
+ "train_runtime": 9964.5835,
6
+ "train_samples": 14167,
7
+ "train_samples_per_second": 1.422,
8
+ "train_steps_per_second": 0.089
9
+ }
train_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 0.9994353472614342,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.6616038349388683,
5
+ "train_runtime": 9964.5835,
6
+ "train_samples": 14167,
7
+ "train_samples_per_second": 1.422,
8
+ "train_steps_per_second": 0.089
9
+ }
trainer_state.json ADDED
@@ -0,0 +1,1377 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.9994353472614342,
5
+ "eval_steps": 1000,
6
+ "global_step": 885,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.001129305477131564,
13
+ "grad_norm": 10.5625,
14
+ "learning_rate": 5.617977528089888e-08,
15
+ "logits/chosen": -2.8912107944488525,
16
+ "logits/rejected": -2.9116690158843994,
17
+ "logps/chosen": -205.43255615234375,
18
+ "logps/rejected": -277.7913818359375,
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": 0.01129305477131564,
28
+ "grad_norm": 11.75,
29
+ "learning_rate": 5.617977528089888e-07,
30
+ "logits/chosen": -2.9238009452819824,
31
+ "logits/rejected": -2.8953051567077637,
32
+ "logps/chosen": -286.31060791015625,
33
+ "logps/rejected": -234.09100341796875,
34
+ "loss": 0.6921,
35
+ "rewards/accuracies": 0.4375,
36
+ "rewards/chosen": 0.0044541251845657825,
37
+ "rewards/margins": 0.0024419513065367937,
38
+ "rewards/rejected": 0.0020121734123677015,
39
+ "step": 10
40
+ },
41
+ {
42
+ "epoch": 0.02258610954263128,
43
+ "grad_norm": 8.0625,
44
+ "learning_rate": 1.1235955056179777e-06,
45
+ "logits/chosen": -2.911529064178467,
46
+ "logits/rejected": -2.8939220905303955,
47
+ "logps/chosen": -242.33349609375,
48
+ "logps/rejected": -230.67916870117188,
49
+ "loss": 0.6967,
50
+ "rewards/accuracies": 0.375,
51
+ "rewards/chosen": -0.00516792107373476,
52
+ "rewards/margins": -0.006779191549867392,
53
+ "rewards/rejected": 0.0016112711746245623,
54
+ "step": 20
55
+ },
56
+ {
57
+ "epoch": 0.03387916431394692,
58
+ "grad_norm": 10.4375,
59
+ "learning_rate": 1.6853932584269663e-06,
60
+ "logits/chosen": -2.933928966522217,
61
+ "logits/rejected": -2.9029603004455566,
62
+ "logps/chosen": -228.00210571289062,
63
+ "logps/rejected": -206.8175506591797,
64
+ "loss": 0.6928,
65
+ "rewards/accuracies": 0.4749999940395355,
66
+ "rewards/chosen": -0.00036367171560414135,
67
+ "rewards/margins": 0.0011087121674790978,
68
+ "rewards/rejected": -0.0014723839703947306,
69
+ "step": 30
70
+ },
71
+ {
72
+ "epoch": 0.04517221908526256,
73
+ "grad_norm": 10.5625,
74
+ "learning_rate": 2.2471910112359554e-06,
75
+ "logits/chosen": -2.9396324157714844,
76
+ "logits/rejected": -2.9021177291870117,
77
+ "logps/chosen": -239.8461456298828,
78
+ "logps/rejected": -223.19216918945312,
79
+ "loss": 0.693,
80
+ "rewards/accuracies": 0.4937500059604645,
81
+ "rewards/chosen": -0.01020533312112093,
82
+ "rewards/margins": 0.0009328271262347698,
83
+ "rewards/rejected": -0.011138159781694412,
84
+ "step": 40
85
+ },
86
+ {
87
+ "epoch": 0.05646527385657821,
88
+ "grad_norm": 11.25,
89
+ "learning_rate": 2.8089887640449444e-06,
90
+ "logits/chosen": -2.9262070655822754,
91
+ "logits/rejected": -2.891216993331909,
92
+ "logps/chosen": -252.71072387695312,
93
+ "logps/rejected": -209.1496124267578,
94
+ "loss": 0.6914,
95
+ "rewards/accuracies": 0.512499988079071,
96
+ "rewards/chosen": -0.02404838800430298,
97
+ "rewards/margins": 0.004404561128467321,
98
+ "rewards/rejected": -0.028452953323721886,
99
+ "step": 50
100
+ },
101
+ {
102
+ "epoch": 0.06775832862789384,
103
+ "grad_norm": 9.125,
104
+ "learning_rate": 3.3707865168539327e-06,
105
+ "logits/chosen": -2.95279860496521,
106
+ "logits/rejected": -2.9229164123535156,
107
+ "logps/chosen": -231.2828826904297,
108
+ "logps/rejected": -223.4114227294922,
109
+ "loss": 0.6893,
110
+ "rewards/accuracies": 0.4937500059604645,
111
+ "rewards/chosen": -0.043369874358177185,
112
+ "rewards/margins": 0.008675651624798775,
113
+ "rewards/rejected": -0.05204553157091141,
114
+ "step": 60
115
+ },
116
+ {
117
+ "epoch": 0.07905138339920949,
118
+ "grad_norm": 16.5,
119
+ "learning_rate": 3.932584269662922e-06,
120
+ "logits/chosen": -2.9178593158721924,
121
+ "logits/rejected": -2.88283109664917,
122
+ "logps/chosen": -249.9050750732422,
123
+ "logps/rejected": -236.652099609375,
124
+ "loss": 0.6814,
125
+ "rewards/accuracies": 0.48750001192092896,
126
+ "rewards/chosen": -0.06348244845867157,
127
+ "rewards/margins": 0.026277879253029823,
128
+ "rewards/rejected": -0.08976032584905624,
129
+ "step": 70
130
+ },
131
+ {
132
+ "epoch": 0.09034443817052512,
133
+ "grad_norm": 11.9375,
134
+ "learning_rate": 4.494382022471911e-06,
135
+ "logits/chosen": -2.893878221511841,
136
+ "logits/rejected": -2.858973979949951,
137
+ "logps/chosen": -244.9037322998047,
138
+ "logps/rejected": -243.09042358398438,
139
+ "loss": 0.674,
140
+ "rewards/accuracies": 0.5687500238418579,
141
+ "rewards/chosen": -0.09704697877168655,
142
+ "rewards/margins": 0.04427679255604744,
143
+ "rewards/rejected": -0.1413237750530243,
144
+ "step": 80
145
+ },
146
+ {
147
+ "epoch": 0.10163749294184077,
148
+ "grad_norm": 6.84375,
149
+ "learning_rate": 4.999980529233026e-06,
150
+ "logits/chosen": -2.9132912158966064,
151
+ "logits/rejected": -2.8923027515411377,
152
+ "logps/chosen": -232.01651000976562,
153
+ "logps/rejected": -232.08056640625,
154
+ "loss": 0.6793,
155
+ "rewards/accuracies": 0.48750001192092896,
156
+ "rewards/chosen": -0.16643305122852325,
157
+ "rewards/margins": 0.03658531233668327,
158
+ "rewards/rejected": -0.20301838219165802,
159
+ "step": 90
160
+ },
161
+ {
162
+ "epoch": 0.11293054771315642,
163
+ "grad_norm": 8.1875,
164
+ "learning_rate": 4.997644404153021e-06,
165
+ "logits/chosen": -2.958055019378662,
166
+ "logits/rejected": -2.9053235054016113,
167
+ "logps/chosen": -232.2222900390625,
168
+ "logps/rejected": -244.8426513671875,
169
+ "loss": 0.6677,
170
+ "rewards/accuracies": 0.550000011920929,
171
+ "rewards/chosen": -0.23467275500297546,
