silviasapora commited on
Commit
9db08f0
·
verified ·
1 Parent(s): edd2438

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 +1050 -0
README.md ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: google/gemma-7b
3
+ library_name: transformers
4
+ model_name: gemma-7b-simpo-basic-5e-5-v4
5
+ tags:
6
+ - generated_from_trainer
7
+ - trl
8
+ - orpo
9
+ licence: license
10
+ ---
11
+
12
+ # Model Card for gemma-7b-simpo-basic-5e-5-v4
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-simpo-basic-5e-5-v4", 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/zkrd5fnp)
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": 38.28643096439422,
5
+ "train_runtime": 9296.3619,
6
+ "train_samples": 6750,
7
+ "train_samples_per_second": 2.178,
8
+ "train_steps_per_second": 0.034
9
+ }
train_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 2.9765925925925925,
3
+ "total_flos": 0.0,
4
+ "train_loss": 38.28643096439422,
5
+ "train_runtime": 9296.3619,
6
+ "train_samples": 6750,
7
+ "train_samples_per_second": 2.178,
8
+ "train_steps_per_second": 0.034
9
+ }
trainer_state.json ADDED
@@ -0,0 +1,1050 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 616.296142578125,
14
+ "learning_rate": 7.8125e-06,
15
+ "log_odds_chosen": -1.0011694431304932,
16
+ "log_odds_ratio": -10.174823760986328,
17
+ "logps/chosen": -21.74160385131836,
18
+ "logps/rejected": -20.740602493286133,
19
+ "loss": 325.5407,
20
+ "nll_loss": 9.713836669921875,
21
+ "rewards/accuracies": 0.4781250059604645,
22
+ "rewards/chosen": -21.74160385131836,
23
+ "rewards/margins": -1.0010027885437012,
24
+ "rewards/rejected": -20.740602493286133,
25
+ "step": 5
26
+ },
27
+ {
28
+ "epoch": 0.09481481481481481,
29
+ "grad_norm": 1159.8155517578125,
30
+ "learning_rate": 1.5625e-05,
31
+ "log_odds_chosen": -2.6088204383850098,
32
+ "log_odds_ratio": -9.970861434936523,
33
+ "logps/chosen": -21.963115692138672,
34
+ "logps/rejected": -19.354297637939453,
35
+ "loss": 319.0449,
36
+ "nll_loss": 9.407869338989258,
37
+ "rewards/accuracies": 0.4625000059604645,
38
+ "rewards/chosen": -21.963115692138672,
39
+ "rewards/margins": -2.6088180541992188,
40
+ "rewards/rejected": -19.354297637939453,
41
+ "step": 10
42
+ },
43
+ {
44
+ "epoch": 0.14222222222222222,
45
+ "grad_norm": 2860.146240234375,
46
+ "learning_rate": 2.34375e-05,
47
+ "log_odds_chosen": 0.5996896624565125,
48
+ "log_odds_ratio": -9.088822364807129,
49
+ "logps/chosen": -39.47287368774414,
50
+ "logps/rejected": -40.07282638549805,
51
+ "loss": 290.8345,
52
+ "nll_loss": 21.445528030395508,
53
+ "rewards/accuracies": 0.5218750238418579,
54
+ "rewards/chosen": -39.47287368774414,
55
+ "rewards/margins": 0.5999561548233032,
56
+ "rewards/rejected": -40.07282638549805,
57
+ "step": 15
58
+ },
59
+ {
60
+ "epoch": 0.18962962962962962,
61
+ "grad_norm": 7156.2275390625,
62
+ "learning_rate": 3.125e-05,
63
+ "log_odds_chosen": 3.607576370239258,
64
+ "log_odds_ratio": -5.140229225158691,
65
+ "logps/chosen": -105.3345947265625,
66
+ "logps/rejected": -108.9421615600586,
67
+ "loss": 164.4873,
68
+ "nll_loss": 73.30210876464844,
69
+ "rewards/accuracies": 0.6656249761581421,
70
+ "rewards/chosen": -105.3345947265625,
71
+ "rewards/margins": 3.607576370239258,
72
+ "rewards/rejected": -108.9421615600586,
73
+ "step": 20
74
+ },
75
+ {
76
+ "epoch": 0.23703703703703705,
77
+ "grad_norm": 2067.25927734375,
78
+ "learning_rate": 3.90625e-05,
79
+ "log_odds_chosen": 3.95326566696167,
80
+ "log_odds_ratio": -2.9588239192962646,
81
+ "logps/chosen": -186.4034423828125,
82
+ "logps/rejected": -190.35671997070312,
83
+ "loss": 94.6824,
84
+ "nll_loss": 151.7024688720703,
85
+ "rewards/accuracies": 0.65625,
86
+ "rewards/chosen": -186.4034423828125,
87
+ "rewards/margins": 3.95326566696167,
88
+ "rewards/rejected": -190.35671997070312,
89
+ "step": 25
90
+ },
91
+ {
92
+ "epoch": 0.28444444444444444,
93
+ "grad_norm": 3878.92724609375,
94
+ "learning_rate": 4.6875e-05,
95
+ "log_odds_chosen": 1.2030792236328125,
96
+ "log_odds_ratio": -1.431229591369629,
97
+ "logps/chosen": -186.4683837890625,
98
+ "logps/rejected": -187.6714630126953,
99
+ "loss": 45.7994,
100
+ "nll_loss": 158.09597778320312,
101
+ "rewards/accuracies": 0.628125011920929,
102
+ "rewards/chosen": -186.4683837890625,
103
+ "rewards/margins": 1.2030792236328125,
104
+ "rewards/rejected": -187.6714630126953,
105
+ "step": 30
106
+ },
107
+ {
108
+ "epoch": 0.33185185185185184,
109
+ "grad_norm": 1238.4620361328125,
110
+ "learning_rate": 4.998613757348784e-05,
111
+ "log_odds_chosen": 1.815678358078003,
112
+ "log_odds_ratio": -2.412240505218506,
113
+ "logps/chosen": -143.62953186035156,
114
+ "logps/rejected": -145.4451904296875,
115
+ "loss": 77.1917,
