silviasapora commited on
Commit
2f79711
·
verified ·
1 Parent(s): 548ddff

Model save

Browse files
Files changed (4) hide show
  1. README.md +67 -0
  2. all_results.json +9 -0
  3. train_results.json +9 -0
  4. trainer_state.json +1113 -0
README.md ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: google/gemma-7b
3
+ library_name: transformers
4
+ model_name: gemma-7b-silvia-basic-5e-5-05-vshp26
5
+ tags:
6
+ - generated_from_trainer
7
+ - trl
8
+ - orpo
9
+ licence: license
10
+ ---
11
+
12
+ # Model Card for gemma-7b-silvia-basic-5e-5-05-vshp26
13
+
14
+ This model is a fine-tuned version of [google/gemma-7b](https://huggingface.co/google/gemma-7b).
15
+ It has been trained using [TRL](https://github.com/huggingface/trl).
16
+
17
+ ## Quick start
18
+
19
+ ```python
20
+ from transformers import pipeline
21
+
22
+ question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
23
+ generator = pipeline("text-generation", model="silviasapora/gemma-7b-silvia-basic-5e-5-05-vshp26", device="cuda")
24
+ output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
25
+ print(output["generated_text"])
26
+ ```
27
+
28
+ ## Training procedure
29
+
30
+ [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/silvias/huggingface/runs/iz0ng2nz)
31
+
32
+
33
+ This model was trained with ORPO, a method introduced in [ORPO: Monolithic Preference Optimization without Reference Model](https://huggingface.co/papers/2403.07691).
34
+
35
+ ### Framework versions
36
+
37
+ - TRL: 0.13.0
38
+ - Transformers: 4.48.1
39
+ - Pytorch: 2.5.1
40
+ - Datasets: 3.2.0
41
+ - Tokenizers: 0.21.0
42
+
43
+ ## Citations
44
+
45
+ Cite ORPO as:
46
+
47
+ ```bibtex
48
+ @article{hong2024orpo,
49
+ title = {{ORPO: Monolithic Preference Optimization without Reference Model}},
50
+ author = {Jiwoo Hong and Noah Lee and James Thorne},
51
+ year = 2024,
52
+ eprint = {arXiv:2403.07691}
53
+ }
54
+ ```
55
+
56
+ Cite TRL as:
57
+
58
+ ```bibtex
59
+ @misc{vonwerra2022trl,
60
+ title = {{TRL: Transformer Reinforcement Learning}},
61
+ author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
62
+ year = 2020,
63
+ journal = {GitHub repository},
64
+ publisher = {GitHub},
65
+ howpublished = {\url{https://github.com/huggingface/trl}}
66
+ }
67
+ ```
all_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 2.9765925925925925,
3
+ "total_flos": 0.0,
4
+ "train_loss": 6.844926292177231,
5
+ "train_runtime": 9518.1794,
6
+ "train_samples": 6750,
7
+ "train_samples_per_second": 2.128,
8
+ "train_steps_per_second": 0.033
9
+ }
train_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 2.9765925925925925,
3
+ "total_flos": 0.0,
4
+ "train_loss": 6.844926292177231,
5
+ "train_runtime": 9518.1794,
6
+ "train_samples": 6750,
7
+ "train_samples_per_second": 2.128,
8
+ "train_steps_per_second": 0.033
9
+ }
trainer_state.json ADDED
@@ -0,0 +1,1113 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 2.9765925925925925,
5
+ "eval_steps": 500,
6
+ "global_step": 315,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.047407407407407405,
13
+ "grad_norm": 1138.854736328125,
14
+ "learning_rate": 7.8125e-06,
15
+ "log_odds_chosen": -1.0697920322418213,
16
+ "log_odds_ratio": -11.011500358581543,
17
+ "logps/chosen": -22.719226837158203,
18
+ "logps/rejected": -21.649660110473633,
19
+ "loss": 450.84,
20
+ "nll_loss": 9.846151351928711,
21
+ "po_loss": 4.242598533630371,
22
+ "rewards/accuracies": 0.48124998807907104,
23
+ "rewards/chosen": -11.359613418579102,
24
+ "rewards/margins": -0.5347837209701538,
25
+ "rewards/rejected": -10.824830055236816,
26
+ "step": 5
27
+ },
28
+ {
29
+ "epoch": 0.09481481481481481,
30
+ "grad_norm": 597.58203125,
31
+ "learning_rate": 1.5625e-05,
32
+ "log_odds_chosen": -3.116546392440796,
33
+ "log_odds_ratio": -11.056388854980469,
34
+ "logps/chosen": -22.023601531982422,
35
+ "logps/rejected": -18.90738868713379,
36
+ "loss": 421.2396,
37
+ "nll_loss": 8.764582633972168,
38
+ "po_loss": 4.39915132522583,
39
+ "rewards/accuracies": 0.44999998807907104,
40
+ "rewards/chosen": -11.011800765991211,
41
+ "rewards/margins": -1.5581079721450806,
42
+ "rewards/rejected": -9.453694343566895,
43
+ "step": 10
44
+ },
45
+ {
46
+ "epoch": 0.14222222222222222,
47
+ "grad_norm": 462.572265625,
48
+ "learning_rate": 2.34375e-05,
49
+ "log_odds_chosen": -1.2942144870758057,
50
+ "log_odds_ratio": -11.329824447631836,
51
+ "logps/chosen": -22.003665924072266,
52
+ "logps/rejected": -20.707988739013672,
53
+ "loss": 393.115,
54
+ "nll_loss": 8.009772300720215,
55
+ "po_loss": 4.2750701904296875,
56
+ "rewards/accuracies": 0.515625,
57
+ "rewards/chosen": -11.001832962036133,
58
+ "rewards/margins": -0.6478384137153625,
59
+ "rewards/rejected": -10.353994369506836,
60
+ "step": 15
61
+ },
62
+ {
63
+ "epoch": 0.18962962962962962,
64
+ "grad_norm": 3452.147705078125,
65
+ "learning_rate": 3.125e-05,
66
+ "log_odds_chosen": -4.19545316696167,
67
+ "log_odds_ratio": -9.310150146484375,
68
+ "logps/chosen": -16.894468307495117,
69
+ "logps/rejected": -12.701272964477539,
70
+ "loss": 308.1191,
71
+ "nll_loss": 6.174555778503418,
72
+ "po_loss": 3.454167127609253,
73
+ "rewards/accuracies": 0.4281249940395355,
74
+ "rewards/chosen": -8.447234153747559,
75
+ "rewards/margins": -2.0965964794158936,