172
+ "rewards/margins": 0.08466038852930069,
173
+ "rewards/rejected": -0.31933316588401794,
174
+ "step": 100
175
+ },
176
+ {
177
+ "epoch": 0.12422360248447205,
178
+ "grad_norm": 9.3125,
179
+ "learning_rate": 4.991418294828522e-06,
180
+ "logits/chosen": -2.896815776824951,
181
+ "logits/rejected": -2.8579037189483643,
182
+ "logps/chosen": -256.8934020996094,
183
+ "logps/rejected": -252.8817138671875,
184
+ "loss": 0.6693,
185
+ "rewards/accuracies": 0.543749988079071,
186
+ "rewards/chosen": -0.33838751912117004,
187
+ "rewards/margins": 0.08679252117872238,
188
+ "rewards/rejected": -0.425180047750473,
189
+ "step": 110
190
+ },
191
+ {
192
+ "epoch": 0.13551665725578768,
193
+ "grad_norm": 8.3125,
194
+ "learning_rate": 4.981311898183214e-06,
195
+ "logits/chosen": -2.9425575733184814,
196
+ "logits/rejected": -2.927373170852661,
197
+ "logps/chosen": -236.5902099609375,
198
+ "logps/rejected": -218.78945922851562,
199
+ "loss": 0.6899,
200
+ "rewards/accuracies": 0.4625000059604645,
201
+ "rewards/chosen": -0.39529091119766235,
202
+ "rewards/margins": 0.03407430276274681,
203
+ "rewards/rejected": -0.42936521768569946,
204
+ "step": 120
205
+ },
206
+ {
207
+ "epoch": 0.14680971202710333,
208
+ "grad_norm": 9.5,
209
+ "learning_rate": 4.967340954537717e-06,
210
+ "logits/chosen": -2.9237935543060303,
211
+ "logits/rejected": -2.896763324737549,
212
+ "logps/chosen": -249.92367553710938,
213
+ "logps/rejected": -257.2637023925781,
214
+ "loss": 0.6441,
215
+ "rewards/accuracies": 0.6625000238418579,
216
+ "rewards/chosen": -0.41192659735679626,
217
+ "rewards/margins": 0.14285169541835785,
218
+ "rewards/rejected": -0.5547782778739929,
219
+ "step": 130
220
+ },
221
+ {
222
+ "epoch": 0.15810276679841898,
223
+ "grad_norm": 7.09375,
224
+ "learning_rate": 4.9495272230946525e-06,
225
+ "logits/chosen": -2.9524216651916504,
226
+ "logits/rejected": -2.915919303894043,
227
+ "logps/chosen": -252.70565795898438,
228
+ "logps/rejected": -242.8680419921875,
229
+ "loss": 0.6634,
230
+ "rewards/accuracies": 0.543749988079071,
231
+ "rewards/chosen": -0.48054733872413635,
232
+ "rewards/margins": 0.10070822387933731,
233
+ "rewards/rejected": -0.5812555551528931,
234
+ "step": 140
235
+ },
236
+ {
237
+ "epoch": 0.16939582156973462,
238
+ "grad_norm": 10.4375,
239
+ "learning_rate": 4.927898448049522e-06,
240
+ "logits/chosen": -2.9190261363983154,
241
+ "logits/rejected": -2.878838062286377,
242
+ "logps/chosen": -262.5068359375,
243
+ "logps/rejected": -233.3218536376953,
244
+ "loss": 0.6398,
245
+ "rewards/accuracies": 0.606249988079071,
246
+ "rewards/chosen": -0.5640592575073242,
247
+ "rewards/margins": 0.16638214886188507,
248
+ "rewards/rejected": -0.7304413914680481,
249
+ "step": 150
250
+ },
251
+ {
252
+ "epoch": 0.18068887634105024,
253
+ "grad_norm": 8.5,
254
+ "learning_rate": 4.9024883153802095e-06,
255
+ "logits/chosen": -2.9224305152893066,
256
+ "logits/rejected": -2.8813347816467285,
257
+ "logps/chosen": -225.17919921875,
258
+ "logps/rejected": -219.84487915039062,
259
+ "loss": 0.64,
260
+ "rewards/accuracies": 0.581250011920929,
261
+ "rewards/chosen": -0.6268054842948914,
262
+ "rewards/margins": 0.16882720589637756,
263
+ "rewards/rejected": -0.7956327199935913,
264
+ "step": 160
265
+ },
266
+ {
267
+ "epoch": 0.1919819311123659,
268
+ "grad_norm": 90.5,
269
+ "learning_rate": 4.873336400382367e-06,
270
+ "logits/chosen": -2.952296733856201,
271
+ "logits/rejected": -2.915030002593994,
272
+ "logps/chosen": -253.6019744873047,
273
+ "logps/rejected": -242.11801147460938,
274
+ "loss": 0.6533,
275
+ "rewards/accuracies": 0.581250011920929,
276
+ "rewards/chosen": -0.6432451009750366,
277
+ "rewards/margins": 0.15737922489643097,
278
+ "rewards/rejected": -0.8006243705749512,
279
+ "step": 170
280
+ },
281
+ {
282
+ "epoch": 0.20327498588368154,
283
+ "grad_norm": 8.5625,
284
+ "learning_rate": 4.8404881060324375e-06,
285
+ "logits/chosen": -2.9133620262145996,
286
+ "logits/rejected": -2.8687403202056885,
287
+ "logps/chosen": -250.570068359375,
288
+ "logps/rejected": -209.63168334960938,
289
+ "loss": 0.6597,
290
+ "rewards/accuracies": 0.6187499761581421,
291
+ "rewards/chosen": -0.7379518747329712,
292
+ "rewards/margins": 0.15293964743614197,
293
+ "rewards/rejected": -0.8908915519714355,
294
+ "step": 180
295
+ },
296
+ {
297
+ "epoch": 0.21456804065499718,
298
+ "grad_norm": 10.1875,
299
+ "learning_rate": 4.803994592274272e-06,
300
+ "logits/chosen": -2.930187225341797,
301
+ "logits/rejected": -2.9065961837768555,
302
+ "logps/chosen": -255.08895874023438,
303
+ "logps/rejected": -240.781982421875,
304
+ "loss": 0.691,
305
+ "rewards/accuracies": 0.5,
306
+ "rewards/chosen": -0.6726095676422119,
307
+ "rewards/margins": 0.09454379975795746,
308
+ "rewards/rejected": -0.7671533226966858,
309
+ "step": 190
310
+ },
311
+ {
312
+ "epoch": 0.22586109542631283,
313
+ "grad_norm": 8.625,
314
+ "learning_rate": 4.763912696339506e-06,
315
+ "logits/chosen": -2.908806800842285,
316
+ "logits/rejected": -2.87739896774292,
317
+ "logps/chosen": -244.31326293945312,
318
+ "logps/rejected": -251.88284301757812,
319
+ "loss": 0.6507,
320
+ "rewards/accuracies": 0.574999988079071,
321
+ "rewards/chosen": -0.6921225786209106,
322
+ "rewards/margins": 0.16923566162586212,
323
+ "rewards/rejected": -0.8613582849502563,
324
+ "step": 200
325
+ },
326
+ {
327
+ "epoch": 0.23715415019762845,
328
+ "grad_norm": 11.5,
329
+ "learning_rate": 4.720304844225781e-06,
330
+ "logits/chosen": -2.9554803371429443,
331
+ "logits/rejected": -2.9158945083618164,
332
+ "logps/chosen": -259.2450866699219,
333
+ "logps/rejected": -220.19070434570312,
334
+ "loss": 0.6435,
335
+ "rewards/accuracies": 0.5687500238418579,
336
+ "rewards/chosen": -0.634341835975647,
337
+ "rewards/margins": 0.1825425922870636,
338
+ "rewards/rejected": -0.8168843388557434,
339
+ "step": 210
340
+ },
341
+ {
342
+ "epoch": 0.2484472049689441,
343
+ "grad_norm": 13.125,
344
+ "learning_rate": 4.6732389534706655e-06,
