Hugging Face's logo

wzhouad
/
zephyr-7b-dpo-full

Text Generation
Transformers
Safetensors
llama
trl
dpo
Generated from Trainer
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
wzhouad commited on
Commit
80b254e
•
1 Parent(s): d556c32

Model save

Browse files
Files changed (7) hide show
  1. README.md +13 -13
  2. all_results.json +4 -4
  3. model-00001-of-00003.safetensors +1 -1
  4. model-00002-of-00003.safetensors +1 -1
  5. model-00003-of-00003.safetensors +1 -1
  6. train_results.json +4 -4
  7. trainer_state.json +472 -472
README.md CHANGED
@@ -17,15 +17,15 @@ should probably proofread and complete it, then remove this comment. -->
17
 
18
  This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on an unknown dataset.
19
  It achieves the following results on the evaluation set:
20
- - Loss: 0.0505
21
- - Rewards/chosen: -2.0818
22
- - Rewards/rejected: -2.5892
23
- - Rewards/accuracies: 0.6641
24
- - Rewards/margins: 0.5074
25
- - Logps/rejected: -516.2719
26
- - Logps/chosen: -465.2220
27
- - Logits/rejected: -2.2332
28
- - Logits/chosen: -2.2543
29
 
30
  ## Model description
31
 
@@ -62,10 +62,10 @@ The following hyperparameters were used during training:
62
 
63
  | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
64
  |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
65
- | 0.0668 | 0.21 | 100 | 0.0940 | -1.3335 | -1.5477 | 0.5625 | 0.2141 | -412.1212 | -390.3937 | -2.3209 | -2.3325 |
66
- | 0.0266 | 0.42 | 200 | 0.0473 | -2.2920 | -2.6617 | 0.6094 | 0.3697 | -523.5247 | -486.2414 | -2.2221 | -2.2375 |
67
- | 0.032 | 0.63 | 300 | 0.0485 | -2.1700 | -2.6020 | 0.6523 | 0.4320 | -517.5544 | -474.0444 | -2.2446 | -2.2653 |
68
- | 0.031 | 0.84 | 400 | 0.0505 | -2.0818 | -2.5892 | 0.6641 | 0.5074 | -516.2719 | -465.2220 | -2.2332 | -2.2543 |
69
 
70
 
71
  ### Framework versions
 
17
 
18
  This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on an unknown dataset.
19
  It achieves the following results on the evaluation set:
20
+ - Loss: 0.5694
21
+ - Rewards/chosen: -2.1945
22
+ - Rewards/rejected: -3.1035
23
+ - Rewards/accuracies: 0.7383
24
+ - Rewards/margins: 0.9091
25
+ - Logps/rejected: -567.7068
26
+ - Logps/chosen: -476.4866
27
+ - Logits/rejected: -2.0444
28
+ - Logits/chosen: -2.0871
29
 
30
  ## Model description
31
 
 
62
 
63
  | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
64
  |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
65
+ | 0.4319 | 0.21 | 100 | 0.6499 | -1.4335 | -1.7747 | 0.6328 | 0.3412 | -434.8237 | -400.3860 | -2.2766 | -2.3001 |
66
+ | 0.3523 | 0.42 | 200 | 0.5954 | -2.0216 | -2.7375 | 0.7188 | 0.7159 | -531.1034 | -459.1997 | -2.0559 | -2.0889 |
67
+ | 0.3611 | 0.63 | 300 | 0.5743 | -1.9913 | -2.8389 | 0.7188 | 0.8476 | -541.2482 | -456.1738 | -2.0873 | -2.1299 |
68
+ | 0.3321 | 0.84 | 400 | 0.5694 | -2.1945 | -3.1035 | 0.7383 | 0.9091 | -567.7068 | -476.4866 | -2.0444 | -2.0871 |
69
 
70
 
71
  ### Framework versions
all_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
  "epoch": 1.0,
3
- "train_loss": 0.05930708255229136,
4
- "train_runtime": 4684.9304,
5
  "train_samples": 61133,
6
- "train_samples_per_second": 13.049,
7
- "train_steps_per_second": 0.102
8
  }
 
1
  {
2
  "epoch": 1.0,
3
+ "train_loss": 0.390623705157675,
4
+ "train_runtime": 4433.9669,
5
  "train_samples": 61133,
6
+ "train_samples_per_second": 13.787,
7
+ "train_steps_per_second": 0.108
8
  }
model-00001-of-00003.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0b0824cd0c719689a14fd99ede449eee62ea6e157813a5493bdfe4738a37976f
3
  size 4943162336
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ac7f5951ba58240da40017294a3f524b60736e2052049a80955d47bb73f6dd7
3
  size 4943162336
model-00002-of-00003.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:00246880b93b5e1f79b01925e98e4bb54659f5c721bb2b2544e35033ecb7a4ac
3
  size 4999819336
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:df1d3db1fc5d58035765a9c4d75823cd49fcd1d84f8706c8ad477d7f7c10c74b
3
  size 4999819336
model-00003-of-00003.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:622ce65388b5a0e396d613d85565d63b5da727522597c1c6cc2d40a1a8f9b646
3
  size 4540516344
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:58534372317f409edd57eefc022e856443567520e3ef5a5df9c6638ecbcfe4ef
3
  size 4540516344
train_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
  "epoch": 1.0,
3
- "train_loss": 0.05930708255229136,
4
- "train_runtime": 4684.9304,
5
  "train_samples": 61133,
6
- "train_samples_per_second": 13.049,
7
- "train_steps_per_second": 0.102
8
  }
 
