alvarobartt HF staff commited on
Commit
537d490
1 Parent(s): 16788c0

Upload trainer_state.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. trainer_state.json +1221 -0
trainer_state.json ADDED
@@ -0,0 +1,1221 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 0.8980213403701782,
3
+ "best_model_checkpoint": "./mistral-7b-v0.1-orpo/checkpoint-211",
4
+ "epoch": 3.0,
5
+ "eval_steps": 500,
6
+ "global_step": 633,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.05,
13
+ "grad_norm": 39.94051543947401,
14
+ "learning_rate": 5.000000000000001e-07,
15
+ "log_odds_chosen": 0.2551603317260742,
16
+ "log_odds_ratio": -0.7050353288650513,
17
+ "logits/chosen": -2.8500335216522217,
18
+ "logits/rejected": -2.8525497913360596,
19
+ "logps/chosen": -1.2754234075546265,
20
+ "logps/rejected": -1.5109716653823853,
21
+ "loss": 1.1661,
22
+ "nll_loss": 1.1492516994476318,
23
+ "rewards/accuracies": 0.5625,
24
+ "rewards/chosen": -0.06377117335796356,
25
+ "rewards/margins": 0.011777402833104134,
26
+ "rewards/rejected": -0.07554857432842255,
27
+ "step": 10
28
+ },
29
+ {
30
+ "epoch": 0.09,
31
+ "grad_norm": 15.606195494341659,
32
+ "learning_rate": 1.0000000000000002e-06,
33
+ "log_odds_chosen": 0.19387400150299072,
34
+ "log_odds_ratio": -0.7067446708679199,
35
+ "logits/chosen": -2.8987910747528076,
36
+ "logits/rejected": -2.9010841846466064,
37
+ "logps/chosen": -0.9522870779037476,
38
+ "logps/rejected": -1.0993469953536987,
39
+ "loss": 0.9734,
40
+ "nll_loss": 0.9094793200492859,
41
+ "rewards/accuracies": 0.5,
42
+ "rewards/chosen": -0.047614358365535736,
43
+ "rewards/margins": 0.0073529938235878944,
44
+ "rewards/rejected": -0.054967351257801056,
45
+ "step": 20
46
+ },
47
+ {
48
+ "epoch": 0.14,
49
+ "grad_norm": 10.36315915115552,
50
+ "learning_rate": 1.5e-06,
51
+ "log_odds_chosen": 0.21015071868896484,
52
+ "log_odds_ratio": -0.6918584108352661,
53
+ "logits/chosen": -2.9974679946899414,
54
+ "logits/rejected": -2.9977316856384277,
55
+ "logps/chosen": -0.9473606944084167,
56
+ "logps/rejected": -1.0642311573028564,
57
+ "loss": 0.9344,
58
+ "nll_loss": 0.8887130618095398,
59
+ "rewards/accuracies": 0.6000000238418579,
60
+ "rewards/chosen": -0.047368038445711136,
61
+ "rewards/margins": 0.005843517370522022,
62
+ "rewards/rejected": -0.05321155861020088,
63
+ "step": 30
64
+ },
65
+ {
66
+ "epoch": 0.19,
67
+ "grad_norm": 12.001709174217318,
68
+ "learning_rate": 2.0000000000000003e-06,
69
+ "log_odds_chosen": 0.10945004224777222,
70
+ "log_odds_ratio": -0.7249928712844849,
71
+ "logits/chosen": -2.8161749839782715,
72
+ "logits/rejected": -2.8172993659973145,
73
+ "logps/chosen": -0.9179791212081909,
74
+ "logps/rejected": -0.9879155158996582,
75
+ "loss": 0.9444,
76
+ "nll_loss": 0.8832181692123413,
77
+ "rewards/accuracies": 0.4749999940395355,
78
+ "rewards/chosen": -0.04589895159006119,
79
+ "rewards/margins": 0.003496820107102394,
80
+ "rewards/rejected": -0.04939577355980873,
81
+ "step": 40
82
+ },
83
+ {
84
+ "epoch": 0.24,
85
+ "grad_norm": 12.504768280715709,
86
+ "learning_rate": 2.5e-06,
87
+ "log_odds_chosen": 0.4096715450286865,
88
+ "log_odds_ratio": -0.6319375038146973,
89
+ "logits/chosen": -2.701756000518799,
