zfz1 commited on
Commit
9cb3e8d
1 Parent(s): 062e283

Model save

Browse files
Files changed (4) hide show
  1. README.md +3 -19
  2. all_results.json +3 -19
  3. train_results.json +3 -3
  4. trainer_state.json +437 -437
README.md CHANGED
@@ -1,11 +1,8 @@
1
  ---
2
  base_model: deepseek-ai/deepseek-math-7b-base
3
- datasets:
4
- - zfz1/my_preference_gsm8k_deepseek
5
  library_name: peft
6
  license: other
7
  tags:
8
- - alignment-handbook
9
  - trl
10
  - orpo
11
  - generated_from_trainer
@@ -17,23 +14,10 @@ model-index:
17
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
18
  should probably proofread and complete it, then remove this comment. -->
19
 
20
- [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/thuzfz1/huggingface/runs/zs68qfsi)
21
  # deepseek-8b-orpo-lora
22
 
23
- This model is a fine-tuned version of [deepseek-ai/deepseek-math-7b-base](https://huggingface.co/deepseek-ai/deepseek-math-7b-base) on the zfz1/my_preference_gsm8k_deepseek dataset.
24
- It achieves the following results on the evaluation set:
25
- - Loss: 1.1456
26
- - Rewards/chosen: -0.0726
27
- - Rewards/rejected: -0.0757
28
- - Rewards/accuracies: 0.5647
29
- - Rewards/margins: 0.0031
30
- - Logps/rejected: -0.7569
31
- - Logps/chosen: -0.7262
32
- - Logits/rejected: 30.3288
33
- - Logits/chosen: 30.4427
34
- - Nll Loss: 1.0719
35
- - Log Odds Ratio: -0.6909
36
- - Log Odds Chosen: 0.0553
37
 
38
  ## Model description
39
 
@@ -52,7 +36,7 @@ More information needed
52
  ### Training hyperparameters
53
 
54
  The following hyperparameters were used during training:
55
- - learning_rate: 3e-06
56
  - train_batch_size: 16
57
  - eval_batch_size: 16
58
  - seed: 43
 
1
  ---
2
  base_model: deepseek-ai/deepseek-math-7b-base
 
 
3
  library_name: peft
4
  license: other
5
  tags:
 
6
  - trl
7
  - orpo
8
  - generated_from_trainer
 
14
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
15
  should probably proofread and complete it, then remove this comment. -->
16
 
17
+ [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/thuzfz1/huggingface/runs/z4q90sz1)
18
  # deepseek-8b-orpo-lora
19
 
20
+ This model is a fine-tuned version of [deepseek-ai/deepseek-math-7b-base](https://huggingface.co/deepseek-ai/deepseek-math-7b-base) on an unknown dataset.
 
 
 
 
 
 
 
 
 
 
 
 
 
21
 
22
  ## Model description
23
 
 
36
  ### Training hyperparameters
37
 
38
  The following hyperparameters were used during training:
39
+ - learning_rate: 0.0001
40
  - train_batch_size: 16
41
  - eval_batch_size: 16
42
  - seed: 43
all_results.json CHANGED
@@ -1,25 +1,9 @@
1
  {
2
  "epoch": 1.9936102236421727,
3
- "eval_log_odds_chosen": 0.05526316165924072,
4
- "eval_log_odds_ratio": -0.6908599138259888,
5
- "eval_logits/chosen": 30.44268798828125,
6
- "eval_logits/rejected": 30.32880210876465,
7
- "eval_logps/chosen": -0.7262152433395386,
8
- "eval_logps/rejected": -0.7568790316581726,
9
- "eval_loss": 1.1456295251846313,
10
- "eval_nll_loss": 1.071852684020996,
11
- "eval_rewards/accuracies": 0.5647059082984924,
12
- "eval_rewards/chosen": -0.07262152433395386,
13
- "eval_rewards/margins": 0.0030663711950182915,
14
- "eval_rewards/rejected": -0.07568789273500443,
15
- "eval_runtime": 247.9026,
16
- "eval_samples": 5398,
17
- "eval_samples_per_second": 21.775,
18
- "eval_steps_per_second": 0.343,
19
  "total_flos": 0.0,
20
- "train_loss": 1.3676127867820935,
21
- "train_runtime": 5524.9246,
22
  "train_samples": 20000,
23
- "train_samples_per_second": 7.24,
24
  "train_steps_per_second": 0.056
25
  }
 
1
  {
2
  "epoch": 1.9936102236421727,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  "total_flos": 0.0,
4
+ "train_loss": 0.7846963420892373,
5
+ "train_runtime": 5577.3844,
6
  "train_samples": 20000,
7
+ "train_samples_per_second": 7.172,
8
  "train_steps_per_second": 0.056
9
  }
train_results.json CHANGED
@@ -1,9 +1,9 @@
1
  {
2
  "epoch": 1.9936102236421727,
3
  "total_flos": 0.0,
4
- "train_loss": 1.3676127867820935,
5
- "train_runtime": 5524.9246,
6
  "train_samples": 20000,
7
- "train_samples_per_second": 7.24,
8
  "train_steps_per_second": 0.056
9
  }
 