1
  {
2
  "epoch": 1.0,
3
+ "train_loss": 0.390623705157675,
4
+ "train_runtime": 4433.9669,
5
  "train_samples": 61133,
6
+ "train_samples_per_second": 13.787,
7
+ "train_steps_per_second": 0.108
8
  }
trainer_state.json CHANGED
@@ -15,7 +15,7 @@
15
  "logits/rejected": -2.710578203201294,
16
  "logps/chosen": -528.6826782226562,
17
  "logps/rejected": -158.67160034179688,
18
- "loss": 0.2812,
19
  "rewards/accuracies": 0.0,
20
  "rewards/chosen": 0.0,
21
  "rewards/margins": 0.0,
@@ -25,733 +25,733 @@
25
  {
26
  "epoch": 0.02,
27
  "learning_rate": 1.0416666666666667e-07,
28
- "logits/chosen": -2.774479866027832,
29
- "logits/rejected": -2.7448859214782715,
30
- "logps/chosen": -372.5478210449219,
31
- "logps/rejected": -129.08352661132812,
32
- "loss": 0.2777,
33
- "rewards/accuracies": 0.4930555522441864,
34
- "rewards/chosen": 0.00046785647282376885,
35
- "rewards/margins": 0.0009927245555445552,
36
- "rewards/rejected": -0.0005248680827207863,
37
  "step": 10
38
  },
39
  {
40
  "epoch": 0.04,
41
  "learning_rate": 2.0833333333333333e-07,
42
- "logits/chosen": -2.7531769275665283,
43
- "logits/rejected": -2.7297403812408447,
44
- "logps/chosen": -367.28875732421875,
45
- "logps/rejected": -137.4302520751953,
46
- "loss": 0.2781,
47
- "rewards/accuracies": 0.7250000238418579,
48
- "rewards/chosen": 0.010312488302588463,
49
- "rewards/margins": 0.020397081971168518,
50
- "rewards/rejected": -0.01008459459990263,
51
  "step": 20
52
  },
53
  {
54
  "epoch": 0.06,
55
  "learning_rate": 3.1249999999999997e-07,
56
- "logits/chosen": -2.6820380687713623,
57
- "logits/rejected": -2.668060779571533,
58
- "logps/chosen": -345.17755126953125,
59
- "logps/rejected": -137.1610870361328,
60
- "loss": 0.2621,
61
- "rewards/accuracies": 0.731249988079071,
62
- "rewards/chosen": 0.03887051343917847,
63
- "rewards/margins": 0.08981995284557343,
64
- "rewards/rejected": -0.050949446856975555,
65
  "step": 30
66
  },
67
  {
68
  "epoch": 0.08,
69
  "learning_rate": 4.1666666666666667e-07,
70
- "logits/chosen": -2.572816848754883,
71
- "logits/rejected": -2.5619583129882812,
72
- "logps/chosen": -314.2635803222656,
73
- "logps/rejected": -128.72232055664062,
74
- "loss": 0.2349,
75
- "rewards/accuracies": 0.75,
76
- "rewards/chosen": 0.040170345455408096,
77
- "rewards/margins": 0.2692191004753113,
78
- "rewards/rejected": -0.2290487289428711,
79
  "step": 40
80
  },
81
  {
82
  "epoch": 0.1,
83
  "learning_rate": 4.999733114418725e-07,
84
- "logits/chosen": -2.520578145980835,
85
- "logits/rejected": -2.509178638458252,
86
- "logps/chosen": -385.6262512207031,
87
- "logps/rejected": -184.83445739746094,
88
- "loss": 0.1791,
89
- "rewards/accuracies": 0.824999988079071,
90
- "rewards/chosen": -0.023466235026717186,
91
- "rewards/margins": 0.5521816611289978,
92
- "rewards/rejected": -0.5756478905677795,
93
  "step": 50
94
  },
95
  {
96
  "epoch": 0.13,
97
  "learning_rate": 4.990398100856366e-07,
98
- "logits/chosen": -2.4466733932495117,
99
- "logits/rejected": -2.431790828704834,
100
- "logps/chosen": -410.63751220703125,
101
- "logps/rejected": -226.5998077392578,
102
- "loss": 0.1296,
103
- "rewards/accuracies": 0.8062499761581421,
104
- "rewards/chosen": -0.08040066063404083,
105
- "rewards/margins": 0.8076855540275574,
106
- "rewards/rejected": -0.888086199760437,
107
  "step": 60
108
  },
109
  {
110
  "epoch": 0.15,
111
  "learning_rate": 4.967775735898179e-07,
112
- "logits/chosen": -2.389977216720581,
113
- "logits/rejected": -2.365199565887451,
114
- "logps/chosen": -381.20111083984375,
115
- "logps/rejected": -252.31393432617188,
116
- "loss": 0.0962,
117
- "rewards/accuracies": 0.7437499761581421,
118
- "rewards/chosen": -0.2990937829017639,
119
- "rewards/margins": 0.871468722820282,
120
- "rewards/rejected": -1.170562505722046,
121
  "step": 70
122
  },
123
  {
124
  "epoch": 0.17,
125
  "learning_rate": 4.931986719649298e-07,
126
- "logits/chosen": -2.3227975368499756,
127
- "logits/rejected": -2.2759556770324707,
128
- "logps/chosen": -366.7736511230469,
129
- "logps/rejected": -246.578857421875,
130
- "loss": 0.0976,
131
- "rewards/accuracies": 0.7437499761581421,
132
- "rewards/chosen": -0.353910893201828,
133
- "rewards/margins": 0.8951172828674316,
134
- "rewards/rejected": -1.249027967453003,
135
  "step": 80
136
  },
137
  {
138
  "epoch": 0.19,
139
  "learning_rate": 4.883222001996351e-07,
140
- "logits/chosen": -2.360783576965332,
141
- "logits/rejected": -2.312258243560791,
142
- "logps/chosen": -430.5379333496094,
143
- "logps/rejected": -304.5018615722656,
144
- "loss": 0.0687,
145
- "rewards/accuracies": 0.7437499761581421,
146
- "rewards/chosen": -0.5686448812484741,
147
- "rewards/margins": 1.149700403213501,
148
- "rewards/rejected": -1.7183454036712646,
149
  "step": 90
150
  },
151
  {
152
  "epoch": 0.21,
153
  "learning_rate": 4.821741763807186e-07,
154
- "logits/chosen": -2.3213393688201904,
155
- "logits/rejected": -2.272404193878174,
156
- "logps/chosen": -410.54107666015625,
157
- "logps/rejected": -300.12994384765625,
158
- "loss": 0.0668,
159
  "rewards/accuracies": 0.8125,
160
- "rewards/chosen": -0.6966512799263,
161
- "rewards/margins": 1.0948442220687866,
162
- "rewards/rejected": -1.791495680809021,
163
  "step": 100
164
  },
165
  {
166
  "epoch": 0.21,
167
- "eval_logits/chosen": -2.3324644565582275,
168
- "eval_logits/rejected": -2.3208560943603516,
169
- "eval_logps/chosen": -390.3937072753906,
170
- "eval_logps/rejected": -412.1212158203125,
171
- "eval_loss": 0.09404966980218887,
172
- "eval_rewards/accuracies": 0.5625,
173
- "eval_rewards/chosen": -1.3335407972335815,
174
- "eval_rewards/margins": 0.21413864195346832,
175
- "eval_rewards/rejected": -1.5476794242858887,
176
- "eval_runtime": 56.479,
177
- "eval_samples_per_second": 35.411,
178
- "eval_steps_per_second": 0.567,
179
  "step": 100
180
  },
181
  {
182
  "epoch": 0.23,
183
  "learning_rate": 4.747874028753375e-07,
184
- "logits/chosen": -2.2716221809387207,
185
- "logits/rejected": -2.219805955886841,
186
- "logps/chosen": -498.3235778808594,
187
- "logps/rejected": -360.07269287109375,
188
- "loss": 0.0486,
189
- "rewards/accuracies": 0.793749988079071,
190
- "rewards/chosen": -0.6510122418403625,
191
- "rewards/margins": 1.5708351135253906,
192
- "rewards/rejected": -2.2218470573425293,
193
  "step": 110
194
  },
195
  {
196
  "epoch": 0.25,
197
  "learning_rate": 4.662012913161997e-07,
198
- "logits/chosen": -2.271559000015259,
199
- "logits/rejected": -2.2487940788269043,
200
- "logps/chosen": -465.41552734375,
201
- "logps/rejected": -385.037841796875,
202
- "loss": 0.036,
203
- "rewards/accuracies": 0.75,
204
- "rewards/chosen": -1.1209948062896729,
205
- "rewards/margins": 1.4643309116363525,
206
- "rewards/rejected": -2.5853257179260254,
207
  "step": 120
208
  },
209
  {
210
  "epoch": 0.27,
211
  "learning_rate": 4.5646165232345103e-07,
212
- "logits/chosen": -2.2621378898620605,
213
- "logits/rejected": -2.1918931007385254,
214
- "logps/chosen": -492.083984375,
215
- "logps/rejected": -379.6644592285156,
216
- "loss": 0.0343,
217
- "rewards/accuracies": 0.75,
218
- "rewards/chosen": -1.2130827903747559,
219
- "rewards/margins": 1.4191052913665771,
220
- "rewards/rejected": -2.632188320159912,
221
  "step": 130
222
  },
223
  {
224
  "epoch": 0.29,
225
  "learning_rate": 4.456204510851956e-07,
226
- "logits/chosen": -2.3005383014678955,
227
- "logits/rejected": -2.2605350017547607,
228
- "logps/chosen": -487.96099853515625,
229
- "logps/rejected": -402.94525146484375,
230
- "loss": 0.0352,
231
- "rewards/accuracies": 0.793749988079071,
232
- "rewards/chosen": -1.0416280031204224,
233
- "rewards/margins": 1.6206719875335693,
234
- "rewards/rejected": -2.662299871444702,
235
  "step": 140
236
  },
237
  {
238
  "epoch": 0.31,
239
  "learning_rate": 4.337355301007335e-07,
240
- "logits/chosen": -2.292916774749756,
241
- "logits/rejected": -2.2345409393310547,
242
- "logps/chosen": -507.41107177734375,
243
- "logps/rejected": -417.5094299316406,
244
- "loss": 0.0363,
245
- "rewards/accuracies": 0.75,
246
- "rewards/chosen": -1.2272781133651733,
247
- "rewards/margins": 1.589632272720337,
248
- "rewards/rejected": -2.8169105052948,
249
  "step": 150
250
  },
251
  {
252
  "epoch": 0.33,
253
  "learning_rate": 4.2087030056579986e-07,
254
- "logits/chosen": -2.272618293762207,
255
- "logits/rejected": -2.2063608169555664,
256
- "logps/chosen": -501.4068908691406,
257
- "logps/rejected": -405.09124755859375,
258
- "loss": 0.0361,
259
- "rewards/accuracies": 0.75,
260