76
+ "rewards/rejected": -6.3506364822387695,
77
+ "step": 20
78
+ },
79
+ {
80
+ "epoch": 0.23703703703703705,
81
+ "grad_norm": 254.29014587402344,
82
+ "learning_rate": 3.90625e-05,
83
+ "log_odds_chosen": -0.08858927339315414,
84
+ "log_odds_ratio": -1.799390435218811,
85
+ "logps/chosen": -3.9111340045928955,
86
+ "logps/rejected": -3.804535388946533,
87
+ "loss": 90.7617,
88
+ "nll_loss": 2.6883437633514404,
89
+ "po_loss": 0.14795991778373718,
90
+ "rewards/accuracies": 0.5562499761581421,
91
+ "rewards/chosen": -1.9555670022964478,
92
+ "rewards/margins": -0.05329931899905205,
93
+ "rewards/rejected": -1.9022676944732666,
94
+ "step": 25
95
+ },
96
+ {
97
+ "epoch": 0.28444444444444444,
98
+ "grad_norm": 269.8670349121094,
99
+ "learning_rate": 4.6875e-05,
100
+ "log_odds_chosen": 0.05695251747965813,
101
+ "log_odds_ratio": -0.9221351742744446,
102
+ "logps/chosen": -1.9418389797210693,
103
+ "logps/rejected": -1.9890735149383545,
104
+ "loss": 65.438,
105
+ "nll_loss": 2.0685558319091797,
106
+ "po_loss": -0.023617293685674667,
107
+ "rewards/accuracies": 0.5406249761581421,
108
+ "rewards/chosen": -0.9709194898605347,
109
+ "rewards/margins": 0.023617293685674667,
110
+ "rewards/rejected": -0.9945367574691772,
111
+ "step": 30
112
+ },
113
+ {
114
+ "epoch": 0.33185185185185184,
115
+ "grad_norm": 265.45452880859375,
116
+ "learning_rate": 4.998613757348784e-05,
117
+ "log_odds_chosen": 0.3491572439670563,
118
+ "log_odds_ratio": -0.8370898962020874,
119
+ "logps/chosen": -1.8574031591415405,
120
+ "logps/rejected": -2.1753742694854736,
121
+ "loss": 57.9215,
122
+ "nll_loss": 1.967643141746521,
123
+ "po_loss": -0.1575944572687149,
124
+ "rewards/accuracies": 0.5406249761581421,
125
+ "rewards/chosen": -0.9287015795707703,
126
+ "rewards/margins": 0.15898552536964417,
127
+ "rewards/rejected": -1.0876871347427368,
128
+ "step": 35
129
+ },
130
+ {
131
+ "epoch": 0.37925925925925924,
132
+ "grad_norm": 194.72329711914062,
133
+ "learning_rate": 4.990147841143462e-05,
134
+ "log_odds_chosen": 0.4601282477378845,
135
+ "log_odds_ratio": -0.6781095266342163,
136
+ "logps/chosen": -1.6383177042007446,
137
+ "logps/rejected": -2.0322861671447754,
138
+ "loss": 54.2986,
139
+ "nll_loss": 1.8938146829605103,
140
+ "po_loss": -0.1969841569662094,
141
+ "rewards/accuracies": 0.643750011920929,
142
+ "rewards/chosen": -0.8191588521003723,
143
+ "rewards/margins": 0.1969841569662094,
144
+ "rewards/rejected": -1.0161430835723877,
145
+ "step": 40
146
+ },
147
+ {
148
+ "epoch": 0.4266666666666667,
149
+ "grad_norm": 421.30731201171875,
150
+ "learning_rate": 4.97401218720448e-05,
151
+ "log_odds_chosen": 1.2004860639572144,
152
+ "log_odds_ratio": -0.6465462446212769,
153
+ "logps/chosen": -1.7445472478866577,
154
+ "logps/rejected": -2.871616840362549,
155
+ "loss": 42.8079,
156
+ "nll_loss": 1.8902437686920166,
157
+ "po_loss": -0.5524969100952148,
158
+ "rewards/accuracies": 0.690625011920929,
159
+ "rewards/chosen": -0.8722736239433289,
160
+ "rewards/margins": 0.5635346174240112,
161
+ "rewards/rejected": -1.4358084201812744,
162
+ "step": 45
163
+ },
164
+ {
165
+ "epoch": 0.4740740740740741,
166
+ "grad_norm": 850.62548828125,
167
+ "learning_rate": 4.9502564938797946e-05,
168
+ "log_odds_chosen": 2.4571447372436523,
169
+ "log_odds_ratio": -1.3941503763198853,
170
+ "logps/chosen": -3.2160325050354004,
171
+ "logps/rejected": -5.586895942687988,
172
+ "loss": 68.2657,
173
+ "nll_loss": 3.1136634349823,
174
+ "po_loss": -0.9803593754768372,
175
+ "rewards/accuracies": 0.706250011920929,
176
+ "rewards/chosen": -1.6080162525177002,
177
+ "rewards/margins": 1.1854315996170044,
178
+ "rewards/rejected": -2.793447971343994,
179
+ "step": 50
180
+ },
181
+ {
182
+ "epoch": 0.5214814814814814,
183
+ "grad_norm": 3458.772216796875,
184
+ "learning_rate": 4.918953929490768e-05,
185
+ "log_odds_chosen": 3.8432812690734863,
186
+ "log_odds_ratio": -0.9803560972213745,
187
+ "logps/chosen": -2.737734317779541,
188
+ "logps/rejected": -6.494544982910156,
189
+ "loss": 47.9796,
190
+ "nll_loss": 2.612356662750244,
191
+ "po_loss": -1.1129950284957886,
192
+ "rewards/accuracies": 0.6875,
193
+ "rewards/chosen": -1.3688671588897705,
194
+ "rewards/margins": 1.8784055709838867,
195
+ "rewards/rejected": -3.247272491455078,
196
+ "step": 55
197
+ },
198
+ {
199
+ "epoch": 0.5688888888888889,
200
+ "grad_norm": 689.9036254882812,
201
+ "learning_rate": 4.88020090697132e-05,
202
+ "log_odds_chosen": 4.405730724334717,
203
+ "log_odds_ratio": -0.9279032945632935,
204
+ "logps/chosen": -2.4398763179779053,
205
+ "logps/rejected": -6.7704877853393555,
206
+ "loss": 34.5555,
207
+ "nll_loss": 2.4739956855773926,
208
+ "po_loss": -1.3941353559494019,
209
+ "rewards/accuracies": 0.675000011920929,
210
+ "rewards/chosen": -1.2199381589889526,
211
+ "rewards/margins": 2.1653053760528564,
212
+ "rewards/rejected": -3.3852438926696777,
213
+ "step": 60
214
+ },
215
+ {
216
+ "epoch": 0.6162962962962963,
217
+ "grad_norm": 627.5538330078125,
218
+ "learning_rate": 4.834116786912897e-05,
219
+ "log_odds_chosen": 3.3358917236328125,
220
+ "log_odds_ratio": -1.2476164102554321,
221
+ "logps/chosen": -3.2438666820526123,
222
+ "logps/rejected": -6.503542423248291,
223
+ "loss": 59.5781,