1
  {
2
  "epoch": 1.9936102236421727,
3
  "total_flos": 0.0,
4
+ "train_loss": 0.7846963420892373,
5
+ "train_runtime": 5577.3844,
6
  "train_samples": 20000,
7
+ "train_samples_per_second": 7.172,
8
  "train_steps_per_second": 0.056
9
  }
trainer_state.json CHANGED
@@ -10,569 +10,569 @@
10
  "log_history": [
11
  {
12
  "epoch": 0.06389776357827476,
13
- "grad_norm": 0.341777060295131,
14
- "learning_rate": 9.375e-07,
15
- "log_odds_chosen": 0.05486620217561722,
16
- "log_odds_ratio": -0.6975381374359131,
17
- "logits/chosen": 33.646114349365234,
18
- "logits/rejected": 33.55038833618164,
19
- "logps/chosen": -0.9772504568099976,
20
- "logps/rejected": -1.0204073190689087,
21
- "loss": 1.6735,
22
- "nll_loss": 1.5879063606262207,
23
- "rewards/accuracies": 0.543749988079071,
24
- "rewards/chosen": -0.09772505611181259,
25
- "rewards/margins": 0.004315672442317009,
26
- "rewards/rejected": -0.10204073041677475,
27
  "step": 10
28
  },
29
  {
30
  "epoch": 0.12779552715654952,
31
- "grad_norm": 0.357474094120966,
32
- "learning_rate": 1.875e-06,
33
- "log_odds_chosen": 0.054913729429244995,
34
- "log_odds_ratio": -0.6985839009284973,
35
- "logits/chosen": 34.386383056640625,
36
- "logits/rejected": 33.98283386230469,
37
- "logps/chosen": -0.9762754440307617,
38
- "logps/rejected": -1.0154238939285278,
39
- "loss": 1.668,
40
- "nll_loss": 1.6016145944595337,
41
- "rewards/accuracies": 0.515625,
42
- "rewards/chosen": -0.09762755781412125,
43
- "rewards/margins": 0.003914830274879932,
44
- "rewards/rejected": -0.10154237598180771,
45
  "step": 20
46
  },
47
  {
48
  "epoch": 0.19169329073482427,
49
- "grad_norm": 0.3765211315354108,
50
- "learning_rate": 2.8125e-06,
51
- "log_odds_chosen": 0.06909728795289993,
52
- "log_odds_ratio": -0.6911064386367798,
53
- "logits/chosen": 33.32851028442383,
54
- "logits/rejected": 32.749427795410156,
55
- "logps/chosen": -0.9813788533210754,
56
- "logps/rejected": -1.0349732637405396,
57
- "loss": 1.6757,
58
- "nll_loss": 1.6085166931152344,
59
- "rewards/accuracies": 0.5249999761581421,
60
- "rewards/chosen": -0.09813789278268814,
61
- "rewards/margins": 0.005359448026865721,
62
- "rewards/rejected": -0.10349734127521515,
63
  "step": 30
64
  },
65
  {
66
  "epoch": 0.25559105431309903,
67
- "grad_norm": 0.39833625263223615,
68
- "learning_rate": 2.993961440992859e-06,
69
- "log_odds_chosen": 0.11841567605733871,
70
- "log_odds_ratio": -0.6757725477218628,
71
- "logits/chosen": 33.839942932128906,
72
- "logits/rejected": 33.162986755371094,
73
- "logps/chosen": -0.9702426791191101,
74
- "logps/rejected": -1.0581519603729248,
75
- "loss": 1.664,
76
- "nll_loss": 1.6022984981536865,
77
- "rewards/accuracies": 0.6031249761581421,
78
- "rewards/chosen": -0.09702426195144653,
79
- "rewards/margins": 0.008790932595729828,
80
- "rewards/rejected": -0.10581519454717636,
81
  "step": 40
82
  },
83
  {
84
  "epoch": 0.3194888178913738,
85
- "grad_norm": 0.45952260963075464,
86
- "learning_rate": 2.9695130976348534e-06,
87
- "log_odds_chosen": 0.09267185628414154,
88
- "log_odds_ratio": -0.684535801410675,
89
- "logits/chosen": 34.052207946777344,
90
- "logits/rejected": 32.802547454833984,
91
- "logps/chosen": -0.9801093935966492,
92
- "logps/rejected": -1.0446739196777344,
93
- "loss": 1.6582,
94
- "nll_loss": 1.5952296257019043,
95
- "rewards/accuracies": 0.565625011920929,
96
- "rewards/chosen": -0.09801094233989716,
97
- "rewards/margins": 0.0064564356580376625,
98
- "rewards/rejected": -0.10446737706661224,
99
  "step": 50
100
  },
101
  {
102
  "epoch": 0.38338658146964855,
103
- "grad_norm": 0.49504068932785444,
104
- "learning_rate": 2.9265847744427307e-06,
105
- "log_odds_chosen": 0.06481704860925674,
106
- "log_odds_ratio": -0.698478102684021,
107
- "logits/chosen": 32.32544708251953,
108
- "logits/rejected": 32.51976776123047,
109
- "logps/chosen": -0.9770153164863586,
110
- "logps/rejected": -1.0272716283798218,
111
- "loss": 1.6387,
112
- "nll_loss": 1.572589635848999,
113
- "rewards/accuracies": 0.5687500238418579,
114
- "rewards/chosen": -0.0977015346288681,
115
- "rewards/margins": 0.0050256275571882725,
116
- "rewards/rejected": -0.10272715240716934,
117
  "step": 60
118
  },
119
  {
120
  "epoch": 0.4472843450479233,
121
- "grad_norm": 0.5262433549192981,
122
- "learning_rate": 2.865716319988224e-06,
123
- "log_odds_chosen": 0.030134931206703186,
124
- "log_odds_ratio": -0.7101849317550659,
125
- "logits/chosen": 33.02894973754883,
126
- "logits/rejected": 33.309715270996094,
127
- "logps/chosen": -0.9551456570625305,
128
- "logps/rejected": -0.9776851534843445,
129
- "loss": 1.6289,
130
- "nll_loss": 1.5476250648498535,
131
- "rewards/accuracies": 0.5531250238418579,
132
- "rewards/chosen": -0.09551456570625305,
133
- "rewards/margins": 0.0022539461497217417,
134
- "rewards/rejected": -0.09776850789785385,
135
  "step": 70
136
  },
137
  {
138
  "epoch": 0.5111821086261981,
139
- "grad_norm": 0.5188886669640671,
140
- "learning_rate": 2.7876731904027993e-06,
141
- "log_odds_chosen": 0.08591620624065399,
142
- "log_odds_ratio": -0.6787055730819702,
143
- "logits/chosen": 34.61581039428711,
144
- "logits/rejected": 34.63213348388672,
145
- "logps/chosen": -0.9702480435371399,
146
- "logps/rejected": -1.0235538482666016,
147
- "loss": 1.5921,
148
- "nll_loss": 1.5600519180297852,
149
- "rewards/accuracies": 0.5625,
150
- "rewards/chosen": -0.0970248132944107,
151
- "rewards/margins": 0.005330582614988089,
152
- "rewards/rejected": -0.10235539823770523,
153
  "step": 80
154
  },
155
  {
156
  "epoch": 0.5750798722044729,
157
- "grad_norm": 0.5939864943475555,
158
- "learning_rate": 2.6934368233226715e-06,
159
- "log_odds_chosen": 0.0763852447271347,
160
- "log_odds_ratio": -0.6891772150993347,
161
- "logits/chosen": 34.269248962402344,
162
- "logits/rejected": 34.00069046020508,
163
- "logps/chosen": -0.9529930949211121,
164
- "logps/rejected": -1.0018848180770874,
165
- "loss": 1.586,
166
- "nll_loss": 1.5087509155273438,
167
- "rewards/accuracies": 0.5687500238418579,
168
- "rewards/chosen": -0.09529931843280792,
169
- "rewards/margins": 0.004889167379587889,
170
- "rewards/rejected": -0.1001884788274765,
171
  "step": 90
172
  },
173
  {
174
  "epoch": 0.6389776357827476,
175
- "grad_norm": 0.6337044866684244,
176
- "learning_rate": 2.584192295741087e-06,
177
- "log_odds_chosen": 0.057517312467098236,
178
- "log_odds_ratio": -0.7008931040763855,
179
- "logits/chosen": 32.44559860229492,
180
- "logits/rejected": 32.7852783203125,
181
- "logps/chosen": -0.9386932253837585,
182
- "logps/rejected": -0.9813023805618286,
183
- "loss": 1.5499,
184
- "nll_loss": 1.4884004592895508,
185
- "rewards/accuracies": 0.5562499761581421,
186
- "rewards/chosen": -0.09386932104825974,
187
- "rewards/margins": 0.004260920453816652,
188
- "rewards/rejected": -0.0981302410364151,
189
  "step": 100
190
  },
191
  {
192
  "epoch": 0.7028753993610224,
193
- "grad_norm": 0.658610355052368,
194
- "learning_rate": 2.461313420977536e-06,
195
- "log_odds_chosen": 0.06078052520751953,
196