- "rewards/chosen": -1.1854524612426758,
261
- "rewards/margins": 1.545490026473999,
262
- "rewards/rejected": -2.730942487716675,
263
  "step": 160
264
  },
265
  {
266
  "epoch": 0.36,
267
  "learning_rate": 4.070934040463998e-07,
268
- "logits/chosen": -2.302668333053589,
269
- "logits/rejected": -2.2190022468566895,
270
- "logps/chosen": -516.4844970703125,
271
- "logps/rejected": -413.7950134277344,
272
- "loss": 0.0335,
273
- "rewards/accuracies": 0.8187500238418579,
274
- "rewards/chosen": -0.9041190147399902,
275
- "rewards/margins": 1.9239938259124756,
276
- "rewards/rejected": -2.828113079071045,
277
  "step": 170
278
  },
279
  {
280
  "epoch": 0.38,
281
  "learning_rate": 3.9247834624635404e-07,
282
- "logits/chosen": -2.283273458480835,
283
- "logits/rejected": -2.216068983078003,
284
- "logps/chosen": -504.46728515625,
285
- "logps/rejected": -436.26104736328125,
286
- "loss": 0.0292,
287
- "rewards/accuracies": 0.7875000238418579,
288
- "rewards/chosen": -1.2864400148391724,
289
- "rewards/margins": 1.6309360265731812,
290
- "rewards/rejected": -2.9173760414123535,
291
  "step": 180
292
  },
293
  {
294
  "epoch": 0.4,
295
  "learning_rate": 3.7710310482256523e-07,
296
- "logits/chosen": -2.2281765937805176,
297
- "logits/rejected": -2.1652708053588867,
298
- "logps/chosen": -598.4683227539062,
299
- "logps/rejected": -481.2908630371094,
300
- "loss": 0.0286,
301
- "rewards/accuracies": 0.75,
302
- "rewards/chosen": -1.5822023153305054,
303
- "rewards/margins": 1.843756914138794,
304
- "rewards/rejected": -3.425959348678589,
305
  "step": 190
306
  },
307
  {
308
  "epoch": 0.42,
309
  "learning_rate": 3.610497133404795e-07,
310
- "logits/chosen": -2.23673152923584,
311
- "logits/rejected": -2.1812241077423096,
312
- "logps/chosen": -581.267822265625,
313
- "logps/rejected": -461.84765625,
314
- "loss": 0.0266,
315
- "rewards/accuracies": 0.7250000238418579,
316
- "rewards/chosen": -1.6656383275985718,
317
- "rewards/margins": 1.6236919164657593,
318
- "rewards/rejected": -3.2893307209014893,
319
  "step": 200
320
  },
321
  {
322
  "epoch": 0.42,
323
- "eval_logits/chosen": -2.2374839782714844,
324
- "eval_logits/rejected": -2.222134828567505,
325
- "eval_logps/chosen": -486.24139404296875,
326
- "eval_logps/rejected": -523.5247192382812,
327
- "eval_loss": 0.047261886298656464,
328
- "eval_rewards/accuracies": 0.609375,
329
- "eval_rewards/chosen": -2.2920174598693848,
330
- "eval_rewards/margins": 0.36969709396362305,
331
- "eval_rewards/rejected": -2.661714553833008,
332
- "eval_runtime": 53.2196,
333
- "eval_samples_per_second": 37.58,
334
- "eval_steps_per_second": 0.601,
335
  "step": 200
336
  },
337
  {
338
  "epoch": 0.44,
339
  "learning_rate": 3.4440382358952115e-07,
340
- "logits/chosen": -2.2590956687927246,
341
- "logits/rejected": -2.193190336227417,
342
- "logps/chosen": -519.4159545898438,
343
- "logps/rejected": -433.34014892578125,
344
- "loss": 0.0326,
345
- "rewards/accuracies": 0.762499988079071,
346
- "rewards/chosen": -1.4323736429214478,
347
- "rewards/margins": 1.4586365222930908,
348
- "rewards/rejected": -2.891010284423828,
349
  "step": 210
350
  },
351
  {
352
  "epoch": 0.46,
353
  "learning_rate": 3.272542485937368e-07,
354
- "logits/chosen": -2.2952065467834473,
355
- "logits/rejected": -2.192499876022339,
356
- "logps/chosen": -534.5960693359375,
357
- "logps/rejected": -413.0035705566406,
358
- "loss": 0.0309,
359
- "rewards/accuracies": 0.800000011920929,
360
- "rewards/chosen": -1.0914585590362549,
361
- "rewards/margins": 1.7417411804199219,
362
- "rewards/rejected": -2.833199977874756,
363
  "step": 220
364
  },
365
  {
366
  "epoch": 0.48,
367
  "learning_rate": 3.096924887558854e-07,
368
- "logits/chosen": -2.2403852939605713,
369
- "logits/rejected": -2.1860156059265137,
370
- "logps/chosen": -512.5582275390625,
371
- "logps/rejected": -427.8997497558594,
372
- "loss": 0.0315,
373
  "rewards/accuracies": 0.8187500238418579,
374
- "rewards/chosen": -1.0830895900726318,
375
- "rewards/margins": 1.6780681610107422,
376
- "rewards/rejected": -2.761157989501953,
377
  "step": 230
378
  },
379
  {
380
  "epoch": 0.5,
381
  "learning_rate": 2.9181224366319943e-07,
382
- "logits/chosen": -2.2761881351470947,
383
- "logits/rejected": -2.2212231159210205,
384
- "logps/chosen": -519.4483032226562,
385
- "logps/rejected": -413.8775939941406,
386
- "loss": 0.0312,
387
- "rewards/accuracies": 0.7250000238418579,
388
- "rewards/chosen": -1.1410013437271118,
389
- "rewards/margins": 1.6820824146270752,
390
- "rewards/rejected": -2.8230838775634766,
391
  "step": 240
392
  },
393
  {
394
  "epoch": 0.52,
395
  "learning_rate": 2.7370891215954565e-07,
396
- "logits/chosen": -2.2177672386169434,
397
- "logits/rejected": -2.1556167602539062,
398
- "logps/chosen": -504.2252502441406,
399
- "logps/rejected": -445.9237365722656,
400
- "loss": 0.03,
401
- "rewards/accuracies": 0.75,
402
- "rewards/chosen": -1.4269623756408691,
403
- "rewards/margins": 1.759594202041626,
404
- "rewards/rejected": -3.186556339263916,
405
  "step": 250
406
  },
407
  {
408
  "epoch": 0.54,
409
  "learning_rate": 2.55479083351317e-07,
410
- "logits/chosen": -2.27811861038208,
411
- "logits/rejected": -2.2151107788085938,
412
- "logps/chosen": -525.9552612304688,
413
- "logps/rejected": -450.84210205078125,
414
- "loss": 0.0257,
415
- "rewards/accuracies": 0.768750011920929,
416
- "rewards/chosen": -1.4494380950927734,
417
- "rewards/margins": 1.6489498615264893,
418
- "rewards/rejected": -3.0983879566192627,
419
  "step": 260
420
  },
421
  {
422
  "epoch": 0.56,
423
  "learning_rate": 2.3722002126275822e-07,
424
- "logits/chosen": -2.254257917404175,
425
- "logits/rejected": -2.1944077014923096,
426
- "logps/chosen": -481.66241455078125,
427
- "logps/rejected": -414.86260986328125,
428
- "loss": 0.0298,
429
- "rewards/accuracies": 0.75,
430
- "rewards/chosen": -1.3056637048721313,
431
- "rewards/margins": 1.7019319534301758,
432
- "rewards/rejected": -3.0075955390930176,
433
  "step": 270
434
  },
435
  {
436
  "epoch": 0.59,
437
  "learning_rate": 2.19029145890313e-07,
438
- "logits/chosen": -2.289529800415039,
439
- "logits/rejected": -2.212759017944336,
440
- "logps/chosen": -460.890380859375,
441
- "logps/rejected": -385.2060852050781,
442
- "loss": 0.0347,
443
- "rewards/accuracies": 0.8187500238418579,
444
- "rewards/chosen": -1.0682761669158936,
445
- "rewards/margins": 1.5219484567642212,
446
- "rewards/rejected": -2.5902247428894043,
447
  "step": 280
448
  },
449
  {
450
  "epoch": 0.61,
451
  "learning_rate": 2.0100351342479216e-07,
452
- "logits/chosen": -2.257993221282959,
453
- "logits/rejected": -2.18015718460083,
454
- "logps/chosen": -504.76708984375,
455
- "logps/rejected": -437.1534729003906,
456
- "loss": 0.0374,
457
- "rewards/accuracies": 0.793749988079071,
458
- "rewards/chosen": -0.9555547833442688,
459
- "rewards/margins": 1.9995323419570923,
460
- "rewards/rejected": -2.9550869464874268,
461
  "step": 290
462
  },
463
  {
464
  "epoch": 0.63,
465
  "learning_rate": 1.8323929841460178e-07,
466
- "logits/chosen": -2.2877249717712402,
467
- "logits/rejected": -2.2307815551757812,
468
- "logps/chosen": -466.857421875,
469
- "logps/rejected": -423.18994140625,
470
- "loss": 0.032,
471
  "rewards/accuracies": 0.824999988079071,
472
- "rewards/chosen": -1.2322758436203003,
473
- "rewards/margins": 1.7422376871109009,
474
- "rewards/rejected": -2.974513530731201,
475
  "step": 300
476
  },
477
  {
478
  "epoch": 0.63,
479
- "eval_logits/chosen": -2.2653496265411377,
480
- "eval_logits/rejected": -2.244642734527588,
481
- "eval_logps/chosen": -474.0444030761719,
482
- "eval_logps/rejected": -517.554443359375,
483
- "eval_loss": 0.0485130250453949,
484
- "eval_rewards/accuracies": 0.65234375,
485
- "eval_rewards/chosen": -2.1700472831726074,
486
- "eval_rewards/margins": 0.43196412920951843,
487
- "eval_rewards/rejected": -2.6020116806030273,
488
- "eval_runtime": 53.1461,
489
- "eval_samples_per_second": 37.632,
490
- "eval_steps_per_second": 0.602,
491
  "step": 300
492
  },
493
  {
494
  "epoch": 0.65,
495
  "learning_rate": 1.6583128063291573e-07,
496
- "logits/chosen": -2.2729740142822266,
497
- "logits/rejected": -2.205904483795166,
498
- "logps/chosen": -542.3432006835938,
499
- "logps/rejected": -463.81005859375,
500
- "loss": 0.0232,
501
- "rewards/accuracies": 0.84375,
502
- "rewards/chosen": -1.397203803062439,
503
- "rewards/margins": 1.827301025390625,
504
- "rewards/rejected": -3.2245049476623535,
505
  "step": 310
506
  },
507
  {
