silviasapora commited on
Commit
a2a488a
·
verified ·
1 Parent(s): 7431a6c

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 +1176 -0
README.md ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: mistralai/Mistral-7B-v0.3
3
+ library_name: transformers
4
+ model_name: mistral-7b-borpo-noisy-5e-5-01-norm2
5
+ tags:
6
+ - generated_from_trainer
7
+ - trl
8
+ - orpo
9
+ licence: license
10
+ ---
11
+
12
+ # Model Card for mistral-7b-borpo-noisy-5e-5-01-norm2
13
+
14
+ This model is a fine-tuned version of [mistralai/Mistral-7B-v0.3](https://huggingface.co/mistralai/Mistral-7B-v0.3).
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/mistral-7b-borpo-noisy-5e-5-01-norm2", 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/8uq3wri5)
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": 26.468634469168528,
5
+ "train_runtime": 6925.9574,
6
+ "train_samples": 6750,
7
+ "train_samples_per_second": 2.924,
8
+ "train_steps_per_second": 0.045
9
+ }
train_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 2.9765925925925925,
3
+ "total_flos": 0.0,
4
+ "train_loss": 26.468634469168528,
5
+ "train_runtime": 6925.9574,
6
+ "train_samples": 6750,
7
+ "train_samples_per_second": 2.924,
8
+ "train_steps_per_second": 0.045
9
+ }
trainer_state.json ADDED
@@ -0,0 +1,1176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 107.5957260131836,
14
+ "learning_rate": 7.8125e-06,
15
+ "log_odds_chosen": 0.26999545097351074,
16
+ "log_odds_ratio": -0.6788457632064819,
17
+ "logits/chosen": -2.960033655166626,
18
+ "logits/rejected": -2.8769211769104004,
19
+ "logps/chosen": -1.1197593212127686,
20
+ "logps/rejected": -1.3242356777191162,
21
+ "loss": 47.1372,
22
+ "nll_loss": 1.4214212894439697,
23
+ "rewards/accuracies": 0.574999988079071,
24
+ "rewards/chosen": -0.11197594553232193,
25
+ "rewards/margins": 0.020447632297873497,
26
+ "rewards/rejected": -0.13242356479167938,
27
+ "step": 5
28
+ },
29
+ {
30
+ "epoch": 0.09481481481481481,
31
+ "grad_norm": 59.963645935058594,
32
+ "learning_rate": 1.5625e-05,
33
+ "log_odds_chosen": 0.17921434342861176,
34
+ "log_odds_ratio": -0.6778210401535034,
35
+ "logits/chosen": -2.9659476280212402,
36
+ "logits/rejected": -2.8464667797088623,
37
+ "logps/chosen": -0.9941017031669617,
38
+ "logps/rejected": -1.1148567199707031,
39
+ "loss": 41.0221,
40
+ "nll_loss": 1.266223669052124,
41
+ "rewards/accuracies": 0.6000000238418579,
42
+ "rewards/chosen": -0.09941016137599945,
43
+ "rewards/margins": 0.012075509876012802,
44
+ "rewards/rejected": -0.11148568242788315,
45
+ "step": 10
46
+ },
47
+ {
48
+ "epoch": 0.14222222222222222,
49
+ "grad_norm": 54.22554016113281,
50
+ "learning_rate": 2.34375e-05,
51
+ "log_odds_chosen": 0.19334463775157928,
52
+ "log_odds_ratio": -0.6705215573310852,
53
+ "logits/chosen": -2.8042972087860107,
54
+ "logits/rejected": -2.72182035446167,
55
+ "logps/chosen": -0.7811649441719055,
56
+ "logps/rejected": -0.8896393775939941,
57
+ "loss": 38.9195,
58
+ "nll_loss": 1.2349504232406616,
59
+ "rewards/accuracies": 0.581250011920929,
60
+ "rewards/chosen": -0.07811649888753891,
61
+ "rewards/margins": 0.01084743719547987,
62
+ "rewards/rejected": -0.08896392583847046,
63
+ "step": 15
64
+ },
65
+ {
66
+ "epoch": 0.18962962962962962,
67
+ "grad_norm": 46.67384338378906,
68
+ "learning_rate": 3.125e-05,
69
+ "log_odds_chosen": 0.16569766402244568,
70
+ "log_odds_ratio": -0.6850992441177368,
71
+ "logits/chosen": -2.8097779750823975,
72
+ "logits/rejected": -2.758016347885132,
73
+ "logps/chosen": -0.8054255247116089,
74
+ "logps/rejected": -0.9155207872390747,
75
+ "loss": 36.5441,
76
+ "nll_loss": 1.1204253435134888,
77
+ "rewards/accuracies": 0.581250011920929,
78
+ "rewards/chosen": -0.08054256439208984,
79
+ "rewards/margins": 0.011009521782398224,
80
+ "rewards/rejected": -0.09155207872390747,
81
+ "step": 20
82
+ },
83
+ {
84
+ "epoch": 0.23703703703703705,
85
+ "grad_norm": 32.77006530761719,
86
+ "learning_rate": 3.90625e-05,
87
+ "log_odds_chosen": 0.11824393272399902,
88
+ "log_odds_ratio": -0.7369982004165649,
89
+ "logits/chosen": -2.8846487998962402,
90
+ "logits/rejected": -2.8371191024780273,
91
+ "logps/chosen": -0.7971025705337524,
92
+ "logps/rejected": -0.8806389570236206,
93
+ "loss": 35.5873,
94
+ "nll_loss": 1.1432932615280151,
95
+ "rewards/accuracies": 0.518750011920929,
96
+ "rewards/chosen": -0.07971025258302689,
97
+ "rewards/margins": 0.008353643119335175,
98
+ "rewards/rejected": -0.08806388825178146,
99
+ "step": 25
100
+ },
101
+ {
102
+ "epoch": 0.28444444444444444,
103
+ "grad_norm": 44.49482345581055,
104
+ "learning_rate": 4.6875e-05,
105
+ "log_odds_chosen": 0.11238924413919449,
106
+ "log_odds_ratio": -0.7262482643127441,
107
+ "logits/chosen": -2.8552589416503906,
108
+ "logits/rejected": -2.7313339710235596,
109
+ "logps/chosen": -0.760868489742279,
110
+ "logps/rejected": -0.8262151479721069,
111
+ "loss": 34.6256,
112
+ "nll_loss": 1.0859296321868896,
113
+ "rewards/accuracies": 0.6000000238418579,
114
+ "rewards/chosen": -0.0760868489742279,
115
+ "rewards/margins": 0.006534676067531109,
116
+ "rewards/rejected": -0.08262152969837189,
117
+ "step": 30
118
+ },
119
+ {
120
+ "epoch": 0.33185185185185184,
121
+ "grad_norm": 35.2032356262207,
122
+ "learning_rate": 4.998613757348784e-05,
123
+ "log_odds_chosen": 0.11870062351226807,
124
+ "log_odds_ratio": -0.7236028909683228,
125
+ "logits/chosen": -2.8732895851135254,
126
+ "logits/rejected": -2.9246368408203125,
127
+ "logps/chosen": -0.7803806066513062,
128