345
+ "logits/chosen": -2.941028118133545,
346
+ "logits/rejected": -2.9083430767059326,
347
+ "logps/chosen": -279.50592041015625,
348
+ "logps/rejected": -249.6971893310547,
349
+ "loss": 0.6602,
350
+ "rewards/accuracies": 0.5249999761581421,
351
+ "rewards/chosen": -0.7295074462890625,
352
+ "rewards/margins": 0.16120071709156036,
353
+ "rewards/rejected": -0.8907082676887512,
354
+ "step": 220
355
+ },
356
+ {
357
+ "epoch": 0.2597402597402597,
358
+ "grad_norm": 10.25,
359
+ "learning_rate": 4.622788327372736e-06,
360
+ "logits/chosen": -2.9610531330108643,
361
+ "logits/rejected": -2.9274752140045166,
362
+ "logps/chosen": -277.1114196777344,
363
+ "logps/rejected": -249.110107421875,
364
+ "loss": 0.6483,
365
+ "rewards/accuracies": 0.543749988079071,
366
+ "rewards/chosen": -0.7016745209693909,
367
+ "rewards/margins": 0.18496869504451752,
368
+ "rewards/rejected": -0.8866432309150696,
369
+ "step": 230
370
+ },
371
+ {
372
+ "epoch": 0.27103331451157536,
373
+ "grad_norm": 12.8125,
374
+ "learning_rate": 4.569031540824526e-06,
375
+ "logits/chosen": -2.933371067047119,
376
+ "logits/rejected": -2.890542984008789,
377
+ "logps/chosen": -237.04415893554688,
378
+ "logps/rejected": -225.23086547851562,
379
+ "loss": 0.6613,
380
+ "rewards/accuracies": 0.53125,
381
+ "rewards/chosen": -0.6652556657791138,
382
+ "rewards/margins": 0.1700500100851059,
383
+ "rewards/rejected": -0.8353056907653809,
384
+ "step": 240
385
+ },
386
+ {
387
+ "epoch": 0.282326369282891,
388
+ "grad_norm": 8.0625,
389
+ "learning_rate": 4.512052317935184e-06,
390
+ "logits/chosen": -2.958394765853882,
391
+ "logits/rejected": -2.9169325828552246,
392
+ "logps/chosen": -239.5146942138672,
393
+ "logps/rejected": -229.6261749267578,
394
+ "loss": 0.6111,
395
+ "rewards/accuracies": 0.637499988079071,
396
+ "rewards/chosen": -0.7014538049697876,
397
+ "rewards/margins": 0.305302232503891,
398
+ "rewards/rejected": -1.006756067276001,
399
+ "step": 250
400
+ },
401
+ {
402
+ "epoch": 0.29361942405420666,
403
+ "grad_norm": 11.375,
404
+ "learning_rate": 4.45193940163342e-06,
405
+ "logits/chosen": -2.9349873065948486,
406
+ "logits/rejected": -2.9028127193450928,
407
+ "logps/chosen": -242.2169189453125,
408
+ "logps/rejected": -246.52584838867188,
409
+ "loss": 0.6804,
410
+ "rewards/accuracies": 0.625,
411
+ "rewards/chosen": -0.7285553812980652,
412
+ "rewards/margins": 0.13187848031520844,
413
+ "rewards/rejected": -0.8604338765144348,
414
+ "step": 260
415
+ },
416
+ {
417
+ "epoch": 0.3049124788255223,
418
+ "grad_norm": 9.9375,
419
+ "learning_rate": 4.3887864154538426e-06,
420
+ "logits/chosen": -2.948988437652588,
421
+ "logits/rejected": -2.934638500213623,
422
+ "logps/chosen": -241.4674072265625,
423
+ "logps/rejected": -244.40072631835938,
424
+ "loss": 0.6575,
425
+ "rewards/accuracies": 0.5625,
426
+ "rewards/chosen": -0.7441710233688354,
427
+ "rewards/margins": 0.2095833122730255,
428
+ "rewards/rejected": -0.9537544250488281,
429
+ "step": 270
430
+ },
431
+ {
432
+ "epoch": 0.31620553359683795,
433
+ "grad_norm": 10.0625,
434
+ "learning_rate": 4.322691717721927e-06,
435
+ "logits/chosen": -2.9385104179382324,
436
+ "logits/rejected": -2.9133810997009277,
437
+ "logps/chosen": -264.28955078125,
438
+ "logps/rejected": -279.1156311035156,
439
+ "loss": 0.6609,
440
+ "rewards/accuracies": 0.5625,
441
+ "rewards/chosen": -0.7867005467414856,
442
+ "rewards/margins": 0.16506877541542053,
443
+ "rewards/rejected": -0.9517693519592285,
444
+ "step": 280
445
+ },
446
+ {
447
+ "epoch": 0.3274985883681536,
448
+ "grad_norm": 14.125,
449
+ "learning_rate": 4.253758248364747e-06,
450
+ "logits/chosen": -2.9210855960845947,
451
+ "logits/rejected": -2.913820743560791,
452
+ "logps/chosen": -258.10015869140625,
453
+ "logps/rejected": -252.4558563232422,
454
+ "loss": 0.7288,
455
+ "rewards/accuracies": 0.5249999761581421,
456
+ "rewards/chosen": -0.8864612579345703,
457
+ "rewards/margins": 0.033281516283750534,
458
+ "rewards/rejected": -0.9197427034378052,
459
+ "step": 290
460
+ },
461
+ {
462
+ "epoch": 0.33879164313946925,
463
+ "grad_norm": 11.1875,
464
+ "learning_rate": 4.182093368586034e-06,
465
+ "logits/chosen": -2.935784101486206,
466
+ "logits/rejected": -2.9176745414733887,
467
+ "logps/chosen": -277.13153076171875,
468
+ "logps/rejected": -258.02166748046875,
469
+ "loss": 0.6835,
470
+ "rewards/accuracies": 0.550000011920929,
471
+ "rewards/chosen": -0.6932904124259949,
472
+ "rewards/margins": 0.10739298164844513,
473
+ "rewards/rejected": -0.8006833791732788,
474
+ "step": 300
475
+ },
476
+ {
477
+ "epoch": 0.3500846979107849,
478
+ "grad_norm": 8.9375,
479
+ "learning_rate": 4.107808693655262e-06,
480
+ "logits/chosen": -2.9573862552642822,
481
+ "logits/rejected": -2.9266855716705322,
482
+ "logps/chosen": -260.0476989746094,
483
+ "logps/rejected": -220.8186492919922,
484
+ "loss": 0.656,
485
+ "rewards/accuracies": 0.581250011920929,
486
+ "rewards/chosen": -0.6015889644622803,
487
+ "rewards/margins": 0.16243910789489746,
488
+ "rewards/rejected": -0.7640281319618225,
489
+ "step": 310
490
+ },
491
+ {
492
+ "epoch": 0.3613777526821005,
493
+ "grad_norm": 9.4375,
494
+ "learning_rate": 4.0310199190712016e-06,
495
+ "logits/chosen": -2.90745210647583,
496
+ "logits/rejected": -2.877584934234619,
497
+ "logps/chosen": -238.2447509765625,
498
+ "logps/rejected": -260.91278076171875,
499
+ "loss": 0.6433,
500
+ "rewards/accuracies": 0.6312500238418579,
501
+ "rewards/chosen": -0.6627223491668701,
502
+ "rewards/margins": 0.18875077366828918,
503
+ "rewards/rejected": -0.8514731526374817,
504
+ "step": 320
505
+ },
506
+ {
507
+ "epoch": 0.37267080745341613,
508
+ "grad_norm": 9.375,
509
+ "learning_rate": 3.951846640370666e-06,
510
+ "logits/chosen": -2.9316892623901367,
511
+ "logits/rejected": -2.8965978622436523,
512
+ "logps/chosen": -249.91738891601562,
513
+ "logps/rejected": -269.43255615234375,
514
+ "loss": 0.6525,
515
+ "rewards/accuracies": 0.512499988079071,
516
+ "rewards/chosen": -0.5791417360305786,