116
+ "nll_loss": 110.61494445800781,
117
+ "rewards/accuracies": 0.606249988079071,
118
+ "rewards/chosen": -143.62953186035156,
119
+ "rewards/margins": 1.815678358078003,
120
+ "rewards/rejected": -145.4451904296875,
121
+ "step": 35
122
+ },
123
+ {
124
+ "epoch": 0.37925925925925924,
125
+ "grad_norm": 4717.11474609375,
126
+ "learning_rate": 4.990147841143462e-05,
127
+ "log_odds_chosen": 1.1096925735473633,
128
+ "log_odds_ratio": -1.2363064289093018,
129
+ "logps/chosen": -129.36184692382812,
130
+ "logps/rejected": -130.47154235839844,
131
+ "loss": 39.5618,
132
+ "nll_loss": 100.73094940185547,
133
+ "rewards/accuracies": 0.6156250238418579,
134
+ "rewards/chosen": -129.36184692382812,
135
+ "rewards/margins": 1.1096925735473633,
136
+ "rewards/rejected": -130.47154235839844,
137
+ "step": 40
138
+ },
139
+ {
140
+ "epoch": 0.4266666666666667,
141
+ "grad_norm": 4185.7666015625,
142
+ "learning_rate": 4.97401218720448e-05,
143
+ "log_odds_chosen": 2.159066915512085,
144
+ "log_odds_ratio": -1.3062574863433838,
145
+ "logps/chosen": -150.41696166992188,
146
+ "logps/rejected": -152.57603454589844,
147
+ "loss": 41.8002,
148
+ "nll_loss": 120.55268859863281,
149
+ "rewards/accuracies": 0.671875,
150
+ "rewards/chosen": -150.41696166992188,
151
+ "rewards/margins": 2.159066915512085,
152
+ "rewards/rejected": -152.57603454589844,
153
+ "step": 45
154
+ },
155
+ {
156
+ "epoch": 0.4740740740740741,
157
+ "grad_norm": 3920.476318359375,
158
+ "learning_rate": 4.9502564938797946e-05,
159
+ "log_odds_chosen": 0.8013877868652344,
160
+ "log_odds_ratio": -1.4890391826629639,
161
+ "logps/chosen": -151.86148071289062,
162
+ "logps/rejected": -152.66287231445312,
163
+ "loss": 47.6492,
164
+ "nll_loss": 136.52732849121094,
165
+ "rewards/accuracies": 0.581250011920929,
166
+ "rewards/chosen": -151.86148071289062,
167
+ "rewards/margins": 0.8013877868652344,
168
+ "rewards/rejected": -152.66287231445312,
169
+ "step": 50
170
+ },
171
+ {
172
+ "epoch": 0.5214814814814814,
173
+ "grad_norm": 346.2166442871094,
174
+ "learning_rate": 4.918953929490768e-05,
175
+ "log_odds_chosen": 0.22884111106395721,
176
+ "log_odds_ratio": -1.2772390842437744,
177
+ "logps/chosen": -142.4754180908203,
178
+ "logps/rejected": -142.70425415039062,
179
+ "loss": 40.8717,
180
+ "nll_loss": 129.07229614257812,
181
+ "rewards/accuracies": 0.59375,
182
+ "rewards/chosen": -142.4754180908203,
183
+ "rewards/margins": 0.22884111106395721,
184
+ "rewards/rejected": -142.70425415039062,
185
+ "step": 55
186
+ },
187
+ {
188
+ "epoch": 0.5688888888888889,
189
+ "grad_norm": 168.2239990234375,
190
+ "learning_rate": 4.88020090697132e-05,
191
+ "log_odds_chosen": 0.3460945785045624,
192
+ "log_odds_ratio": -1.070421576499939,
193
+ "logps/chosen": -115.83624267578125,
194
+ "logps/rejected": -116.18232727050781,
195
+ "loss": 34.2535,
196
+ "nll_loss": 104.88349914550781,
197
+ "rewards/accuracies": 0.578125,
198
+ "rewards/chosen": -115.83624267578125,
199
+ "rewards/margins": 0.3460945785045624,
200
+ "rewards/rejected": -116.18232727050781,
201
+ "step": 60
202
+ },
203
+ {
204
+ "epoch": 0.6162962962962963,
205
+ "grad_norm": 612.6487426757812,
206
+ "learning_rate": 4.834116786912897e-05,
207
+ "log_odds_chosen": 0.39803346991539,
208
+ "log_odds_ratio": -1.389655590057373,
209
+ "logps/chosen": -152.57472229003906,
210
+ "logps/rejected": -152.97276306152344,
211
+ "loss": 44.469,
212
+ "nll_loss": 142.9998016357422,
213
+ "rewards/accuracies": 0.590624988079071,
214
+ "rewards/chosen": -152.57472229003906,
215
+ "rewards/margins": 0.39803346991539,
216
+ "rewards/rejected": -152.97276306152344,
217
+ "step": 65
218
+ },
219
+ {
220
+ "epoch": 0.6637037037037037,
221
+ "grad_norm": 277.798583984375,
222
+ "learning_rate": 4.7808435099299045e-05,
223
+ "log_odds_chosen": 0.5616430044174194,
224
+ "log_odds_ratio": -0.66581791639328,
225
+ "logps/chosen": -134.55548095703125,
226
+ "logps/rejected": -135.1171417236328,
227
+ "loss": 21.3062,
228
+ "nll_loss": 128.04086303710938,
229
+ "rewards/accuracies": 0.659375011920929,
230
+ "rewards/chosen": -134.55548095703125,
231
+ "rewards/margins": 0.5616430044174194,
232
+ "rewards/rejected": -135.1171417236328,
233
+ "step": 70
234
+ },
235
+ {
236
+ "epoch": 0.7111111111111111,
237
+ "grad_norm": 325.87396240234375,
238
+ "learning_rate": 4.720545159477922e-05,
239
+ "log_odds_chosen": 0.8940374255180359,
240
+ "log_odds_ratio": -0.6344600915908813,
241
+ "logps/chosen": -130.94586181640625,
242
+ "logps/rejected": -131.8398895263672,
243
+ "loss": 20.3027,
244
+ "nll_loss": 125.71907806396484,
245
+ "rewards/accuracies": 0.6468750238418579,
246
+ "rewards/chosen": -130.94586181640625,
247