+ "logps/rejected": -0.8395193219184875,
129
+ "loss": 33.8984,
130
+ "nll_loss": 1.0961161851882935,
131
+ "rewards/accuracies": 0.574999988079071,
132
+ "rewards/chosen": -0.0780380591750145,
133
+ "rewards/margins": 0.005913878325372934,
134
+ "rewards/rejected": -0.083951935172081,
135
+ "step": 35
136
+ },
137
+ {
138
+ "epoch": 0.37925925925925924,
139
+ "grad_norm": 31.074262619018555,
140
+ "learning_rate": 4.990147841143462e-05,
141
+ "log_odds_chosen": 0.16566646099090576,
142
+ "log_odds_ratio": -0.6969919204711914,
143
+ "logits/chosen": -2.8940587043762207,
144
+ "logits/rejected": -2.846470832824707,
145
+ "logps/chosen": -0.7750462293624878,
146
+ "logps/rejected": -0.8606101870536804,
147
+ "loss": 33.9215,
148
+ "nll_loss": 1.0579229593276978,
149
+ "rewards/accuracies": 0.5687500238418579,
150
+ "rewards/chosen": -0.07750462740659714,
151
+ "rewards/margins": 0.008556395769119263,
152
+ "rewards/rejected": -0.086061030626297,
153
+ "step": 40
154
+ },
155
+ {
156
+ "epoch": 0.4266666666666667,
157
+ "grad_norm": 26.24691390991211,
158
+ "learning_rate": 4.97401218720448e-05,
159
+ "log_odds_chosen": 0.15471526980400085,
160
+ "log_odds_ratio": -0.724504828453064,
161
+ "logits/chosen": -2.8084166049957275,
162
+ "logits/rejected": -2.8175244331359863,
163
+ "logps/chosen": -0.7267907857894897,
164
+ "logps/rejected": -0.8097001314163208,
165
+ "loss": 33.6756,
166
+ "nll_loss": 1.0580939054489136,
167
+ "rewards/accuracies": 0.512499988079071,
168
+ "rewards/chosen": -0.07267908751964569,
169
+ "rewards/margins": 0.00829092413187027,
170
+ "rewards/rejected": -0.08097001165151596,
171
+ "step": 45
172
+ },
173
+ {
174
+ "epoch": 0.4740740740740741,
175
+ "grad_norm": 33.051918029785156,
176
+ "learning_rate": 4.9502564938797946e-05,
177
+ "log_odds_chosen": 0.11409230530261993,
178
+ "log_odds_ratio": -0.7364336848258972,
179
+ "logits/chosen": -2.795748472213745,
180
+ "logits/rejected": -2.805166006088257,
181
+ "logps/chosen": -0.7951005697250366,
182
+ "logps/rejected": -0.8444055318832397,
183
+ "loss": 35.2252,
184
+ "nll_loss": 1.1173629760742188,
185
+ "rewards/accuracies": 0.53125,
186
+ "rewards/chosen": -0.07951006293296814,
187
+ "rewards/margins": 0.004930487833917141,
188
+ "rewards/rejected": -0.08444055169820786,
189
+ "step": 50
190
+ },
191
+ {
192
+ "epoch": 0.5214814814814814,
193
+ "grad_norm": 25.374826431274414,
194
+ "learning_rate": 4.918953929490768e-05,
195
+ "log_odds_chosen": 0.09201577305793762,
196
+ "log_odds_ratio": -0.7416759729385376,
197
+ "logits/chosen": -2.7740275859832764,
198
+ "logits/rejected": -2.732271194458008,
199
+ "logps/chosen": -0.7545334100723267,
200
+ "logps/rejected": -0.8210805058479309,
201
+ "loss": 32.9599,
202
+ "nll_loss": 1.0351916551589966,
203
+ "rewards/accuracies": 0.550000011920929,
204
+ "rewards/chosen": -0.07545334100723267,
205
+ "rewards/margins": 0.006654707249253988,
206
+ "rewards/rejected": -0.0821080431342125,
207
+ "step": 55
208
+ },
209
+ {
210
+ "epoch": 0.5688888888888889,
211
+ "grad_norm": 27.035547256469727,
212
+ "learning_rate": 4.88020090697132e-05,
213
+ "log_odds_chosen": 0.17979878187179565,
214
+ "log_odds_ratio": -0.6883889436721802,
215
+ "logits/chosen": -2.917654037475586,
216
+ "logits/rejected": -2.8189804553985596,
217
+ "logps/chosen": -0.7465445399284363,
218
+ "logps/rejected": -0.8283771276473999,
219
+ "loss": 32.1903,
220
+ "nll_loss": 1.0064656734466553,
221
+ "rewards/accuracies": 0.5562499761581421,
222
+ "rewards/chosen": -0.0746544599533081,
223
+ "rewards/margins": 0.008183250203728676,
224
+ "rewards/rejected": -0.08283771574497223,
225
+ "step": 60
226
+ },
227
+ {
228
+ "epoch": 0.6162962962962963,
229
+ "grad_norm": 29.824989318847656,
230
+ "learning_rate": 4.834116786912897e-05,
231
+ "log_odds_chosen": 0.14272919297218323,
232
+ "log_odds_ratio": -0.7043147087097168,
233
+ "logits/chosen": -2.823390007019043,
234
+ "logits/rejected": -2.815788984298706,
235
+ "logps/chosen": -0.7772132754325867,
236
+ "logps/rejected": -0.8756623268127441,
237
+ "loss": 32.3031,
238
+ "nll_loss": 1.006659984588623,
239
+ "rewards/accuracies": 0.5687500238418579,
240
+ "rewards/chosen": -0.07772133499383926,
241
+ "rewards/margins": 0.00984489731490612,
242
+ "rewards/rejected": -0.08756622672080994,
243
+ "step": 65
244
+ },
245
+ {
246
+ "epoch": 0.6637037037037037,
247
+ "grad_norm": 25.48737907409668,
248
+ "learning_rate": 4.7808435099299045e-05,
249
+ "log_odds_chosen": 0.10067082941532135,
250
+ "log_odds_ratio": -0.7191606760025024,
251
+ "logits/chosen": -2.91566801071167,
252
+ "logits/rejected": -2.7849342823028564,
253
+ "logps/chosen": -0.7710874676704407,
254
+ "logps/rejected": -0.8389760255813599,
255
+ "loss": 31.6335,
256
+ "nll_loss": 1.0086121559143066,
257
+ "rewards/accuracies": 0.574999988079071,
258
+ "rewards/chosen": -0.07710875570774078,
259
+ "rewards/margins": 0.0067888544872403145,
260
+ "rewards/rejected": -0.08389760553836823,
261
+ "step": 70
262
+ },
263
+ {
264
+ "epoch": 0.7111111111111111,
265
+ "grad_norm": 26.371871948242188,
266
+ "learning_rate": 4.720545159477922e-05,
267
+ "log_odds_chosen": 0.32483047246932983,
268
+ "log_odds_ratio": -0.6155678033828735,
269
+ "logits/chosen": -2.8636772632598877,
270
+ "logits/rejected": -2.8909075260162354,
271
+ "logps/chosen": -0.692024827003479,
272
+ "logps/rejected": -0.8595916628837585,
273
+ "loss": 32.0981,
274
+ "nll_loss": 1.0088964700698853,
275
+ "rewards/accuracies": 0.6812499761581421,
276
+ "rewards/chosen": -0.0692024827003479,