517
+ "rewards/margins": 0.18505547940731049,
518
+ "rewards/rejected": -0.7641971707344055,
519
+ "step": 330
520
+ },
521
+ {
522
+ "epoch": 0.3839638622247318,
523
+ "grad_norm": 10.0,
524
+ "learning_rate": 3.870412166863107e-06,
525
+ "logits/chosen": -2.9794411659240723,
526
+ "logits/rejected": -2.947648525238037,
527
+ "logps/chosen": -259.78289794921875,
528
+ "logps/rejected": -249.4364776611328,
529
+ "loss": 0.6399,
530
+ "rewards/accuracies": 0.581250011920929,
531
+ "rewards/chosen": -0.7674423456192017,
532
+ "rewards/margins": 0.24263504147529602,
533
+ "rewards/rejected": -1.0100773572921753,
534
+ "step": 340
535
+ },
536
+ {
537
+ "epoch": 0.3952569169960474,
538
+ "grad_norm": 13.1875,
539
+ "learning_rate": 3.786843329581147e-06,
540
+ "logits/chosen": -2.9657294750213623,
541
+ "logits/rejected": -2.944413900375366,
542
+ "logps/chosen": -254.1212158203125,
543
+ "logps/rejected": -251.4638214111328,
544
+ "loss": 0.6688,
545
+ "rewards/accuracies": 0.5625,
546
+ "rewards/chosen": -0.9390102624893188,
547
+ "rewards/margins": 0.23788292706012726,
548
+ "rewards/rejected": -1.1768931150436401,
549
+ "step": 350
550
+ },
551
+ {
552
+ "epoch": 0.40654997176736307,
553
+ "grad_norm": 10.25,
554
+ "learning_rate": 3.701270283746168e-06,
555
+ "logits/chosen": -2.997607707977295,
556
+ "logits/rejected": -2.9588074684143066,
557
+ "logps/chosen": -250.17568969726562,
558
+ "logps/rejected": -234.97073364257812,
559
+ "loss": 0.6788,
560
+ "rewards/accuracies": 0.5062500238418579,
561
+ "rewards/chosen": -0.9304245114326477,
562
+ "rewards/margins": 0.18224455416202545,
563
+ "rewards/rejected": -1.1126692295074463,
564
+ "step": 360
565
+ },
566
+ {
567
+ "epoch": 0.4178430265386787,
568
+ "grad_norm": 10.875,
569
+ "learning_rate": 3.613826306056607e-06,
570
+ "logits/chosen": -2.8931360244750977,
571
+ "logits/rejected": -2.8573451042175293,
572
+ "logps/chosen": -274.30328369140625,
573
+ "logps/rejected": -245.84512329101562,
574
+ "loss": 0.6967,
575
+ "rewards/accuracies": 0.543749988079071,
576
+ "rewards/chosen": -1.037747859954834,
577
+ "rewards/margins": 0.18897393345832825,
578
+ "rewards/rejected": -1.2267218828201294,
579
+ "step": 370
580
+ },
581
+ {
582
+ "epoch": 0.42913608130999437,
583
+ "grad_norm": 12.0625,
584
+ "learning_rate": 3.5246475871146646e-06,
585
+ "logits/chosen": -2.9629368782043457,
586
+ "logits/rejected": -2.920045852661133,
587
+ "logps/chosen": -286.22943115234375,
588
+ "logps/rejected": -249.3096160888672,
589
+ "loss": 0.6428,
590
+ "rewards/accuracies": 0.59375,
591
+ "rewards/chosen": -0.9461029767990112,
592
+ "rewards/margins": 0.26600000262260437,
593
+ "rewards/rejected": -1.2121028900146484,
594
+ "step": 380
595
+ },
596
+ {
597
+ "epoch": 0.44042913608131,
598
+ "grad_norm": 8.5,
599
+ "learning_rate": 3.433873019314731e-06,
600
+ "logits/chosen": -2.905508279800415,
601
+ "logits/rejected": -2.8779773712158203,
602
+ "logps/chosen": -249.64663696289062,
603
+ "logps/rejected": -215.81161499023438,
604
+ "loss": 0.6645,
605
+ "rewards/accuracies": 0.5562499761581421,
606
+ "rewards/chosen": -0.9739023447036743,
607
+ "rewards/margins": 0.22194626927375793,
608
+ "rewards/rejected": -1.1958487033843994,
609
+ "step": 390
610
+ },
611
+ {
612
+ "epoch": 0.45172219085262566,
613
+ "grad_norm": 8.5,
614
+ "learning_rate": 3.341643980523871e-06,
615
+ "logits/chosen": -2.905945062637329,
616
+ "logits/rejected": -2.8710389137268066,
617
+ "logps/chosen": -292.928955078125,
618
+ "logps/rejected": -241.6216278076172,
619
+ "loss": 0.6466,
620
+ "rewards/accuracies": 0.5562499761581421,
621
+ "rewards/chosen": -0.7866181135177612,
622
+ "rewards/margins": 0.2684091627597809,
623
+ "rewards/rejected": -1.0550272464752197,
624
+ "step": 400
625
+ },
626
+ {
627
+ "epoch": 0.46301524562394125,
628
+ "grad_norm": 12.8125,
629
+ "learning_rate": 3.2481041138912784e-06,
630
+ "logits/chosen": -2.8903775215148926,
631
+ "logits/rejected": -2.861661672592163,
632
+ "logps/chosen": -260.47174072265625,
633
+ "logps/rejected": -255.26931762695312,
634
+ "loss": 0.6732,
635
+ "rewards/accuracies": 0.543749988079071,
636
+ "rewards/chosen": -1.0014512538909912,
637
+ "rewards/margins": 0.2172473967075348,
638
+ "rewards/rejected": -1.2186987400054932,
639
+ "step": 410
640
+ },
641
+ {
642
+ "epoch": 0.4743083003952569,
643
+ "grad_norm": 9.75,
644
+ "learning_rate": 3.1533991041296484e-06,
645
+ "logits/chosen": -2.962191104888916,
646
+ "logits/rejected": -2.932969331741333,
647
+ "logps/chosen": -259.97283935546875,
648
+ "logps/rejected": -248.1492919921875,
649
+ "loss": 0.6506,
650
+ "rewards/accuracies": 0.5625,
651
+ "rewards/chosen": -0.9169241189956665,
652
+ "rewards/margins": 0.20076242089271545,
653
+ "rewards/rejected": -1.1176865100860596,
654
+ "step": 420
655
+ },
656
+ {
657
+ "epoch": 0.48560135516657255,
658
+ "grad_norm": 11.375,
659
+ "learning_rate": 3.0576764506168893e-06,
660
+ "logits/chosen": -2.9573709964752197,
661
+ "logits/rejected": -2.9248433113098145,
662
+ "logps/chosen": -281.449462890625,
663
+ "logps/rejected": -258.242431640625,
664
+ "loss": 0.6653,
665
+ "rewards/accuracies": 0.581250011920929,
666
+ "rewards/chosen": -0.8910619616508484,
667
+ "rewards/margins": 0.1914868950843811,
668
+ "rewards/rejected": -1.0825488567352295,
669
+ "step": 430
670
+ },
671
+ {
672
+ "epoch": 0.4968944099378882,
673
+ "grad_norm": 8.5,
674
+ "learning_rate": 2.9610852376715625e-06,
675
+ "logits/chosen": -2.9228568077087402,
676
+ "logits/rejected": -2.8933017253875732,
677
+ "logps/chosen": -258.58245849609375,
678
+ "logps/rejected": -248.20846557617188,
679
+ "loss": 0.6594,
680
+ "rewards/accuracies": 0.574999988079071,
681
+ "rewards/chosen": -0.9236515164375305,
682
+ "rewards/margins": 0.2783660292625427,
683
+ "rewards/rejected": -1.2020175457000732,
684
+ "step": 440
685
+ },
686
+ {
687
+ "epoch": 0.5081874647092038,
688
+ "grad_norm": 7.84375,
689
+ "learning_rate": 2.8637759023598467e-06,
690