- "log_odds_ratio": -0.694057822227478,
197
- "logits/chosen": 34.87937927246094,
198
- "logits/rejected": 34.10810470581055,
199
- "logps/chosen": -0.9328826665878296,
200
- "logps/rejected": -0.9768407940864563,
201
- "loss": 1.5083,
202
- "nll_loss": 1.4501432180404663,
203
- "rewards/accuracies": 0.543749988079071,
204
- "rewards/chosen": -0.09328827261924744,
205
- "rewards/margins": 0.004395808558911085,
206
- "rewards/rejected": -0.09768407791852951,
207
  "step": 110
208
  },
209
  {
210
  "epoch": 0.7667731629392971,
211
- "grad_norm": 0.7021451570051785,
212
- "learning_rate": 2.3263454721781537e-06,
213
- "log_odds_chosen": 0.012948527932167053,
214
- "log_odds_ratio": -0.7222181558609009,
215
- "logits/chosen": 32.433719635009766,
216
- "logits/rejected": 32.53590774536133,
217
- "logps/chosen": -0.9092488288879395,
218
- "logps/rejected": -0.9225506782531738,
219
- "loss": 1.4673,
220
- "nll_loss": 1.4067564010620117,
221
- "rewards/accuracies": 0.4937500059604645,
222
- "rewards/chosen": -0.09092487394809723,
223
- "rewards/margins": 0.0013302009319886565,
224
- "rewards/rejected": -0.09225507825613022,
225
  "step": 120
226
  },
227
  {
228
  "epoch": 0.8306709265175719,
229
- "grad_norm": 0.6227085414802311,
230
- "learning_rate": 2.18098574960932e-06,
231
- "log_odds_chosen": 0.08671535551548004,
232
- "log_odds_ratio": -0.6804493069648743,
233
- "logits/chosen": 33.88345718383789,
234
- "logits/rejected": 33.001914978027344,
235
- "logps/chosen": -0.8342811465263367,
236
- "logps/rejected": -0.8885458707809448,
237
- "loss": 1.3999,
238
- "nll_loss": 1.3158425092697144,
239
- "rewards/accuracies": 0.581250011920929,
240
- "rewards/chosen": -0.08342811465263367,
241
- "rewards/margins": 0.005426469258964062,
242
- "rewards/rejected": -0.0888545885682106,
243
  "step": 130
244
  },
245
  {
246
  "epoch": 0.8945686900958466,
247
- "grad_norm": 0.6264864351444569,
248
- "learning_rate": 2.027062236122014e-06,
249
- "log_odds_chosen": 0.04714164510369301,
250
- "log_odds_ratio": -0.6999929547309875,
251
- "logits/chosen": 31.832406997680664,
252
- "logits/rejected": 31.350921630859375,
253
- "logps/chosen": -0.8387459516525269,
254
- "logps/rejected": -0.876442551612854,
255
- "loss": 1.3491,
256
- "nll_loss": 1.2940082550048828,
257
- "rewards/accuracies": 0.534375011920929,
258
- "rewards/chosen": -0.08387459814548492,
259
- "rewards/margins": 0.0037696503568440676,
260
- "rewards/rejected": -0.08764425665140152,
261
  "step": 140
262
  },
263
  {
264
  "epoch": 0.9584664536741214,
265
- "grad_norm": 0.5655314348380543,
266
- "learning_rate": 1.866510609206841e-06,
267
- "log_odds_chosen": 0.044148243963718414,
268
- "log_odds_ratio": -0.7019084692001343,
269
- "logits/chosen": 33.6451416015625,
270
- "logits/rejected": 33.04177474975586,
271
- "logps/chosen": -0.8360700607299805,
272
- "logps/rejected": -0.8655373454093933,
273
- "loss": 1.3242,
274
- "nll_loss": 1.2609388828277588,
275
- "rewards/accuracies": 0.53125,
276
- "rewards/chosen": -0.08360700309276581,
277
- "rewards/margins": 0.0029467367567121983,
278
- "rewards/rejected": -0.08655373752117157,
279
  "step": 150
280
  },
281
  {
282
  "epoch": 1.0223642172523961,
283
- "grad_norm": 0.5571454673635674,
284
- "learning_rate": 1.7013498987264833e-06,
285
- "log_odds_chosen": 0.06723789870738983,
286
- "log_odds_ratio": -0.6888397932052612,
287
- "logits/chosen": 33.27342987060547,
288
- "logits/rejected": 32.30434036254883,
289
- "logps/chosen": -0.7994817495346069,
290
- "logps/rejected": -0.8369095921516418,
291
- "loss": 1.2899,
292
- "nll_loss": 1.211586594581604,
293
- "rewards/accuracies": 0.565625011920929,
294
- "rewards/chosen": -0.07994817197322845,
295
- "rewards/margins": 0.0037427886854857206,
296
- "rewards/rejected": -0.08369095623493195,
297
  "step": 160
298
  },
299
  {
300
  "epoch": 1.0862619808306708,
301
- "grad_norm": 0.5333966997862959,
302
- "learning_rate": 1.5336570964437077e-06,
303
- "log_odds_chosen": 0.038995109498500824,
304
- "log_odds_ratio": -0.6969125866889954,
305
- "logits/chosen": 32.00867462158203,
306
- "logits/rejected": 31.789836883544922,
307
- "logps/chosen": -0.8085753321647644,
308
- "logps/rejected": -0.8331602215766907,
309
- "loss": 1.2863,
310
- "nll_loss": 1.2334020137786865,
311
- "rewards/accuracies": 0.503125011920929,
312
- "rewards/chosen": -0.0808575376868248,
313
- "rewards/margins": 0.0024584876373410225,
314
- "rewards/rejected": -0.08331602811813354,
315
  "step": 170
316
  },
317
  {
318
  "epoch": 1.1501597444089458,
319
- "grad_norm": 0.5317464882215207,
320
- "learning_rate": 1.3655410366448499e-06,
321
- "log_odds_chosen": 0.10975570976734161,
322
- "log_odds_ratio": -0.6656275987625122,
323
- "logits/chosen": 32.76184844970703,
324
- "logits/rejected": 32.081581115722656,
325
- "logps/chosen": -0.7589792013168335,
326
- "logps/rejected": -0.8160526156425476,
327
- "loss": 1.2451,
328
- "nll_loss": 1.1476625204086304,
329
- "rewards/accuracies": 0.6000000238418579,
330
- "rewards/chosen": -0.0758979320526123,
331
- "rewards/margins": 0.0057073310017585754,
332
- "rewards/rejected": -0.08160526305437088,
333
  "step": 180
334
  },
335
  {
336
  "epoch": 1.2140575079872205,
337
- "grad_norm": 0.5186911330716555,
338
- "learning_rate": 1.199115876325091e-06,
339
- "log_odds_chosen": 0.04539128392934799,
340
- "log_odds_ratio": -0.6990563869476318,
341
- "logits/chosen": 33.35085678100586,
342
- "logits/rejected": 33.3908805847168,
343
- "logps/chosen": -0.7675826549530029,
344
- "logps/rejected": -0.7923057675361633,
345
- "loss": 1.2217,
346
- "nll_loss": 1.161102294921875,
347
- "rewards/accuracies": 0.53125,
348
- "rewards/chosen": -0.07675826549530029,
349
- "rewards/margins": 0.0024723131209611893,
350
- "rewards/rejected": -0.07923058420419693,
351
  "step": 190
352
  },
353
  {
354
  "epoch": 1.2779552715654952,
355
- "grad_norm": 0.5411315465201,
356
- "learning_rate": 1.036474508437579e-06,
357
- "log_odds_chosen": 0.07649532705545425,
358
- "log_odds_ratio": -0.6858269572257996,
359
- "logits/chosen": 30.826059341430664,
360
- "logits/rejected": 30.824636459350586,
361
- "logps/chosen": -0.786870002746582,
362
- "logps/rejected": -0.8327391743659973,
363
- "loss": 1.215,
364
- "nll_loss": 1.165637493133545,
365
- "rewards/accuracies": 0.578125,
366
- "rewards/chosen": -0.07868699729442596,
367
- "rewards/margins": 0.004586914554238319,
368
- "rewards/rejected": -0.08327391743659973,
369
  "step": 200
370
  },
371
  {
372
  "epoch": 1.34185303514377,
373
- "grad_norm": 0.5587119516125117,
374
- "learning_rate": 8.796622425502193e-07,
375
- "log_odds_chosen": 0.04922567307949066,
376
- "log_odds_ratio": -0.6928130984306335,
377
- "logits/chosen": 32.06999969482422,
378
- "logits/rejected": 31.100805282592773,
379
- "logps/chosen": -0.760046124458313,
380
- "logps/rejected": -0.7879078388214111,
381
- "loss": 1.2018,
382
- "nll_loss": 1.1223350763320923,
383
- "rewards/accuracies": 0.5218750238418579,
384
- "rewards/chosen": -0.07600460201501846,
385