277
+ "rewards/margins": 0.016756689175963402,
278
+ "rewards/rejected": -0.08595915883779526,
279
+ "step": 75
280
+ },
281
+ {
282
+ "epoch": 0.7585185185185185,
283
+ "grad_norm": 25.50819206237793,
284
+ "learning_rate": 4.653407456471222e-05,
285
+ "log_odds_chosen": 0.19090376794338226,
286
+ "log_odds_ratio": -0.6875820159912109,
287
+ "logits/chosen": -2.970930814743042,
288
+ "logits/rejected": -2.879568338394165,
289
+ "logps/chosen": -0.784750759601593,
290
+ "logps/rejected": -0.8799096941947937,
291
+ "loss": 32.0035,
292
+ "nll_loss": 1.0407371520996094,
293
+ "rewards/accuracies": 0.637499988079071,
294
+ "rewards/chosen": -0.07847508043050766,
295
+ "rewards/margins": 0.009515885263681412,
296
+ "rewards/rejected": -0.08799097687005997,
297
+ "step": 80
298
+ },
299
+ {
300
+ "epoch": 0.8059259259259259,
301
+ "grad_norm": 21.803564071655273,
302
+ "learning_rate": 4.579637187256222e-05,
303
+ "log_odds_chosen": 0.15773186087608337,
304
+ "log_odds_ratio": -0.6912397742271423,
305
+ "logits/chosen": -2.9440784454345703,
306
+ "logits/rejected": -2.7553608417510986,
307
+ "logps/chosen": -0.7397053241729736,
308
+ "logps/rejected": -0.8279611468315125,
309
+ "loss": 31.9112,
310
+ "nll_loss": 1.0011701583862305,
311
+ "rewards/accuracies": 0.581250011920929,
312
+ "rewards/chosen": -0.07397053390741348,
313
+ "rewards/margins": 0.00882558710873127,
314
+ "rewards/rejected": -0.0827961266040802,
315
+ "step": 85
316
+ },
317
+ {
318
+ "epoch": 0.8533333333333334,
319
+ "grad_norm": 23.784290313720703,
320
+ "learning_rate": 4.499461566702685e-05,
321
+ "log_odds_chosen": 0.22803974151611328,
322
+ "log_odds_ratio": -0.6771184206008911,
323
+ "logits/chosen": -2.9111487865448,
324
+ "logits/rejected": -2.8253490924835205,
325
+ "logps/chosen": -0.7062662839889526,
326
+ "logps/rejected": -0.8198903799057007,
327
+ "loss": 32.583,
328
+ "nll_loss": 0.9986779093742371,
329
+ "rewards/accuracies": 0.612500011920929,
330
+ "rewards/chosen": -0.0706266313791275,
331
+ "rewards/margins": 0.011362412944436073,
332
+ "rewards/rejected": -0.08198904246091843,
333
+ "step": 90
334
+ },
335
+ {
336
+ "epoch": 0.9007407407407407,
337
+ "grad_norm": 24.204864501953125,
338
+ "learning_rate": 4.413127538374411e-05,
339
+ "log_odds_chosen": 0.25104185938835144,
340
+ "log_odds_ratio": -0.6683576703071594,
341
+ "logits/chosen": -2.8897457122802734,
342
+ "logits/rejected": -2.7754015922546387,
343
+ "logps/chosen": -0.6870729923248291,
344
+ "logps/rejected": -0.8187867403030396,
345
+ "loss": 31.9808,
346
+ "nll_loss": 0.9533413052558899,
347
+ "rewards/accuracies": 0.6000000238418579,
348
+ "rewards/chosen": -0.06870730221271515,
349
+ "rewards/margins": 0.013171376660466194,
350
+ "rewards/rejected": -0.08187868446111679,
351
+ "step": 95
352
+ },
353
+ {
354
+ "epoch": 0.9481481481481482,
355
+ "grad_norm": 25.084274291992188,
356
+ "learning_rate": 4.320901013934887e-05,
357
+ "log_odds_chosen": 0.054086942225694656,
358
+ "log_odds_ratio": -0.7604089379310608,
359
+ "logits/chosen": -2.862680435180664,
360
+ "logits/rejected": -2.7483227252960205,
361
+ "logps/chosen": -0.7729731798171997,
362
+ "logps/rejected": -0.8076101541519165,
363
+ "loss": 32.02,
364
+ "nll_loss": 1.0369198322296143,
365
+ "rewards/accuracies": 0.53125,
366
+ "rewards/chosen": -0.07729730755090714,
367
+ "rewards/margins": 0.003463698085397482,
368
+ "rewards/rejected": -0.08076101541519165,
369
+ "step": 100
370
+ },
371
+ {
372
+ "epoch": 0.9955555555555555,
373
+ "grad_norm": 24.100692749023438,
374
+ "learning_rate": 4.223066054130568e-05,
375
+ "log_odds_chosen": 0.15314075350761414,
376
+ "log_odds_ratio": -0.70216304063797,
377
+ "logits/chosen": -2.9259018898010254,
378
+ "logits/rejected": -2.785470724105835,
379
+ "logps/chosen": -0.7330957055091858,
380
+ "logps/rejected": -0.851188063621521,
381
+ "loss": 31.8368,
382
+ "nll_loss": 0.9923515319824219,
383
+ "rewards/accuracies": 0.6000000238418579,
384
+ "rewards/chosen": -0.07330957055091858,
385
+ "rewards/margins": 0.011809232644736767,
386
+ "rewards/rejected": -0.08511880785226822,
387
+ "step": 105
388
+ },
389
+ {
390
+ "epoch": 1.037925925925926,
391
+ "grad_norm": 24.946685791015625,
392
+ "learning_rate": 4.1199239938743797e-05,
393
+ "log_odds_chosen": 0.4960819184780121,
394
+ "log_odds_ratio": -0.5754141807556152,
395
+ "logits/chosen": -2.847261667251587,
396
+ "logits/rejected": -2.7735049724578857,
397
+ "logps/chosen": -0.5829983949661255,
398
+ "logps/rejected": -0.8058205246925354,
399
+ "loss": 24.0306,
400
+ "nll_loss": 0.8102088570594788,
401
+ "rewards/accuracies": 0.7132866978645325,
402
+ "rewards/chosen": -0.05829983949661255,
403
+ "rewards/margins": 0.022282211109995842,
404
+ "rewards/rejected": -0.08058205991983414,
405
+ "step": 110
406
+ },
407
+ {
408
+ "epoch": 1.0853333333333333,
409
+ "grad_norm": 26.976266860961914,
410
+ "learning_rate": 4.0117925141242174e-05,
411
+ "log_odds_chosen": 0.654797375202179,
412
+ "log_odds_ratio": -0.5052435994148254,
413
+ "logits/chosen": -2.877814769744873,
414
+ "logits/rejected": -2.680081844329834,
415
+ "logps/chosen": -0.5720535516738892,
416
+ "logps/rejected": -0.8802263140678406,
417
+ "loss": 25.9472,
418
+ "nll_loss": 0.7921805381774902,
419
+ "rewards/accuracies": 0.793749988079071,
420
+ "rewards/chosen": -0.057205356657505035,
421
+ "rewards/margins": 0.03081728145480156,
422
+ "rewards/rejected": -0.0880226343870163,
423
+ "step": 115
424
+ },
425
+ {
426
+ "epoch": 1.1327407407407408,
427