+ "logits/chosen": -2.9429585933685303,
691
+ "logits/rejected": -2.9177327156066895,
692
+ "logps/chosen": -246.7409210205078,
693
+ "logps/rejected": -252.89218139648438,
694
+ "loss": 0.6842,
695
+ "rewards/accuracies": 0.4937500059604645,
696
+ "rewards/chosen": -0.8663239479064941,
697
+ "rewards/margins": 0.17053480446338654,
698
+ "rewards/rejected": -1.0368586778640747,
699
+ "step": 450
700
+ },
701
+ {
702
+ "epoch": 0.5194805194805194,
703
+ "grad_norm": 9.25,
704
+ "learning_rate": 2.7659000001956376e-06,
705
+ "logits/chosen": -2.9368577003479004,
706
+ "logits/rejected": -2.899815559387207,
707
+ "logps/chosen": -258.90765380859375,
708
+ "logps/rejected": -234.8472137451172,
709
+ "loss": 0.6159,
710
+ "rewards/accuracies": 0.625,
711
+ "rewards/chosen": -0.8144370317459106,
712
+ "rewards/margins": 0.30742961168289185,
713
+ "rewards/rejected": -1.1218667030334473,
714
+ "step": 460
715
+ },
716
+ {
717
+ "epoch": 0.5307735742518351,
718
+ "grad_norm": 9.8125,
719
+ "learning_rate": 2.6676099690987227e-06,
720
+ "logits/chosen": -2.938596248626709,
721
+ "logits/rejected": -2.9030938148498535,
722
+ "logps/chosen": -251.0491485595703,
723
+ "logps/rejected": -227.3994140625,
724
+ "loss": 0.6612,
725
+ "rewards/accuracies": 0.5625,
726
+ "rewards/chosen": -0.7350276708602905,
727
+ "rewards/margins": 0.21316580474376678,
728
+ "rewards/rejected": -0.9481935501098633,
729
+ "step": 470
730
+ },
731
+ {
732
+ "epoch": 0.5420666290231507,
733
+ "grad_norm": 8.625,
734
+ "learning_rate": 2.5690588919786307e-06,
735
+ "logits/chosen": -2.9533700942993164,
736
+ "logits/rejected": -2.9271883964538574,
737
+ "logps/chosen": -265.0854187011719,
738
+ "logps/rejected": -239.2405242919922,
739
+ "loss": 0.6447,
740
+ "rewards/accuracies": 0.5562499761581421,
741
+ "rewards/chosen": -0.8996723294258118,
742
+ "rewards/margins": 0.29516300559043884,
743
+ "rewards/rejected": -1.1948354244232178,
744
+ "step": 480
745
+ },
746
+ {
747
+ "epoch": 0.5533596837944664,
748
+ "grad_norm": 12.6875,
749
+ "learning_rate": 2.4704002583139426e-06,
750
+ "logits/chosen": -2.9021379947662354,
751
+ "logits/rejected": -2.8771567344665527,
752
+ "logps/chosen": -264.7264709472656,
753
+ "logps/rejected": -246.2521209716797,
754
+ "loss": 0.6077,
755
+ "rewards/accuracies": 0.606249988079071,
756
+ "rewards/chosen": -0.8024009466171265,
757
+ "rewards/margins": 0.2971910834312439,
758
+ "rewards/rejected": -1.0995919704437256,
759
+ "step": 490
760
+ },
761
+ {
762
+ "epoch": 0.564652738565782,
763
+ "grad_norm": 8.9375,
764
+ "learning_rate": 2.3717877250983864e-06,
765
+ "logits/chosen": -2.938613176345825,
766
+ "logits/rejected": -2.904693841934204,
767
+ "logps/chosen": -250.9031219482422,
768
+ "logps/rejected": -249.9896240234375,
769
+ "loss": 0.6574,
770
+ "rewards/accuracies": 0.5249999761581421,
771
+ "rewards/chosen": -0.845720648765564,
772
+ "rewards/margins": 0.24266941845417023,
773
+ "rewards/rejected": -1.0883901119232178,
774
+ "step": 500
775
+ },
776
+ {
777
+ "epoch": 0.5759457933370977,
778
+ "grad_norm": 8.8125,
779
+ "learning_rate": 2.2733748775260256e-06,
780
+ "logits/chosen": -2.8921456336975098,
781
+ "logits/rejected": -2.871410846710205,
782
+ "logps/chosen": -226.85903930664062,
783
+ "logps/rejected": -219.52587890625,
784
+ "loss": 0.6478,
785
+ "rewards/accuracies": 0.5687500238418579,
786
+ "rewards/chosen": -0.7570312023162842,
787
+ "rewards/margins": 0.20835983753204346,
788
+ "rewards/rejected": -0.9653909802436829,
789
+ "step": 510
790
+ },
791
+ {
792
+ "epoch": 0.5872388481084133,
793
+ "grad_norm": 10.25,
794
+ "learning_rate": 2.1753149897882847e-06,
795
+ "logits/chosen": -2.9432761669158936,
796
+ "logits/rejected": -2.892080545425415,
797
+ "logps/chosen": -238.2821044921875,
798
+ "logps/rejected": -222.26644897460938,
799
+ "loss": 0.5833,
800
+ "rewards/accuracies": 0.6875,
801
+ "rewards/chosen": -0.8168408274650574,
802
+ "rewards/margins": 0.41468945145606995,
803
+ "rewards/rejected": -1.2315301895141602,
804
+ "step": 520
805
+ },
806
+ {
807
+ "epoch": 0.598531902879729,
808
+ "grad_norm": 10.0,
809
+ "learning_rate": 2.0777607863553423e-06,
810
+ "logits/chosen": -2.9561731815338135,
811
+ "logits/rejected": -2.9202933311462402,
812
+ "logps/chosen": -236.2799835205078,
813
+ "logps/rejected": -216.89901733398438,
814
+ "loss": 0.673,
815
+ "rewards/accuracies": 0.518750011920929,
816
+ "rewards/chosen": -0.8298661112785339,
817
+ "rewards/margins": 0.18440715968608856,
818
+ "rewards/rejected": -1.0142732858657837,
819
+ "step": 530
820
+ },
821
+ {
822
+ "epoch": 0.6098249576510446,
823
+ "grad_norm": 12.5,
824
+ "learning_rate": 1.9808642041136923e-06,
825
+ "logits/chosen": -2.9524600505828857,
826
+ "logits/rejected": -2.9057111740112305,
827
+ "logps/chosen": -239.9007110595703,
828
+ "logps/rejected": -232.8452606201172,
829
+ "loss": 0.6733,
830
+ "rewards/accuracies": 0.550000011920929,
831
+ "rewards/chosen": -0.8775818943977356,
832
+ "rewards/margins": 0.22822308540344238,
833
+ "rewards/rejected": -1.1058050394058228,
834
+ "step": 540
835
+ },
836
+ {
837
+ "epoch": 0.6211180124223602,
838
+ "grad_norm": 12.0625,
839
+ "learning_rate": 1.884776155730342e-06,
840
+ "logits/chosen": -2.935317277908325,
841
+ "logits/rejected": -2.9047186374664307,
842
+ "logps/chosen": -250.3077850341797,
843
+ "logps/rejected": -248.65652465820312,
844
+ "loss": 0.7337,
845
+ "rewards/accuracies": 0.5062500238418579,
846
+ "rewards/chosen": -0.9318073391914368,
847
+ "rewards/margins": 0.11215513944625854,
848
+ "rewards/rejected": -1.0439624786376953,
849
+ "step": 550
850
+ },
851
+ {
852
+ "epoch": 0.6324110671936759,
853
+ "grad_norm": 11.5625,
854
+ "learning_rate": 1.7896462946121873e-06,
855
+ "logits/chosen": -2.951495885848999,
856
+ "logits/rejected": -2.9132513999938965,
857
+ "logps/chosen": -269.6780700683594,
858
+ "logps/rejected": -256.81597900390625,
859
+ "loss": 0.6249,
860
+ "rewards/accuracies": 0.6000000238418579,
861
+ "rewards/chosen": -0.8714116215705872,