- "rewards/margins": 0.00278617930598557,
386
- "rewards/rejected": -0.07879078388214111,
387
  "step": 210
388
  },
389
  {
390
  "epoch": 1.4057507987220448,
391
- "grad_norm": 0.5532006875711314,
392
- "learning_rate": 7.30651083891141e-07,
393
- "log_odds_chosen": 0.09409201890230179,
394
- "log_odds_ratio": -0.6719040870666504,
395
- "logits/chosen": 31.775470733642578,
396
- "logits/rejected": 31.210159301757812,
397
- "logps/chosen": -0.7408252954483032,
398
- "logps/rejected": -0.7944933772087097,
399
- "loss": 1.1848,
400
- "nll_loss": 1.1082121133804321,
401
- "rewards/accuracies": 0.6000000238418579,
402
- "rewards/chosen": -0.07408253848552704,
403
- "rewards/margins": 0.005366811528801918,
404
- "rewards/rejected": -0.07944934070110321,
405
  "step": 220
406
  },
407
  {
408
  "epoch": 1.4696485623003195,
409
- "grad_norm": 0.5478964489714558,
410
- "learning_rate": 5.913149342387704e-07,
411
- "log_odds_chosen": 0.022008871659636497,
412
- "log_odds_ratio": -0.7079219818115234,
413
- "logits/chosen": 30.458974838256836,
414
- "logits/rejected": 30.637874603271484,
415
- "logps/chosen": -0.7661860585212708,
416
- "logps/rejected": -0.783053994178772,
417
- "loss": 1.1753,
418
- "nll_loss": 1.1201671361923218,
419
- "rewards/accuracies": 0.5406249761581421,
420
- "rewards/chosen": -0.07661859691143036,
421
- "rewards/margins": 0.00168679840862751,
422
- "rewards/rejected": -0.07830540090799332,
423
  "step": 230
424
  },
425
  {
426
  "epoch": 1.5335463258785942,
427
- "grad_norm": 0.5614636317534148,
428
- "learning_rate": 4.63406026519703e-07,
429
- "log_odds_chosen": 0.12717078626155853,
430
- "log_odds_ratio": -0.6581142544746399,
431
- "logits/chosen": 31.024799346923828,
432
- "logits/rejected": 30.477502822875977,
433
- "logps/chosen": -0.7121531367301941,
434
- "logps/rejected": -0.7798753380775452,
435
- "loss": 1.1701,
436
- "nll_loss": 1.073432207107544,
437
- "rewards/accuracies": 0.596875011920929,
438
- "rewards/chosen": -0.07121531665325165,
439
- "rewards/margins": 0.006772211752831936,
440
- "rewards/rejected": -0.07798753678798676,
441
  "step": 240
442
  },
443
  {
444
  "epoch": 1.5974440894568689,
445
- "grad_norm": 0.5532326330060124,
446
- "learning_rate": 3.4853288946298335e-07,
447
- "log_odds_chosen": 0.031235849484801292,
448
- "log_odds_ratio": -0.7052245140075684,
449
- "logits/chosen": 31.759140014648438,
450
- "logits/rejected": 31.2039737701416,
451
- "logps/chosen": -0.7514272332191467,
452
- "logps/rejected": -0.7712705135345459,
453
- "loss": 1.1628,
454
- "nll_loss": 1.0986436605453491,
455
- "rewards/accuracies": 0.518750011920929,
456
- "rewards/chosen": -0.07514272630214691,
457
- "rewards/margins": 0.0019843343179672956,
458
- "rewards/rejected": -0.07712705433368683,
459
  "step": 250
460
  },
461
  {
462
  "epoch": 1.6613418530351438,
463
- "grad_norm": 0.5579681347407102,
464
- "learning_rate": 2.48140119418046e-07,
465
- "log_odds_chosen": 0.09361070394515991,
466
- "log_odds_ratio": -0.6759673953056335,
467
- "logits/chosen": 30.127365112304688,
468
- "logits/rejected": 29.955150604248047,
469
- "logps/chosen": -0.7377598881721497,
470
- "logps/rejected": -0.789117157459259,
471
- "loss": 1.1595,
472
- "nll_loss": 1.0882426500320435,
473
- "rewards/accuracies": 0.581250011920929,
474
- "rewards/chosen": -0.0737759917974472,
475
- "rewards/margins": 0.0051357327029109,
476
- "rewards/rejected": -0.07891170680522919,
477
  "step": 260
478
  },
479
  {
480
  "epoch": 1.7252396166134185,
481
- "grad_norm": 0.5388500509178578,
482
- "learning_rate": 1.634902137174483e-07,
483
- "log_odds_chosen": 0.06861741840839386,
484
- "log_odds_ratio": -0.6897167563438416,
485
- "logits/chosen": 31.801860809326172,
486
- "logits/rejected": 30.827133178710938,
487
- "logps/chosen": -0.7297223210334778,
488
- "logps/rejected": -0.7675324082374573,
489
- "loss": 1.15,
490
- "nll_loss": 1.091435194015503,
491
- "rewards/accuracies": 0.5562499761581421,
492
- "rewards/chosen": -0.07297223806381226,
493
- "rewards/margins": 0.003781010629609227,
494
- "rewards/rejected": -0.07675323635339737,
495
  "step": 270
496
  },
497
  {
498
  "epoch": 1.7891373801916934,
499
- "grad_norm": 0.5597110218019701,
500
- "learning_rate": 9.564769404039419e-08,
501
- "log_odds_chosen": 0.10912013053894043,
502
- "log_odds_ratio": -0.6706396341323853,
503
- "logits/chosen": 30.64389419555664,
504
- "logits/rejected": 30.014019012451172,
505
- "logps/chosen": -0.7340375185012817,
506
- "logps/rejected": -0.7948423624038696,
507
- "loss": 1.1518,
508
- "nll_loss": 1.0911478996276855,
509
- "rewards/accuracies": 0.590624988079071,
510
- "rewards/chosen": -0.0734037533402443,
511
- "rewards/margins": 0.00608047703281045,
512
- "rewards/rejected": -0.0794842392206192,
513
  "step": 280
514
  },
515
  {
516
  "epoch": 1.8530351437699681,
517
- "grad_norm": 0.5432963139659418,
518
- "learning_rate": 4.546571943496969e-08,
519
- "log_odds_chosen": 0.071634940803051,
520
- "log_odds_ratio": -0.6880494356155396,
521
- "logits/chosen": 31.523670196533203,
522
- "logits/rejected": 31.051345825195312,
523
- "logps/chosen": -0.7210476398468018,
524
- "logps/rejected": -0.7614242434501648,
525
- "loss": 1.1457,
526
- "nll_loss": 1.078147292137146,
527
- "rewards/accuracies": 0.53125,
528
- "rewards/chosen": -0.07210476696491241,
529
- "rewards/margins": 0.0040376619435846806,
530
- "rewards/rejected": -0.07614242285490036,
531
  "step": 290
532
  },
533
  {
534
  "epoch": 1.9169329073482428,
535
- "grad_norm": 0.5550233900085839,
536
- "learning_rate": 1.357535734809795e-08,
537
- "log_odds_chosen": 0.08968226611614227,
538
- "log_odds_ratio": -0.6760362386703491,
539
- "logits/chosen": 30.686954498291016,
540
- "logits/rejected": 29.28194236755371,
541
- "logps/chosen": -0.7248884439468384,
542
- "logps/rejected": -0.7753577828407288,
543
- "loss": 1.1474,
544
- "nll_loss": 1.0696710348129272,
545
- "rewards/accuracies": 0.5531250238418579,
546
- "rewards/chosen": -0.07248884439468384,
547
- "rewards/margins": 0.005046932026743889,
548
- "rewards/rejected": -0.07753578573465347,
549
  "step": 300
550
  },
551
  {
552
  "epoch": 1.9808306709265175,
553
- "grad_norm": 0.49758643306678957,
554
- "learning_rate": 3.77647586240204e-10,
555
- "log_odds_chosen": 0.048125725239515305,
556
- "log_odds_ratio": -0.6949858665466309,
557
- "logits/chosen": 30.8554744720459,
558
- "logits/rejected": 30.65484046936035,
559
- "logps/chosen": -0.7389064431190491,
560
- "logps/rejected": -0.7689411640167236,
561
- "loss": 1.1435,
562
- "nll_loss": 1.0751014947891235,
563
- "rewards/accuracies": 0.534375011920929,
564
- "rewards/chosen": -0.07389064878225327,
565
- "rewards/margins": 0.0030034759547561407,
566
- "rewards/rejected": -0.0768941268324852,
567
  "step": 310
568
  },
569
  {
570
  "epoch": 1.9936102236421727,
571
  "step": 312,
572
  "total_flos": 0.0,
573
- "train_loss": 1.3676127867820935,
574
- "train_runtime": 5524.9246,
575
- "train_samples_per_second": 7.24,
576
  "train_steps_per_second": 0.056
577
  }
578
  ],
 