+ "grad_norm": 24.159303665161133,
428
+ "learning_rate": 3.899004663415084e-05,
429
+ "log_odds_chosen": 0.6985219120979309,
430
+ "log_odds_ratio": -0.5000964403152466,
431
+ "logits/chosen": -2.8363966941833496,
432
+ "logits/rejected": -2.77884840965271,
433
+ "logps/chosen": -0.5254793167114258,
434
+ "logps/rejected": -0.8278259038925171,
435
+ "loss": 25.8696,
436
+ "nll_loss": 0.7871843576431274,
437
+ "rewards/accuracies": 0.762499988079071,
438
+ "rewards/chosen": -0.05254793167114258,
439
+ "rewards/margins": 0.03023465909063816,
440
+ "rewards/rejected": -0.08278258144855499,
441
+ "step": 120
442
+ },
443
+ {
444
+ "epoch": 1.1801481481481482,
445
+ "grad_norm": 25.176586151123047,
446
+ "learning_rate": 3.781907832058587e-05,
447
+ "log_odds_chosen": 0.7422643303871155,
448
+ "log_odds_ratio": -0.49143218994140625,
449
+ "logits/chosen": -2.8330469131469727,
450
+ "logits/rejected": -2.794886827468872,
451
+ "logps/chosen": -0.5751325488090515,
452
+ "logps/rejected": -0.9076019525527954,
453
+ "loss": 24.3083,
454
+ "nll_loss": 0.7706891298294067,
455
+ "rewards/accuracies": 0.7875000238418579,
456
+ "rewards/chosen": -0.05751325935125351,
457
+ "rewards/margins": 0.033246930688619614,
458
+ "rewards/rejected": -0.09076018631458282,
459
+ "step": 125
460
+ },
461
+ {
462
+ "epoch": 1.2275555555555555,
463
+ "grad_norm": 24.129467010498047,
464
+ "learning_rate": 3.660862682169282e-05,
465
+ "log_odds_chosen": 0.6279364824295044,
466
+ "log_odds_ratio": -0.52507483959198,
467
+ "logits/chosen": -2.8697004318237305,
468
+ "logits/rejected": -2.7813589572906494,
469
+ "logps/chosen": -0.5734705328941345,
470
+ "logps/rejected": -0.8611089587211609,
471
+ "loss": 24.637,
472
+ "nll_loss": 0.7875975966453552,
473
+ "rewards/accuracies": 0.737500011920929,
474
+ "rewards/chosen": -0.05734705924987793,
475
+ "rewards/margins": 0.028763839974999428,
476
+ "rewards/rejected": -0.0861109048128128,
477
+ "step": 130
478
+ },
479
+ {
480
+ "epoch": 1.274962962962963,
481
+ "grad_norm": 22.866924285888672,
482
+ "learning_rate": 3.5362420368134356e-05,
483
+ "log_odds_chosen": 0.6941056251525879,
484
+ "log_odds_ratio": -0.52723228931427,
485
+ "logits/chosen": -2.845369577407837,
486
+ "logits/rejected": -2.7874202728271484,
487
+ "logps/chosen": -0.508094310760498,
488
+ "logps/rejected": -0.7907238006591797,
489
+ "loss": 24.905,
490
+ "nll_loss": 0.7590480446815491,
491
+ "rewards/accuracies": 0.7250000238418579,
492
+ "rewards/chosen": -0.050809431821107864,
493
+ "rewards/margins": 0.028262948617339134,
494
+ "rewards/rejected": -0.07907238602638245,
495
+ "step": 135
496
+ },
497
+ {
498
+ "epoch": 1.3223703703703704,
499
+ "grad_norm": 26.00486183166504,
500
+ "learning_rate": 3.408429731701635e-05,
501
+ "log_odds_chosen": 0.6756368279457092,
502
+ "log_odds_ratio": -0.5294371247291565,
503
+ "logits/chosen": -2.7922794818878174,
504
+ "logits/rejected": -2.8006014823913574,
505
+ "logps/chosen": -0.6041140556335449,
506
+ "logps/rejected": -0.9101554751396179,
507
+ "loss": 26.2335,
508
+ "nll_loss": 0.8219060897827148,
509
+ "rewards/accuracies": 0.7437499761581421,
510
+ "rewards/chosen": -0.06041140481829643,
511
+ "rewards/margins": 0.03060414455831051,
512
+ "rewards/rejected": -0.09101554751396179,
513
+ "step": 140
514
+ },
515
+ {
516
+ "epoch": 1.3697777777777778,
517
+ "grad_norm": 25.439064025878906,
518
+ "learning_rate": 3.2778194329621104e-05,
519
+ "log_odds_chosen": 0.5306147933006287,
520
+ "log_odds_ratio": -0.5561264157295227,
521
+ "logits/chosen": -2.901402235031128,
522
+ "logits/rejected": -2.760891914367676,
523
+ "logps/chosen": -0.6078349351882935,
524
+ "logps/rejected": -0.8788266181945801,
525
+ "loss": 25.9841,
526
+ "nll_loss": 0.8179038166999817,
527
+ "rewards/accuracies": 0.6812499761581421,
528
+ "rewards/chosen": -0.06078348681330681,
529
+ "rewards/margins": 0.027099166065454483,
530
+ "rewards/rejected": -0.08788266032934189,
531
+ "step": 145
532
+ },
533
+ {
534
+ "epoch": 1.417185185185185,
535
+ "grad_norm": 29.468650817871094,
536
+ "learning_rate": 3.144813424636031e-05,
537
+ "log_odds_chosen": 0.5843684673309326,
538
+ "log_odds_ratio": -0.549953043460846,
539
+ "logits/chosen": -2.8449625968933105,
540
+ "logits/rejected": -2.8639121055603027,
541
+ "logps/chosen": -0.5666799545288086,
542
+ "logps/rejected": -0.8146460652351379,
543
+ "loss": 25.1356,
544
+ "nll_loss": 0.7769128084182739,
545
+ "rewards/accuracies": 0.6875,
546
+ "rewards/chosen": -0.0566679947078228,
547
+ "rewards/margins": 0.024796610698103905,
548
+ "rewards/rejected": -0.08146460354328156,
549
+ "step": 150
550
+ },
551
+ {
552
+ "epoch": 1.4645925925925927,
553
+ "grad_norm": 25.020612716674805,
554
+ "learning_rate": 3.0098213696293542e-05,
555
+ "log_odds_chosen": 0.6784299612045288,
556
+ "log_odds_ratio": -0.5230305790901184,
557
+ "logits/chosen": -2.895751476287842,
558
+ "logits/rejected": -2.7628910541534424,
559
+ "logps/chosen": -0.5524539351463318,
560
+ "logps/rejected": -0.8418664932250977,
561
+ "loss": 25.5094,
562
+ "nll_loss": 0.8113977313041687,
563
+ "rewards/accuracies": 0.737500011920929,
564
+ "rewards/chosen": -0.055245399475097656,
565
+ "rewards/margins": 0.028941255062818527,
566
+ "rewards/rejected": -0.08418664336204529,
567
+ "step": 155
568
+ },
569
+ {
570
+ "epoch": 1.512,
571
+ "grad_norm": 26.339784622192383,
572
+ "learning_rate": 2.8732590479375165e-05,
573
+ "log_odds_chosen": 0.5665224194526672,
574
+ "log_odds_ratio": -0.5301405191421509,
575
+ "logits/chosen": -2.8651890754699707,