90
+ "logits/rejected": -2.713608980178833,
91
+ "logps/chosen": -0.8992551565170288,
92
+ "logps/rejected": -1.1893055438995361,
93
+ "loss": 0.9242,
94
+ "nll_loss": 0.9320483207702637,
95
+ "rewards/accuracies": 0.637499988079071,
96
+ "rewards/chosen": -0.04496275633573532,
97
+ "rewards/margins": 0.014502523466944695,
98
+ "rewards/rejected": -0.059465277940034866,
99
+ "step": 50
100
+ },
101
+ {
102
+ "epoch": 0.28,
103
+ "grad_norm": 9.02387893392391,
104
+ "learning_rate": 3e-06,
105
+ "log_odds_chosen": 0.2635055184364319,
106
+ "log_odds_ratio": -0.6361549496650696,
107
+ "logits/chosen": -2.6746716499328613,
108
+ "logits/rejected": -2.6736464500427246,
109
+ "logps/chosen": -0.7554010152816772,
110
+ "logps/rejected": -0.9083536863327026,
111
+ "loss": 0.8533,
112
+ "nll_loss": 0.7264706492424011,
113
+ "rewards/accuracies": 0.6499999761581421,
114
+ "rewards/chosen": -0.03777005523443222,
115
+ "rewards/margins": 0.007647632621228695,
116
+ "rewards/rejected": -0.04541768878698349,
117
+ "step": 60
118
+ },
119
+ {
120
+ "epoch": 0.33,
121
+ "grad_norm": 10.698033738270377,
122
+ "learning_rate": 3.5e-06,
123
+ "log_odds_chosen": 0.30966854095458984,
124
+ "log_odds_ratio": -0.654030442237854,
125
+ "logits/chosen": -2.7319040298461914,
126
+ "logits/rejected": -2.737243175506592,
127
+ "logps/chosen": -0.84112948179245,
128
+ "logps/rejected": -1.0554386377334595,
129
+ "loss": 0.9031,
130
+ "nll_loss": 0.8661483526229858,
131
+ "rewards/accuracies": 0.6000000238418579,
132
+ "rewards/chosen": -0.042056478559970856,
133
+ "rewards/margins": 0.010715454816818237,
134
+ "rewards/rejected": -0.05277193710207939,
135
+ "step": 70
136
+ },
137
+ {
138
+ "epoch": 0.38,
139
+ "grad_norm": 9.996385073585538,
140
+ "learning_rate": 4.000000000000001e-06,
141
+ "log_odds_chosen": -0.006746725644916296,
142
+ "log_odds_ratio": -0.775671124458313,
143
+ "logits/chosen": -2.8110978603363037,
144
+ "logits/rejected": -2.812554121017456,
145
+ "logps/chosen": -0.8807821273803711,
146
+ "logps/rejected": -0.883065402507782,
147
+ "loss": 0.8813,
148
+ "nll_loss": 0.9144401550292969,
149
+ "rewards/accuracies": 0.4749999940395355,
150
+ "rewards/chosen": -0.044039104133844376,
151
+ "rewards/margins": 0.00011416729830671102,
152
+ "rewards/rejected": -0.04415327310562134,
153
+ "step": 80
154
+ },
155
+ {
156
+ "epoch": 0.43,
157
+ "grad_norm": 9.119674780883251,
158
+ "learning_rate": 4.5e-06,
159
+ "log_odds_chosen": 0.2785256505012512,
160
+ "log_odds_ratio": -0.6412050127983093,
161
+ "logits/chosen": -2.8578667640686035,
162
+ "logits/rejected": -2.843656063079834,
163
+ "logps/chosen": -0.8698889017105103,
164
+ "logps/rejected": -1.0659993886947632,
165
+ "loss": 0.8932,
166
+ "nll_loss": 0.8667906522750854,
167
+ "rewards/accuracies": 0.5874999761581421,
168
+ "rewards/chosen": -0.04349444434046745,
169
+ "rewards/margins": 0.009805524721741676,
170
+ "rewards/rejected": -0.05329997092485428,
171
+ "step": 90
172
+ },
173
+ {
174
+ "epoch": 0.47,
175
+ "grad_norm": 12.306871572257608,
176
+ "learning_rate": 5e-06,
177
+ "log_odds_chosen": 0.3695891201496124,
178
+ "log_odds_ratio": -0.6482647061347961,
179
+ "logits/chosen": -2.7652063369750977,