+ "rewards/margins": 0.8940374255180359,
248
+ "rewards/rejected": -131.8398895263672,
249
+ "step": 75
250
+ },
251
+ {
252
+ "epoch": 0.7585185185185185,
253
+ "grad_norm": 137.472900390625,
254
+ "learning_rate": 4.653407456471222e-05,
255
+ "log_odds_chosen": 0.40505266189575195,
256
+ "log_odds_ratio": -0.6848157644271851,
257
+ "logps/chosen": -125.63932037353516,
258
+ "logps/rejected": -126.04437255859375,
259
+ "loss": 21.9141,
260
+ "nll_loss": 121.97508239746094,
261
+ "rewards/accuracies": 0.612500011920929,
262
+ "rewards/chosen": -125.63932037353516,
263
+ "rewards/margins": 0.40505266189575195,
264
+ "rewards/rejected": -126.04437255859375,
265
+ "step": 80
266
+ },
267
+ {
268
+ "epoch": 0.8059259259259259,
269
+ "grad_norm": 603.01025390625,
270
+ "learning_rate": 4.579637187256222e-05,
271
+ "log_odds_chosen": 0.24384291470050812,
272
+ "log_odds_ratio": -0.7465418577194214,
273
+ "logps/chosen": -122.43513488769531,
274
+ "logps/rejected": -122.6789779663086,
275
+ "loss": 23.8893,
276
+ "nll_loss": 119.52433776855469,
277
+ "rewards/accuracies": 0.6000000238418579,
278
+ "rewards/chosen": -122.43513488769531,
279
+ "rewards/margins": 0.24384291470050812,
280
+ "rewards/rejected": -122.6789779663086,
281
+ "step": 85
282
+ },
283
+ {
284
+ "epoch": 0.8533333333333334,
285
+ "grad_norm": 98.68675231933594,
286
+ "learning_rate": 4.499461566702685e-05,
287
+ "log_odds_chosen": 0.4335232675075531,
288
+ "log_odds_ratio": -0.6551752686500549,
289
+ "logps/chosen": -113.51853942871094,
290
+ "logps/rejected": -113.9520492553711,
291
+ "loss": 20.9656,
292
+ "nll_loss": 110.78584289550781,
293
+ "rewards/accuracies": 0.668749988079071,
294
+ "rewards/chosen": -113.51853942871094,
295
+ "rewards/margins": 0.4335232675075531,
296
+ "rewards/rejected": -113.9520492553711,
297
+ "step": 90
298
+ },
299
+ {
300
+ "epoch": 0.9007407407407407,
301
+ "grad_norm": 486.6390380859375,
302
+ "learning_rate": 4.413127538374411e-05,
303
+ "log_odds_chosen": 0.5088523030281067,
304
+ "log_odds_ratio": -0.64421546459198,
305
+ "logps/chosen": -113.52534484863281,
306
+ "logps/rejected": -114.0342025756836,
307
+ "loss": 20.6149,
308
+ "nll_loss": 111.275634765625,
309
+ "rewards/accuracies": 0.6875,
310
+ "rewards/chosen": -113.52534484863281,
311
+ "rewards/margins": 0.5088523030281067,
312
+ "rewards/rejected": -114.0342025756836,
313
+ "step": 95
314
+ },
315
+ {
316
+ "epoch": 0.9481481481481482,
317
+ "grad_norm": 259.378173828125,
318
+ "learning_rate": 4.320901013934887e-05,
319
+ "log_odds_chosen": 0.5669993162155151,
320
+ "log_odds_ratio": -0.8858685493469238,
321
+ "logps/chosen": -106.97605895996094,
322
+ "logps/rejected": -107.5430679321289,
323
+ "loss": 28.3478,
324
+ "nll_loss": 104.57115173339844,
325
+ "rewards/accuracies": 0.621874988079071,
326
+ "rewards/chosen": -106.97605895996094,
327
+ "rewards/margins": 0.5669993162155151,
328
+ "rewards/rejected": -107.5430679321289,
329
+ "step": 100
330
+ },
331
+ {
332
+ "epoch": 0.9955555555555555,
333
+ "grad_norm": 564.1270141601562,
334
+ "learning_rate": 4.223066054130568e-05,
335
+ "log_odds_chosen": 0.6227962374687195,
336
+ "log_odds_ratio": -0.9325195550918579,
337
+ "logps/chosen": -92.55345916748047,
338
+ "logps/rejected": -93.17625427246094,
339
+ "loss": 29.8406,
340
+ "nll_loss": 89.38184356689453,
341
+ "rewards/accuracies": 0.6187499761581421,
342
+ "rewards/chosen": -92.55345916748047,
343
+ "rewards/margins": 0.6227962374687195,
344
+ "rewards/rejected": -93.17625427246094,
345
+ "step": 105
346
+ },
347
+ {
348
+ "epoch": 1.037925925925926,
349
+ "grad_norm": 287.26513671875,
350
+ "learning_rate": 4.1199239938743797e-05,
351
+ "log_odds_chosen": 0.5149707794189453,
352
+ "log_odds_ratio": -0.6010006070137024,
353
+ "logps/chosen": -74.13626861572266,
354
+ "logps/rejected": -74.6512451171875,
355
+ "loss": 17.1886,
356
+ "nll_loss": 71.68141174316406,
357
+ "rewards/accuracies": 0.7027971744537354,
358
+ "rewards/chosen": -74.13626861572266,
359
+ "rewards/margins": 0.5149707794189453,
360
+ "rewards/rejected": -74.6512451171875,
361
+ "step": 110
362
+ },
363
+ {
364
+ "epoch": 1.0853333333333333,
365
+ "grad_norm": 430.4316711425781,
366
+ "learning_rate": 4.0117925141242174e-05,
367
+ "log_odds_chosen": 0.6114441156387329,
368
+ "log_odds_ratio": -0.6317542195320129,
369
+ "logps/chosen": -70.51701354980469,
370
+ "logps/rejected": -71.12846374511719,
371
+ "loss": 20.2161,
372
+ "nll_loss": 68.18811798095703,
373
+ "rewards/accuracies": 0.653124988079071,
374
+ "rewards/chosen": -70.51701354980469,
375
+ "rewards/margins": 0.6114441156387329,
376
+ "rewards/rejected": -71.12846374511719,
377
+ "step": 115
378
+ },
379
+ {
380
+ "epoch": 1.1327407407407408,