224
+ "nll_loss": 3.0055854320526123,
225
+ "po_loss": -1.143770456314087,
226
+ "rewards/accuracies": 0.65625,
227
+ "rewards/chosen": -1.6219333410263062,
228
+ "rewards/margins": 1.6298377513885498,
229
+ "rewards/rejected": -3.2517712116241455,
230
+ "step": 65
231
+ },
232
+ {
233
+ "epoch": 0.6637037037037037,
234
+ "grad_norm": 1038.5926513671875,
235
+ "learning_rate": 4.7808435099299045e-05,
236
+ "log_odds_chosen": 5.5194196701049805,
237
+ "log_odds_ratio": -1.1829580068588257,
238
+ "logps/chosen": -3.315161943435669,
239
+ "logps/rejected": -8.737940788269043,
240
+ "loss": 18.1436,
241
+ "nll_loss": 2.795347213745117,
242
+ "po_loss": -2.228358030319214,
243
+ "rewards/accuracies": 0.7437499761581421,
244
+ "rewards/chosen": -1.6575809717178345,
245
+ "rewards/margins": 2.7113900184631348,
246
+ "rewards/rejected": -4.3689703941345215,
247
+ "step": 70
248
+ },
249
+ {
250
+ "epoch": 0.7111111111111111,
251
+ "grad_norm": 957.5834350585938,
252
+ "learning_rate": 4.720545159477922e-05,
253
+ "log_odds_chosen": 3.6360058784484863,
254
+ "log_odds_ratio": -1.2177813053131104,
255
+ "logps/chosen": -2.6733596324920654,
256
+ "logps/rejected": -6.228394031524658,
257
+ "loss": 39.0704,
258
+ "nll_loss": 2.3507373332977295,
259
+ "po_loss": -1.1297887563705444,
260
+ "rewards/accuracies": 0.6812499761581421,
261
+ "rewards/chosen": -1.3366798162460327,
262
+ "rewards/margins": 1.7775170803070068,
263
+ "rewards/rejected": -3.114197015762329,
264
+ "step": 75
265
+ },
266
+ {
267
+ "epoch": 0.7585185185185185,
268
+ "grad_norm": 391.593505859375,
269
+ "learning_rate": 4.653407456471222e-05,
270
+ "log_odds_chosen": 4.495142459869385,
271
+ "log_odds_ratio": -1.3136035203933716,
272
+ "logps/chosen": -2.7727952003479004,
273
+ "logps/rejected": -7.165794372558594,
274
+ "loss": 31.409,
275
+ "nll_loss": 2.5395021438598633,
276
+ "po_loss": -1.5579702854156494,
277
+ "rewards/accuracies": 0.628125011920929,
278
+ "rewards/chosen": -1.3863976001739502,
279
+ "rewards/margins": 2.196500062942505,
280
+ "rewards/rejected": -3.582897186279297,
281
+ "step": 80
282
+ },
283
+ {
284
+ "epoch": 0.8059259259259259,
285
+ "grad_norm": 2327.972900390625,
286
+ "learning_rate": 4.579637187256222e-05,
287
+ "log_odds_chosen": 1.418734073638916,
288
+ "log_odds_ratio": -0.8357670903205872,
289
+ "logps/chosen": -1.7513229846954346,
290
+ "logps/rejected": -3.0980544090270996,
291
+ "loss": 41.072,
292
+ "nll_loss": 1.9310455322265625,
293
+ "po_loss": -0.6475440859794617,
294
+ "rewards/accuracies": 0.625,
295
+ "rewards/chosen": -0.8756614923477173,
296
+ "rewards/margins": 0.6733657121658325,
297
+ "rewards/rejected": -1.5490272045135498,
298
+ "step": 85
299
+ },
300
+ {
301
+ "epoch": 0.8533333333333334,
302
+ "grad_norm": 778.7096557617188,
303
+ "learning_rate": 4.499461566702685e-05,
304
+ "log_odds_chosen": 4.9186859130859375,
305
+ "log_odds_ratio": -0.8927431106567383,
306
+ "logps/chosen": -3.104053497314453,
307
+ "logps/rejected": -7.928048610687256,
308
+ "loss": 26.255,
309
+ "nll_loss": 2.6208713054656982,
310
+ "po_loss": -1.8004029989242554,
311
+ "rewards/accuracies": 0.706250011920929,
312
+ "rewards/chosen": -1.5520267486572266,
313
+ "rewards/margins": 2.4119973182678223,
314
+ "rewards/rejected": -3.964024305343628,
315
+ "step": 90
316
+ },
317
+ {
318
+ "epoch": 0.9007407407407407,
319
+ "grad_norm": 1113.090576171875,
320
+ "learning_rate": 4.413127538374411e-05,
321
+ "log_odds_chosen": 4.415733337402344,
322
+ "log_odds_ratio": -1.0066527128219604,
323
+ "logps/chosen": -2.505484104156494,
324
+ "logps/rejected": -6.809667110443115,
325
+ "loss": 15.2194,
326
+ "nll_loss": 2.1575310230255127,
327
+ "po_loss": -1.6819244623184204,
328
+ "rewards/accuracies": 0.6937500238418579,
329
+ "rewards/chosen": -1.252742052078247,
330
+ "rewards/margins": 2.1520915031433105,
331
+ "rewards/rejected": -3.4048335552215576,
332
+ "step": 95
333
+ },
334
+ {
335
+ "epoch": 0.9481481481481482,
336
+ "grad_norm": 5262.9931640625,
337
+ "learning_rate": 4.320901013934887e-05,
338
+ "log_odds_chosen": 4.908814430236816,
339
+ "log_odds_ratio": -1.2427480220794678,
340
+ "logps/chosen": -2.8461787700653076,
341
+ "logps/rejected": -7.642964839935303,
342
+ "loss": 17.6349,
343
+ "nll_loss": 2.5891518592834473,
344
+ "po_loss": -2.0380616188049316,
345
+ "rewards/accuracies": 0.699999988079071,
346
+ "rewards/chosen": -1.4230893850326538,
347
+ "rewards/margins": 2.398393154144287,
348
+ "rewards/rejected": -3.8214824199676514,
349
+ "step": 100
350
+ },
351
+ {
352
+ "epoch": 0.9955555555555555,
353
+ "grad_norm": 465.037841796875,
354
+ "learning_rate": 4.223066054130568e-05,
355
+ "log_odds_chosen": 2.4661428928375244,
356
+ "log_odds_ratio": -0.8429304957389832,
357
+ "logps/chosen": -1.9539477825164795,
358
+ "logps/rejected": -4.3218865394592285,
359
+ "loss": 27.8368,
360
+ "nll_loss": 2.0108437538146973,
361
+ "po_loss": -1.1409438848495483,
362
+ "rewards/accuracies": 0.690625011920929,
363
+ "rewards/chosen": -0.9769738912582397,
364
+ "rewards/margins": 1.1839690208435059,
365
+ "rewards/rejected": -2.1609432697296143,
366
+ "step": 105
367
+ },
368
+ {
369
+ "epoch": 1.037925925925926,
370
+ "grad_norm": 1398.3836669921875,
371