508
  "epoch": 0.67,
509
  "learning_rate": 1.488723393865766e-07,
510
- "logits/chosen": -2.15993595123291,
511
- "logits/rejected": -2.093684673309326,
512
- "logps/chosen": -533.7662963867188,
513
- "logps/rejected": -466.84649658203125,
514
- "loss": 0.0231,
515
- "rewards/accuracies": 0.75,
516
- "rewards/chosen": -1.5950443744659424,
517
- "rewards/margins": 1.6558326482772827,
518
- "rewards/rejected": -3.2508769035339355,
519
  "step": 320
520
  },
521
  {
522
  "epoch": 0.69,
523
  "learning_rate": 1.3245295796480788e-07,
524
- "logits/chosen": -2.320767879486084,
525
- "logits/rejected": -2.234102249145508,
526
- "logps/chosen": -549.3599853515625,
527
- "logps/rejected": -460.04962158203125,
528
- "loss": 0.0205,
529
- "rewards/accuracies": 0.8187500238418579,
530
- "rewards/chosen": -1.3181811571121216,
531
- "rewards/margins": 2.0199782848358154,
532
- "rewards/rejected": -3.3381595611572266,
533
  "step": 330
534
  },
535
  {
536
  "epoch": 0.71,
537
  "learning_rate": 1.1666074087171627e-07,
538
- "logits/chosen": -2.2680351734161377,
539
- "logits/rejected": -2.188194751739502,
540
- "logps/chosen": -516.3128662109375,
541
- "logps/rejected": -424.4224548339844,
542
- "loss": 0.0281,
543
- "rewards/accuracies": 0.768750011920929,
544
- "rewards/chosen": -1.3223750591278076,
545
- "rewards/margins": 1.6930872201919556,
546
- "rewards/rejected": -3.0154619216918945,
547
  "step": 340
548
  },
549
  {
550
  "epoch": 0.73,
551
  "learning_rate": 1.0157994641835734e-07,
552
- "logits/chosen": -2.258295774459839,
553
- "logits/rejected": -2.1769368648529053,
554
- "logps/chosen": -512.5427856445312,
555
- "logps/rejected": -446.8404846191406,
556
- "loss": 0.0305,
557
- "rewards/accuracies": 0.78125,
558
- "rewards/chosen": -1.2598600387573242,
559
- "rewards/margins": 1.8013684749603271,
560
- "rewards/rejected": -3.0612287521362305,
561
  "step": 350
562
  },
563
  {
564
  "epoch": 0.75,
565
  "learning_rate": 8.729103716819111e-08,
566
- "logits/chosen": -2.295018196105957,
567
- "logits/rejected": -2.2004973888397217,
568
- "logps/chosen": -546.9755859375,
569
- "logps/rejected": -439.086669921875,
570
- "loss": 0.0263,
571
- "rewards/accuracies": 0.8062499761581421,
572
- "rewards/chosen": -1.1137566566467285,
573
- "rewards/margins": 2.123899221420288,
574
- "rewards/rejected": -3.2376561164855957,
575
  "step": 360
576
  },
577
  {
578
  "epoch": 0.77,
579
  "learning_rate": 7.387025063449081e-08,
580
- "logits/chosen": -2.259816884994507,
581
- "logits/rejected": -2.1902153491973877,
582
- "logps/chosen": -499.126220703125,
583
- "logps/rejected": -439.12835693359375,
584
- "loss": 0.0238,
585
- "rewards/accuracies": 0.7749999761581421,
586
- "rewards/chosen": -1.3282043933868408,
587
- "rewards/margins": 1.8397128582000732,
588
- "rewards/rejected": -3.167917490005493,
589
  "step": 370
590
  },
591
  {
592
  "epoch": 0.79,
593
  "learning_rate": 6.138919252022435e-08,
594
- "logits/chosen": -2.247947931289673,
595
- "logits/rejected": -2.1640267372131348,
596
- "logps/chosen": -516.7667846679688,
597
- "logps/rejected": -443.81805419921875,
598
- "loss": 0.0275,
599
- "rewards/accuracies": 0.8187500238418579,
600
- "rewards/chosen": -1.3331043720245361,
601
- "rewards/margins": 2.0034899711608887,
602
- "rewards/rejected": -3.3365941047668457,
603
  "step": 380
604
  },
605
  {
606
  "epoch": 0.82,
607
  "learning_rate": 4.991445467064689e-08,
608
- "logits/chosen": -2.2855653762817383,
609
- "logits/rejected": -2.1994881629943848,
610
- "logps/chosen": -521.7694091796875,
611
- "logps/rejected": -434.9959411621094,
612
- "loss": 0.0275,
613
  "rewards/accuracies": 0.824999988079071,
614
- "rewards/chosen": -1.1784518957138062,
615
- "rewards/margins": 1.7896277904510498,
616
- "rewards/rejected": -2.9680798053741455,
617
  "step": 390
618
  },
619
  {
620
  "epoch": 0.84,
621
  "learning_rate": 3.9507259776993954e-08,
622
- "logits/chosen": -2.2101545333862305,
623
- "logits/rejected": -2.121650218963623,
624
- "logps/chosen": -535.8075561523438,
625
- "logps/rejected": -441.65594482421875,
626
- "loss": 0.031,
627
- "rewards/accuracies": 0.8187500238418579,
628
- "rewards/chosen": -1.1732594966888428,
629
- "rewards/margins": 1.917687177658081,
630
- "rewards/rejected": -3.090946674346924,
631
  "step": 400
632
  },
633
  {
634
  "epoch": 0.84,
635
- "eval_logits/chosen": -2.254281520843506,
636
- "eval_logits/rejected": -2.2331910133361816,
637
- "eval_logps/chosen": -465.22198486328125,
638
- "eval_logps/rejected": -516.2718505859375,
639
- "eval_loss": 0.050464928150177,
640
- "eval_rewards/accuracies": 0.6640625,
641
- "eval_rewards/chosen": -2.0818235874176025,
642
- "eval_rewards/margins": 0.5073620080947876,
643
- "eval_rewards/rejected": -2.5891854763031006,
644
- "eval_runtime": 53.2766,
645
- "eval_samples_per_second": 37.54,
646
- "eval_steps_per_second": 0.601,
647
  "step": 400
648
  },
649
  {
650
  "epoch": 0.86,
651
  "learning_rate": 3.022313472693447e-08,
652
- "logits/chosen": -2.2890207767486572,
653
- "logits/rejected": -2.1932108402252197,
654
- "logps/chosen": -567.7844848632812,
655
- "logps/rejected": -450.72021484375,
656
- "loss": 0.0286,
657
- "rewards/accuracies": 0.8500000238418579,
658
- "rewards/chosen": -1.1856052875518799,
659
- "rewards/margins": 1.9726451635360718,
660
- "rewards/rejected": -3.158250331878662,
661
  "step": 410
662
  },
663
  {
664
  "epoch": 0.88,
665
  "learning_rate": 2.2111614344599684e-08,
666
- "logits/chosen": -2.1596083641052246,
667
- "logits/rejected": -2.1195709705352783,
668
- "logps/chosen": -500.33770751953125,
669
- "logps/rejected": -440.6878967285156,
670
- "loss": 0.031,
671
- "rewards/accuracies": 0.793749988079071,
672
- "rewards/chosen": -1.4034124612808228,
673
- "rewards/margins": 1.7072322368621826,
674
- "rewards/rejected": -3.110644817352295,
675
  "step": 420
676
  },
677
  {
678
  "epoch": 0.9,
679
  "learning_rate": 1.521597710086439e-08,
680
- "logits/chosen": -2.2506420612335205,
681
- "logits/rejected": -2.1721880435943604,
682
- "logps/chosen": -486.343994140625,
683
- "logps/rejected": -427.8528747558594,
684
- "loss": 0.0246,
685
- "rewards/accuracies": 0.7749999761581421,
686
- "rewards/chosen": -1.4625972509384155,
687
- "rewards/margins": 1.6511366367340088,
688
- "rewards/rejected": -3.1137337684631348,
689
  "step": 430
690
  },
691
  {
692
  "epoch": 0.92,
693
  "learning_rate": 9.57301420397924e-09,
694
- "logits/chosen": -2.2469868659973145,
695
- "logits/rejected": -2.18215012550354,
696
- "logps/chosen": -516.0849609375,
697
- "logps/rejected": -474.19281005859375,
698
- "loss": 0.0247,
699
- "rewards/accuracies": 0.84375,
700
- "rewards/chosen": -1.4236122369766235,
701
- "rewards/margins": 1.9111779928207397,
702
- "rewards/rejected": -3.3347904682159424,
703
  "step": 440
704
  },
705
  {
706
  "epoch": 0.94,
707
  "learning_rate": 5.212833302556258e-09,
708
- "logits/chosen": -2.247234582901001,
709
- "logits/rejected": -2.169130563735962,
710
- "logps/chosen": -509.833251953125,
711
- "logps/rejected": -449.7959899902344,
712
- "loss": 0.0235,
713
- "rewards/accuracies": 0.824999988079071,
714
- "rewards/chosen": -1.250127911567688,
715
- "rewards/margins": 1.9727777242660522,
716
- "rewards/rejected": -3.2229056358337402,
717
  "step": 450
718
  },
719
  {
720
  "epoch": 0.96,
721
  "learning_rate": 2.158697848236607e-09,
722
- "logits/chosen": -2.219299793243408,
723
- "logits/rejected": -2.157402276992798,
724
- "logps/chosen": -526.9745483398438,
725
- "logps/rejected": -457.9435119628906,
726
- "loss": 0.0296,
727
- "rewards/accuracies": 0.8062499761581421,
728
- "rewards/chosen": -1.3694864511489868,
729
- "rewards/margins": 1.7717416286468506,
730
- "rewards/rejected": -3.141228199005127,
731
  "step": 460
732
  },
733
  {
734
  "epoch": 0.98,
735
  "learning_rate": 4.269029751107489e-10,
736
- "logits/chosen": -2.2496750354766846,
737
- "logits/rejected": -2.155200242996216,
738
- "logps/chosen": -525.7666015625,
739
- "logps/rejected": -456.72076416015625,
740
- "loss": 0.0282,
741
- "rewards/accuracies": 0.84375,
742
- "rewards/chosen": -1.1328816413879395,
743
- "rewards/margins": 2.134422779083252,
744
- "rewards/rejected": -3.267303943634033,
745
  "step": 470
746
  },
747
  {
748
  "epoch": 1.0,
749
  "step": 478,
750
  "total_flos": 0.0,
751
- "train_loss": 0.05930708255229136,
752
- "train_runtime": 4684.9304,
753
- "train_samples_per_second": 13.049,
754
- "train_steps_per_second": 0.102
755
  }
756
  ],
757
  "logging_steps": 10,
 