381
+ "grad_norm": 225.65744018554688,
382
+ "learning_rate": 3.899004663415084e-05,
383
+ "log_odds_chosen": 0.8829742670059204,
384
+ "log_odds_ratio": -0.7018040418624878,
385
+ "logps/chosen": -72.43450927734375,
386
+ "logps/rejected": -73.3174819946289,
387
+ "loss": 22.4577,
388
+ "nll_loss": 70.06659698486328,
389
+ "rewards/accuracies": 0.6343749761581421,
390
+ "rewards/chosen": -72.43450927734375,
391
+ "rewards/margins": 0.8829742670059204,
392
+ "rewards/rejected": -73.3174819946289,
393
+ "step": 120
394
+ },
395
+ {
396
+ "epoch": 1.1801481481481482,
397
+ "grad_norm": 605.541748046875,
398
+ "learning_rate": 3.781907832058587e-05,
399
+ "log_odds_chosen": 0.7452532052993774,
400
+ "log_odds_ratio": -0.6702367663383484,
401
+ "logps/chosen": -74.36035919189453,
402
+ "logps/rejected": -75.1056137084961,
403
+ "loss": 21.4476,
404
+ "nll_loss": 71.98149108886719,
405
+ "rewards/accuracies": 0.6656249761581421,
406
+ "rewards/chosen": -74.36035919189453,
407
+ "rewards/margins": 0.7452532052993774,
408
+ "rewards/rejected": -75.1056137084961,
409
+ "step": 125
410
+ },
411
+ {
412
+ "epoch": 1.2275555555555555,
413
+ "grad_norm": 292.6062927246094,
414
+ "learning_rate": 3.660862682169282e-05,
415
+ "log_odds_chosen": 0.5303805470466614,
416
+ "log_odds_ratio": -0.6513002514839172,
417
+ "logps/chosen": -67.90737915039062,
418
+ "logps/rejected": -68.43775939941406,
419
+ "loss": 20.8416,
420
+ "nll_loss": 66.057861328125,
421
+ "rewards/accuracies": 0.668749988079071,
422
+ "rewards/chosen": -67.90737915039062,
423
+ "rewards/margins": 0.5303805470466614,
424
+ "rewards/rejected": -68.43775939941406,
425
+ "step": 130
426
+ },
427
+ {
428
+ "epoch": 1.274962962962963,
429
+ "grad_norm": 118.44042205810547,
430
+ "learning_rate": 3.5362420368134356e-05,
431
+ "log_odds_chosen": 0.702187716960907,
432
+ "log_odds_ratio": -0.6650616526603699,
433
+ "logps/chosen": -78.17137145996094,
434
+ "logps/rejected": -78.8735580444336,
435
+ "loss": 21.282,
436
+ "nll_loss": 76.62179565429688,
437
+ "rewards/accuracies": 0.7281249761581421,
438
+ "rewards/chosen": -78.17137145996094,
439
+ "rewards/margins": 0.702187716960907,
440
+ "rewards/rejected": -78.8735580444336,
441
+ "step": 135
442
+ },
443
+ {
444
+ "epoch": 1.3223703703703704,
445
+ "grad_norm": 196.71144104003906,
446
+ "learning_rate": 3.408429731701635e-05,
447
+ "log_odds_chosen": 0.7111908793449402,
448
+ "log_odds_ratio": -0.662623405456543,
449
+ "logps/chosen": -64.79576873779297,
450
+ "logps/rejected": -65.5069580078125,
451
+ "loss": 21.2039,
452
+ "nll_loss": 63.509056091308594,
453
+ "rewards/accuracies": 0.659375011920929,
454
+ "rewards/chosen": -64.79576873779297,
455
+ "rewards/margins": 0.7111908793449402,
456
+ "rewards/rejected": -65.5069580078125,
457
+ "step": 140
458
+ },
459
+ {
460
+ "epoch": 1.3697777777777778,
461
+ "grad_norm": 76.86389923095703,
462
+ "learning_rate": 3.2778194329621104e-05,
463
+ "log_odds_chosen": 1.3437374830245972,
464
+ "log_odds_ratio": -0.5445761680603027,
465
+ "logps/chosen": -76.33842468261719,
466
+ "logps/rejected": -77.68216705322266,
467
+ "loss": 17.4264,
468
+ "nll_loss": 74.82538604736328,
469
+ "rewards/accuracies": 0.715624988079071,
470
+ "rewards/chosen": -76.33842468261719,
471
+ "rewards/margins": 1.3437374830245972,
472
+ "rewards/rejected": -77.68216705322266,
473
+ "step": 145
474
+ },
475
+ {
476
+ "epoch": 1.417185185185185,
477
+ "grad_norm": 379.3780212402344,
478
+ "learning_rate": 3.144813424636031e-05,
479
+ "log_odds_chosen": 0.8894694447517395,
480
+ "log_odds_ratio": -0.60655677318573,
481
+ "logps/chosen": -73.5786361694336,
482
+ "logps/rejected": -74.46809387207031,
483
+ "loss": 19.4098,
484
+ "nll_loss": 71.83946228027344,
485
+ "rewards/accuracies": 0.6875,
486
+ "rewards/chosen": -73.5786361694336,
487
+ "rewards/margins": 0.8894694447517395,
488
+ "rewards/rejected": -74.46809387207031,
489
+ "step": 150
490
+ },
491
+ {
492
+ "epoch": 1.4645925925925927,
493
+ "grad_norm": 194.71551513671875,
494
+ "learning_rate": 3.0098213696293542e-05,
495
+ "log_odds_chosen": 0.7581270933151245,
496
+ "log_odds_ratio": -0.6102460622787476,
497
+ "logps/chosen": -78.3839111328125,
498
+ "logps/rejected": -79.14203643798828,
499
+ "loss": 19.5279,
500
+ "nll_loss": 76.60259246826172,
501
+ "rewards/accuracies": 0.6625000238418579,
502
+ "rewards/chosen": -78.3839111328125,
503
+ "rewards/margins": 0.7581270933151245,
504
+ "rewards/rejected": -79.14203643798828,
505
+ "step": 155
506
+ },
507
+ {
508
+ "epoch": 1.512,
509
+ "grad_norm": 181.2035369873047,
510
+ "learning_rate": 2.8732590479375165e-05,
511
+ "log_odds_chosen": 0.659544825553894,
512
+ "log_odds_ratio": -0.5649217367172241,