862
+ "rewards/margins": 0.29088258743286133,
863
+ "rewards/rejected": -1.1622941493988037,
864
+ "step": 560
865
+ },
866
+ {
867
+ "epoch": 0.6437041219649915,
868
+ "grad_norm": 9.75,
869
+ "learning_rate": 1.695622781826638e-06,
870
+ "logits/chosen": -2.954216718673706,
871
+ "logits/rejected": -2.906550884246826,
872
+ "logps/chosen": -257.61883544921875,
873
+ "logps/rejected": -228.7353973388672,
874
+ "loss": 0.6324,
875
+ "rewards/accuracies": 0.5874999761581421,
876
+ "rewards/chosen": -0.7798657417297363,
877
+ "rewards/margins": 0.3046206533908844,
878
+ "rewards/rejected": -1.0844862461090088,
879
+ "step": 570
880
+ },
881
+ {
882
+ "epoch": 0.6549971767363072,
883
+ "grad_norm": 11.1875,
884
+ "learning_rate": 1.6028520553465053e-06,
885
+ "logits/chosen": -2.943326473236084,
886
+ "logits/rejected": -2.907252788543701,
887
+ "logps/chosen": -255.4546356201172,
888
+ "logps/rejected": -229.12185668945312,
889
+ "loss": 0.6212,
890
+ "rewards/accuracies": 0.581250011920929,
891
+ "rewards/chosen": -1.125480055809021,
892
+ "rewards/margins": 0.27197134494781494,
893
+ "rewards/rejected": -1.3974515199661255,
894
+ "step": 580
895
+ },
896
+ {
897
+ "epoch": 0.6662902315076228,
898
+ "grad_norm": 10.8125,
899
+ "learning_rate": 1.5114786019785416e-06,
900
+ "logits/chosen": -2.9562203884124756,
901
+ "logits/rejected": -2.9087963104248047,
902
+ "logps/chosen": -244.3138427734375,
903
+ "logps/rejected": -240.51870727539062,
904
+ "loss": 0.6511,
905
+ "rewards/accuracies": 0.5874999761581421,
906
+ "rewards/chosen": -0.9466648101806641,
907
+ "rewards/margins": 0.2708016335964203,
908
+ "rewards/rejected": -1.2174664735794067,
909
+ "step": 590
910
+ },
911
+ {
912
+ "epoch": 0.6775832862789385,
913
+ "grad_norm": 11.0,
914
+ "learning_rate": 1.4216447323308444e-06,
915
+ "logits/chosen": -2.9042696952819824,
916
+ "logits/rejected": -2.883103609085083,
917
+ "logps/chosen": -240.0981903076172,
918
+ "logps/rejected": -246.4265899658203,
919
+ "loss": 0.6582,
920
+ "rewards/accuracies": 0.6000000238418579,
921
+ "rewards/chosen": -1.1731393337249756,
922
+ "rewards/margins": 0.262151300907135,
923
+ "rewards/rejected": -1.4352905750274658,
924
+ "step": 600
925
+ },
926
+ {
927
+ "epoch": 0.6888763410502541,
928
+ "grad_norm": 7.78125,
929
+ "learning_rate": 1.3334903591696055e-06,
930
+ "logits/chosen": -2.924773931503296,
931
+ "logits/rejected": -2.8905885219573975,
932
+ "logps/chosen": -243.14501953125,
933
+ "logps/rejected": -215.37881469726562,
934
+ "loss": 0.6368,
935
+ "rewards/accuracies": 0.53125,
936
+ "rewards/chosen": -0.9755071401596069,
937
+ "rewards/margins": 0.27910715341567993,
938
+ "rewards/rejected": -1.2546144723892212,
939
+ "step": 610
940
+ },
941
+ {
942
+ "epoch": 0.7001693958215698,
943
+ "grad_norm": 13.875,
944
+ "learning_rate": 1.247152779510411e-06,
945
+ "logits/chosen": -2.941826820373535,
946
+ "logits/rejected": -2.906125783920288,
947
+ "logps/chosen": -248.43362426757812,
948
+ "logps/rejected": -236.13525390625,
949
+ "loss": 0.7073,
950
+ "rewards/accuracies": 0.4625000059604645,
951
+ "rewards/chosen": -1.108798623085022,
952
+ "rewards/margins": 0.21407607197761536,
953
+ "rewards/rejected": -1.3228747844696045,
954
+ "step": 620
955
+ },
956
+ {
957
+ "epoch": 0.7114624505928854,
958
+ "grad_norm": 11.625,
959
+ "learning_rate": 1.1627664607834593e-06,
960
+ "logits/chosen": -2.946981191635132,
961
+ "logits/rejected": -2.9171371459960938,
962
+ "logps/chosen": -251.35324096679688,
963
+ "logps/rejected": -261.75396728515625,
964
+ "loss": 0.6583,
965
+ "rewards/accuracies": 0.581250011920929,
966
+ "rewards/chosen": -0.9292078018188477,
967
+ "rewards/margins": 0.23983097076416016,
968
+ "rewards/rejected": -1.1690387725830078,
969
+ "step": 630
970
+ },
971
+ {
972
+ "epoch": 0.722755505364201,
973
+ "grad_norm": 14.5625,
974
+ "learning_rate": 1.0804628314057659e-06,
975
+ "logits/chosen": -2.920762538909912,
976
+ "logits/rejected": -2.8985981941223145,
977
+ "logps/chosen": -235.6101837158203,
978
+ "logps/rejected": -207.3476104736328,
979
+ "loss": 0.6607,
980
+ "rewards/accuracies": 0.5687500238418579,
981
+ "rewards/chosen": -1.0142499208450317,
982
+ "rewards/margins": 0.2305765599012375,
983
+ "rewards/rejected": -1.2448265552520752,
984
+ "step": 640
985
+ },
986
+ {
987
+ "epoch": 0.7340485601355167,
988
+ "grad_norm": 16.0,
989
+ "learning_rate": 1.0003700760864931e-06,
990
+ "logits/chosen": -2.902686595916748,
991
+ "logits/rejected": -2.8842315673828125,
992
+ "logps/chosen": -252.89035034179688,
993
+ "logps/rejected": -250.73623657226562,
994
+ "loss": 0.6813,
995
+ "rewards/accuracies": 0.5562499761581421,
996
+ "rewards/chosen": -1.051636815071106,
997
+ "rewards/margins": 0.21588905155658722,
998
+ "rewards/rejected": -1.2675259113311768,
999
+ "step": 650
1000
+ },
1001
+ {
1002
+ "epoch": 0.7453416149068323,
1003
+ "grad_norm": 12.3125,
1004
+ "learning_rate": 9.226129361842498e-07,
1005
+ "logits/chosen": -2.9301822185516357,
1006
+ "logits/rejected": -2.881704807281494,
1007
+ "logps/chosen": -236.1824951171875,
1008
+ "logps/rejected": -207.68685913085938,
1009
+ "loss": 0.6363,
1010
+ "rewards/accuracies": 0.574999988079071,
1011
+ "rewards/chosen": -1.060085415840149,
1012
+ "rewards/margins": 0.3055773675441742,
1013
+ "rewards/rejected": -1.365662693977356,
1014
+ "step": 660
1015
+ },
1016
+ {
1017
+ "epoch": 0.756634669678148,
1018
+ "grad_norm": 11.3125,
1019
+ "learning_rate": 8.473125154272563e-07,
1020
+ "logits/chosen": -2.8978400230407715,
1021
+ "logits/rejected": -2.845043659210205,
1022
+ "logps/chosen": -245.31027221679688,
1023
+ "logps/rejected": -251.3234100341797,
1024
+ "loss": 0.6158,
1025
+ "rewards/accuracies": 0.5874999761581421,
1026
+ "rewards/chosen": -0.980219841003418,
1027
+ "rewards/margins": 0.3760405480861664,
1028
+ "rewards/rejected": -1.3562604188919067,
1029
+ "step": 670
1030
+ },
1031
+ {
1032
+ "epoch": 0.7679277244494636,
1033
+ "grad_norm": 11.4375,
1034