15
  "logits/rejected": -2.710578203201294,
16
  "logps/chosen": -528.6826782226562,
17
  "logps/rejected": -158.67160034179688,
18
+ "loss": 0.6931,
19
  "rewards/accuracies": 0.0,
20
  "rewards/chosen": 0.0,
21
  "rewards/margins": 0.0,
 
25
  {
26
  "epoch": 0.02,
27
  "learning_rate": 1.0416666666666667e-07,
28
+ "logits/chosen": -2.7741899490356445,
29
+ "logits/rejected": -2.7445850372314453,
30
+ "logps/chosen": -372.50982666015625,
31
+ "logps/rejected": -129.0567169189453,
32
+ "loss": 0.6928,
33
+ "rewards/accuracies": 0.5486111044883728,
34
+ "rewards/chosen": 0.0008477116352878511,
35
+ "rewards/margins": 0.0011047045700252056,
36
+ "rewards/rejected": -0.00025699284742586315,
37
  "step": 10
38
  },
39
  {
40
  "epoch": 0.04,
41
  "learning_rate": 2.0833333333333333e-07,
42
+ "logits/chosen": -2.7527709007263184,
43
+ "logits/rejected": -2.729275703430176,
44
+ "logps/chosen": -367.230712890625,
45
+ "logps/rejected": -137.4402313232422,
46
+ "loss": 0.683,
47
+ "rewards/accuracies": 0.762499988079071,
48
+ "rewards/chosen": 0.010893024504184723,
49
+ "rewards/margins": 0.02107730321586132,
50
+ "rewards/rejected": -0.010184278711676598,
51
  "step": 20
52
  },
53
  {
54
  "epoch": 0.06,
55
  "learning_rate": 3.1249999999999997e-07,
56
+ "logits/chosen": -2.6821541786193848,
57
+ "logits/rejected": -2.6682209968566895,
58
+ "logps/chosen": -345.0670166015625,
59
+ "logps/rejected": -137.2961883544922,
60
+ "loss": 0.6356,
61
+ "rewards/accuracies": 0.7562500238418579,
62
+ "rewards/chosen": 0.0399758443236351,
63
+ "rewards/margins": 0.09227657318115234,
64
+ "rewards/rejected": -0.05230072885751724,
65
  "step": 30
66
  },
67
  {
68
  "epoch": 0.08,
69
  "learning_rate": 4.1666666666666667e-07,
70
+ "logits/chosen": -2.5740535259246826,
71
+ "logits/rejected": -2.5633654594421387,
72
+ "logps/chosen": -314.03118896484375,
73
+ "logps/rejected": -128.52047729492188,
74
+ "loss": 0.5788,
75
+ "rewards/accuracies": 0.737500011920929,
76
+ "rewards/chosen": 0.0424940325319767,
77
+ "rewards/margins": 0.26952430605888367,
78
+ "rewards/rejected": -0.22703024744987488,
79
  "step": 40
80
  },
81
  {
82
  "epoch": 0.1,
83
  "learning_rate": 4.999733114418725e-07,
84
+ "logits/chosen": -2.5243449211120605,
85
+ "logits/rejected": -2.512850046157837,
86
+ "logps/chosen": -384.26727294921875,
87
+ "logps/rejected": -183.08828735351562,
88
+ "loss": 0.5176,
89
+ "rewards/accuracies": 0.831250011920929,
90
+ "rewards/chosen": -0.009876979514956474,
91
+ "rewards/margins": 0.5483093857765198,
92
+ "rewards/rejected": -0.5581863522529602,
93
  "step": 50
94
  },
95
  {
96
  "epoch": 0.13,
97
  "learning_rate": 4.990398100856366e-07,
98
+ "logits/chosen": -2.452897548675537,
99
+ "logits/rejected": -2.436417818069458,
100
+ "logps/chosen": -404.2002868652344,
101
+ "logps/rejected": -219.8968505859375,
102
+ "loss": 0.4859,
103
+ "rewards/accuracies": 0.824999988079071,
104
+ "rewards/chosen": -0.01602819748222828,
105
+ "rewards/margins": 0.8050289154052734,
106
+ "rewards/rejected": -0.8210570216178894,
107
  "step": 60
108
  },
109
  {
110
  "epoch": 0.15,
111
  "learning_rate": 4.967775735898179e-07,
112
+ "logits/chosen": -2.3969461917877197,
113
+ "logits/rejected": -2.367985248565674,
114
+ "logps/chosen": -372.53216552734375,
115
+ "logps/rejected": -244.3639373779297,
116
+ "loss": 0.4452,
117
+ "rewards/accuracies": 0.768750011920929,
118
+ "rewards/chosen": -0.21240440011024475,
119
+ "rewards/margins": 0.8786581754684448,
120
+ "rewards/rejected": -1.0910624265670776,
121
  "step": 70
122
  },
123
  {
124
  "epoch": 0.17,
125
  "learning_rate": 4.931986719649298e-07,
126
+ "logits/chosen": -2.3214633464813232,
127
+ "logits/rejected": -2.2642316818237305,
128
+ "logps/chosen": -382.7598571777344,
129
+ "logps/rejected": -267.8367004394531,
130
+ "loss": 0.4444,
131
+ "rewards/accuracies": 0.75,
132
+ "rewards/chosen": -0.5137729048728943,
133
+ "rewards/margins": 0.947833240032196,
134
+ "rewards/rejected": -1.4616062641143799,
135
  "step": 80
136
  },
137
  {
138
  "epoch": 0.19,
139
  "learning_rate": 4.883222001996351e-07,
140
+ "logits/chosen": -2.374678134918213,
141
+ "logits/rejected": -2.3153045177459717,
142
+ "logps/chosen": -416.29443359375,
143
+ "logps/rejected": -304.565673828125,
144
+ "loss": 0.4029,
145
+ "rewards/accuracies": 0.7749999761581421,
146
+ "rewards/chosen": -0.4262094497680664,
147
+ "rewards/margins": 1.292773723602295,
148
+ "rewards/rejected": -1.7189830541610718,
149
  "step": 90
150
  },
151
  {
152
  "epoch": 0.21,
153
  "learning_rate": 4.821741763807186e-07,
154
+ "logits/chosen": -2.3450849056243896,
155
+ "logits/rejected": -2.2891693115234375,
156
+ "logps/chosen": -364.70013427734375,
157
+ "logps/rejected": -251.80361938476562,
158
+ "loss": 0.4319,
159
  "rewards/accuracies": 0.8125,
160
+ "rewards/chosen": -0.2382417917251587,
161
+ "rewards/margins": 1.069990873336792,
162
+ "rewards/rejected": -1.3082325458526611,
163
  "step": 100
164
  },
165
  {
166
  "epoch": 0.21,
167
+ "eval_logits/chosen": -2.3001105785369873,
168
+ "eval_logits/rejected": -2.2766027450561523,
169
+ "eval_logps/chosen": -400.3859558105469,
170
+ "eval_logps/rejected": -434.82366943359375,
171
+ "eval_loss": 0.6498657464981079,
172
+ "eval_rewards/accuracies": 0.6328125,
173
+ "eval_rewards/chosen": -1.433463454246521,
174
+ "eval_rewards/margins": 0.3412409722805023,
175
+ "eval_rewards/rejected": -1.7747045755386353,
176
+ "eval_runtime": 54.2454,
177
+ "eval_samples_per_second": 36.869,
178
+ "eval_steps_per_second": 0.59,
179
  "step": 100
180
  },
181
  {
182
  "epoch": 0.23,
183
  "learning_rate": 4.747874028753375e-07,
184
+ "logits/chosen": -2.233649730682373,
185
+ "logits/rejected": -2.1401028633117676,
186
+ "logps/chosen": -503.4501953125,
187
+ "logps/rejected": -361.48541259765625,
188
+ "loss": 0.4161,
189
+ "rewards/accuracies": 0.800000011920929,
190
+ "rewards/chosen": -0.7022782564163208,
191
+ "rewards/margins": 1.533695936203003,
192
+ "rewards/rejected": -2.2359743118286133,
193
  "step": 110
194
  },
195
  {
196
  "epoch": 0.25,
197
  "learning_rate": 4.662012913161997e-07,
198
+ "logits/chosen": -2.273224353790283,
199
+ "logits/rejected": -2.2325010299682617,
200
+ "logps/chosen": -407.72186279296875,
201
+ "logps/rejected": -321.96258544921875,
202
+ "loss": 0.3851,
203