513
+ "logps/chosen": -81.12464904785156,
514
+ "logps/rejected": -81.7842025756836,
515
+ "loss": 18.0775,
516
+ "nll_loss": 79.08438873291016,
517
+ "rewards/accuracies": 0.7093750238418579,
518
+ "rewards/chosen": -81.12464904785156,
519
+ "rewards/margins": 0.659544825553894,
520
+ "rewards/rejected": -81.7842025756836,
521
+ "step": 160
522
+ },
523
+ {
524
+ "epoch": 1.5594074074074074,
525
+ "grad_norm": 220.28268432617188,
526
+ "learning_rate": 2.7355470760292956e-05,
527
+ "log_odds_chosen": 0.9221769571304321,
528
+ "log_odds_ratio": -0.5502141118049622,
529
+ "logps/chosen": -79.66305541992188,
530
+ "logps/rejected": -80.58523559570312,
531
+ "loss": 17.6069,
532
+ "nll_loss": 77.8115463256836,
533
+ "rewards/accuracies": 0.7124999761581421,
534
+ "rewards/chosen": -79.66305541992188,
535
+ "rewards/margins": 0.9221769571304321,
536
+ "rewards/rejected": -80.58523559570312,
537
+ "step": 165
538
+ },
539
+ {
540
+ "epoch": 1.6068148148148147,
541
+ "grad_norm": 65.46253967285156,
542
+ "learning_rate": 2.597109611334169e-05,
543
+ "log_odds_chosen": 1.040714979171753,
544
+ "log_odds_ratio": -0.5623927116394043,
545
+ "logps/chosen": -79.55305480957031,
546
+ "logps/rejected": -80.5937728881836,
547
+ "loss": 17.9966,
548
+ "nll_loss": 77.59423065185547,
549
+ "rewards/accuracies": 0.7093750238418579,
550
+ "rewards/chosen": -79.55305480957031,
551
+ "rewards/margins": 1.040714979171753,
552
+ "rewards/rejected": -80.5937728881836,
553
+ "step": 170
554
+ },
555
+ {
556
+ "epoch": 1.6542222222222223,
557
+ "grad_norm": 224.69976806640625,
558
+ "learning_rate": 2.458373045823404e-05,
559
+ "log_odds_chosen": 0.8248542547225952,
560
+ "log_odds_ratio": -0.5441884994506836,
561
+ "logps/chosen": -75.96087646484375,
562
+ "logps/rejected": -76.78572845458984,
563
+ "loss": 17.414,
564
+ "nll_loss": 73.72982025146484,
565
+ "rewards/accuracies": 0.731249988079071,
566
+ "rewards/chosen": -75.96087646484375,
567
+ "rewards/margins": 0.8248542547225952,
568
+ "rewards/rejected": -76.78572845458984,
569
+ "step": 175
570
+ },
571
+ {
572
+ "epoch": 1.7016296296296296,
573
+ "grad_norm": 123.06717681884766,
574
+ "learning_rate": 2.3197646927086697e-05,
575
+ "log_odds_chosen": 0.8531917333602905,
576
+ "log_odds_ratio": -0.502475380897522,
577
+ "logps/chosen": -77.56637573242188,
578
+ "logps/rejected": -78.41956329345703,
579
+ "loss": 16.0792,
580
+ "nll_loss": 75.34583282470703,
581
+ "rewards/accuracies": 0.768750011920929,
582
+ "rewards/chosen": -77.56637573242188,
583
+ "rewards/margins": 0.8531917333602905,
584
+ "rewards/rejected": -78.41956329345703,
585
+ "step": 180
586
+ },
587
+ {
588
+ "epoch": 1.749037037037037,
589
+ "grad_norm": 90.936767578125,
590
+ "learning_rate": 2.1817114703032176e-05,
591
+ "log_odds_chosen": 0.8035793304443359,
592
+ "log_odds_ratio": -0.5467336773872375,
593
+ "logps/chosen": -74.6766357421875,
594
+ "logps/rejected": -75.48020935058594,
595
+ "loss": 17.4955,
596
+ "nll_loss": 72.79801940917969,
597
+ "rewards/accuracies": 0.7093750238418579,
598
+ "rewards/chosen": -74.6766357421875,
599
+ "rewards/margins": 0.8035793304443359,
600
+ "rewards/rejected": -75.48020935058594,
601
+ "step": 185
602
+ },
603
+ {
604
+ "epoch": 1.7964444444444445,
605
+ "grad_norm": 161.3369140625,
606
+ "learning_rate": 2.0446385870993467e-05,
607
+ "log_odds_chosen": 1.057072401046753,
608
+ "log_odds_ratio": -0.4894322454929352,
609
+ "logps/chosen": -69.64981079101562,
610
+ "logps/rejected": -70.70687103271484,
611
+ "loss": 15.6618,
612
+ "nll_loss": 67.64393615722656,
613
+ "rewards/accuracies": 0.7718750238418579,
614
+ "rewards/chosen": -69.64981079101562,
615
+ "rewards/margins": 1.057072401046753,
616
+ "rewards/rejected": -70.70687103271484,
617
+ "step": 190
618
+ },
619
+ {
620
+ "epoch": 1.8438518518518519,
621
+ "grad_norm": 349.56439208984375,
622
+ "learning_rate": 1.9089682321121834e-05,
623
+ "log_odds_chosen": 1.5846078395843506,
624
+ "log_odds_ratio": -0.5456617474555969,
625
+ "logps/chosen": -80.89094543457031,
626
+ "logps/rejected": -82.4755630493164,
627
+ "loss": 17.4612,
628
+ "nll_loss": 77.97457885742188,
629
+ "rewards/accuracies": 0.762499988079071,
630
+ "rewards/chosen": -80.89094543457031,
631
+ "rewards/margins": 1.5846078395843506,
632
+ "rewards/rejected": -82.4755630493164,
633
+ "step": 195
634
+ },
635
+ {
636
+ "epoch": 1.8912592592592592,
637
+ "grad_norm": 73.02623748779297,
638
+ "learning_rate": 1.775118274523545e-05,
639
+ "log_odds_chosen": 1.1135013103485107,
640
+ "log_odds_ratio": -0.532755970954895,
641
+ "logps/chosen": -84.60154724121094,
642
+ "logps/rejected": -85.71505737304688,
643
+ "loss": 17.0482,
644