180
+ "logits/rejected": -2.789933681488037,
181
+ "logps/chosen": -0.878996729850769,
182
+ "logps/rejected": -1.082092046737671,
183
+ "loss": 0.9352,
184
+ "nll_loss": 0.9471464157104492,
185
+ "rewards/accuracies": 0.6625000238418579,
186
+ "rewards/chosen": -0.04394983872771263,
187
+ "rewards/margins": 0.01015476230531931,
188
+ "rewards/rejected": -0.054104603826999664,
189
+ "step": 100
190
+ },
191
+ {
192
+ "epoch": 0.52,
193
+ "grad_norm": 8.587658638329193,
194
+ "learning_rate": 4.995658601978859e-06,
195
+ "log_odds_chosen": 0.3208955228328705,
196
+ "log_odds_ratio": -0.624093234539032,
197
+ "logits/chosen": -2.75911808013916,
198
+ "logits/rejected": -2.7485742568969727,
199
+ "logps/chosen": -0.7625306844711304,
200
+ "logps/rejected": -0.950202465057373,
201
+ "loss": 0.9039,
202
+ "nll_loss": 0.8189178705215454,
203
+ "rewards/accuracies": 0.675000011920929,
204
+ "rewards/chosen": -0.03812653571367264,
205
+ "rewards/margins": 0.009383587166666985,
206
+ "rewards/rejected": -0.04751012474298477,
207
+ "step": 110
208
+ },
209
+ {
210
+ "epoch": 0.57,
211
+ "grad_norm": 9.573968736057266,
212
+ "learning_rate": 4.982649486104858e-06,
213
+ "log_odds_chosen": 0.2796480059623718,
214
+ "log_odds_ratio": -0.6660342216491699,
215
+ "logits/chosen": -2.8325819969177246,
216
+ "logits/rejected": -2.830364942550659,
217
+ "logps/chosen": -0.8180511593818665,
218
+ "logps/rejected": -0.9945104718208313,
219
+ "loss": 0.9465,
220
+ "nll_loss": 0.8578864932060242,
221
+ "rewards/accuracies": 0.550000011920929,
222
+ "rewards/chosen": -0.040902554988861084,
223
+ "rewards/margins": 0.008822967298328876,
224
+ "rewards/rejected": -0.049725525081157684,
225
+ "step": 120
226
+ },
227
+ {
228
+ "epoch": 0.62,
229
+ "grad_norm": 10.453345314275992,
230
+ "learning_rate": 4.961017834577927e-06,
231
+ "log_odds_chosen": 0.2720031440258026,
232
+ "log_odds_ratio": -0.656061053276062,
233
+ "logits/chosen": -2.846191644668579,
234
+ "logits/rejected": -2.8502097129821777,
235
+ "logps/chosen": -0.8261640667915344,
236
+ "logps/rejected": -0.9698032140731812,
237
+ "loss": 0.9689,
238
+ "nll_loss": 0.9089719653129578,
239
+ "rewards/accuracies": 0.574999988079071,
240
+ "rewards/chosen": -0.0413082018494606,
241
+ "rewards/margins": 0.007181968539953232,
242
+ "rewards/rejected": -0.048490170389413834,
243
+ "step": 130
244
+ },
245
+ {
246
+ "epoch": 0.66,
247
+ "grad_norm": 8.571550283980045,
248
+ "learning_rate": 4.930838776685373e-06,
249
+ "log_odds_chosen": 0.30521708726882935,
250
+ "log_odds_ratio": -0.6360923647880554,
251
+ "logits/chosen": -2.8390443325042725,
252
+ "logits/rejected": -2.8424527645111084,
253
+ "logps/chosen": -0.8498956561088562,
254
+ "logps/rejected": -1.071276068687439,
255
+ "loss": 0.9233,
256
+ "nll_loss": 0.8490797281265259,
257
+ "rewards/accuracies": 0.699999988079071,
258
+ "rewards/chosen": -0.04249478504061699,
259
+ "rewards/margins": 0.011069012805819511,
260
+ "rewards/rejected": -0.05356380343437195,
261
+ "step": 140
262
+ },
263
+ {
264
+ "epoch": 0.71,
265
+ "grad_norm": 9.244452429837853,
266
+ "learning_rate": 4.8922171278689655e-06,
267
+ "log_odds_chosen": 0.38359910249710083,