+ "rewards/accuracies": 0.8187500238418579,
204
+ "rewards/chosen": -0.5440581440925598,
205
+ "rewards/margins": 1.4105150699615479,
206
+ "rewards/rejected": -1.9545730352401733,
207
  "step": 120
208
  },
209
  {
210
  "epoch": 0.27,
211
  "learning_rate": 4.5646165232345103e-07,
212
+ "logits/chosen": -2.2552895545959473,
213
+ "logits/rejected": -2.164522647857666,
214
+ "logps/chosen": -436.8758239746094,
215
+ "logps/rejected": -313.2684631347656,
216
+ "loss": 0.3915,
217
+ "rewards/accuracies": 0.800000011920929,
218
+ "rewards/chosen": -0.6610016822814941,
219
+ "rewards/margins": 1.3072267770767212,
220
+ "rewards/rejected": -1.9682285785675049,
221
  "step": 130
222
  },
223
  {
224
  "epoch": 0.29,
225
  "learning_rate": 4.456204510851956e-07,
226
+ "logits/chosen": -2.194746494293213,
227
+ "logits/rejected": -2.107067108154297,
228
+ "logps/chosen": -472.96380615234375,
229
+ "logps/rejected": -399.2961730957031,
230
+ "loss": 0.3906,
231
+ "rewards/accuracies": 0.831250011920929,
232
+ "rewards/chosen": -0.8916561007499695,
233
+ "rewards/margins": 1.7341525554656982,
234
+ "rewards/rejected": -2.6258087158203125,
235
  "step": 140
236
  },
237
  {
238
  "epoch": 0.31,
239
  "learning_rate": 4.337355301007335e-07,
240
+ "logits/chosen": -2.1862947940826416,
241
+ "logits/rejected": -2.0885581970214844,
242
+ "logps/chosen": -470.4183654785156,
243
+ "logps/rejected": -381.9734191894531,
244
+ "loss": 0.377,
245
+ "rewards/accuracies": 0.8125,
246
+ "rewards/chosen": -0.8573510050773621,
247
+ "rewards/margins": 1.6041991710662842,
248
+ "rewards/rejected": -2.461550235748291,
249
  "step": 150
250
  },
251
  {
252
  "epoch": 0.33,
253
  "learning_rate": 4.2087030056579986e-07,
254
+ "logits/chosen": -2.149928092956543,
255
+ "logits/rejected": -2.0291361808776855,
256
+ "logps/chosen": -473.9505310058594,
257
+ "logps/rejected": -384.2547607421875,
258
+ "loss": 0.3711,
259
+ "rewards/accuracies": 0.8125,
260
+ "rewards/chosen": -0.9108888506889343,
261
+ "rewards/margins": 1.6116889715194702,
262
+ "rewards/rejected": -2.5225777626037598,
263
  "step": 160
264
  },
265
  {
266
  "epoch": 0.36,
267
  "learning_rate": 4.070934040463998e-07,
268
+ "logits/chosen": -2.1031107902526855,
269
+ "logits/rejected": -1.9537290334701538,
270
+ "logps/chosen": -552.788330078125,
271
+ "logps/rejected": -459.04718017578125,
272
+ "loss": 0.3672,
273
+ "rewards/accuracies": 0.831250011920929,
274
+ "rewards/chosen": -1.2671574354171753,
275
+ "rewards/margins": 2.013477087020874,
276
+ "rewards/rejected": -3.2806344032287598,
277
  "step": 170
278
  },
279
  {
280
  "epoch": 0.38,
281
  "learning_rate": 3.9247834624635404e-07,
282
+ "logits/chosen": -2.116580009460449,
283
+ "logits/rejected": -2.005300283432007,
284
+ "logps/chosen": -485.86346435546875,
285
+ "logps/rejected": -424.8414611816406,
286
+ "loss": 0.3752,
287
+ "rewards/accuracies": 0.8062499761581421,
288
+ "rewards/chosen": -1.1004022359848022,
289
+ "rewards/margins": 1.7027782201766968,
290
+ "rewards/rejected": -2.803180456161499,
291
  "step": 180
292
  },
293
  {
294
  "epoch": 0.4,
295
  "learning_rate": 3.7710310482256523e-07,
296
+ "logits/chosen": -2.0973212718963623,
297
+ "logits/rejected": -1.9831724166870117,
298
+ "logps/chosen": -553.8712158203125,
299
+ "logps/rejected": -454.5481872558594,
300
+ "loss": 0.3748,
301
+ "rewards/accuracies": 0.8062499761581421,
302
+ "rewards/chosen": -1.1362320184707642,
303
+ "rewards/margins": 2.0223002433776855,
304
+ "rewards/rejected": -3.1585323810577393,
305
  "step": 190
306
  },
307
  {
308
  "epoch": 0.42,
309
  "learning_rate": 3.610497133404795e-07,
310
+ "logits/chosen": -2.0881752967834473,
311
+ "logits/rejected": -1.980513572692871,
312
+ "logps/chosen": -542.2076416015625,
313
+ "logps/rejected": -463.62890625,
314
+ "loss": 0.3523,
315
+ "rewards/accuracies": 0.8187500238418579,
316
+ "rewards/chosen": -1.2750364542007446,
317
+ "rewards/margins": 2.0321054458618164,
318
+ "rewards/rejected": -3.3071417808532715,
319
  "step": 200
320
  },
321
  {
322
  "epoch": 0.42,
323
+ "eval_logits/chosen": -2.0889179706573486,
324
+ "eval_logits/rejected": -2.0558736324310303,
325
+ "eval_logps/chosen": -459.19970703125,
326
+ "eval_logps/rejected": -531.1033935546875,
327
+ "eval_loss": 0.5953767895698547,
328
+ "eval_rewards/accuracies": 0.71875,
329
+ "eval_rewards/chosen": -2.0216004848480225,
330
+ "eval_rewards/margins": 0.715901255607605,
331
+ "eval_rewards/rejected": -2.737501621246338,
332
+ "eval_runtime": 53.4074,
333
+ "eval_samples_per_second": 37.448,
334
+ "eval_steps_per_second": 0.599,
335
  "step": 200
336
  },
337
  {
338
  "epoch": 0.44,
339
  "learning_rate": 3.4440382358952115e-07,
340
+ "logits/chosen": -2.1334011554718018,
341
+ "logits/rejected": -2.029670476913452,
342
+ "logps/chosen": -484.1700744628906,
343
+ "logps/rejected": -433.7367248535156,
344
+ "loss": 0.3484,
345
+ "rewards/accuracies": 0.831250011920929,
346
+ "rewards/chosen": -1.0799142122268677,
347
+ "rewards/margins": 1.8150622844696045,
348
+ "rewards/rejected": -2.894976854324341,
349
  "step": 210
350
  },
351
  {
352
  "epoch": 0.46,
353
  "learning_rate": 3.272542485937368e-07,
354
+ "logits/chosen": -2.18733549118042,
355
+ "logits/rejected": -2.02341890335083,
356
+ "logps/chosen": -521.8484497070312,
357
+ "logps/rejected": -446.3534240722656,
358
+ "loss": 0.3472,
359
+ "rewards/accuracies": 0.8812500238418579,
360
+ "rewards/chosen": -0.9639819264411926,
361
+ "rewards/margins": 2.202716827392578,
362
+ "rewards/rejected": -3.166698932647705,
363
  "step": 220
364
  },
365
  {
366
  "epoch": 0.48,
367
  "learning_rate": 3.096924887558854e-07,
368
+ "logits/chosen": -2.1084365844726562,
369
+ "logits/rejected": -2.022747278213501,
370
+ "logps/chosen": -504.220947265625,
371
+ "logps/rejected": -441.27667236328125,
372
+ "loss": 0.3541,
373
  "rewards/accuracies": 0.8187500238418579,
374
+ "rewards/chosen": -0.9997168779373169,
375
+ "rewards/margins": 1.8952099084854126,
376
+ "rewards/rejected": -2.8949267864227295,
377
  "step": 230
378
  },
379
  {
380
  "epoch": 0.5,
381
  "learning_rate": 2.9181224366319943e-07,
382
+ "logits/chosen": -2.111999034881592,
383
+ "logits/rejected": -2.023984432220459,
384
+ "logps/chosen": -526.4149169921875,
385
+ "logps/rejected": -442.402099609375,
386
+ "loss": 0.3521,
387