+ "nll_loss": 81.52278900146484,
645
+ "rewards/accuracies": 0.7593749761581421,
646
+ "rewards/chosen": -84.60154724121094,
647
+ "rewards/margins": 1.1135013103485107,
648
+ "rewards/rejected": -85.71505737304688,
649
+ "step": 200
650
+ },
651
+ {
652
+ "epoch": 1.9386666666666668,
653
+ "grad_norm": 68.42233276367188,
654
+ "learning_rate": 1.643500976631037e-05,
655
+ "log_odds_chosen": 0.717894971370697,
656
+ "log_odds_ratio": -0.5447430610656738,
657
+ "logps/chosen": -80.26295471191406,
658
+ "logps/rejected": -80.9808578491211,
659
+ "loss": 17.4318,
660
+ "nll_loss": 77.52245330810547,
661
+ "rewards/accuracies": 0.721875011920929,
662
+ "rewards/chosen": -80.26295471191406,
663
+ "rewards/margins": 0.717894971370697,
664
+ "rewards/rejected": -80.9808578491211,
665
+ "step": 205
666
+ },
667
+ {
668
+ "epoch": 1.986074074074074,
669
+ "grad_norm": 114.36587524414062,
670
+ "learning_rate": 1.514521724066537e-05,
671
+ "log_odds_chosen": 0.6508203744888306,
672
+ "log_odds_ratio": -0.6152015328407288,
673
+ "logps/chosen": -80.61943817138672,
674
+ "logps/rejected": -81.27025604248047,
675
+ "loss": 19.6865,
676
+ "nll_loss": 77.93927001953125,
677
+ "rewards/accuracies": 0.671875,
678
+ "rewards/chosen": -80.61943817138672,
679
+ "rewards/margins": 0.6508203744888306,
680
+ "rewards/rejected": -81.27025604248047,
681
+ "step": 210
682
+ },
683
+ {
684
+ "epoch": 2.0284444444444443,
685
+ "grad_norm": 152.35540771484375,
686
+ "learning_rate": 1.3885777771950348e-05,
687
+ "log_odds_chosen": 1.0508326292037964,
688
+ "log_odds_ratio": -0.5209993720054626,
689
+ "logps/chosen": -80.12591552734375,
690
+ "logps/rejected": -81.1767578125,
691
+ "loss": 14.9006,
692
+ "nll_loss": 77.8501968383789,
693
+ "rewards/accuracies": 0.751748263835907,
694
+ "rewards/chosen": -80.12591552734375,
695
+ "rewards/margins": 1.0508326292037964,
696
+ "rewards/rejected": -81.1767578125,
697
+ "step": 215
698
+ },
699
+ {
700
+ "epoch": 2.075851851851852,
701
+ "grad_norm": 67.25732421875,
702
+ "learning_rate": 1.2660570475395683e-05,
703
+ "log_odds_chosen": 1.0816318988800049,
704
+ "log_odds_ratio": -0.4839634299278259,
705
+ "logps/chosen": -74.36226654052734,
706
+ "logps/rejected": -75.44390106201172,
707
+ "loss": 15.4868,
708
+ "nll_loss": 72.31621551513672,
709
+ "rewards/accuracies": 0.7562500238418579,
710
+ "rewards/chosen": -74.36226654052734,
711
+ "rewards/margins": 1.0816318988800049,
712
+ "rewards/rejected": -75.44390106201172,
713
+ "step": 220
714
+ },
715
+ {
716
+ "epoch": 2.1232592592592594,
717
+ "grad_norm": 121.6506118774414,
718
+ "learning_rate": 1.1473369030008974e-05,
719
+ "log_odds_chosen": 1.3746577501296997,
720
+ "log_odds_ratio": -0.41175857186317444,
721
+ "logps/chosen": -75.92201232910156,
722
+ "logps/rejected": -77.29666137695312,
723
+ "loss": 13.1763,
724
+ "nll_loss": 73.99418640136719,
725
+ "rewards/accuracies": 0.824999988079071,
726
+ "rewards/chosen": -75.92201232910156,
727
+ "rewards/margins": 1.3746577501296997,
728
+ "rewards/rejected": -77.29666137695312,
729
+ "step": 225
730
+ },
731
+ {
732
+ "epoch": 2.1706666666666665,
733
+ "grad_norm": 91.61468505859375,
734
+ "learning_rate": 1.0327830055518842e-05,
735
+ "log_odds_chosen": 1.8022880554199219,
736
+ "log_odds_ratio": -0.4180576205253601,
737
+ "logps/chosen": -76.81086730957031,
738
+ "logps/rejected": -78.61314392089844,
739
+ "loss": 13.3778,
740
+ "nll_loss": 74.9874038696289,
741
+ "rewards/accuracies": 0.809374988079071,
742
+ "rewards/chosen": -76.81086730957031,
743
+ "rewards/margins": 1.8022880554199219,
744
+ "rewards/rejected": -78.61314392089844,
745
+ "step": 230
746
+ },
747
+ {
748
+ "epoch": 2.218074074074074,
749
+ "grad_norm": 185.319580078125,
750
+ "learning_rate": 9.227481849865235e-06,
751
+ "log_odds_chosen": 1.7025684118270874,
752
+ "log_odds_ratio": -0.43913546204566956,
753
+ "logps/chosen": -78.4276351928711,
754
+ "logps/rejected": -80.13020324707031,
755
+ "loss": 14.0523,
756
+ "nll_loss": 76.0723648071289,
757
+ "rewards/accuracies": 0.8125,
758
+ "rewards/chosen": -78.4276351928711,
759
+ "rewards/margins": 1.7025684118270874,
760
+ "rewards/rejected": -80.13020324707031,
761
+ "step": 235
762
+ },
763
+ {
764
+ "epoch": 2.2654814814814817,
765
+ "grad_norm": 235.93431091308594,
766
+ "learning_rate": 8.175713521924978e-06,
767
+ "log_odds_chosen": 1.7359157800674438,
768
+ "log_odds_ratio": -0.3843821883201599,
769
+ "logps/chosen": -74.28846740722656,
770
+ "logps/rejected": -76.02437591552734,
771
+ "loss": 12.3002,
772
+ "nll_loss": 72.41877746582031,
773
+ "rewards/accuracies": 0.831250011920929,
774
+ "rewards/chosen": -74.28846740722656,
775
+ "rewards/margins": 1.7359157800674438,
776