576
+ "logits/rejected": -2.8648974895477295,
577
+ "logps/chosen": -0.628059983253479,
578
+ "logps/rejected": -0.9070053100585938,
579
+ "loss": 26.3028,
580
+ "nll_loss": 0.8419073224067688,
581
+ "rewards/accuracies": 0.7250000238418579,
582
+ "rewards/chosen": -0.06280599534511566,
583
+ "rewards/margins": 0.02789454162120819,
584
+ "rewards/rejected": -0.09070053696632385,
585
+ "step": 160
586
+ },
587
+ {
588
+ "epoch": 1.5594074074074074,
589
+ "grad_norm": 24.364530563354492,
590
+ "learning_rate": 2.7355470760292956e-05,
591
+ "log_odds_chosen": 0.8036432266235352,
592
+ "log_odds_ratio": -0.4671328663825989,
593
+ "logits/chosen": -2.807288408279419,
594
+ "logits/rejected": -2.860252857208252,
595
+ "logps/chosen": -0.5591052174568176,
596
+ "logps/rejected": -0.9000210762023926,
597
+ "loss": 25.9433,
598
+ "nll_loss": 0.7956122756004333,
599
+ "rewards/accuracies": 0.793749988079071,
600
+ "rewards/chosen": -0.05591052025556564,
601
+ "rewards/margins": 0.03409158065915108,
602
+ "rewards/rejected": -0.09000210464000702,
603
+ "step": 165
604
+ },
605
+ {
606
+ "epoch": 1.6068148148148147,
607
+ "grad_norm": 26.519641876220703,
608
+ "learning_rate": 2.597109611334169e-05,
609
+ "log_odds_chosen": 0.7449638843536377,
610
+ "log_odds_ratio": -0.4777608811855316,
611
+ "logits/chosen": -2.812837600708008,
612
+ "logits/rejected": -2.7875418663024902,
613
+ "logps/chosen": -0.5420358180999756,
614
+ "logps/rejected": -0.8753792643547058,
615
+ "loss": 25.1331,
616
+ "nll_loss": 0.7836862802505493,
617
+ "rewards/accuracies": 0.793749988079071,
618
+ "rewards/chosen": -0.054203588515520096,
619
+ "rewards/margins": 0.03333434462547302,
620
+ "rewards/rejected": -0.08753792196512222,
621
+ "step": 170
622
+ },
623
+ {
624
+ "epoch": 1.6542222222222223,
625
+ "grad_norm": 24.914644241333008,
626
+ "learning_rate": 2.458373045823404e-05,
627
+ "log_odds_chosen": 0.6300590634346008,
628
+ "log_odds_ratio": -0.5320771932601929,
629
+ "logits/chosen": -2.8087375164031982,
630
+ "logits/rejected": -2.713311195373535,
631
+ "logps/chosen": -0.5621276497840881,
632
+ "logps/rejected": -0.8645404577255249,
633
+ "loss": 25.1251,
634
+ "nll_loss": 0.8012740015983582,
635
+ "rewards/accuracies": 0.75,
636
+ "rewards/chosen": -0.056212764233350754,
637
+ "rewards/margins": 0.030241286382079124,
638
+ "rewards/rejected": -0.08645405620336533,
639
+ "step": 175
640
+ },
641
+ {
642
+ "epoch": 1.7016296296296296,
643
+ "grad_norm": 24.421783447265625,
644
+ "learning_rate": 2.3197646927086697e-05,
645
+ "log_odds_chosen": 0.511154294013977,
646
+ "log_odds_ratio": -0.58769690990448,
647
+ "logits/chosen": -2.733468770980835,
648
+ "logits/rejected": -2.773117780685425,
649
+ "logps/chosen": -0.5480560064315796,
650
+ "logps/rejected": -0.7600927352905273,
651
+ "loss": 25.3565,
652
+ "nll_loss": 0.7882987260818481,
653
+ "rewards/accuracies": 0.675000011920929,
654
+ "rewards/chosen": -0.054805606603622437,
655
+ "rewards/margins": 0.021203670650720596,
656
+ "rewards/rejected": -0.07600927352905273,
657
+ "step": 180
658
+ },
659
+ {
660
+ "epoch": 1.749037037037037,
661
+ "grad_norm": 24.164962768554688,
662
+ "learning_rate": 2.1817114703032176e-05,
663
+ "log_odds_chosen": 0.6179710030555725,
664
+ "log_odds_ratio": -0.5148657560348511,
665
+ "logits/chosen": -2.845136880874634,
666
+ "logits/rejected": -2.8214142322540283,
667
+ "logps/chosen": -0.5659202337265015,
668
+ "logps/rejected": -0.8389551043510437,
669
+ "loss": 25.83,
670
+ "nll_loss": 0.8041189312934875,
671
+ "rewards/accuracies": 0.762499988079071,
672
+ "rewards/chosen": -0.056592028588056564,
673
+ "rewards/margins": 0.027303475886583328,
674
+ "rewards/rejected": -0.08389550447463989,
675
+ "step": 185
676
+ },
677
+ {
678
+ "epoch": 1.7964444444444445,
679
+ "grad_norm": 24.201217651367188,
680
+ "learning_rate": 2.0446385870993467e-05,
681
+ "log_odds_chosen": 0.7095943689346313,
682
+ "log_odds_ratio": -0.4986342489719391,
683
+ "logits/chosen": -2.864686965942383,
684
+ "logits/rejected": -2.7122364044189453,
685
+ "logps/chosen": -0.5929415225982666,
686
+ "logps/rejected": -0.9149606823921204,
687
+ "loss": 25.6013,
688
+ "nll_loss": 0.7932403087615967,
689
+ "rewards/accuracies": 0.800000011920929,
690
+ "rewards/chosen": -0.05929415300488472,
691
+ "rewards/margins": 0.032201915979385376,
692
+ "rewards/rejected": -0.0914960727095604,
693
+ "step": 190
694
+ },
695
+ {
696
+ "epoch": 1.8438518518518519,
697
+ "grad_norm": 25.97276496887207,
698
+ "learning_rate": 1.9089682321121834e-05,
699
+ "log_odds_chosen": 0.7197151184082031,
700
+ "log_odds_ratio": -0.5103024840354919,
701
+ "logits/chosen": -2.8166863918304443,
702
+ "logits/rejected": -2.6799378395080566,
703
+ "logps/chosen": -0.5752886533737183,
704
+ "logps/rejected": -0.8961560130119324,
705
+ "loss": 25.1918,
706
+ "nll_loss": 0.7882518172264099,
707
+ "rewards/accuracies": 0.768750011920929,
708
+ "rewards/chosen": -0.05752886086702347,
709
+ "rewards/margins": 0.03208674117922783,
710
+ "rewards/rejected": -0.089615598320961,
711
+ "step": 195
712
+ },
713
+ {
714
+ "epoch": 1.8912592592592592,
715
+ "grad_norm": 24.435470581054688,
716
+ "learning_rate": 1.775118274523545e-05,
717
+ "log_odds_chosen": 0.6891900300979614,
718
+ "log_odds_ratio": -0.5204278230667114,
719
+ "logits/chosen": -2.885361909866333,
720
+ "logits/rejected": -2.828627109527588,
721
+ "logps/chosen": -0.5744878053665161,
722
+ "logps/rejected": -0.892475962638855,
723
+ "loss": 26.1889,
724