10
  "log_history": [
11
  {
12
  "epoch": 0.06389776357827476,
13
+ "grad_norm": 0.4242139016276009,
14
+ "learning_rate": 3.125e-05,
15
+ "log_odds_chosen": 0.05670202523469925,
16
+ "log_odds_ratio": -0.6963189840316772,
17
+ "logits/chosen": 33.65892791748047,
18
+ "logits/rejected": 33.56386947631836,
19
+ "logps/chosen": -0.9726333618164062,
20
+ "logps/rejected": -1.0167349576950073,
21
+ "loss": 1.6695,
22
+ "nll_loss": 1.5828170776367188,
23
+ "rewards/accuracies": 0.5406249761581421,
24
+ "rewards/chosen": -0.09726335108280182,
25
+ "rewards/margins": 0.004410145338624716,
26
+ "rewards/rejected": -0.10167349874973297,
27
  "step": 10
28
  },
29
  {
30
  "epoch": 0.12779552715654952,
31
+ "grad_norm": 0.527757575829653,
32
+ "learning_rate": 6.25e-05,
33
+ "log_odds_chosen": 0.060249220579862595,
34
+ "log_odds_ratio": -0.6938394904136658,
35
+ "logits/chosen": 34.7052001953125,
36
+ "logits/rejected": 34.30677032470703,
37
+ "logps/chosen": -0.8907906413078308,
38
+ "logps/rejected": -0.9303584098815918,
39
+ "loss": 1.4944,
40
+ "nll_loss": 1.4315097332000732,
41
+ "rewards/accuracies": 0.546875,
42
+ "rewards/chosen": -0.08907906711101532,
43
+ "rewards/margins": 0.003956770058721304,
44
+ "rewards/rejected": -0.09303583949804306,
45
  "step": 20
46
  },
47
  {
48
  "epoch": 0.19169329073482427,
49
+ "grad_norm": 0.4014606132018809,
50
+ "learning_rate": 9.375e-05,
51
+ "log_odds_chosen": 0.04292842000722885,
52
+ "log_odds_ratio": -0.698261022567749,
53
+ "logits/chosen": 30.987014770507812,
54
+ "logits/rejected": 30.439056396484375,
55
+ "logps/chosen": -0.6581910252571106,
56
+ "logps/rejected": -0.6830389499664307,
57
+ "loss": 1.0721,
58
+ "nll_loss": 1.006190538406372,
59
+ "rewards/accuracies": 0.5375000238418579,
60
+ "rewards/chosen": -0.06581910699605942,
61
+ "rewards/margins": 0.002484795870259404,
62
+ "rewards/rejected": -0.0683038979768753,
63
  "step": 30
64
  },
65
  {
66
  "epoch": 0.25559105431309903,
67
+ "grad_norm": 0.09410150706534143,
68
+ "learning_rate": 9.979871469976196e-05,
69
+ "log_odds_chosen": 0.07667034864425659,
70
+ "log_odds_ratio": -0.6961523294448853,
71
+ "logits/chosen": 28.151453018188477,
72
+ "logits/rejected": 27.650014877319336,
73
+ "logps/chosen": -0.43666666746139526,
74
+ "logps/rejected": -0.46118512749671936,
75
+ "loss": 0.8277,
76
+ "nll_loss": 0.7572250366210938,
77
+ "rewards/accuracies": 0.5687500238418579,
78
+ "rewards/chosen": -0.043666668236255646,
79
+ "rewards/margins": 0.0024518456775695086,
80
+ "rewards/rejected": -0.046118512749671936,
81
  "step": 40
82
  },
83
  {
84
  "epoch": 0.3194888178913738,
85
+ "grad_norm": 0.09768125665941661,
86
+ "learning_rate": 9.898376992116179e-05,
87
+ "log_odds_chosen": 0.06605090945959091,
88
+ "log_odds_ratio": -0.7164387106895447,
89
+ "logits/chosen": 29.047494888305664,
90
+ "logits/rejected": 27.90814781188965,
91
+ "logps/chosen": -0.37932097911834717,
92
+ "logps/rejected": -0.39148497581481934,
93
+ "loss": 0.7778,
94
+ "nll_loss": 0.7044206261634827,
95
+ "rewards/accuracies": 0.5375000238418579,
96
+ "rewards/chosen": -0.03793209791183472,
97
+ "rewards/margins": 0.0012163992505520582,
98
+ "rewards/rejected": -0.03914849832653999,
99
  "step": 50
100
  },
101
  {
102
  "epoch": 0.38338658146964855,
103
+ "grad_norm": 0.07343907441931204,
104
+ "learning_rate": 9.755282581475769e-05,
105
+ "log_odds_chosen": 0.08987905830144882,
106
+ "log_odds_ratio": -0.7063173055648804,
107
+ "logits/chosen": 28.47494125366211,
108
+ "logits/rejected": 28.794668197631836,
109
+ "logps/chosen": -0.3645591139793396,
110
+ "logps/rejected": -0.38652682304382324,
111
+ "loss": 0.7576,
112
+ "nll_loss": 0.6854395866394043,
113
+ "rewards/accuracies": 0.559374988079071,
114
+ "rewards/chosen": -0.0364559181034565,
115
+ "rewards/margins": 0.0021967687644064426,
116
+ "rewards/rejected": -0.038652680814266205,
117
  "step": 60
118
  },
119
  {
120
  "epoch": 0.4472843450479233,
121
+ "grad_norm": 0.06951733417216101,
122
+ "learning_rate": 9.552387733294081e-05,
123
+ "log_odds_chosen": 0.1109728068113327,
124
+ "log_odds_ratio": -0.7004402875900269,
125
+ "logits/chosen": 29.863407135009766,
126
+ "logits/rejected": 30.27783203125,
127
+ "logps/chosen": -0.35166820883750916,
128
+ "logps/rejected": -0.3710673451423645,
129
+ "loss": 0.7621,
130
+ "nll_loss": 0.6829724907875061,
131
+ "rewards/accuracies": 0.559374988079071,
132
+ "rewards/chosen": -0.035166822373867035,
133
+ "rewards/margins": 0.0019399106968194246,
134
+ "rewards/rejected": -0.03710673004388809,
135
  "step": 70
136
  },
137
  {
138
  "epoch": 0.5111821086261981,
139
+ "grad_norm": 0.06675882825106033,
140
+ "learning_rate": 9.292243968009331e-05,
141
+ "log_odds_chosen": 0.22688157856464386,
142
+ "log_odds_ratio": -0.6433924436569214,
143
+ "logits/chosen": 31.850915908813477,
144
+ "logits/rejected": 32.051448822021484,
145
+ "logps/chosen": -0.35137858986854553,
146
+ "logps/rejected": -0.4119800925254822,
147
+ "loss": 0.7548,
148
+ "nll_loss": 0.6953305602073669,
149
+ "rewards/accuracies": 0.640625,
150
+ "rewards/chosen": -0.03513786196708679,
151
+ "rewards/margins": 0.006060150917619467,
152
+ "rewards/rejected": -0.0411980114877224,