+ "rewards/accuracies": 0.824999988079071,
388
+ "rewards/chosen": -1.2106670141220093,
389
+ "rewards/margins": 1.8976614475250244,
390
+ "rewards/rejected": -3.108328342437744,
391
  "step": 240
392
  },
393
  {
394
  "epoch": 0.52,
395
  "learning_rate": 2.7370891215954565e-07,
396
+ "logits/chosen": -2.089487075805664,
397
+ "logits/rejected": -2.001037120819092,
398
+ "logps/chosen": -484.54559326171875,
399
+ "logps/rejected": -447.81463623046875,
400
+ "loss": 0.3631,
401
+ "rewards/accuracies": 0.831250011920929,
402
+ "rewards/chosen": -1.2301665544509888,
403
+ "rewards/margins": 1.9752994775772095,
404
+ "rewards/rejected": -3.2054660320281982,
405
  "step": 250
406
  },
407
  {
408
  "epoch": 0.54,
409
  "learning_rate": 2.55479083351317e-07,
410
+ "logits/chosen": -2.2092173099517822,
411
+ "logits/rejected": -2.1207237243652344,
412
+ "logps/chosen": -484.1170959472656,
413
+ "logps/rejected": -414.44134521484375,
414
+ "loss": 0.3554,
415
+ "rewards/accuracies": 0.8187500238418579,
416
+ "rewards/chosen": -1.0310567617416382,
417
+ "rewards/margins": 1.7033233642578125,
418
+ "rewards/rejected": -2.7343802452087402,
419
  "step": 260
420
  },
421
  {
422
  "epoch": 0.56,
423
  "learning_rate": 2.3722002126275822e-07,
424
+ "logits/chosen": -2.128089427947998,
425
+ "logits/rejected": -2.0141799449920654,
426
+ "logps/chosen": -475.9017028808594,
427
+ "logps/rejected": -440.27142333984375,
428
+ "loss": 0.3495,
429
+ "rewards/accuracies": 0.8500000238418579,
430
+ "rewards/chosen": -1.2480560541152954,
431
+ "rewards/margins": 2.013627529144287,
432
+ "rewards/rejected": -3.261683940887451,
433
  "step": 270
434
  },
435
  {
436
  "epoch": 0.59,
437
  "learning_rate": 2.19029145890313e-07,
438
+ "logits/chosen": -2.1510426998138428,
439
+ "logits/rejected": -2.0308613777160645,
440
+ "logps/chosen": -480.66485595703125,
441
+ "logps/rejected": -439.12530517578125,
442
+ "loss": 0.3473,
443
+ "rewards/accuracies": 0.8374999761581421,
444
+ "rewards/chosen": -1.2660208940505981,
445
+ "rewards/margins": 1.8633956909179688,
446
+ "rewards/rejected": -3.1294167041778564,
447
  "step": 280
448
  },
449
  {
450
  "epoch": 0.61,
451
  "learning_rate": 2.0100351342479216e-07,
452
+ "logits/chosen": -2.1496615409851074,
453
+ "logits/rejected": -2.032989025115967,
454
+ "logps/chosen": -499.2440490722656,
455
+ "logps/rejected": -448.85107421875,
456
+ "loss": 0.3588,
457
+ "rewards/accuracies": 0.8374999761581421,
458
+ "rewards/chosen": -0.9003239870071411,
459
+ "rewards/margins": 2.171739101409912,
460
+ "rewards/rejected": -3.0720629692077637,
461
  "step": 290
462
  },
463
  {
464
  "epoch": 0.63,
465
  "learning_rate": 1.8323929841460178e-07,
466
+ "logits/chosen": -2.158599376678467,
467
+ "logits/rejected": -2.0707848072052,
468
+ "logps/chosen": -469.2913513183594,
469
+ "logps/rejected": -447.3359375,
470
+ "loss": 0.3611,
471
  "rewards/accuracies": 0.824999988079071,
472
+ "rewards/chosen": -1.2566159963607788,
473
+ "rewards/margins": 1.9593569040298462,
474
+ "rewards/rejected": -3.215972900390625,
475
  "step": 300
476
  },
477
  {
478
  "epoch": 0.63,
479
+ "eval_logits/chosen": -2.1299195289611816,
480
+ "eval_logits/rejected": -2.0872669219970703,
481
+ "eval_logps/chosen": -456.173828125,
482
+ "eval_logps/rejected": -541.2481689453125,
483
+ "eval_loss": 0.5743047595024109,
484
+ "eval_rewards/accuracies": 0.71875,
485
+ "eval_rewards/chosen": -1.9913413524627686,
486
+ "eval_rewards/margins": 0.847608208656311,
487
+ "eval_rewards/rejected": -2.838949680328369,
488
+ "eval_runtime": 53.3455,
489
+ "eval_samples_per_second": 37.491,
490
+ "eval_steps_per_second": 0.6,
491
  "step": 300
492
  },
493
  {
494
  "epoch": 0.65,
495
  "learning_rate": 1.6583128063291573e-07,
496
+ "logits/chosen": -2.1583058834075928,
497
+ "logits/rejected": -2.0590124130249023,
498
+ "logps/chosen": -517.3380737304688,
499
+ "logps/rejected": -457.45928955078125,
500
+ "loss": 0.3338,
501
+ "rewards/accuracies": 0.8687499761581421,
502
+ "rewards/chosen": -1.147152066230774,
503
+ "rewards/margins": 2.0138449668884277,
504
+ "rewards/rejected": -3.160996913909912,
505
  "step": 310
506
  },
507
  {
508
  "epoch": 0.67,
509
  "learning_rate": 1.488723393865766e-07,
510
+ "logits/chosen": -2.0327324867248535,
511
+ "logits/rejected": -1.9210631847381592,
512
+ "logps/chosen": -525.4856567382812,
513
+ "logps/rejected": -502.88934326171875,
514
+ "loss": 0.3473,
515
+ "rewards/accuracies": 0.824999988079071,
516
+ "rewards/chosen": -1.5122380256652832,
517
+ "rewards/margins": 2.099067449569702,
518
+ "rewards/rejected": -3.6113052368164062,
519
  "step": 320
520
  },
521
  {
522
  "epoch": 0.69,
523
  "learning_rate": 1.3245295796480788e-07,
524
+ "logits/chosen": -2.1876847743988037,
525
+ "logits/rejected": -2.0464909076690674,
526
+ "logps/chosen": -545.4427490234375,
527
+ "logps/rejected": -492.4483947753906,
528
+ "loss": 0.3278,
529
+ "rewards/accuracies": 0.84375,
530
+ "rewards/chosen": -1.2790089845657349,
531
+ "rewards/margins": 2.3831381797790527,
532
+ "rewards/rejected": -3.662147045135498,
533
  "step": 330
534
  },
535
  {
536
  "epoch": 0.71,
537
  "learning_rate": 1.1666074087171627e-07,
538
+ "logits/chosen": -2.108828544616699,
539
+ "logits/rejected": -1.9729408025741577,
540
+ "logps/chosen": -533.9615478515625,
541
+ "logps/rejected": -490.66851806640625,
542
+ "loss": 0.3636,
543
+ "rewards/accuracies": 0.8374999761581421,
544
+ "rewards/chosen": -1.4988614320755005,
545
+ "rewards/margins": 2.179060459136963,
546
+ "rewards/rejected": -3.677922010421753,
547
  "step": 340
548
  },
549
  {
550
  "epoch": 0.73,
551
  "learning_rate": 1.0157994641835734e-07,
552
+ "logits/chosen": -2.1237919330596924,
553
+ "logits/rejected": -1.9951225519180298,
554
+ "logps/chosen": -511.7247619628906,
555
+ "logps/rejected": -494.952880859375,
556
+ "loss": 0.3334,
557
+ "rewards/accuracies": 0.84375,
558
+ "rewards/chosen": -1.251680612564087,
559
+ "rewards/margins": 2.2906720638275146,
560
+ "rewards/rejected": -3.5423526763916016,
561
  "step": 350
562
  },
563
  {
564
  "epoch": 0.75,
565
  "learning_rate": 8.729103716819111e-08,
566
+ "logits/chosen": -2.1752729415893555,
567
+ "logits/rejected": -2.0284740924835205,
568
+ "logps/chosen": -543.2025146484375,
569
+ "logps/rejected": -459.12493896484375,
570
+ "loss": 0.3591,
571