+ "rewards/rejected": -76.02437591552734,
777
+ "step": 240
778
+ },
779
+ {
780
+ "epoch": 2.3128888888888888,
781
+ "grad_norm": 96.19325256347656,
782
+ "learning_rate": 7.1757645529443665e-06,
783
+ "log_odds_chosen": 1.8641507625579834,
784
+ "log_odds_ratio": -0.39463645219802856,
785
+ "logps/chosen": -72.28008270263672,
786
+ "logps/rejected": -74.14424133300781,
787
+ "loss": 12.6284,
788
+ "nll_loss": 70.3760986328125,
789
+ "rewards/accuracies": 0.831250011920929,
790
+ "rewards/chosen": -72.28008270263672,
791
+ "rewards/margins": 1.8641507625579834,
792
+ "rewards/rejected": -74.14424133300781,
793
+ "step": 245
794
+ },
795
+ {
796
+ "epoch": 2.3602962962962963,
797
+ "grad_norm": 111.45867919921875,
798
+ "learning_rate": 6.230714818829733e-06,
799
+ "log_odds_chosen": 1.782758116722107,
800
+ "log_odds_ratio": -0.3957170844078064,
801
+ "logps/chosen": -71.14763641357422,
802
+ "logps/rejected": -72.93038177490234,
803
+ "loss": 12.6629,
804
+ "nll_loss": 69.536376953125,
805
+ "rewards/accuracies": 0.815625011920929,
806
+ "rewards/chosen": -71.14763641357422,
807
+ "rewards/margins": 1.782758116722107,
808
+ "rewards/rejected": -72.93038177490234,
809
+ "step": 250
810
+ },
811
+ {
812
+ "epoch": 2.407703703703704,
813
+ "grad_norm": 96.2086181640625,
814
+ "learning_rate": 5.343475104027743e-06,
815
+ "log_odds_chosen": 1.9097713232040405,
816
+ "log_odds_ratio": -0.3789903521537781,
817
+ "logps/chosen": -72.61198425292969,
818
+ "logps/rejected": -74.5217514038086,
819
+ "loss": 12.1277,
820
+ "nll_loss": 70.6400146484375,
821
+ "rewards/accuracies": 0.84375,
822
+ "rewards/chosen": -72.61198425292969,
823
+ "rewards/margins": 1.9097713232040405,
824
+ "rewards/rejected": -74.5217514038086,
825
+ "step": 255
826
+ },
827
+ {
828
+ "epoch": 2.455111111111111,
829
+ "grad_norm": 107.21746063232422,
830
+ "learning_rate": 4.516778136213037e-06,
831
+ "log_odds_chosen": 2.123696804046631,
832
+ "log_odds_ratio": -0.33625128865242004,
833
+ "logps/chosen": -75.09449768066406,
834
+ "logps/rejected": -77.21818542480469,
835
+ "loss": 10.76,
836
+ "nll_loss": 72.9236068725586,
837
+ "rewards/accuracies": 0.8374999761581421,
838
+ "rewards/chosen": -75.09449768066406,
839
+ "rewards/margins": 2.123696804046631,
840
+ "rewards/rejected": -77.21818542480469,
841
+ "step": 260
842
+ },
843
+ {
844
+ "epoch": 2.5025185185185186,
845
+ "grad_norm": 140.90371704101562,
846
+ "learning_rate": 3.7531701693965554e-06,
847
+ "log_odds_chosen": 2.315692663192749,
848
+ "log_odds_ratio": -0.3569856882095337,
849
+ "logps/chosen": -75.63587951660156,
850
+ "logps/rejected": -77.95157623291016,
851
+ "loss": 11.4235,
852
+ "nll_loss": 73.40586853027344,
853
+ "rewards/accuracies": 0.859375,
854
+ "rewards/chosen": -75.63587951660156,
855
+ "rewards/margins": 2.315692663192749,
856
+ "rewards/rejected": -77.95157623291016,
857
+ "step": 265
858
+ },
859
+ {
860
+ "epoch": 2.549925925925926,
861
+ "grad_norm": 116.43041229248047,
862
+ "learning_rate": 3.055003141378948e-06,
863
+ "log_odds_chosen": 2.3956801891326904,
864
+ "log_odds_ratio": -0.3444186747074127,
865
+ "logps/chosen": -76.45235443115234,
866
+ "logps/rejected": -78.84803771972656,
867
+ "loss": 11.0214,
868
+ "nll_loss": 74.2215576171875,
869
+ "rewards/accuracies": 0.856249988079071,
870
+ "rewards/chosen": -76.45235443115234,
871
+ "rewards/margins": 2.3956801891326904,
872
+ "rewards/rejected": -78.84803771972656,
873
+ "step": 270
874
+ },
875
+ {
876
+ "epoch": 2.5973333333333333,
877
+ "grad_norm": 111.95805358886719,
878
+ "learning_rate": 2.424427429704365e-06,
879
+ "log_odds_chosen": 2.0682857036590576,
880
+ "log_odds_ratio": -0.40607601404190063,
881
+ "logps/chosen": -76.42601776123047,
882
+ "logps/rejected": -78.49430847167969,
883
+ "loss": 12.9944,
884
+ "nll_loss": 74.2015380859375,
885
+ "rewards/accuracies": 0.815625011920929,
886
+ "rewards/chosen": -76.42601776123047,
887
+ "rewards/margins": 2.0682857036590576,
888
+ "rewards/rejected": -78.49430847167969,
889
+ "step": 275
890
+ },
891
+ {
892
+ "epoch": 2.644740740740741,
893
+ "grad_norm": 117.52398681640625,
894
+ "learning_rate": 1.8633852284264508e-06,
895
+ "log_odds_chosen": 2.284158229827881,
896
+ "log_odds_ratio": -0.3496808409690857,
897
+ "logps/chosen": -76.18301391601562,
898
+ "logps/rejected": -78.4671630859375,
899
+ "loss": 11.1898,
900
+ "nll_loss": 74.24303436279297,
901
+ "rewards/accuracies": 0.840624988079071,
902
+ "rewards/chosen": -76.18301391601562,
903
+ "rewards/margins": 2.284158229827881,
904
+ "rewards/rejected": -78.4671630859375,
905
+ "step": 280
906
+ },
907
+ {
908
+ "epoch": 2.6921481481481484,
909
+ "grad_norm": 145.8434600830078,
910