153
  "step": 80
154
  },
155
  {
156
  "epoch": 0.5750798722044729,
157
+ "grad_norm": 0.07132675829804414,
158
+ "learning_rate": 8.978122744408906e-05,
159
+ "log_odds_chosen": 0.20013344287872314,
160
+ "log_odds_ratio": -0.6580983400344849,
161
+ "logits/chosen": 32.63883972167969,
162
+ "logits/rejected": 32.50263595581055,
163
+ "logps/chosen": -0.35542625188827515,
164
+ "logps/rejected": -0.41074585914611816,
165
+ "loss": 0.7549,
166
+ "nll_loss": 0.6860114932060242,
167
+ "rewards/accuracies": 0.6312500238418579,
168
+ "rewards/chosen": -0.03554262965917587,
169
+ "rewards/margins": 0.0055319564417004585,
170
+ "rewards/rejected": -0.04107458516955376,
171
  "step": 90
172
  },
173
  {
174
  "epoch": 0.6389776357827476,
175
+ "grad_norm": 0.09149445807314589,
176
+ "learning_rate": 8.613974319136958e-05,
177
+ "log_odds_chosen": 0.2682611346244812,
178
+ "log_odds_ratio": -0.6336508989334106,
179
+ "logits/chosen": 32.11505889892578,
180
+ "logits/rejected": 32.7309455871582,
181
+ "logps/chosen": -0.35972970724105835,
182
+ "logps/rejected": -0.4426427483558655,
183
+ "loss": 0.753,
184
+ "nll_loss": 0.6944054365158081,
185
+ "rewards/accuracies": 0.653124988079071,
186
+ "rewards/chosen": -0.035972971469163895,
187
+ "rewards/margins": 0.008291301317512989,
188
+ "rewards/rejected": -0.04426427185535431,
189
  "step": 100
190
  },
191
  {
192
  "epoch": 0.7028753993610224,
193
+ "grad_norm": 0.08926217777865493,
194
+ "learning_rate": 8.20437806992512e-05,
195
+ "log_odds_chosen": 0.2868707776069641,
196
+ "log_odds_ratio": -0.6233163475990295,
197
+ "logits/chosen": 35.566490173339844,
198
+ "logits/rejected": 34.921634674072266,
199
+ "logps/chosen": -0.3582688868045807,
200
+ "logps/rejected": -0.4581901431083679,
201
+ "loss": 0.744,
202
+ "nll_loss": 0.6832225918769836,
203
+ "rewards/accuracies": 0.6343749761581421,
204
+ "rewards/chosen": -0.03582689166069031,
205
+ "rewards/margins": 0.00999213196337223,
206
+ "rewards/rejected": -0.04581902176141739,
207
  "step": 110
208
  },
209
  {
210
  "epoch": 0.7667731629392971,
211
+ "grad_norm": 0.14837704688840048,
212
+ "learning_rate": 7.754484907260513e-05,
213
+ "log_odds_chosen": 0.30330824851989746,
214
+ "log_odds_ratio": -0.6247963905334473,
215
+ "logits/chosen": 34.01176834106445,
216
+ "logits/rejected": 34.39574432373047,
217
+ "logps/chosen": -0.35339340567588806,
218
+ "logps/rejected": -0.457806259393692,
219
+ "loss": 0.7419,
220
+ "nll_loss": 0.6851586103439331,
221
+ "rewards/accuracies": 0.6187499761581421,
222
+ "rewards/chosen": -0.035339340567588806,
223
+ "rewards/margins": 0.01044128555804491,
224
+ "rewards/rejected": -0.045780621469020844,
225
  "step": 120
226
  },
227
  {
228
  "epoch": 0.8306709265175719,
229
+ "grad_norm": 0.10699968939091851,
230
+ "learning_rate": 7.269952498697734e-05,
231
+ "log_odds_chosen": 0.3560214638710022,
232
+ "log_odds_ratio": -0.6051545143127441,
233
+ "logits/chosen": 36.380924224853516,
234
+ "logits/rejected": 35.524024963378906,
235
+ "logps/chosen": -0.355410635471344,
236
+ "logps/rejected": -0.47948652505874634,
237
+ "loss": 0.7296,
238
+ "nll_loss": 0.6667423844337463,
239
+ "rewards/accuracies": 0.668749988079071,
240
+ "rewards/chosen": -0.03554106876254082,
241
+ "rewards/margins": 0.01240758690983057,
242
+ "rewards/rejected": -0.04794865846633911,
243
  "step": 130
244
  },
245
  {
246
  "epoch": 0.8945686900958466,
247
+ "grad_norm": 0.1720595584929413,
248
+ "learning_rate": 6.756874120406714e-05,
249
+ "log_odds_chosen": 0.5141419172286987,
250
+ "log_odds_ratio": -0.5401668548583984,
251
+ "logits/chosen": 34.641624450683594,
252
+ "logits/rejected": 34.34906768798828,
253
+ "logps/chosen": -0.347829669713974,
254
+ "logps/rejected": -0.5514229536056519,
255
+ "loss": 0.7143,
256
+ "nll_loss": 0.6645579934120178,
257
+ "rewards/accuracies": 0.731249988079071,
258
+ "rewards/chosen": -0.03478296846151352,
259
+ "rewards/margins": 0.020359333604574203,
260
+ "rewards/rejected": -0.05514230206608772,
261
  "step": 140
262
  },
263
  {
264
  "epoch": 0.9584664536741214,
265
+ "grad_norm": 0.12630063541107736,
266
+ "learning_rate": 6.22170203068947e-05,
267
+ "log_odds_chosen": 0.5517296195030212,
268
+ "log_odds_ratio": -0.543639600276947,
269
+ "logits/chosen": 36.98326873779297,
270
+ "logits/rejected": 36.52146530151367,
271
+ "logps/chosen": -0.3562212884426117,
272
+ "logps/rejected": -0.578203022480011,
273
+ "loss": 0.7162,
274
+ "nll_loss": 0.6639058589935303,
275
+ "rewards/accuracies": 0.731249988079071,
276
+ "rewards/chosen": -0.03562213480472565,
277
+ "rewards/margins": 0.02219817042350769,
278
+ "rewards/rejected": -0.05782030150294304,
279
  "step": 150
280
  },
281
  {
282
  "epoch": 1.0223642172523961,
283
+ "grad_norm": 0.1403636887974213,
284
+ "learning_rate": 5.6711663290882776e-05,
285
+ "log_odds_chosen": 0.6598159074783325,
286
+ "log_odds_ratio": -0.5024036169052124,
287
+ "logits/chosen": 36.992469787597656,
288
+ "logits/rejected": 35.901676177978516,
289
+ "logps/chosen": -0.33446019887924194,
290
+ "logps/rejected": -0.5835962891578674,
291
+ "loss": 0.7105,
292
+ "nll_loss": 0.657503604888916,
293
+ "rewards/accuracies": 0.7562500238418579,