+ "nll_loss": 0.7850193977355957,
725
+ "rewards/accuracies": 0.7437499761581421,
726
+ "rewards/chosen": -0.05744878575205803,
727
+ "rewards/margins": 0.031798820942640305,
728
+ "rewards/rejected": -0.08924759924411774,
729
+ "step": 200
730
+ },
731
+ {
732
+ "epoch": 1.9386666666666668,
733
+ "grad_norm": 22.373199462890625,
734
+ "learning_rate": 1.643500976631037e-05,
735
+ "log_odds_chosen": 0.554446816444397,
736
+ "log_odds_ratio": -0.5625575184822083,
737
+ "logits/chosen": -2.8687710762023926,
738
+ "logits/rejected": -2.728229284286499,
739
+ "logps/chosen": -0.5928065776824951,
740
+ "logps/rejected": -0.8653701543807983,
741
+ "loss": 25.113,
742
+ "nll_loss": 0.7887372374534607,
743
+ "rewards/accuracies": 0.699999988079071,
744
+ "rewards/chosen": -0.05928065627813339,
745
+ "rewards/margins": 0.027256358414888382,
746
+ "rewards/rejected": -0.08653701841831207,
747
+ "step": 205
748
+ },
749
+ {
750
+ "epoch": 1.986074074074074,
751
+ "grad_norm": 23.288532257080078,
752
+ "learning_rate": 1.514521724066537e-05,
753
+ "log_odds_chosen": 0.6083322763442993,
754
+ "log_odds_ratio": -0.5458379983901978,
755
+ "logits/chosen": -2.854628086090088,
756
+ "logits/rejected": -2.8980631828308105,
757
+ "logps/chosen": -0.5750885605812073,
758
+ "logps/rejected": -0.8365706205368042,
759
+ "loss": 23.8154,
760
+ "nll_loss": 0.7530295252799988,
761
+ "rewards/accuracies": 0.731249988079071,
762
+ "rewards/chosen": -0.05750884860754013,
763
+ "rewards/margins": 0.02614821121096611,
764
+ "rewards/rejected": -0.08365706354379654,
765
+ "step": 210
766
+ },
767
+ {
768
+ "epoch": 2.0284444444444443,
769
+ "grad_norm": 24.61016082763672,
770
+ "learning_rate": 1.3885777771950348e-05,
771
+ "log_odds_chosen": 1.0653361082077026,
772
+ "log_odds_ratio": -0.4205915629863739,
773
+ "logits/chosen": -2.783123254776001,
774
+ "logits/rejected": -2.719882011413574,
775
+ "logps/chosen": -0.4712402820587158,
776
+ "logps/rejected": -0.8792357444763184,
777
+ "loss": 19.6217,
778
+ "nll_loss": 0.6721786856651306,
779
+ "rewards/accuracies": 0.8181818127632141,
780
+ "rewards/chosen": -0.04712402820587158,
781
+ "rewards/margins": 0.040799543261528015,
782
+ "rewards/rejected": -0.08792358636856079,
783
+ "step": 215
784
+ },
785
+ {
786
+ "epoch": 2.075851851851852,
787
+ "grad_norm": 33.874610900878906,
788
+ "learning_rate": 1.2660570475395683e-05,
789
+ "log_odds_chosen": 1.245701789855957,
790
+ "log_odds_ratio": -0.36852705478668213,
791
+ "logits/chosen": -2.7865612506866455,
792
+ "logits/rejected": -2.7448060512542725,
793
+ "logps/chosen": -0.4370403289794922,
794
+ "logps/rejected": -0.9250956773757935,
795
+ "loss": 19.9741,
796
+ "nll_loss": 0.6481715440750122,
797
+ "rewards/accuracies": 0.84375,
798
+ "rewards/chosen": -0.04370403289794922,
799
+ "rewards/margins": 0.04880553111433983,
800
+ "rewards/rejected": -0.09250956773757935,
801
+ "step": 220
802
+ },
803
+ {
804
+ "epoch": 2.1232592592592594,
805
+ "grad_norm": 26.486011505126953,
806
+ "learning_rate": 1.1473369030008974e-05,
807
+ "log_odds_chosen": 1.2085529565811157,
808
+ "log_odds_ratio": -0.38652700185775757,
809
+ "logits/chosen": -2.754307270050049,
810
+ "logits/rejected": -2.792224168777466,
811
+ "logps/chosen": -0.4022933840751648,
812
+ "logps/rejected": -0.8532126545906067,
813
+ "loss": 20.6782,
814
+ "nll_loss": 0.6350698471069336,
815
+ "rewards/accuracies": 0.8500000238418579,
816
+ "rewards/chosen": -0.04022934287786484,
817
+ "rewards/margins": 0.045091934502124786,
818
+ "rewards/rejected": -0.08532126247882843,
819
+ "step": 225
820
+ },
821
+ {
822
+ "epoch": 2.1706666666666665,
823
+ "grad_norm": 26.554258346557617,
824
+ "learning_rate": 1.0327830055518842e-05,
825
+ "log_odds_chosen": 1.3051682710647583,
826
+ "log_odds_ratio": -0.3686615526676178,
827
+ "logits/chosen": -2.7639107704162598,
828
+ "logits/rejected": -2.693751811981201,
829
+ "logps/chosen": -0.41990265250205994,
830
+ "logps/rejected": -0.8988162279129028,
831
+ "loss": 19.327,
832
+ "nll_loss": 0.5870728492736816,
833
+ "rewards/accuracies": 0.8500000238418579,
834
+ "rewards/chosen": -0.04199026897549629,
835
+ "rewards/margins": 0.04789134860038757,
836
+ "rewards/rejected": -0.08988161385059357,
837
+ "step": 230
838
+ },
839
+ {
840
+ "epoch": 2.218074074074074,
841
+ "grad_norm": 31.239700317382812,
842
+ "learning_rate": 9.227481849865235e-06,
843
+ "log_odds_chosen": 1.1520278453826904,
844
+ "log_odds_ratio": -0.41394537687301636,
845
+ "logits/chosen": -2.778120756149292,
846
+ "logits/rejected": -2.7755844593048096,
847
+ "logps/chosen": -0.4497902989387512,
848
+ "logps/rejected": -0.8696093559265137,
849
+ "loss": 20.2714,
850
+ "nll_loss": 0.6394407153129578,
851
+ "rewards/accuracies": 0.8374999761581421,
852
+ "rewards/chosen": -0.044979028403759,
853
+ "rewards/margins": 0.04198190197348595,
854
+ "rewards/rejected": -0.08696093410253525,
855
+ "step": 235
856
+ },
857
+ {
858
+ "epoch": 2.2654814814814817,
859
+ "grad_norm": 26.99662971496582,
860
+ "learning_rate": 8.175713521924978e-06,
861
+ "log_odds_chosen": 1.287632703781128,
862
+ "log_odds_ratio": -0.37112313508987427,
863
+ "logits/chosen": -2.7816336154937744,
864
+ "logits/rejected": -2.8271846771240234,
865
+ "logps/chosen": -0.43638721108436584,
866
+ "logps/rejected": -0.9052692651748657,
867
+ "loss": 20.5114,
868
+ "nll_loss": 0.6593940854072571,
869
+ "rewards/accuracies": 0.862500011920929,
870
+ "rewards/chosen": -0.04363872855901718,
871
+ "rewards/margins": 0.046888213604688644,