+ "learning_rate": 4.1199239938743797e-05,
372
+ "log_odds_chosen": 6.5568623542785645,
373
+ "log_odds_ratio": -1.0280228853225708,
374
+ "logps/chosen": -3.167757511138916,
375
+ "logps/rejected": -9.576342582702637,
376
+ "loss": 6.365,
377
+ "nll_loss": 2.601748466491699,
378
+ "po_loss": -2.3791961669921875,
379
+ "rewards/accuracies": 0.748251736164093,
380
+ "rewards/chosen": -1.583878755569458,
381
+ "rewards/margins": 3.2042922973632812,
382
+ "rewards/rejected": -4.788171291351318,
383
+ "step": 110
384
+ },
385
+ {
386
+ "epoch": 1.0853333333333333,
387
+ "grad_norm": 468.6627502441406,
388
+ "learning_rate": 4.0117925141242174e-05,
389
+ "log_odds_chosen": 7.842375755310059,
390
+ "log_odds_ratio": -0.8314769864082336,
391
+ "logps/chosen": -2.4292473793029785,
392
+ "logps/rejected": -10.092866897583008,
393
+ "loss": -10.5443,
394
+ "nll_loss": 2.201972007751465,
395
+ "po_loss": -2.5314807891845703,
396
+ "rewards/accuracies": 0.75,
397
+ "rewards/chosen": -1.2146236896514893,
398
+ "rewards/margins": 3.8318095207214355,
399
+ "rewards/rejected": -5.046433448791504,
400
+ "step": 115
401
+ },
402
+ {
403
+ "epoch": 1.1327407407407408,
404
+ "grad_norm": 1342.4088134765625,
405
+ "learning_rate": 3.899004663415084e-05,
406
+ "log_odds_chosen": 6.2360992431640625,
407
+ "log_odds_ratio": -0.7881489992141724,
408
+ "logps/chosen": -2.0156655311584473,
409
+ "logps/rejected": -8.078754425048828,
410
+ "loss": -3.799,
411
+ "nll_loss": 2.015702247619629,
412
+ "po_loss": -2.1344211101531982,
413
+ "rewards/accuracies": 0.7562500238418579,
414
+ "rewards/chosen": -1.0078327655792236,
415
+ "rewards/margins": 3.0315444469451904,
416
+ "rewards/rejected": -4.039377212524414,
417
+ "step": 120
418
+ },
419
+ {
420
+ "epoch": 1.1801481481481482,
421
+ "grad_norm": 780.4815673828125,
422
+ "learning_rate": 3.781907832058587e-05,
423
+ "log_odds_chosen": 6.6184797286987305,
424
+ "log_odds_ratio": -1.2134829759597778,
425
+ "logps/chosen": -2.7371468544006348,
426
+ "logps/rejected": -9.186931610107422,
427
+ "loss": 11.0535,
428
+ "nll_loss": 2.5944888591766357,
429
+ "po_loss": -2.249067783355713,
430
+ "rewards/accuracies": 0.75,
431
+ "rewards/chosen": -1.3685734272003174,
432
+ "rewards/margins": 3.2248923778533936,
433
+ "rewards/rejected": -4.593465805053711,
434
+ "step": 125
435
+ },
436
+ {
437
+ "epoch": 1.2275555555555555,
438
+ "grad_norm": 827.787353515625,
439
+ "learning_rate": 3.660862682169282e-05,
440
+ "log_odds_chosen": 8.269857406616211,
441
+ "log_odds_ratio": -1.018838882446289,
442
+ "logps/chosen": -2.7996513843536377,
443
+ "logps/rejected": -10.883502960205078,
444
+ "loss": -11.0712,
445
+ "nll_loss": 2.5594162940979004,
446
+ "po_loss": -2.9053916931152344,
447
+ "rewards/accuracies": 0.753125011920929,
448
+ "rewards/chosen": -1.3998256921768188,
449
+ "rewards/margins": 4.041925430297852,
450
+ "rewards/rejected": -5.441751480102539,
451
+ "step": 130
452
+ },
453
+ {
454
+ "epoch": 1.274962962962963,
455
+ "grad_norm": 602.3118896484375,
456
+ "learning_rate": 3.5362420368134356e-05,
457
+ "log_odds_chosen": 5.271297454833984,
458
+ "log_odds_ratio": -0.828192412853241,
459
+ "logps/chosen": -1.9964498281478882,
460
+ "logps/rejected": -7.077614784240723,
461
+ "loss": 1.6283,
462
+ "nll_loss": 2.024073362350464,
463
+ "po_loss": -1.973189353942871,
464
+ "rewards/accuracies": 0.7437499761581421,
465
+ "rewards/chosen": -0.9982249140739441,
466
+ "rewards/margins": 2.5405819416046143,
467
+ "rewards/rejected": -3.5388073921203613,
468
+ "step": 135
469
+ },
470
+ {
471
+ "epoch": 1.3223703703703704,
472
+ "grad_norm": 926.779052734375,
473
+ "learning_rate": 3.408429731701635e-05,
474
+ "log_odds_chosen": 5.869942665100098,
475
+ "log_odds_ratio": -0.8702658414840698,
476
+ "logps/chosen": -2.663573980331421,
477
+ "logps/rejected": -8.38515853881836,
478
+ "loss": 0.999,
479
+ "nll_loss": 2.1175601482391357,
480
+ "po_loss": -2.0863418579101562,
481
+ "rewards/accuracies": 0.7437499761581421,
482
+ "rewards/chosen": -1.3317869901657104,
483
+ "rewards/margins": 2.8607921600341797,
484
+ "rewards/rejected": -4.19257926940918,
485
+ "step": 140
486
+ },
487
+ {
488
+ "epoch": 1.3697777777777778,
489
+ "grad_norm": 1278.476806640625,
490
+ "learning_rate": 3.2778194329621104e-05,
491
+ "log_odds_chosen": 9.726934432983398,
492
+ "log_odds_ratio": -1.0185357332229614,
493
+ "logps/chosen": -3.2876574993133545,
494
+ "logps/rejected": -12.812705993652344,
495
+ "loss": -19.5231,
496
+ "nll_loss": 2.5718603134155273,
497
+ "po_loss": -3.181957483291626,
498
+ "rewards/accuracies": 0.793749988079071,
499
+ "rewards/chosen": -1.6438287496566772,
500
+ "rewards/margins": 4.762524604797363,
501
+ "rewards/rejected": -6.406352996826172,
502
+ "step": 145
503
+ },
504
+ {
505
+ "epoch": 1.417185185185185,
506
+ "grad_norm": 1008.686279296875,
507
+ "learning_rate": 3.144813424636031e-05,
508
+ "log_odds_chosen": 7.952484130859375,
509
+ "log_odds_ratio": -0.8371334075927734,
510
+ "logps/chosen": -2.3361740112304688,
511
+ "logps/rejected": -10.076719284057617,
512
+ "loss": -22.1519,
513
+ "nll_loss": 1.983311653137207,
514
+ "po_loss": -2.6755595207214355,
515
+ "rewards/accuracies": 0.7875000238418579,
516
+ "rewards/chosen": -1.1680870056152344,
517