294
+ "rewards/chosen": -0.033446017652750015,
295
+ "rewards/margins": 0.0249136071652174,
296
+ "rewards/rejected": -0.05835963040590286,
297
  "step": 160
298
  },
299
  {
300
  "epoch": 1.0862619808306708,
301
+ "grad_norm": 0.14799945317167665,
302
+ "learning_rate": 5.112190321479026e-05,
303
+ "log_odds_chosen": 0.6551031470298767,
304
+ "log_odds_ratio": -0.5045552849769592,
305
+ "logits/chosen": 35.59492874145508,
306
+ "logits/rejected": 35.479576110839844,
307
+ "logps/chosen": -0.36466851830482483,
308
+ "logps/rejected": -0.6384136080741882,
309
+ "loss": 0.7194,
310
+ "nll_loss": 0.6874456405639648,
311
+ "rewards/accuracies": 0.765625,
312
+ "rewards/chosen": -0.03646685555577278,
313
+ "rewards/margins": 0.027374515309929848,
314
+ "rewards/rejected": -0.06384135782718658,
315
  "step": 170
316
  },
317
  {
318
  "epoch": 1.1501597444089458,
319
+ "grad_norm": 0.4144450852693044,
320
+ "learning_rate": 4.551803455482833e-05,
321
+ "log_odds_chosen": 0.8569015264511108,
322
+ "log_odds_ratio": -0.43813252449035645,
323
+ "logits/chosen": 36.478363037109375,
324
+ "logits/rejected": 35.625038146972656,
325
+ "logps/chosen": -0.3283715844154358,
326
+ "logps/rejected": -0.6759421229362488,
327
+ "loss": 0.7011,
328
+ "nll_loss": 0.6358424425125122,
329
+ "rewards/accuracies": 0.809374988079071,
330
+ "rewards/chosen": -0.0328371599316597,
331
+ "rewards/margins": 0.03475705534219742,
332
+ "rewards/rejected": -0.06759421527385712,
333
  "step": 180
334
  },
335
  {
336
  "epoch": 1.2140575079872205,
337
+ "grad_norm": 0.1393106135709619,
338
+ "learning_rate": 3.9970529210836366e-05,
339
+ "log_odds_chosen": 0.7694636583328247,
340
+ "log_odds_ratio": -0.48203492164611816,
341
+ "logits/chosen": 37.65850067138672,
342
+ "logits/rejected": 37.32286071777344,
343
+ "logps/chosen": -0.34162312746047974,
344
+ "logps/rejected": -0.6649552583694458,
345
+ "loss": 0.696,
346
+ "nll_loss": 0.6590245962142944,
347
+ "rewards/accuracies": 0.78125,
348
+ "rewards/chosen": -0.034162312746047974,
349
+ "rewards/margins": 0.032333213835954666,
350
+ "rewards/rejected": -0.06649552285671234,
351
  "step": 190
352
  },
353
  {
354
  "epoch": 1.2779552715654952,
355
+ "grad_norm": 0.18676893862042468,
356
+ "learning_rate": 3.4549150281252636e-05,
357
+ "log_odds_chosen": 0.8638985753059387,
358
+ "log_odds_ratio": -0.4615907073020935,
359
+ "logits/chosen": 34.856327056884766,
360
+ "logits/rejected": 34.688533782958984,
361
+ "logps/chosen": -0.3576621413230896,
362
+ "logps/rejected": -0.74553382396698,
363
+ "loss": 0.6981,
364
+ "nll_loss": 0.6673237085342407,
365
+ "rewards/accuracies": 0.78125,
366
+ "rewards/chosen": -0.03576621413230896,
367
+ "rewards/margins": 0.03878717124462128,
368
+ "rewards/rejected": -0.07455337792634964,
369
  "step": 200
370
  },
371
  {
372
  "epoch": 1.34185303514377,
373
+ "grad_norm": 0.2564156817901977,
374
+ "learning_rate": 2.932207475167398e-05,
375
+ "log_odds_chosen": 0.8325332403182983,
376
+ "log_odds_ratio": -0.45953792333602905,
377
+ "logits/chosen": 36.798072814941406,
378
+ "logits/rejected": 35.09117889404297,
379
+ "logps/chosen": -0.3448064923286438,
380
+ "logps/rejected": -0.7093031406402588,
381
+ "loss": 0.6982,
382
+ "nll_loss": 0.6432278752326965,
383
+ "rewards/accuracies": 0.8062499761581421,
384
+ "rewards/chosen": -0.03448065370321274,
385
+ "rewards/margins": 0.03644966334104538,
386
+ "rewards/rejected": -0.07093031704425812,
387
  "step": 210
388
  },
389
  {
390
  "epoch": 1.4057507987220448,
391
+ "grad_norm": 0.1531688356816753,
392
+ "learning_rate": 2.43550361297047e-05,
393
+ "log_odds_chosen": 1.0281397104263306,
394
+ "log_odds_ratio": -0.4170387387275696,
395
+ "logits/chosen": 35.599212646484375,
396
+ "logits/rejected": 34.3337287902832,
397
+ "logps/chosen": -0.3171806335449219,
398
+ "logps/rejected": -0.7794741988182068,
399
+ "loss": 0.6881,
400
+ "nll_loss": 0.6331400275230408,
401
+ "rewards/accuracies": 0.84375,
402
+ "rewards/chosen": -0.03171805664896965,
403
+ "rewards/margins": 0.04622935503721237,
404
+ "rewards/rejected": -0.07794742286205292,
405
  "step": 220
406
  },
407
  {
408
  "epoch": 1.4696485623003195,
409
+ "grad_norm": 0.13237811516695533,
410
+ "learning_rate": 1.971049780795901e-05,
411
+ "log_odds_chosen": 0.9918270111083984,
412
+ "log_odds_ratio": -0.440161794424057,
413
+ "logits/chosen": 34.80529022216797,
414
+ "logits/rejected": 34.433349609375,
415
+ "logps/chosen": -0.3528688848018646,
416
+ "logps/rejected": -0.8333184123039246,
417
+ "loss": 0.684,
418
+ "nll_loss": 0.653908908367157,
419
+ "rewards/accuracies": 0.8125,
420
+ "rewards/chosen": -0.03528688848018646,
421
+ "rewards/margins": 0.04804495349526405,
422
+ "rewards/rejected": -0.08333183825016022,
423
  "step": 230
424
  },
425
  {
426
  "epoch": 1.5335463258785942,
427
+ "grad_norm": 0.17618952870677607,
428
+ "learning_rate": 1.544686755065677e-05,
429
+ "log_odds_chosen": 1.110446810722351,
430
+ "log_odds_ratio": -0.39400768280029297,
431
+ "logits/chosen": 35.90544891357422,
432
+ "logits/rejected": 34.70677185058594,
433
+ "logps/chosen": -0.3103191554546356,
434
+ "logps/rejected": -0.8037627935409546,
435