872
+ "rewards/rejected": -0.09052693098783493,
873
+ "step": 240
874
+ },
875
+ {
876
+ "epoch": 2.3128888888888888,
877
+ "grad_norm": 27.01542854309082,
878
+ "learning_rate": 7.1757645529443665e-06,
879
+ "log_odds_chosen": 1.3749712705612183,
880
+ "log_odds_ratio": -0.3259935975074768,
881
+ "logits/chosen": -2.788048028945923,
882
+ "logits/rejected": -2.701140880584717,
883
+ "logps/chosen": -0.4096066951751709,
884
+ "logps/rejected": -0.91987144947052,
885
+ "loss": 19.5069,
886
+ "nll_loss": 0.62156742811203,
887
+ "rewards/accuracies": 0.8999999761581421,
888
+ "rewards/chosen": -0.04096066951751709,
889
+ "rewards/margins": 0.05102648213505745,
890
+ "rewards/rejected": -0.09198714792728424,
891
+ "step": 245
892
+ },
893
+ {
894
+ "epoch": 2.3602962962962963,
895
+ "grad_norm": 27.530044555664062,
896
+ "learning_rate": 6.230714818829733e-06,
897
+ "log_odds_chosen": 1.3079187870025635,
898
+ "log_odds_ratio": -0.33900976181030273,
899
+ "logits/chosen": -2.7988991737365723,
900
+ "logits/rejected": -2.7709498405456543,
901
+ "logps/chosen": -0.4449501931667328,
902
+ "logps/rejected": -0.9620338678359985,
903
+ "loss": 19.171,
904
+ "nll_loss": 0.6168783903121948,
905
+ "rewards/accuracies": 0.8812500238418579,
906
+ "rewards/chosen": -0.04449502006173134,
907
+ "rewards/margins": 0.05170837789773941,
908
+ "rewards/rejected": -0.09620338678359985,
909
+ "step": 250
910
+ },
911
+ {
912
+ "epoch": 2.407703703703704,
913
+ "grad_norm": 28.68503189086914,
914
+ "learning_rate": 5.343475104027743e-06,
915
+ "log_odds_chosen": 1.423537254333496,
916
+ "log_odds_ratio": -0.328805148601532,
917
+ "logits/chosen": -2.7419135570526123,
918
+ "logits/rejected": -2.7263638973236084,
919
+ "logps/chosen": -0.3865131735801697,
920
+ "logps/rejected": -0.9120645523071289,
921
+ "loss": 19.2,
922
+ "nll_loss": 0.6022146344184875,
923
+ "rewards/accuracies": 0.887499988079071,
924
+ "rewards/chosen": -0.03865131735801697,
925
+ "rewards/margins": 0.0525551438331604,
926
+ "rewards/rejected": -0.09120644629001617,
927
+ "step": 255
928
+ },
929
+ {
930
+ "epoch": 2.455111111111111,
931
+ "grad_norm": 26.74730110168457,
932
+ "learning_rate": 4.516778136213037e-06,
933
+ "log_odds_chosen": 1.4979491233825684,
934
+ "log_odds_ratio": -0.3211529850959778,
935
+ "logits/chosen": -2.7689156532287598,
936
+ "logits/rejected": -2.6690585613250732,
937
+ "logps/chosen": -0.3924681544303894,
938
+ "logps/rejected": -0.9140318632125854,
939
+ "loss": 19.3033,
940
+ "nll_loss": 0.594567060470581,
941
+ "rewards/accuracies": 0.875,
942
+ "rewards/chosen": -0.039246816188097,
943
+ "rewards/margins": 0.052156370133161545,
944
+ "rewards/rejected": -0.09140319377183914,
945
+ "step": 260
946
+ },
947
+ {
948
+ "epoch": 2.5025185185185186,
949
+ "grad_norm": 28.366756439208984,
950
+ "learning_rate": 3.7531701693965554e-06,
951
+ "log_odds_chosen": 1.323599100112915,
952
+ "log_odds_ratio": -0.37130218744277954,
953
+ "logits/chosen": -2.7441766262054443,
954
+ "logits/rejected": -2.7618346214294434,
955
+ "logps/chosen": -0.4224226474761963,
956
+ "logps/rejected": -0.9127100110054016,
957
+ "loss": 19.0019,
958
+ "nll_loss": 0.5947035551071167,
959
+ "rewards/accuracies": 0.8687499761581421,
960
+ "rewards/chosen": -0.04224226251244545,
961
+ "rewards/margins": 0.049028728157281876,
962
+ "rewards/rejected": -0.09127099066972733,
963
+ "step": 265
964
+ },
965
+ {
966
+ "epoch": 2.549925925925926,
967
+ "grad_norm": 27.086530685424805,
968
+ "learning_rate": 3.055003141378948e-06,
969
+ "log_odds_chosen": 1.311583399772644,
970
+ "log_odds_ratio": -0.3499976694583893,
971
+ "logits/chosen": -2.8330342769622803,
972
+ "logits/rejected": -2.7891921997070312,
973
+ "logps/chosen": -0.43048733472824097,
974
+ "logps/rejected": -0.9034746885299683,
975
+ "loss": 19.0972,
976
+ "nll_loss": 0.6235347986221313,
977
+ "rewards/accuracies": 0.8812500238418579,
978
+ "rewards/chosen": -0.043048735707998276,
979
+ "rewards/margins": 0.04729873687028885,
980
+ "rewards/rejected": -0.09034748375415802,
981
+ "step": 270
982
+ },
983
+ {
984
+ "epoch": 2.5973333333333333,
985
+ "grad_norm": 26.073270797729492,
986
+ "learning_rate": 2.424427429704365e-06,
987
+ "log_odds_chosen": 1.382707118988037,
988
+ "log_odds_ratio": -0.3385299742221832,
989
+ "logits/chosen": -2.8038322925567627,
990
+ "logits/rejected": -2.691887378692627,
991
+ "logps/chosen": -0.43798762559890747,
992
+ "logps/rejected": -0.9750789403915405,
993
+ "loss": 20.113,
994
+ "nll_loss": 0.6334654688835144,
995
+ "rewards/accuracies": 0.8687499761581421,
996
+ "rewards/chosen": -0.04379875957965851,
997
+ "rewards/margins": 0.05370912700891495,
998
+ "rewards/rejected": -0.09750787913799286,
999
+ "step": 275
1000
+ },
1001
+ {
1002
+ "epoch": 2.644740740740741,
1003
+ "grad_norm": 28.34626007080078,
1004
+ "learning_rate": 1.8633852284264508e-06,
1005
+ "log_odds_chosen": 1.2504308223724365,
1006
+ "log_odds_ratio": -0.367613822221756,
1007
+ "logits/chosen": -2.7782769203186035,
1008
+ "logits/rejected": -2.7843925952911377,
1009
+ "logps/chosen": -0.4142681062221527,
1010
+ "logps/rejected": -0.8895042538642883,
1011
+ "loss": 19.2085,
1012
+ "nll_loss": 0.591576874256134,
1013
+ "rewards/accuracies": 0.887499988079071,
1014
+ "rewards/chosen": -0.04142681136727333,
1015
+ "rewards/margins": 0.047523610293865204,
1016
+ "rewards/rejected": -0.08895042538642883,
1017
+ "step": 280
1018
+ },
1019
+ {
1020
+ "epoch": 2.6921481481481484,
1021
+ "grad_norm": 30.690038681030273,