+ "rewards/margins": 3.8702728748321533,
518
+ "rewards/rejected": -5.038359642028809,
519
+ "step": 150
520
+ },
521
+ {
522
+ "epoch": 1.4645925925925927,
523
+ "grad_norm": 789.317138671875,
524
+ "learning_rate": 3.0098213696293542e-05,
525
+ "log_odds_chosen": 8.040031433105469,
526
+ "log_odds_ratio": -1.153850793838501,
527
+ "logps/chosen": -2.9109256267547607,
528
+ "logps/rejected": -10.789738655090332,
529
+ "loss": 0.2885,
530
+ "nll_loss": 2.3872437477111816,
531
+ "po_loss": -2.378229856491089,
532
+ "rewards/accuracies": 0.71875,
533
+ "rewards/chosen": -1.4554628133773804,
534
+ "rewards/margins": 3.939406156539917,
535
+ "rewards/rejected": -5.394869327545166,
536
+ "step": 155
537
+ },
538
+ {
539
+ "epoch": 1.512,
540
+ "grad_norm": 951.3047485351562,
541
+ "learning_rate": 2.8732590479375165e-05,
542
+ "log_odds_chosen": 7.912862300872803,
543
+ "log_odds_ratio": -0.7059909105300903,
544
+ "logps/chosen": -2.1726765632629395,
545
+ "logps/rejected": -9.871898651123047,
546
+ "loss": -31.1045,
547
+ "nll_loss": 2.01442289352417,
548
+ "po_loss": -2.9864375591278076,
549
+ "rewards/accuracies": 0.768750011920929,
550
+ "rewards/chosen": -1.0863382816314697,
551
+ "rewards/margins": 3.8496108055114746,
552
+ "rewards/rejected": -4.935949325561523,
553
+ "step": 160
554
+ },
555
+ {
556
+ "epoch": 1.5594074074074074,
557
+ "grad_norm": 2454.81591796875,
558
+ "learning_rate": 2.7355470760292956e-05,
559
+ "log_odds_chosen": 9.038829803466797,
560
+ "log_odds_ratio": -1.0712902545928955,
561
+ "logps/chosen": -3.265695095062256,
562
+ "logps/rejected": -12.09699821472168,
563
+ "loss": -28.5012,
564
+ "nll_loss": 2.4790894985198975,
565
+ "po_loss": -3.3697502613067627,
566
+ "rewards/accuracies": 0.778124988079071,
567
+ "rewards/chosen": -1.632847547531128,
568
+ "rewards/margins": 4.415651798248291,
569
+ "rewards/rejected": -6.04849910736084,
570
+ "step": 165
571
+ },
572
+ {
573
+ "epoch": 1.6068148148148147,
574
+ "grad_norm": 1764.2366943359375,
575
+ "learning_rate": 2.597109611334169e-05,
576
+ "log_odds_chosen": 9.091514587402344,
577
+ "log_odds_ratio": -1.6616445779800415,
578
+ "logps/chosen": -4.427676200866699,
579
+ "logps/rejected": -13.332109451293945,
580
+ "loss": -3.2606,
581
+ "nll_loss": 3.0899510383605957,
582
+ "po_loss": -3.191845417022705,
583
+ "rewards/accuracies": 0.7718750238418579,
584
+ "rewards/chosen": -2.2138381004333496,
585
+ "rewards/margins": 4.452216625213623,
586
+ "rewards/rejected": -6.666054725646973,
587
+ "step": 170
588
+ },
589
+ {
590
+ "epoch": 1.6542222222222223,
591
+ "grad_norm": 1194.388916015625,
592
+ "learning_rate": 2.458373045823404e-05,
593
+ "log_odds_chosen": 7.1326398849487305,
594
+ "log_odds_ratio": -0.9163221120834351,
595
+ "logps/chosen": -2.580127716064453,
596
+ "logps/rejected": -9.50422477722168,
597
+ "loss": -18.0293,
598
+ "nll_loss": 2.2237651348114014,
599
+ "po_loss": -2.787180185317993,
600
+ "rewards/accuracies": 0.762499988079071,
601
+ "rewards/chosen": -1.2900638580322266,
602
+ "rewards/margins": 3.462048292160034,
603
+ "rewards/rejected": -4.75211238861084,
604
+ "step": 175
605
+ },
606
+ {
607
+ "epoch": 1.7016296296296296,
608
+ "grad_norm": 2008.02197265625,
609
+ "learning_rate": 2.3197646927086697e-05,
610
+ "log_odds_chosen": 7.623831748962402,
611
+ "log_odds_ratio": -0.8227715492248535,
612
+ "logps/chosen": -2.170476198196411,
613
+ "logps/rejected": -9.566916465759277,
614
+ "loss": -24.3356,
615
+ "nll_loss": 2.127624988555908,
616
+ "po_loss": -2.888112783432007,
617
+ "rewards/accuracies": 0.768750011920929,
618
+ "rewards/chosen": -1.0852380990982056,
619
+ "rewards/margins": 3.698220729827881,
620
+ "rewards/rejected": -4.783458232879639,
621
+ "step": 180
622
+ },
623
+ {
624
+ "epoch": 1.749037037037037,
625
+ "grad_norm": 2212.9755859375,
626
+ "learning_rate": 2.1817114703032176e-05,
627
+ "log_odds_chosen": 11.049670219421387,
628
+ "log_odds_ratio": -1.2048530578613281,
629
+ "logps/chosen": -3.1207997798919678,
630
+ "logps/rejected": -13.965868949890137,
631
+ "loss": -42.3188,
632
+ "nll_loss": 2.493201732635498,
633
+ "po_loss": -3.8156638145446777,
634
+ "rewards/accuracies": 0.78125,
635
+ "rewards/chosen": -1.5603998899459839,
636
+ "rewards/margins": 5.422535419464111,
637
+ "rewards/rejected": -6.982934474945068,
638
+ "step": 185
639
+ },
640
+ {
641
+ "epoch": 1.7964444444444445,
642
+ "grad_norm": 1651.638671875,
643
+ "learning_rate": 2.0446385870993467e-05,
644
+ "log_odds_chosen": 9.86381721496582,
645
+ "log_odds_ratio": -0.9923731088638306,
646
+ "logps/chosen": -2.9157769680023193,
647
+ "logps/rejected": -12.55068588256836,
648
+ "loss": -36.9979,
649
+ "nll_loss": 2.172974109649658,
650
+ "po_loss": -3.329157590866089,
651
+ "rewards/accuracies": 0.796875,
652
+ "rewards/chosen": -1.4578884840011597,
653
+ "rewards/margins": 4.8174543380737305,
654
+ "rewards/rejected": -6.27534294128418,
655
+ "step": 190
656
+ },
657
+ {
658
+ "epoch": 1.8438518518518519,
659
+ "grad_norm": 5688.65869140625,
660
+ "learning_rate": 1.9089682321121834e-05,
661
+ "log_odds_chosen": 8.133454322814941,
662
+ "log_odds_ratio": -1.3896712064743042,
663
+ "logps/chosen": -3.334374189376831,
664
+ "logps/rejected": -11.257940292358398,
665