268
+ "log_odds_ratio": -0.6358404159545898,
269
+ "logits/chosen": -2.866654396057129,
270
+ "logits/rejected": -2.858891487121582,
271
+ "logps/chosen": -0.8682330250740051,
272
+ "logps/rejected": -1.1270641088485718,
273
+ "loss": 0.9085,
274
+ "nll_loss": 0.8730288743972778,
275
+ "rewards/accuracies": 0.6000000238418579,
276
+ "rewards/chosen": -0.04341164976358414,
277
+ "rewards/margins": 0.01294155977666378,
278
+ "rewards/rejected": -0.056353211402893066,
279
+ "step": 150
280
+ },
281
+ {
282
+ "epoch": 0.76,
283
+ "grad_norm": 10.379571331754194,
284
+ "learning_rate": 4.845287025688503e-06,
285
+ "log_odds_chosen": 0.29743850231170654,
286
+ "log_odds_ratio": -0.650119423866272,
287
+ "logits/chosen": -2.8795416355133057,
288
+ "logits/rejected": -2.881042957305908,
289
+ "logps/chosen": -0.8429490923881531,
290
+ "logps/rejected": -1.049843668937683,
291
+ "loss": 0.8767,
292
+ "nll_loss": 0.8780015110969543,
293
+ "rewards/accuracies": 0.612500011920929,
294
+ "rewards/chosen": -0.04214745759963989,
295
+ "rewards/margins": 0.010344731621444225,
296
+ "rewards/rejected": -0.052492182701826096,
297
+ "step": 160
298
+ },
299
+ {
300
+ "epoch": 0.81,
301
+ "grad_norm": 8.62227952312482,
302
+ "learning_rate": 4.790211463946174e-06,
303
+ "log_odds_chosen": 0.10580413043498993,
304
+ "log_odds_ratio": -0.7094690203666687,
305
+ "logits/chosen": -2.745680093765259,
306
+ "logits/rejected": -2.751802921295166,
307
+ "logps/chosen": -0.8849539756774902,
308
+ "logps/rejected": -0.9533231854438782,
309
+ "loss": 0.8963,
310
+ "nll_loss": 0.8941303491592407,
311
+ "rewards/accuracies": 0.5625,
312
+ "rewards/chosen": -0.044247694313526154,
313
+ "rewards/margins": 0.0034184574615210295,
314
+ "rewards/rejected": -0.04766615480184555,
315
+ "step": 170
316
+ },
317
+ {
318
+ "epoch": 0.85,
319
+ "grad_norm": 10.04192028153657,
320
+ "learning_rate": 4.727181726589789e-06,
321
+ "log_odds_chosen": 0.1737401783466339,
322
+ "log_odds_ratio": -0.7261314988136292,
323
+ "logits/chosen": -2.7415287494659424,
324
+ "logits/rejected": -2.7601096630096436,
325
+ "logps/chosen": -0.8834375143051147,
326
+ "logps/rejected": -1.0135564804077148,
327
+ "loss": 0.9263,
328
+ "nll_loss": 0.9143431782722473,
329
+ "rewards/accuracies": 0.5375000238418579,
330
+ "rewards/chosen": -0.044171880930662155,
331
+ "rewards/margins": 0.006505946163088083,
332
+ "rewards/rejected": -0.0506778284907341,
333
+ "step": 180
334
+ },
335
+ {
336
+ "epoch": 0.9,
337
+ "grad_norm": 9.615534174630943,
338
+ "learning_rate": 4.656416723360974e-06,
339
+ "log_odds_chosen": 0.24925272166728973,
340
+ "log_odds_ratio": -0.6496003866195679,
341
+ "logits/chosen": -2.6970863342285156,
342
+ "logits/rejected": -2.6921534538269043,
343
+ "logps/chosen": -0.8308185338973999,
344
+ "logps/rejected": -0.9803632497787476,
345
+ "loss": 0.9089,
346
+ "nll_loss": 0.8415037393569946,
347
+ "rewards/accuracies": 0.675000011920929,
348
+ "rewards/chosen": -0.04154093191027641,
349
+ "rewards/margins": 0.0074772341176867485,
350
+ "rewards/rejected": -0.04901815950870514,
351
+ "step": 190
352
+ },
353
+ {
354
+ "epoch": 0.95,
355
+ "grad_norm": 9.799715635757147,
356