+ "learning_rate": 7.74586091298995e-07,
1035
+ "logits/chosen": -2.9392664432525635,
1036
+ "logits/rejected": -2.8935275077819824,
1037
+ "logps/chosen": -263.9918518066406,
1038
+ "logps/rejected": -226.14920043945312,
1039
+ "loss": 0.6485,
1040
+ "rewards/accuracies": 0.574999988079071,
1041
+ "rewards/chosen": -1.0862810611724854,
1042
+ "rewards/margins": 0.3031725287437439,
1043
+ "rewards/rejected": -1.389453649520874,
1044
+ "step": 680
1045
+ },
1046
+ {
1047
+ "epoch": 0.7792207792207793,
1048
+ "grad_norm": 10.4375,
1049
+ "learning_rate": 7.045469323830731e-07,
1050
+ "logits/chosen": -2.908051013946533,
1051
+ "logits/rejected": -2.877911329269409,
1052
+ "logps/chosen": -247.0584259033203,
1053
+ "logps/rejected": -250.07669067382812,
1054
+ "loss": 0.677,
1055
+ "rewards/accuracies": 0.5249999761581421,
1056
+ "rewards/chosen": -1.146324872970581,
1057
+ "rewards/margins": 0.17685498297214508,
1058
+ "rewards/rejected": -1.3231797218322754,
1059
+ "step": 690
1060
+ },
1061
+ {
1062
+ "epoch": 0.7905138339920948,
1063
+ "grad_norm": 9.625,
1064
+ "learning_rate": 6.373041219518059e-07,
1065
+ "logits/chosen": -2.9572408199310303,
1066
+ "logits/rejected": -2.937887191772461,
1067
+ "logps/chosen": -252.33755493164062,
1068
+ "logps/rejected": -229.02926635742188,
1069
+ "loss": 0.6706,
1070
+ "rewards/accuracies": 0.5375000238418579,
1071
+ "rewards/chosen": -1.0239284038543701,
1072
+ "rewards/margins": 0.20092003047466278,
1073
+ "rewards/rejected": -1.2248482704162598,
1074
+ "step": 700
1075
+ },
1076
+ {
1077
+ "epoch": 0.8018068887634106,
1078
+ "grad_norm": 9.5625,
1079
+ "learning_rate": 5.729623880732455e-07,
1080
+ "logits/chosen": -2.9367566108703613,
1081
+ "logits/rejected": -2.9192299842834473,
1082
+ "logps/chosen": -245.69668579101562,
1083
+ "logps/rejected": -225.72848510742188,
1084
+ "loss": 0.6588,
1085
+ "rewards/accuracies": 0.612500011920929,
1086
+ "rewards/chosen": -1.0508708953857422,
1087
+ "rewards/margins": 0.26678308844566345,
1088
+ "rewards/rejected": -1.3176538944244385,
1089
+ "step": 710
1090
+ },
1091
+ {
1092
+ "epoch": 0.8130999435347261,
1093
+ "grad_norm": 8.5,
1094
+ "learning_rate": 5.116219405012865e-07,
1095
+ "logits/chosen": -2.9278311729431152,
1096
+ "logits/rejected": -2.8987011909484863,
1097
+ "logps/chosen": -246.24691772460938,
1098
+ "logps/rejected": -206.0987548828125,
1099
+ "loss": 0.6105,
1100
+ "rewards/accuracies": 0.581250011920929,
1101
+ "rewards/chosen": -1.0415557622909546,
1102
+ "rewards/margins": 0.3364638686180115,
1103
+ "rewards/rejected": -1.3780196905136108,
1104
+ "step": 720
1105
+ },
1106
+ {
1107
+ "epoch": 0.8243929983060417,
1108
+ "grad_norm": 7.75,
1109
+ "learning_rate": 4.533783146028542e-07,
1110
+ "logits/chosen": -2.9237890243530273,
1111
+ "logits/rejected": -2.8972859382629395,
1112
+ "logps/chosen": -234.02517700195312,
1113
+ "logps/rejected": -260.5685119628906,
1114
+ "loss": 0.6314,
1115
+ "rewards/accuracies": 0.6000000238418579,
1116
+ "rewards/chosen": -1.0762670040130615,
1117
+ "rewards/margins": 0.31192946434020996,
1118
+ "rewards/rejected": -1.388196587562561,
1119
+ "step": 730
1120
+ },
1121
+ {
1122
+ "epoch": 0.8356860530773574,
1123
+ "grad_norm": 7.71875,
1124
+ "learning_rate": 3.9832222256528633e-07,
1125
+ "logits/chosen": -2.9241185188293457,
1126
+ "logits/rejected": -2.8836278915405273,
1127
+ "logps/chosen": -238.7316131591797,
1128
+ "logps/rejected": -258.7846984863281,
1129
+ "loss": 0.625,
1130
+ "rewards/accuracies": 0.581250011920929,
1131
+ "rewards/chosen": -1.1147111654281616,
1132
+ "rewards/margins": 0.31473425030708313,
1133
+ "rewards/rejected": -1.4294453859329224,
1134
+ "step": 740
1135
+ },
1136
+ {
1137
+ "epoch": 0.846979107848673,
1138
+ "grad_norm": 7.8125,
1139
+ "learning_rate": 3.4653941211561265e-07,
1140
+ "logits/chosen": -2.918570041656494,
1141
+ "logits/rejected": -2.8599822521209717,
1142
+ "logps/chosen": -228.1241912841797,
1143
+ "logps/rejected": -200.6188507080078,
1144
+ "loss": 0.591,
1145
+ "rewards/accuracies": 0.581250011920929,
1146
+ "rewards/chosen": -1.1377301216125488,
1147
+ "rewards/margins": 0.45671454071998596,
1148
+ "rewards/rejected": -1.5944445133209229,
1149
+ "step": 750
1150
+ },
1151
+ {
1152
+ "epoch": 0.8582721626199887,
1153
+ "grad_norm": 11.9375,
1154
+ "learning_rate": 2.9811053297180454e-07,
1155
+ "logits/chosen": -2.9596495628356934,
1156
+ "logits/rejected": -2.938786268234253,
1157
+ "logps/chosen": -231.0944061279297,
1158
+ "logps/rejected": -241.2841339111328,
1159
+ "loss": 0.6626,
1160
+ "rewards/accuracies": 0.5687500238418579,
1161
+ "rewards/chosen": -1.0585596561431885,
1162
+ "rewards/margins": 0.2590670883655548,
1163
+ "rewards/rejected": -1.317626714706421,
1164
+ "step": 760
1165
+ },
1166
+ {
1167
+ "epoch": 0.8695652173913043,
1168
+ "grad_norm": 9.6875,
1169
+ "learning_rate": 2.531110112339638e-07,
1170
+ "logits/chosen": -2.933401584625244,
1171
+ "logits/rejected": -2.920301914215088,
1172
+ "logps/chosen": -237.66879272460938,
1173
+ "logps/rejected": -238.02297973632812,
1174
+ "loss": 0.6987,
1175
+ "rewards/accuracies": 0.543749988079071,
1176
+ "rewards/chosen": -1.0770819187164307,
1177
+ "rewards/margins": 0.14950065314769745,
1178
+ "rewards/rejected": -1.2265825271606445,
1179
+ "step": 770
1180
+ },
1181
+ {
1182
+ "epoch": 0.88085827216262,
1183
+ "grad_norm": 12.5,
1184
+ "learning_rate": 2.116109319111015e-07,
1185
+ "logits/chosen": -2.912548065185547,
1186
+ "logits/rejected": -2.8936960697174072,
1187
+ "logps/chosen": -235.7461395263672,
1188
+ "logps/rejected": -246.876708984375,
1189
+ "loss": 0.7254,
1190
+ "rewards/accuracies": 0.518750011920929,
1191
+ "rewards/chosen": -1.1536673307418823,
1192
+ "rewards/margins": 0.14096426963806152,
1193
+ "rewards/rejected": -1.2946317195892334,
1194
+ "step": 780
1195
+ },
1196
+ {
1197
+ "epoch": 0.8921513269339356,
1198
+ "grad_norm": 7.34375,
1199
+ "learning_rate": 1.736749297664539e-07,
1200
+ "logits/chosen": -2.9254114627838135,