+ "loss": -14.291,
666
+ "nll_loss": 2.494044303894043,
667
+ "po_loss": -2.9406371116638184,
668
+ "rewards/accuracies": 0.7906249761581421,
669
+ "rewards/chosen": -1.6671870946884155,
670
+ "rewards/margins": 3.961782932281494,
671
+ "rewards/rejected": -5.628970146179199,
672
+ "step": 195
673
+ },
674
+ {
675
+ "epoch": 1.8912592592592592,
676
+ "grad_norm": 4854.55810546875,
677
+ "learning_rate": 1.775118274523545e-05,
678
+ "log_odds_chosen": 8.46328353881836,
679
+ "log_odds_ratio": -2.0774197578430176,
680
+ "logps/chosen": -4.338870525360107,
681
+ "logps/rejected": -12.631368637084961,
682
+ "loss": 0.4968,
683
+ "nll_loss": 3.0006825923919678,
684
+ "po_loss": -2.985157012939453,
685
+ "rewards/accuracies": 0.737500011920929,
686
+ "rewards/chosen": -2.1694352626800537,
687
+ "rewards/margins": 4.1462483406066895,
688
+ "rewards/rejected": -6.3156843185424805,
689
+ "step": 200
690
+ },
691
+ {
692
+ "epoch": 1.9386666666666668,
693
+ "grad_norm": 1195.15283203125,
694
+ "learning_rate": 1.643500976631037e-05,
695
+ "log_odds_chosen": 8.545049667358398,
696
+ "log_odds_ratio": -1.2959668636322021,
697
+ "logps/chosen": -3.3477349281311035,
698
+ "logps/rejected": -11.675881385803223,
699
+ "loss": -24.4773,
700
+ "nll_loss": 2.649782419204712,
701
+ "po_loss": -3.414698362350464,
702
+ "rewards/accuracies": 0.784375011920929,
703
+ "rewards/chosen": -1.6738674640655518,
704
+ "rewards/margins": 4.1640729904174805,
705
+ "rewards/rejected": -5.837940692901611,
706
+ "step": 205
707
+ },
708
+ {
709
+ "epoch": 1.986074074074074,
710
+ "grad_norm": 1435.34326171875,
711
+ "learning_rate": 1.514521724066537e-05,
712
+ "log_odds_chosen": 7.497690677642822,
713
+ "log_odds_ratio": -1.6079471111297607,
714
+ "logps/chosen": -3.8709499835968018,
715
+ "logps/rejected": -11.176200866699219,
716
+ "loss": -8.5711,
717
+ "nll_loss": 2.686655282974243,
718
+ "po_loss": -2.9545013904571533,
719
+ "rewards/accuracies": 0.762499988079071,
720
+ "rewards/chosen": -1.9354749917984009,
721
+ "rewards/margins": 3.6526246070861816,
722
+ "rewards/rejected": -5.588100433349609,
723
+ "step": 210
724
+ },
725
+ {
726
+ "epoch": 2.0284444444444443,
727
+ "grad_norm": 784.1499633789062,
728
+ "learning_rate": 1.3885777771950348e-05,
729
+ "log_odds_chosen": 8.465600967407227,
730
+ "log_odds_ratio": -0.868893027305603,
731
+ "logps/chosen": -2.0467655658721924,
732
+ "logps/rejected": -10.241917610168457,
733
+ "loss": -37.4481,
734
+ "nll_loss": 1.8826793432235718,
735
+ "po_loss": -3.1920523643493652,
736
+ "rewards/accuracies": 0.7692307829856873,
737
+ "rewards/chosen": -1.0233827829360962,
738
+ "rewards/margins": 4.097576141357422,
739
+ "rewards/rejected": -5.1209588050842285,
740
+ "step": 215
741
+ },
742
+ {
743
+ "epoch": 2.075851851851852,
744
+ "grad_norm": 1654.237548828125,
745
+ "learning_rate": 1.2660570475395683e-05,
746
+ "log_odds_chosen": 8.654611587524414,
747
+ "log_odds_ratio": -1.1223597526550293,
748
+ "logps/chosen": -2.459672212600708,
749
+ "logps/rejected": -10.875265121459961,
750
+ "loss": -34.0465,
751
+ "nll_loss": 2.0625672340393066,
752
+ "po_loss": -3.126519203186035,
753
+ "rewards/accuracies": 0.765625,
754
+ "rewards/chosen": -1.229836106300354,
755
+ "rewards/margins": 4.207796573638916,
756
+ "rewards/rejected": -5.4376325607299805,
757
+ "step": 220
758
+ },
759
+ {
760
+ "epoch": 2.1232592592592594,
761
+ "grad_norm": 2589.6494140625,
762
+ "learning_rate": 1.1473369030008974e-05,
763
+ "log_odds_chosen": 15.148641586303711,
764
+ "log_odds_ratio": -1.0744845867156982,
765
+ "logps/chosen": -3.3443615436553955,
766
+ "logps/rejected": -18.203109741210938,
767
+ "loss": -73.6899,
768
+ "nll_loss": 2.520362377166748,
769
+ "po_loss": -4.8231706619262695,
770
+ "rewards/accuracies": 0.840624988079071,
771
+ "rewards/chosen": -1.6721807718276978,
772
+ "rewards/margins": 7.429374694824219,
773
+ "rewards/rejected": -9.101554870605469,
774
+ "step": 225
775
+ },
776
+ {
777
+ "epoch": 2.1706666666666665,
778
+ "grad_norm": 1232.9119873046875,
779
+ "learning_rate": 1.0327830055518842e-05,
780
+ "log_odds_chosen": 14.175605773925781,
781
+ "log_odds_ratio": -0.9750823974609375,
782
+ "logps/chosen": -3.1031196117401123,
783
+ "logps/rejected": -17.016685485839844,
784
+ "loss": -66.4024,
785
+ "nll_loss": 2.539062976837158,
786
+ "po_loss": -4.614138603210449,
787
+ "rewards/accuracies": 0.840624988079071,
788
+ "rewards/chosen": -1.5515598058700562,
789
+ "rewards/margins": 6.956783294677734,
790
+ "rewards/rejected": -8.508342742919922,
791
+ "step": 230
792
+ },
793
+ {
794
+ "epoch": 2.218074074074074,
795
+ "grad_norm": 2281.852294921875,
796
+ "learning_rate": 9.227481849865235e-06,
797
+ "log_odds_chosen": 13.702494621276855,
798
+ "log_odds_ratio": -0.6508721113204956,
799
+ "logps/chosen": -2.776512861251831,
800
+ "logps/rejected": -16.18694496154785,
801
+ "loss": -76.9486,
802
+ "nll_loss": 2.2473607063293457,
803
+ "po_loss": -4.652005672454834,
804
+ "rewards/accuracies": 0.831250011920929,
805
+ "rewards/chosen": -1.3882564306259155,
806
+ "rewards/margins": 6.705216407775879,
807
+ "rewards/rejected": -8.093472480773926,
808
+ "step": 235
809
+ },
810
+ {
811
+ "epoch": 2.2654814814814817,
812
+ "grad_norm": 1797.7962646484375,