+ "learning_rate": 4.578162229495714e-06,
357
+ "log_odds_chosen": 0.17407485842704773,
358
+ "log_odds_ratio": -0.6980692148208618,
359
+ "logits/chosen": -2.762956142425537,
360
+ "logits/rejected": -2.768340826034546,
361
+ "logps/chosen": -0.8834056854248047,
362
+ "logps/rejected": -1.005025863647461,
363
+ "loss": 0.9565,
364
+ "nll_loss": 0.9161331057548523,
365
+ "rewards/accuracies": 0.6000000238418579,
366
+ "rewards/chosen": -0.044170282781124115,
367
+ "rewards/margins": 0.006081007421016693,
368
+ "rewards/rejected": -0.05025129392743111,
369
+ "step": 200
370
+ },
371
+ {
372
+ "epoch": 1.0,
373
+ "grad_norm": 9.940266165525735,
374
+ "learning_rate": 4.492690032117859e-06,
375
+ "log_odds_chosen": 0.1925085484981537,
376
+ "log_odds_ratio": -0.6957636475563049,
377
+ "logits/chosen": -2.753004550933838,
378
+ "logits/rejected": -2.742924690246582,
379
+ "logps/chosen": -0.9696134328842163,
380
+ "logps/rejected": -1.1040910482406616,
381
+ "loss": 0.974,
382
+ "nll_loss": 0.9876837730407715,
383
+ "rewards/accuracies": 0.574999988079071,
384
+ "rewards/chosen": -0.04848066717386246,
385
+ "rewards/margins": 0.006723886821419001,
386
+ "rewards/rejected": -0.05520455911755562,
387
+ "step": 210
388
+ },
389
+ {
390
+ "epoch": 1.0,
391
+ "eval_log_odds_chosen": 0.35508599877357483,
392
+ "eval_log_odds_ratio": -0.6352247595787048,
393
+ "eval_logits/chosen": -2.751180410385132,
394
+ "eval_logits/rejected": -2.754164934158325,
395
+ "eval_logps/chosen": -0.8390052914619446,
396
+ "eval_logps/rejected": -1.0409260988235474,
397
+ "eval_loss": 0.8980213403701782,
398
+ "eval_nll_loss": 0.8887932896614075,
399
+ "eval_rewards/accuracies": 0.6145833134651184,
400
+ "eval_rewards/chosen": -0.04195027053356171,
401
+ "eval_rewards/margins": 0.01009603776037693,
402
+ "eval_rewards/rejected": -0.05204630270600319,
403
+ "eval_runtime": 57.3212,
404
+ "eval_samples_per_second": 13.084,
405
+ "eval_steps_per_second": 0.419,
406
+ "step": 211
407
+ },
408
+ {
409
+ "epoch": 1.04,
410
+ "grad_norm": 8.638509296818858,
411
+ "learning_rate": 4.400296986290258e-06,
412
+ "log_odds_chosen": 1.3975646495819092,
413
+ "log_odds_ratio": -0.32538530230522156,
414
+ "logits/chosen": -2.7828710079193115,
415
+ "logits/rejected": -2.7821409702301025,
416
+ "logps/chosen": -0.446835994720459,
417
+ "logps/rejected": -1.1457499265670776,
418
+ "loss": 0.5455,
419
+ "nll_loss": 0.4835132956504822,
420
+ "rewards/accuracies": 0.8999999761581421,
421
+ "rewards/chosen": -0.02234180085361004,
422
+ "rewards/margins": 0.03494570404291153,
423
+ "rewards/rejected": -0.05728749558329582,
424
+ "step": 220
425
+ },
426
+ {
427
+ "epoch": 1.09,
428
+ "grad_norm": 9.543274613242575,
429
+ "learning_rate": 4.3013039840019675e-06,
430
+ "log_odds_chosen": 1.6737674474716187,
431
+ "log_odds_ratio": -0.23729205131530762,
432
+ "logits/chosen": -2.860888957977295,
433
+ "logits/rejected": -2.84743332862854,
434
+ "logps/chosen": -0.4813949167728424,
435
+ "logps/rejected": -1.4018938541412354,
436
+ "loss": 0.5041,
437
+ "nll_loss": 0.5257013440132141,
438
+ "rewards/accuracies": 0.9750000238418579,
439
+ "rewards/chosen": -0.02406974695622921,
440
+ "rewards/margins": 0.04602494835853577,