1022
+ "learning_rate": 1.3736045660864034e-06,
1023
+ "log_odds_chosen": 1.2426550388336182,
1024
+ "log_odds_ratio": -0.36959585547447205,
1025
+ "logits/chosen": -2.8215603828430176,
1026
+ "logits/rejected": -2.7462406158447266,
1027
+ "logps/chosen": -0.4342527389526367,
1028
+ "logps/rejected": -0.8976672887802124,
1029
+ "loss": 19.7584,
1030
+ "nll_loss": 0.6197787523269653,
1031
+ "rewards/accuracies": 0.8374999761581421,
1032
+ "rewards/chosen": -0.04342527687549591,
1033
+ "rewards/margins": 0.04634145647287369,
1034
+ "rewards/rejected": -0.0897667333483696,
1035
+ "step": 285
1036
+ },
1037
+ {
1038
+ "epoch": 2.7395555555555555,
1039
+ "grad_norm": 28.228126525878906,
1040
+ "learning_rate": 9.565939833279192e-07,
1041
+ "log_odds_chosen": 1.3652734756469727,
1042
+ "log_odds_ratio": -0.3423106074333191,
1043
+ "logits/chosen": -2.7341599464416504,
1044
+ "logits/rejected": -2.7509799003601074,
1045
+ "logps/chosen": -0.38983267545700073,
1046
+ "logps/rejected": -0.8893811106681824,
1047
+ "loss": 19.743,
1048
+ "nll_loss": 0.6161855459213257,
1049
+ "rewards/accuracies": 0.8687499761581421,
1050
+ "rewards/chosen": -0.038983266800642014,
1051
+ "rewards/margins": 0.0499548502266407,
1052
+ "rewards/rejected": -0.08893812447786331,
1053
+ "step": 290
1054
+ },
1055
+ {
1056
+ "epoch": 2.786962962962963,
1057
+ "grad_norm": 27.338224411010742,
1058
+ "learning_rate": 6.136378865420872e-07,
1059
+ "log_odds_chosen": 1.2006913423538208,
1060
+ "log_odds_ratio": -0.3960389494895935,
1061
+ "logits/chosen": -2.686972141265869,
1062
+ "logits/rejected": -2.7599165439605713,
1063
+ "logps/chosen": -0.45548802614212036,
1064
+ "logps/rejected": -0.915076732635498,
1065
+ "loss": 19.6585,
1066
+ "nll_loss": 0.6340048909187317,
1067
+ "rewards/accuracies": 0.8500000238418579,
1068
+ "rewards/chosen": -0.04554879665374756,
1069
+ "rewards/margins": 0.04595886915922165,
1070
+ "rewards/rejected": -0.09150765836238861,
1071
+ "step": 295
1072
+ },
1073
+ {
1074
+ "epoch": 2.83437037037037,
1075
+ "grad_norm": 29.504423141479492,
1076
+ "learning_rate": 3.45792591853214e-07,
1077
+ "log_odds_chosen": 1.189578890800476,
1078
+ "log_odds_ratio": -0.3797909617424011,
1079
+ "logits/chosen": -2.7965073585510254,
1080
+ "logits/rejected": -2.6686172485351562,
1081
+ "logps/chosen": -0.4425443112850189,
1082
+ "logps/rejected": -0.9239686131477356,
1083
+ "loss": 19.6148,
1084
+ "nll_loss": 0.6315287947654724,
1085
+ "rewards/accuracies": 0.8687499761581421,
1086
+ "rewards/chosen": -0.04425443336367607,
1087
+ "rewards/margins": 0.04814242199063301,
1088
+ "rewards/rejected": -0.09239685535430908,
1089
+ "step": 300
1090
+ },
1091
+ {
1092
+ "epoch": 2.8817777777777778,
1093
+ "grad_norm": 26.830537796020508,
1094
+ "learning_rate": 1.538830716302092e-07,
1095
+ "log_odds_chosen": 1.2787503004074097,
1096
+ "log_odds_ratio": -0.34538131952285767,
1097
+ "logits/chosen": -2.859900951385498,
1098
+ "logits/rejected": -2.7875828742980957,
1099
+ "logps/chosen": -0.458304226398468,
1100
+ "logps/rejected": -0.9439491033554077,
1101
+ "loss": 19.4322,
1102
+ "nll_loss": 0.6039472222328186,
1103
+ "rewards/accuracies": 0.856249988079071,
1104
+ "rewards/chosen": -0.04583042114973068,
1105
+ "rewards/margins": 0.04856449365615845,
1106
+ "rewards/rejected": -0.09439491480588913,
1107
+ "step": 305
1108
+ },
1109
+ {
1110
+ "epoch": 2.9291851851851853,
1111
+ "grad_norm": 28.568008422851562,
1112
+ "learning_rate": 3.8500413544415025e-08,
1113
+ "log_odds_chosen": 1.291078805923462,
1114
+ "log_odds_ratio": -0.36775314807891846,
1115
+ "logits/chosen": -2.760559558868408,
1116
+ "logits/rejected": -2.6593780517578125,
1117
+ "logps/chosen": -0.41886311769485474,
1118
+ "logps/rejected": -0.8871288299560547,
1119
+ "loss": 19.394,
1120
+ "nll_loss": 0.6148114204406738,
1121
+ "rewards/accuracies": 0.8374999761581421,
1122
+ "rewards/chosen": -0.041886307299137115,
1123
+ "rewards/margins": 0.04682657867670059,
1124
+ "rewards/rejected": -0.08871288597583771,
1125
+ "step": 310
1126
+ },
1127
+ {
1128
+ "epoch": 2.9765925925925925,
1129
+ "grad_norm": 26.362131118774414,
1130
+ "learning_rate": 0.0,
1131
+ "log_odds_chosen": 1.38863205909729,
1132
+ "log_odds_ratio": -0.3244430124759674,
1133
+ "logits/chosen": -2.7366228103637695,
1134
+ "logits/rejected": -2.6630187034606934,
1135
+ "logps/chosen": -0.4075457453727722,
1136
+ "logps/rejected": -0.9129207730293274,
1137
+ "loss": 18.6997,
1138
+ "nll_loss": 0.5973876118659973,
1139
+ "rewards/accuracies": 0.90625,
1140
+ "rewards/chosen": -0.04075457900762558,
1141
+ "rewards/margins": 0.05053749680519104,
1142
+ "rewards/rejected": -0.09129206836223602,
1143
+ "step": 315
1144
+ },
1145
+ {
1146
+ "epoch": 2.9765925925925925,
1147
+ "step": 315,
1148
+ "total_flos": 0.0,
1149
+ "train_loss": 26.468634469168528,
1150
+ "train_runtime": 6925.9574,
1151
+ "train_samples_per_second": 2.924,
1152
+ "train_steps_per_second": 0.045
1153
+ }
1154
+ ],
1155
+ "logging_steps": 5,
1156
+ "max_steps": 315,
1157
+ "num_input_tokens_seen": 0,
1158
+ "num_train_epochs": 3,
1159
+ "save_steps": 100000,
1160
+ "stateful_callbacks": {
1161
+ "TrainerControl": {
1162
+ "args": {
1163
+ "should_epoch_stop": false,
1164
+ "should_evaluate": false,
1165
+ "should_log": false,
1166
+ "should_save": true,
1167
+ "should_training_stop": true
1168
+ },
1169
+ "attributes": {}
1170
+ }
1171
+ },
1172
+ "total_flos": 0.0,
1173
+ "train_batch_size": 1,
1174
+ "trial_name": null,
1175
+ "trial_params": null
1176
+ }