813
+ "learning_rate": 8.175713521924978e-06,
814
+ "log_odds_chosen": 14.10627555847168,
815
+ "log_odds_ratio": -0.9793645739555359,
816
+ "logps/chosen": -3.2738196849823,
817
+ "logps/rejected": -17.115571975708008,
818
+ "loss": -71.8213,
819
+ "nll_loss": 2.362848997116089,
820
+ "po_loss": -4.607265472412109,
821
+ "rewards/accuracies": 0.859375,
822
+ "rewards/chosen": -1.63690984249115,
823
+ "rewards/margins": 6.92087459564209,
824
+ "rewards/rejected": -8.557785987854004,
825
+ "step": 240
826
+ },
827
+ {
828
+ "epoch": 2.3128888888888888,
829
+ "grad_norm": 2065.531005859375,
830
+ "learning_rate": 7.1757645529443665e-06,
831
+ "log_odds_chosen": 14.409799575805664,
832
+ "log_odds_ratio": -0.6093363165855408,
833
+ "logps/chosen": -2.691865921020508,
834
+ "logps/rejected": -16.832874298095703,
835
+ "loss": -86.8692,
836
+ "nll_loss": 2.2543296813964844,
837
+ "po_loss": -4.968992710113525,
838
+ "rewards/accuracies": 0.8374999761581421,
839
+ "rewards/chosen": -1.345932960510254,
840
+ "rewards/margins": 7.070504188537598,
841
+ "rewards/rejected": -8.416437149047852,
842
+ "step": 245
843
+ },
844
+ {
845
+ "epoch": 2.3602962962962963,
846
+ "grad_norm": 1674.9385986328125,
847
+ "learning_rate": 6.230714818829733e-06,
848
+ "log_odds_chosen": 14.006881713867188,
849
+ "log_odds_ratio": -0.6611822247505188,
850
+ "logps/chosen": -2.298327684402466,
851
+ "logps/rejected": -16.025455474853516,
852
+ "loss": -73.411,
853
+ "nll_loss": 2.0404696464538574,
854
+ "po_loss": -4.3345627784729,
855
+ "rewards/accuracies": 0.784375011920929,
856
+ "rewards/chosen": -1.149163842201233,
857
+ "rewards/margins": 6.863563537597656,
858
+ "rewards/rejected": -8.012727737426758,
859
+ "step": 250
860
+ },
861
+ {
862
+ "epoch": 2.407703703703704,
863
+ "grad_norm": 1458.6246337890625,
864
+ "learning_rate": 5.343475104027743e-06,
865
+ "log_odds_chosen": 14.005816459655762,
866
+ "log_odds_ratio": -0.6005424857139587,
867
+ "logps/chosen": -2.2070460319519043,
868
+ "logps/rejected": -15.93396282196045,
869
+ "loss": -76.6993,
870
+ "nll_loss": 1.9984962940216064,
871
+ "po_loss": -4.395350456237793,
872
+ "rewards/accuracies": 0.856249988079071,
873
+ "rewards/chosen": -1.1035230159759521,
874
+ "rewards/margins": 6.863458156585693,
875
+ "rewards/rejected": -7.966981410980225,
876
+ "step": 255
877
+ },
878
+ {
879
+ "epoch": 2.455111111111111,
880
+ "grad_norm": 1062.482666015625,
881
+ "learning_rate": 4.516778136213037e-06,
882
+ "log_odds_chosen": 13.041534423828125,
883
+ "log_odds_ratio": -0.6385210752487183,
884
+ "logps/chosen": -2.050607204437256,
885
+ "logps/rejected": -14.810098648071289,
886
+ "loss": -73.7325,
887
+ "nll_loss": 1.8901704549789429,
888
+ "po_loss": -4.19431209564209,
889
+ "rewards/accuracies": 0.862500011920929,
890
+ "rewards/chosen": -1.025303602218628,
891
+ "rewards/margins": 6.379745006561279,
892
+ "rewards/rejected": -7.4050493240356445,
893
+ "step": 260
894
+ },
895
+ {
896
+ "epoch": 2.5025185185185186,
897
+ "grad_norm": 2426.10986328125,
898
+ "learning_rate": 3.7531701693965554e-06,
899
+ "log_odds_chosen": 16.418107986450195,
900
+ "log_odds_ratio": -0.8002341985702515,
901
+ "logps/chosen": -2.38873028755188,
902
+ "logps/rejected": -18.518217086791992,
903
+ "loss": -90.9132,
904
+ "nll_loss": 2.055996894836426,
905
+ "po_loss": -4.897036075592041,
906
+ "rewards/accuracies": 0.84375,
907
+ "rewards/chosen": -1.19436514377594,
908
+ "rewards/margins": 8.064743995666504,
909
+ "rewards/rejected": -9.259108543395996,
910
+ "step": 265
911
+ },
912
+ {
913
+ "epoch": 2.549925925925926,
914
+ "grad_norm": 2285.034423828125,
915
+ "learning_rate": 3.055003141378948e-06,
916
+ "log_odds_chosen": 18.129314422607422,
917
+ "log_odds_ratio": -1.1312395334243774,
918
+ "logps/chosen": -3.370629072189331,
919
+ "logps/rejected": -21.226877212524414,
920
+ "loss": -80.7968,
921
+ "nll_loss": 2.6373114585876465,
922
+ "po_loss": -5.162210941314697,
923
+ "rewards/accuracies": 0.828125,
924
+ "rewards/chosen": -1.6853145360946655,
925
+ "rewards/margins": 8.928122520446777,
926
+ "rewards/rejected": -10.613438606262207,
927
+ "step": 270
928
+ },
929
+ {
930
+ "epoch": 2.5973333333333333,
931
+ "grad_norm": 2841.40380859375,
932
+ "learning_rate": 2.424427429704365e-06,
933
+ "log_odds_chosen": 16.357511520385742,
934
+ "log_odds_ratio": -1.1468102931976318,
935
+ "logps/chosen": -3.206864595413208,
936
+ "logps/rejected": -19.272470474243164,
937
+ "loss": -74.3951,
938
+ "nll_loss": 2.3466782569885254,
939
+ "po_loss": -4.671523571014404,
940
+ "rewards/accuracies": 0.856249988079071,
941
+ "rewards/chosen": -1.603432297706604,
942
+ "rewards/margins": 8.032803535461426,
943
+ "rewards/rejected": -9.636235237121582,
944
+ "step": 275
945
+ },
946
+ {
947
+ "epoch": 2.644740740740741,
948
+ "grad_norm": 1928.48876953125,
949
+ "learning_rate": 1.8633852284264508e-06,
950
+ "log_odds_chosen": 16.074771881103516,
951
+ "log_odds_ratio": -0.9569836854934692,
952
+ "logps/chosen": -2.4576973915100098,
953
+ "logps/rejected": -18.23931884765625,
954
+ "loss": -88.0982,
955
+ "nll_loss": 2.244922161102295,
956
+ "po_loss": -4.997992515563965,
957
+ "rewards/accuracies": 0.828125,
958
+ "rewards/chosen": -1.2288486957550049,
959