1201
+ "logits/rejected": -2.89033842086792,
1202
+ "logps/chosen": -263.29071044921875,
1203
+ "logps/rejected": -274.1188659667969,
1204
+ "loss": 0.6559,
1205
+ "rewards/accuracies": 0.5625,
1206
+ "rewards/chosen": -0.9891961216926575,
1207
+ "rewards/margins": 0.2855607271194458,
1208
+ "rewards/rejected": -1.2747570276260376,
1209
+ "step": 790
1210
+ },
1211
+ {
1212
+ "epoch": 0.9034443817052513,
1213
+ "grad_norm": 9.4375,
1214
+ "learning_rate": 1.3936208865134666e-07,
1215
+ "logits/chosen": -2.930879831314087,
1216
+ "logits/rejected": -2.8965582847595215,
1217
+ "logps/chosen": -254.17044067382812,
1218
+ "logps/rejected": -273.36724853515625,
1219
+ "loss": 0.6567,
1220
+ "rewards/accuracies": 0.5687500238418579,
1221
+ "rewards/chosen": -1.0911872386932373,
1222
+ "rewards/margins": 0.27449628710746765,
1223
+ "rewards/rejected": -1.3656837940216064,
1224
+ "step": 800
1225
+ },
1226
+ {
1227
+ "epoch": 0.9147374364765669,
1228
+ "grad_norm": 8.125,
1229
+ "learning_rate": 1.0872584948439208e-07,
1230
+ "logits/chosen": -2.9673569202423096,
1231
+ "logits/rejected": -2.938894748687744,
1232
+ "logps/chosen": -264.1458435058594,
1233
+ "logps/rejected": -253.3695068359375,
1234
+ "loss": 0.6348,
1235
+ "rewards/accuracies": 0.574999988079071,
1236
+ "rewards/chosen": -0.9376649856567383,
1237
+ "rewards/margins": 0.28761720657348633,
1238
+ "rewards/rejected": -1.225282073020935,
1239
+ "step": 810
1240
+ },
1241
+ {
1242
+ "epoch": 0.9260304912478825,
1243
+ "grad_norm": 10.75,
1244
+ "learning_rate": 8.181392701932745e-08,
1245
+ "logits/chosen": -2.9556472301483154,
1246
+ "logits/rejected": -2.930018663406372,
1247
+ "logps/chosen": -244.6985321044922,
1248
+ "logps/rejected": -241.3518524169922,
1249
+ "loss": 0.6962,
1250
+ "rewards/accuracies": 0.5375000238418579,
1251
+ "rewards/chosen": -1.143431305885315,
1252
+ "rewards/margins": 0.20309944450855255,
1253
+ "rewards/rejected": -1.3465306758880615,
1254
+ "step": 820
1255
+ },
1256
+ {
1257
+ "epoch": 0.9373235460191982,
1258
+ "grad_norm": 11.1875,
1259
+ "learning_rate": 5.866823553114187e-08,
1260
+ "logits/chosen": -2.9636921882629395,
1261
+ "logits/rejected": -2.934422254562378,
1262
+ "logps/chosen": -249.5702667236328,
1263
+ "logps/rejected": -229.45053100585938,
1264
+ "loss": 0.6749,
1265
+ "rewards/accuracies": 0.5375000238418579,
1266
+ "rewards/chosen": -1.1629018783569336,
1267
+ "rewards/margins": 0.19255365431308746,
1268
+ "rewards/rejected": -1.3554556369781494,
1269
+ "step": 830
1270
+ },
1271
+ {
1272
+ "epoch": 0.9486166007905138,
1273
+ "grad_norm": 12.6875,
1274
+ "learning_rate": 3.932482353621536e-08,
1275
+ "logits/chosen": -2.9121766090393066,
1276
+ "logits/rejected": -2.8778061866760254,
1277
+ "logps/chosen": -252.2318115234375,
1278
+ "logps/rejected": -242.0477294921875,
1279
+ "loss": 0.7142,
1280
+ "rewards/accuracies": 0.4375,
1281
+ "rewards/chosen": -1.1486766338348389,
1282
+ "rewards/margins": 0.10505317151546478,
1283
+ "rewards/rejected": -1.253730058670044,
1284
+ "step": 840
1285
+ },
1286
+ {
1287
+ "epoch": 0.9599096555618295,
1288
+ "grad_norm": 10.6875,
1289
+ "learning_rate": 2.381381764815671e-08,
1290
+ "logits/chosen": -2.9521381855010986,
1291
+ "logits/rejected": -2.9332849979400635,
1292
+ "logps/chosen": -248.31655883789062,
1293
+ "logps/rejected": -244.4917449951172,
1294
+ "loss": 0.7399,
1295
+ "rewards/accuracies": 0.53125,
1296
+ "rewards/chosen": -1.1000101566314697,
1297
+ "rewards/margins": 0.10928668826818466,
1298
+ "rewards/rejected": -1.2092968225479126,
1299
+ "step": 850
1300
+ },
1301
+ {
1302
+ "epoch": 0.9712027103331451,
1303
+ "grad_norm": 11.125,
1304
+ "learning_rate": 1.2159375656770056e-08,
1305
+ "logits/chosen": -2.964635133743286,
1306
+ "logits/rejected": -2.9422874450683594,
1307
+ "logps/chosen": -266.625732421875,
1308
+ "logps/rejected": -265.8684997558594,
1309
+ "loss": 0.693,
1310
+ "rewards/accuracies": 0.518750011920929,
1311
+ "rewards/chosen": -1.1866648197174072,
1312
+ "rewards/margins": 0.21592363715171814,
1313
+ "rewards/rejected": -1.4025886058807373,
1314
+ "step": 860
1315
+ },
1316
+ {
1317
+ "epoch": 0.9824957651044608,
1318
+ "grad_norm": 11.5,
1319
+ "learning_rate": 4.379648903234668e-09,
1320
+ "logits/chosen": -2.930752754211426,
1321
+ "logits/rejected": -2.8959178924560547,
1322
+ "logps/chosen": -268.1014404296875,
1323
+ "logps/rejected": -252.4002227783203,
1324
+ "loss": 0.6735,
1325
+ "rewards/accuracies": 0.5375000238418579,
1326
+ "rewards/chosen": -0.995762825012207,
1327
+ "rewards/margins": 0.236598402261734,
1328
+ "rewards/rejected": -1.2323614358901978,
1329
+ "step": 870
1330
+ },
1331
+ {
1332
+ "epoch": 0.9937888198757764,
1333
+ "grad_norm": 9.75,
1334
+ "learning_rate": 4.867540100961244e-10,
1335
+ "logits/chosen": -2.928917646408081,
1336
+ "logits/rejected": -2.881701946258545,
1337
+ "logps/chosen": -294.4600524902344,
1338
+ "logps/rejected": -246.2449493408203,
1339
+ "loss": 0.6221,
1340
+ "rewards/accuracies": 0.5687500238418579,
1341
+ "rewards/chosen": -1.1883199214935303,
1342
+ "rewards/margins": 0.35472264885902405,
1343
+ "rewards/rejected": -1.543042540550232,
1344
+ "step": 880
1345
+ },
1346
+ {
1347
+ "epoch": 0.9994353472614342,
1348
+ "step": 885,
1349
+ "total_flos": 0.0,
1350
+ "train_loss": 0.6616038349388683,
1351
+ "train_runtime": 9964.5835,
1352
+ "train_samples_per_second": 1.422,
1353
+ "train_steps_per_second": 0.089
1354
+ }
1355
+ ],
1356
+ "logging_steps": 10,
1357
+ "max_steps": 885,
1358
+ "num_input_tokens_seen": 0,
1359
+ "num_train_epochs": 1,
1360
+ "save_steps": 500,
1361
+ "stateful_callbacks": {
1362
+ "TrainerControl": {
1363
+ "args": {
1364
+ "should_epoch_stop": false,
1365
+ "should_evaluate": false,
1366
+ "should_log": false,
1367
+ "should_save": true,
1368
+ "should_training_stop": true
1369
+ },
1370
+ "attributes": {}
1371
+ }
1372
+ },
1373
+ "total_flos": 0.0,
1374
+ "train_batch_size": 8,
1375
+ "trial_name": null,
1376
+ "trial_params": null
1377
+ }