441
+ "rewards/rejected": -0.07009468972682953,
442
+ "step": 230
443
+ },
444
+ {
445
+ "epoch": 1.14,
446
+ "grad_norm": 9.413079564715025,
447
+ "learning_rate": 4.196054839672382e-06,
448
+ "log_odds_chosen": 1.4506713151931763,
449
+ "log_odds_ratio": -0.31961891055107117,
450
+ "logits/chosen": -2.848227024078369,
451
+ "logits/rejected": -2.849867820739746,
452
+ "logps/chosen": -0.46944475173950195,
453
+ "logps/rejected": -1.153623342514038,
454
+ "loss": 0.4801,
455
+ "nll_loss": 0.4581332802772522,
456
+ "rewards/accuracies": 0.925000011920929,
457
+ "rewards/chosen": -0.023472238332033157,
458
+ "rewards/margins": 0.03420892730355263,
459
+ "rewards/rejected": -0.057681165635585785,
460
+ "step": 240
461
+ },
462
+ {
463
+ "epoch": 1.18,
464
+ "grad_norm": 9.405658606526101,
465
+ "learning_rate": 4.084915096043035e-06,
466
+ "log_odds_chosen": 1.3965736627578735,
467
+ "log_odds_ratio": -0.3138691484928131,
468
+ "logits/chosen": -2.8238320350646973,
469
+ "logits/rejected": -2.811922311782837,
470
+ "logps/chosen": -0.48123639822006226,
471
+ "logps/rejected": -1.1604976654052734,
472
+ "loss": 0.5396,
473
+ "nll_loss": 0.5218371748924255,
474
+ "rewards/accuracies": 0.887499988079071,
475
+ "rewards/chosen": -0.024061817675828934,
476
+ "rewards/margins": 0.03396306931972504,
477
+ "rewards/rejected": -0.05802489444613457,
478
+ "step": 250
479
+ },
480
+ {
481
+ "epoch": 1.23,
482
+ "grad_norm": 9.830258381762862,
483
+ "learning_rate": 3.9682707546043785e-06,
484
+ "log_odds_chosen": 1.5788991451263428,
485
+ "log_odds_ratio": -0.28013017773628235,
486
+ "logits/chosen": -2.7972917556762695,
487
+ "logits/rejected": -2.7948007583618164,
488
+ "logps/chosen": -0.4086516797542572,
489
+ "logps/rejected": -1.1518056392669678,
490
+ "loss": 0.4785,
491
+ "nll_loss": 0.4458548128604889,
492
+ "rewards/accuracies": 0.9125000238418579,
493
+ "rewards/chosen": -0.02043258398771286,
494
+ "rewards/margins": 0.03715769201517105,
495
+ "rewards/rejected": -0.05759027600288391,
496
+ "step": 260
497
+ },
498
+ {
499
+ "epoch": 1.28,
500
+ "grad_norm": 8.095849996024462,
501
+ "learning_rate": 3.846526934966891e-06,
502
+ "log_odds_chosen": 1.360436201095581,
503
+ "log_odds_ratio": -0.33352774381637573,
504
+ "logits/chosen": -2.7501392364501953,
505
+ "logits/rejected": -2.7644224166870117,
506
+ "logps/chosen": -0.43927937746047974,
507
+ "logps/rejected": -1.0804184675216675,
508
+ "loss": 0.4998,
509
+ "nll_loss": 0.4916819632053375,
510
+ "rewards/accuracies": 0.9125000238418579,
511
+ "rewards/chosen": -0.021963967010378838,
512
+ "rewards/margins": 0.03205695003271103,
513
+ "rewards/rejected": -0.054020922631025314,
514
+ "step": 270
515
+ },
516
+ {
517
+ "epoch": 1.33,
518
+ "grad_norm": 8.39958498646755,
519
+ "learning_rate": 3.7201064678327008e-06,
520
+ "log_odds_chosen": 1.620120644569397,
521
+ "log_odds_ratio": -0.26243138313293457,
522
+ "logits/chosen": -2.7666475772857666,
523
+ "logits/rejected": -2.756240129470825,
524
+ "logps/chosen": -0.4216586947441101,
525
+ "logps/rejected": -1.1700552701950073,
526
+ "loss": 0.4808,
527
+ "nll_loss": 0.4861266016960144,
528
+ "rewards/accuracies": 0.949999988079071,
529