+ "learning_rate": 1.3736045660864034e-06,
911
+ "log_odds_chosen": 2.0426528453826904,
912
+ "log_odds_ratio": -0.4473814070224762,
913
+ "logps/chosen": -76.21638488769531,
914
+ "logps/rejected": -78.259033203125,
915
+ "loss": 14.3162,
916
+ "nll_loss": 73.87956237792969,
917
+ "rewards/accuracies": 0.7718750238418579,
918
+ "rewards/chosen": -76.21638488769531,
919
+ "rewards/margins": 2.0426528453826904,
920
+ "rewards/rejected": -78.259033203125,
921
+ "step": 285
922
+ },
923
+ {
924
+ "epoch": 2.7395555555555555,
925
+ "grad_norm": 262.74298095703125,
926
+ "learning_rate": 9.565939833279192e-07,
927
+ "log_odds_chosen": 2.02114200592041,
928
+ "log_odds_ratio": -0.38766786456108093,
929
+ "logps/chosen": -76.1156997680664,
930
+ "logps/rejected": -78.13684844970703,
931
+ "loss": 12.4054,
932
+ "nll_loss": 74.00138854980469,
933
+ "rewards/accuracies": 0.815625011920929,
934
+ "rewards/chosen": -76.1156997680664,
935
+ "rewards/margins": 2.02114200592041,
936
+ "rewards/rejected": -78.13684844970703,
937
+ "step": 290
938
+ },
939
+ {
940
+ "epoch": 2.786962962962963,
941
+ "grad_norm": 126.94654083251953,
942
+ "learning_rate": 6.136378865420872e-07,
943
+ "log_odds_chosen": 1.9735645055770874,
944
+ "log_odds_ratio": -0.37328869104385376,
945
+ "logps/chosen": -75.25834655761719,
946
+ "logps/rejected": -77.23191833496094,
947
+ "loss": 11.9452,
948
+ "nll_loss": 73.13874053955078,
949
+ "rewards/accuracies": 0.8500000238418579,
950
+ "rewards/chosen": -75.25834655761719,
951
+ "rewards/margins": 1.9735645055770874,
952
+ "rewards/rejected": -77.23191833496094,
953
+ "step": 295
954
+ },
955
+ {
956
+ "epoch": 2.83437037037037,
957
+ "grad_norm": 99.43417358398438,
958
+ "learning_rate": 3.45792591853214e-07,
959
+ "log_odds_chosen": 2.100620746612549,
960
+ "log_odds_ratio": -0.403224378824234,
961
+ "logps/chosen": -75.48963165283203,
962
+ "logps/rejected": -77.59024810791016,
963
+ "loss": 12.9032,
964
+ "nll_loss": 73.26192474365234,
965
+ "rewards/accuracies": 0.815625011920929,
966
+ "rewards/chosen": -75.48963165283203,
967
+ "rewards/margins": 2.100620746612549,
968
+ "rewards/rejected": -77.59024810791016,
969
+ "step": 300
970
+ },
971
+ {
972
+ "epoch": 2.8817777777777778,
973
+ "grad_norm": 115.87845611572266,
974
+ "learning_rate": 1.538830716302092e-07,
975
+ "log_odds_chosen": 2.1068568229675293,
976
+ "log_odds_ratio": -0.3900986313819885,
977
+ "logps/chosen": -75.33139038085938,
978
+ "logps/rejected": -77.43824005126953,
979
+ "loss": 12.4832,
980
+ "nll_loss": 73.32325744628906,
981
+ "rewards/accuracies": 0.8062499761581421,
982
+ "rewards/chosen": -75.33139038085938,
983
+ "rewards/margins": 2.1068568229675293,
984
+ "rewards/rejected": -77.43824005126953,
985
+ "step": 305
986
+ },
987
+ {
988
+ "epoch": 2.9291851851851853,
989
+ "grad_norm": 101.96614074707031,
990
+ "learning_rate": 3.8500413544415025e-08,
991
+ "log_odds_chosen": 1.9942286014556885,
992
+ "log_odds_ratio": -0.3553644120693207,
993
+ "logps/chosen": -74.625244140625,
994
+ "logps/rejected": -76.6194839477539,
995
+ "loss": 11.3717,
996
+ "nll_loss": 72.78034210205078,
997
+ "rewards/accuracies": 0.8374999761581421,
998
+ "rewards/chosen": -74.625244140625,
999
+ "rewards/margins": 1.9942286014556885,
1000
+ "rewards/rejected": -76.6194839477539,
1001
+ "step": 310
1002
+ },
1003
+ {
1004
+ "epoch": 2.9765925925925925,
1005
+ "grad_norm": 176.18894958496094,
1006
+ "learning_rate": 0.0,
1007
+ "log_odds_chosen": 2.045795202255249,
1008
+ "log_odds_ratio": -0.3809313178062439,
1009
+ "logps/chosen": -75.89002990722656,
1010
+ "logps/rejected": -77.93583679199219,
1011
+ "loss": 12.1898,
1012
+ "nll_loss": 73.94648742675781,
1013
+ "rewards/accuracies": 0.840624988079071,
1014
+ "rewards/chosen": -75.89002990722656,
1015
+ "rewards/margins": 2.045795202255249,
1016
+ "rewards/rejected": -77.93583679199219,
1017
+ "step": 315
1018
+ },
1019
+ {
1020
+ "epoch": 2.9765925925925925,
1021
+ "step": 315,
1022
+ "total_flos": 0.0,
1023
+ "train_loss": 38.28643096439422,
1024
+ "train_runtime": 9296.3619,
1025
+ "train_samples_per_second": 2.178,
1026
+ "train_steps_per_second": 0.034
1027
+ }
1028
+ ],
1029
+ "logging_steps": 5,
1030
+ "max_steps": 315,
1031
+ "num_input_tokens_seen": 0,
1032
+ "num_train_epochs": 3,
1033
+ "save_steps": 100000,
1034
+ "stateful_callbacks": {
1035
+ "TrainerControl": {
1036
+ "args": {
1037
+ "should_epoch_stop": false,
1038
+ "should_evaluate": false,
1039
+ "should_log": false,
1040
+ "should_save": true,
1041
+ "should_training_stop": true
1042
+ },
1043
+ "attributes": {}
1044
+ }
1045
+ },
1046
+ "total_flos": 0.0,
1047
+ "train_batch_size": 1,
1048
+ "trial_name": null,
1049
+ "trial_params": null
1050
+ }