+ "rewards/accuracies": 0.8500000238418579,
572
+ "rewards/chosen": -1.0760266780853271,
573
+ "rewards/margins": 2.3620123863220215,
574
+ "rewards/rejected": -3.4380390644073486,
575
  "step": 360
576
  },
577
  {
578
  "epoch": 0.77,
579
  "learning_rate": 7.387025063449081e-08,
580
+ "logits/chosen": -2.1319737434387207,
581
+ "logits/rejected": -2.020498275756836,
582
+ "logps/chosen": -488.7276306152344,
583
+ "logps/rejected": -446.2596740722656,
584
+ "loss": 0.3274,
585
+ "rewards/accuracies": 0.8187500238418579,
586
+ "rewards/chosen": -1.2242181301116943,
587
+ "rewards/margins": 2.015012502670288,
588
+ "rewards/rejected": -3.2392303943634033,
589
  "step": 370
590
  },
591
  {
592
  "epoch": 0.79,
593
  "learning_rate": 6.138919252022435e-08,
594
+ "logits/chosen": -2.1206445693969727,
595
+ "logits/rejected": -1.98409104347229,
596
+ "logps/chosen": -502.968994140625,
597
+ "logps/rejected": -463.2781677246094,
598
+ "loss": 0.3248,
599
+ "rewards/accuracies": 0.8500000238418579,
600
+ "rewards/chosen": -1.1951261758804321,
601
+ "rewards/margins": 2.3360695838928223,
602
+ "rewards/rejected": -3.531195878982544,
603
  "step": 380
604
  },
605
  {
606
  "epoch": 0.82,
607
  "learning_rate": 4.991445467064689e-08,
608
+ "logits/chosen": -2.160815715789795,
609
+ "logits/rejected": -2.0336337089538574,
610
+ "logps/chosen": -521.4240112304688,
611
+ "logps/rejected": -455.9097595214844,
612
+ "loss": 0.3419,
613
  "rewards/accuracies": 0.824999988079071,
614
+ "rewards/chosen": -1.1749980449676514,
615
+ "rewards/margins": 2.0022196769714355,
616
+ "rewards/rejected": -3.177217721939087,
617
  "step": 390
618
  },
619
  {
620
  "epoch": 0.84,
621
  "learning_rate": 3.9507259776993954e-08,
622
+ "logits/chosen": -2.0740323066711426,
623
+ "logits/rejected": -1.9285573959350586,
624
+ "logps/chosen": -533.530029296875,
625
+ "logps/rejected": -476.70166015625,
626
+ "loss": 0.3321,
627
+ "rewards/accuracies": 0.8687499761581421,
628
+ "rewards/chosen": -1.1504842042922974,
629
+ "rewards/margins": 2.290919780731201,
630
+ "rewards/rejected": -3.441403865814209,
631
  "step": 400
632
  },
633
  {
634
  "epoch": 0.84,
635
+ "eval_logits/chosen": -2.0871171951293945,
636
+ "eval_logits/rejected": -2.04443621635437,
637
+ "eval_logps/chosen": -476.4866027832031,
638
+ "eval_logps/rejected": -567.706787109375,
639
+ "eval_loss": 0.5693503618240356,
640
+ "eval_rewards/accuracies": 0.73828125,
641
+ "eval_rewards/chosen": -2.194469451904297,
642
+ "eval_rewards/margins": 0.909065306186676,
643
+ "eval_rewards/rejected": -3.103534698486328,
644
+ "eval_runtime": 53.3645,
645
+ "eval_samples_per_second": 37.478,
646
+ "eval_steps_per_second": 0.6,
647
  "step": 400
648
  },
649
  {
650
  "epoch": 0.86,
651
  "learning_rate": 3.022313472693447e-08,
652
+ "logits/chosen": -2.1464505195617676,
653
+ "logits/rejected": -1.9985647201538086,
654
+ "logps/chosen": -569.2177124023438,
655
+ "logps/rejected": -501.0960388183594,
656
+ "loss": 0.3418,
657
+ "rewards/accuracies": 0.8812500238418579,
658
+ "rewards/chosen": -1.199937343597412,
659
+ "rewards/margins": 2.462070941925049,
660
+ "rewards/rejected": -3.662008762359619,
661
  "step": 410
662
  },
663
  {
664
  "epoch": 0.88,
665
  "learning_rate": 2.2111614344599684e-08,
666
+ "logits/chosen": -2.01700758934021,
667
+ "logits/rejected": -1.9420589208602905,
668
+ "logps/chosen": -496.57281494140625,
669
+ "logps/rejected": -473.21502685546875,
670
+ "loss": 0.3415,
671
+ "rewards/accuracies": 0.8187500238418579,
672
+ "rewards/chosen": -1.3657633066177368,
673
+ "rewards/margins": 2.07015323638916,
674
+ "rewards/rejected": -3.4359164237976074,
675
  "step": 420
676
  },
677
  {
678
  "epoch": 0.9,
679
  "learning_rate": 1.521597710086439e-08,
680
+ "logits/chosen": -2.0969324111938477,
681
+ "logits/rejected": -1.9792184829711914,
682
+ "logps/chosen": -480.86492919921875,
683
+ "logps/rejected": -453.40216064453125,
684
+ "loss": 0.3243,
685
+ "rewards/accuracies": 0.8374999761581421,
686
+ "rewards/chosen": -1.4078073501586914,
687
+ "rewards/margins": 1.961418867111206,
688
+ "rewards/rejected": -3.3692264556884766,
689
  "step": 430
690
  },
691
  {
692
  "epoch": 0.92,
693
  "learning_rate": 9.57301420397924e-09,
694
+ "logits/chosen": -2.0999059677124023,
695
+ "logits/rejected": -1.9967210292816162,
696
+ "logps/chosen": -514.0860595703125,
697
+ "logps/rejected": -497.3169860839844,
698
+ "loss": 0.3191,
699
+ "rewards/accuracies": 0.8812500238418579,
700
+ "rewards/chosen": -1.4036227464675903,
701
+ "rewards/margins": 2.1624085903167725,
702
+ "rewards/rejected": -3.5660316944122314,
703
  "step": 440
704
  },
705
  {
706
  "epoch": 0.94,
707
  "learning_rate": 5.212833302556258e-09,
708
+ "logits/chosen": -2.1073479652404785,
709
+ "logits/rejected": -1.98806631565094,
710
+ "logps/chosen": -508.578125,
711
+ "logps/rejected": -465.4120178222656,
712
+ "loss": 0.3281,
713
+ "rewards/accuracies": 0.862500011920929,
714
+ "rewards/chosen": -1.237577199935913,
715
+ "rewards/margins": 2.141489028930664,
716
+ "rewards/rejected": -3.379065990447998,
717
  "step": 450
718
  },
719
  {
720
  "epoch": 0.96,
721
  "learning_rate": 2.158697848236607e-09,
722
+ "logits/chosen": -2.0812013149261475,
723
+ "logits/rejected": -1.983912706375122,
724
+ "logps/chosen": -520.6010131835938,
725
+ "logps/rejected": -473.6297302246094,
726
+ "loss": 0.3436,
727
+ "rewards/accuracies": 0.8500000238418579,
728
+ "rewards/chosen": -1.305751085281372,
729
+ "rewards/margins": 1.9923391342163086,
730
+ "rewards/rejected": -3.2980899810791016,
731
  "step": 460
732
  },
733
  {
734
  "epoch": 0.98,
735
  "learning_rate": 4.269029751107489e-10,
736
+ "logits/chosen": -2.1194050312042236,
737
+ "logits/rejected": -1.9787203073501587,
738
+ "logps/chosen": -526.4293212890625,
739
+ "logps/rejected": -478.50537109375,
740
+ "loss": 0.3464,
741
+ "rewards/accuracies": 0.893750011920929,
742
+ "rewards/chosen": -1.139508843421936,
743
+ "rewards/margins": 2.3456408977508545,
744
+ "rewards/rejected": -3.485149383544922,
745
  "step": 470
746
  },
747
  {
748
  "epoch": 1.0,
749
  "step": 478,
750
  "total_flos": 0.0,
751
+ "train_loss": 0.390623705157675,
752
+ "train_runtime": 4433.9669,
753
+ "train_samples_per_second": 13.787,
754
+ "train_steps_per_second": 0.108
755
  }
756
  ],
757
  "logging_steps": 10,