+ "loss": 0.6899,
436
+ "nll_loss": 0.6227988004684448,
437
+ "rewards/accuracies": 0.824999988079071,
438
+ "rewards/chosen": -0.03103191778063774,
439
+ "rewards/margins": 0.04934436455368996,
440
+ "rewards/rejected": -0.0803762823343277,
441
  "step": 240
442
  },
443
  {
444
  "epoch": 1.5974440894568689,
445
+ "grad_norm": 0.17643350651397863,
446
+ "learning_rate": 1.1617762982099446e-05,
447
+ "log_odds_chosen": 1.0278904438018799,
448
+ "log_odds_ratio": -0.40579158067703247,
449
+ "logits/chosen": 36.86786651611328,
450
+ "logits/rejected": 35.534324645996094,
451
+ "logps/chosen": -0.34132060408592224,
452
+ "logps/rejected": -0.8093164563179016,
453
+ "loss": 0.6839,
454
+ "nll_loss": 0.6438942551612854,
455
+ "rewards/accuracies": 0.8218749761581421,
456
+ "rewards/chosen": -0.034132059663534164,
457
+ "rewards/margins": 0.04679957777261734,
458
+ "rewards/rejected": -0.0809316486120224,
459
  "step": 250
460
  },
461
  {
462
  "epoch": 1.6613418530351438,
463
+ "grad_norm": 0.20335250536853633,
464
+ "learning_rate": 8.271337313934869e-06,
465
+ "log_odds_chosen": 1.0964257717132568,
466
+ "log_odds_ratio": -0.40502405166625977,
467
+ "logits/chosen": 35.01730728149414,
468
+ "logits/rejected": 33.87272262573242,
469
+ "logps/chosen": -0.35952043533325195,
470
+ "logps/rejected": -0.8931809663772583,
471
+ "loss": 0.6875,
472
+ "nll_loss": 0.6542503833770752,
473
+ "rewards/accuracies": 0.8500000238418579,
474
+ "rewards/chosen": -0.035952042788267136,
475
+ "rewards/margins": 0.0533660463988781,
476
+ "rewards/rejected": -0.08931808918714523,
477
  "step": 260
478
  },
479
  {
480
  "epoch": 1.7252396166134185,
481
+ "grad_norm": 0.1417159132341433,
482
+ "learning_rate": 5.449673790581611e-06,
483
+ "log_odds_chosen": 1.1632859706878662,
484
+ "log_odds_ratio": -0.39327362179756165,
485
+ "logits/chosen": 37.34235763549805,
486
+ "logits/rejected": 35.14636993408203,
487
+ "logps/chosen": -0.3363361358642578,
488
+ "logps/rejected": -0.8959217071533203,
489
+ "loss": 0.6837,
490
+ "nll_loss": 0.6559703350067139,
491
+ "rewards/accuracies": 0.8531249761581421,
492
+ "rewards/chosen": -0.03363361582159996,
493
+ "rewards/margins": 0.055958546698093414,
494
+ "rewards/rejected": -0.08959217369556427,
495
  "step": 270
496
  },
497
  {
498
  "epoch": 1.7891373801916934,
499
+ "grad_norm": 0.39828151397493416,
500
+ "learning_rate": 3.18825646801314e-06,
501
+ "log_odds_chosen": 1.0620207786560059,
502
+ "log_odds_ratio": -0.41589412093162537,
503
+ "logits/chosen": 35.572288513183594,
504
+ "logits/rejected": 34.087669372558594,
505
+ "logps/chosen": -0.3491634726524353,
506
+ "logps/rejected": -0.8419939875602722,
507
+ "loss": 0.6856,
508
+ "nll_loss": 0.6521117091178894,
509
+ "rewards/accuracies": 0.8500000238418579,
510
+ "rewards/chosen": -0.03491634875535965,
511
+ "rewards/margins": 0.04928305745124817,
512
+ "rewards/rejected": -0.08419940620660782,
513
  "step": 280
514
  },
515
  {
516
  "epoch": 1.8530351437699681,
517
+ "grad_norm": 0.17410175622655502,
518
+ "learning_rate": 1.5155239811656563e-06,
519
+ "log_odds_chosen": 1.1792383193969727,
520
+ "log_odds_ratio": -0.3946213126182556,
521
+ "logits/chosen": 36.629905700683594,
522
+ "logits/rejected": 35.32917404174805,
523
+ "logps/chosen": -0.33119866251945496,
524
+ "logps/rejected": -0.899772047996521,
525
+ "loss": 0.6837,
526
+ "nll_loss": 0.6425603628158569,
527
+ "rewards/accuracies": 0.815625011920929,
528
+ "rewards/chosen": -0.033119868487119675,
529
+ "rewards/margins": 0.05685734748840332,
530
+ "rewards/rejected": -0.0899772122502327,
531
  "step": 290
532
  },
533
  {
534
  "epoch": 1.9169329073482428,
535
+ "grad_norm": 0.13374493386406422,
536
+ "learning_rate": 4.52511911603265e-07,
537
+ "log_odds_chosen": 1.0675084590911865,
538
+ "log_odds_ratio": -0.42029696702957153,
539
+ "logits/chosen": 36.18767547607422,
540
+ "logits/rejected": 33.20973205566406,
541
+ "logps/chosen": -0.34383276104927063,
542
+ "logps/rejected": -0.8607551455497742,
543
+ "loss": 0.6844,
544
+ "nll_loss": 0.6432604193687439,
545
+ "rewards/accuracies": 0.809374988079071,
546
+ "rewards/chosen": -0.034383274614810944,
547
+ "rewards/margins": 0.05169224739074707,
548
+ "rewards/rejected": -0.08607552200555801,
549
  "step": 300
550
  },
551
  {
552
  "epoch": 1.9808306709265175,
553
+ "grad_norm": 0.1377772193573712,
554
+ "learning_rate": 1.2588252874673468e-08,
555
+ "log_odds_chosen": 1.0935267210006714,
556
+ "log_odds_ratio": -0.41092976927757263,
557
+ "logits/chosen": 36.207176208496094,
558
+ "logits/rejected": 34.82603073120117,
559
+ "logps/chosen": -0.344975084066391,
560
+ "logps/rejected": -0.8865535855293274,
561
+ "loss": 0.6781,
562
+ "nll_loss": 0.6345787644386292,
563
+ "rewards/accuracies": 0.8062499761581421,
564
+ "rewards/chosen": -0.03449751064181328,
565
+ "rewards/margins": 0.05415785312652588,
566
+ "rewards/rejected": -0.08865536749362946,
567
  "step": 310
568
  },
569
  {
570
  "epoch": 1.9936102236421727,
571
  "step": 312,
572
  "total_flos": 0.0,
573
+ "train_loss": 0.7846963420892373,
574
+ "train_runtime": 5577.3844,
575
+ "train_samples_per_second": 7.172,
576
  "train_steps_per_second": 0.056
577
  }
578
  ],