+ "rewards/margins": 7.890810489654541,
960
+ "rewards/rejected": -9.119659423828125,
961
+ "step": 280
962
+ },
963
+ {
964
+ "epoch": 2.6921481481481484,
965
+ "grad_norm": 1673.14990234375,
966
+ "learning_rate": 1.3736045660864034e-06,
967
+ "log_odds_chosen": 15.582837104797363,
968
+ "log_odds_ratio": -0.819076657295227,
969
+ "logps/chosen": -2.702998638153076,
970
+ "logps/rejected": -18.022306442260742,
971
+ "loss": -88.1281,
972
+ "nll_loss": 2.126950740814209,
973
+ "po_loss": -4.880953788757324,
974
+ "rewards/accuracies": 0.8374999761581421,
975
+ "rewards/chosen": -1.351499319076538,
976
+ "rewards/margins": 7.659654140472412,
977
+ "rewards/rejected": -9.011153221130371,
978
+ "step": 285
979
+ },
980
+ {
981
+ "epoch": 2.7395555555555555,
982
+ "grad_norm": 1918.1820068359375,
983
+ "learning_rate": 9.565939833279192e-07,
984
+ "log_odds_chosen": 14.058061599731445,
985
+ "log_odds_ratio": -1.0425951480865479,
986
+ "logps/chosen": -3.089294910430908,
987
+ "logps/rejected": -16.88263511657715,
988
+ "loss": -59.2928,
989
+ "nll_loss": 2.3471436500549316,
990
+ "po_loss": -4.200043678283691,
991
+ "rewards/accuracies": 0.8343750238418579,
992
+ "rewards/chosen": -1.544647455215454,
993
+ "rewards/margins": 6.896668434143066,
994
+ "rewards/rejected": -8.441317558288574,
995
+ "step": 290
996
+ },
997
+ {
998
+ "epoch": 2.786962962962963,
999
+ "grad_norm": 1549.3990478515625,
1000
+ "learning_rate": 6.136378865420872e-07,
1001
+ "log_odds_chosen": 14.449419975280762,
1002
+ "log_odds_ratio": -0.8233845829963684,
1003
+ "logps/chosen": -2.452683448791504,
1004
+ "logps/rejected": -16.653139114379883,
1005
+ "loss": -82.7587,
1006
+ "nll_loss": 1.9743688106536865,
1007
+ "po_loss": -4.560579299926758,
1008
+ "rewards/accuracies": 0.8031250238418579,
1009
+ "rewards/chosen": -1.226341724395752,
1010
+ "rewards/margins": 7.100228309631348,
1011
+ "rewards/rejected": -8.326569557189941,
1012
+ "step": 295
1013
+ },
1014
+ {
1015
+ "epoch": 2.83437037037037,
1016
+ "grad_norm": 1157.6502685546875,
1017
+ "learning_rate": 3.45792591853214e-07,
1018
+ "log_odds_chosen": 15.727914810180664,
1019
+ "log_odds_ratio": -0.5168970227241516,
1020
+ "logps/chosen": -2.1314849853515625,
1021
+ "logps/rejected": -17.58350944519043,
1022
+ "loss": -96.6356,
1023
+ "nll_loss": 1.8993088006973267,
1024
+ "po_loss": -4.9191718101501465,
1025
+ "rewards/accuracies": 0.8374999761581421,
1026
+ "rewards/chosen": -1.0657424926757812,
1027
+ "rewards/margins": 7.72601318359375,
1028
+ "rewards/rejected": -8.791754722595215,
1029
+ "step": 300
1030
+ },
1031
+ {
1032
+ "epoch": 2.8817777777777778,
1033
+ "grad_norm": 3823.998291015625,
1034
+ "learning_rate": 1.538830716302092e-07,
1035
+ "log_odds_chosen": 16.228670120239258,
1036
+ "log_odds_ratio": -0.9718038439750671,
1037
+ "logps/chosen": -2.4307305812835693,
1038
+ "logps/rejected": -18.350786209106445,
1039
+ "loss": -87.9496,
1040
+ "nll_loss": 2.063523769378662,
1041
+ "po_loss": -4.811949253082275,
1042
+ "rewards/accuracies": 0.828125,
1043
+ "rewards/chosen": -1.2153652906417847,
1044
+ "rewards/margins": 7.960028171539307,
1045
+ "rewards/rejected": -9.175393104553223,
1046
+ "step": 305
1047
+ },
1048
+ {
1049
+ "epoch": 2.9291851851851853,
1050
+ "grad_norm": 2266.873779296875,
1051
+ "learning_rate": 3.8500413544415025e-08,
1052
+ "log_odds_chosen": 16.433456420898438,
1053
+ "log_odds_ratio": -0.6722828149795532,
1054
+ "logps/chosen": -2.7718887329101562,
1055
+ "logps/rejected": -18.93569564819336,
1056
+ "loss": -83.3821,
1057
+ "nll_loss": 2.255375623703003,
1058
+ "po_loss": -4.861066818237305,
1059
+ "rewards/accuracies": 0.828125,
1060
+ "rewards/chosen": -1.3859443664550781,
1061
+ "rewards/margins": 8.081903457641602,
1062
+ "rewards/rejected": -9.46784782409668,
1063
+ "step": 310
1064
+ },
1065
+ {
1066
+ "epoch": 2.9765925925925925,
1067
+ "grad_norm": 1334.365234375,
1068
+ "learning_rate": 0.0,
1069
+ "log_odds_chosen": 15.443585395812988,
1070
+ "log_odds_ratio": -0.6007210612297058,
1071
+ "logps/chosen": -1.9952449798583984,
1072
+ "logps/rejected": -17.120744705200195,
1073
+ "loss": -98.7664,
1074
+ "nll_loss": 1.7973524332046509,
1075
+ "po_loss": -4.8838019371032715,
1076
+ "rewards/accuracies": 0.8500000238418579,
1077
+ "rewards/chosen": -0.9976224899291992,
1078
+ "rewards/margins": 7.562750339508057,
1079
+ "rewards/rejected": -8.560372352600098,
1080
+ "step": 315
1081
+ },
1082
+ {
1083
+ "epoch": 2.9765925925925925,
1084
+ "step": 315,
1085
+ "total_flos": 0.0,
1086
+ "train_loss": 6.844926292177231,
1087
+ "train_runtime": 9518.1794,
1088
+ "train_samples_per_second": 2.128,
1089
+ "train_steps_per_second": 0.033
1090
+ }
1091
+ ],
1092
+ "logging_steps": 5,
1093
+ "max_steps": 315,
1094
+ "num_input_tokens_seen": 0,
1095
+ "num_train_epochs": 3,
1096
+ "save_steps": 100000,
1097
+ "stateful_callbacks": {
1098
+ "TrainerControl": {
1099
+ "args": {
1100
+ "should_epoch_stop": false,
1101
+ "should_evaluate": false,
1102
+ "should_log": false,
1103
+ "should_save": true,
1104
+ "should_training_stop": true
1105
+ },
1106
+ "attributes": {}
1107
+ }
1108
+ },
1109
+ "total_flos": 0.0,
1110
+ "train_batch_size": 1,
1111
+ "trial_name": null,
1112
+ "trial_params": null
1113
+ }