+ "rewards/chosen": -0.021082933992147446,
530
+ "rewards/margins": 0.03741983324289322,
531
+ "rewards/rejected": -0.058502763509750366,
532
+ "step": 280
533
+ },
534
+ {
535
+ "epoch": 1.37,
536
+ "grad_norm": 9.702907898591407,
537
+ "learning_rate": 3.589448426454486e-06,
538
+ "log_odds_chosen": 1.5312963724136353,
539
+ "log_odds_ratio": -0.26756590604782104,
540
+ "logits/chosen": -2.716158390045166,
541
+ "logits/rejected": -2.7521204948425293,
542
+ "logps/chosen": -0.4387860894203186,
543
+ "logps/rejected": -1.1713616847991943,
544
+ "loss": 0.4853,
545
+ "nll_loss": 0.479191392660141,
546
+ "rewards/accuracies": 0.9375,
547
+ "rewards/chosen": -0.02193930372595787,
548
+ "rewards/margins": 0.036628786474466324,
549
+ "rewards/rejected": -0.0585680827498436,
550
+ "step": 290
551
+ },
552
+ {
553
+ "epoch": 1.42,
554
+ "grad_norm": 9.595382828818373,
555
+ "learning_rate": 3.4550066016820755e-06,
556
+ "log_odds_chosen": 1.752833604812622,
557
+ "log_odds_ratio": -0.2656395733356476,
558
+ "logits/chosen": -2.7655932903289795,
559
+ "logits/rejected": -2.7632648944854736,
560
+ "logps/chosen": -0.39629921317100525,
561
+ "logps/rejected": -1.1531693935394287,
562
+ "loss": 0.4794,
563
+ "nll_loss": 0.4017585217952728,
564
+ "rewards/accuracies": 0.925000011920929,
565
+ "rewards/chosen": -0.01981496252119541,
566
+ "rewards/margins": 0.037843503057956696,
567
+ "rewards/rejected": -0.05765846371650696,
568
+ "step": 300
569
+ },
570
+ {
571
+ "epoch": 1.47,
572
+ "grad_norm": 8.842462954214794,
573
+ "learning_rate": 3.317247925893089e-06,
574
+ "log_odds_chosen": 1.7586959600448608,
575
+ "log_odds_ratio": -0.24750974774360657,
576
+ "logits/chosen": -2.7276859283447266,
577
+ "logits/rejected": -2.7263400554656982,
578
+ "logps/chosen": -0.40662258863449097,
579
+ "logps/rejected": -1.34078848361969,
580
+ "loss": 0.4871,
581
+ "nll_loss": 0.43305450677871704,
582
+ "rewards/accuracies": 0.987500011920929,
583
+ "rewards/chosen": -0.020331131294369698,
584
+ "rewards/margins": 0.04670828953385353,
585
+ "rewards/rejected": -0.06703941524028778,
586
+ "step": 310
587
+ },
588
+ {
589
+ "epoch": 1.52,
590
+ "grad_norm": 8.573977166367447,
591
+ "learning_rate": 3.1766508512814997e-06,
592
+ "log_odds_chosen": 1.6913954019546509,
593
+ "log_odds_ratio": -0.23067903518676758,
594
+ "logits/chosen": -2.7266640663146973,
595
+ "logits/rejected": -2.694443702697754,
596
+ "logps/chosen": -0.4407864212989807,
597
+ "logps/rejected": -1.3253118991851807,
598
+ "loss": 0.4717,
599
+ "nll_loss": 0.4706190228462219,
600
+ "rewards/accuracies": 0.9750000238418579,
601
+ "rewards/chosen": -0.022039320319890976,
602
+ "rewards/margins": 0.044226277619600296,
603
+ "rewards/rejected": -0.06626559793949127,
604
+ "step": 320
605
+ },
606
+ {
607
+ "epoch": 1.56,
608
+ "grad_norm": 9.155832857330214,
609
+ "learning_rate": 3.0337036881365045e-06,
610
+ "log_odds_chosen": 1.499769926071167,
611
+ "log_odds_ratio": -0.2850986421108246,
612
+ "logits/chosen": -2.7671217918395996,
613
+ "logits/rejected": -2.7773778438568115,
614
+ "logps/chosen": -0.44799503684043884,
615
+ "logps/rejected": -1.2192200422286987,
616
+ "loss": 0.4549,
617
+ "nll_loss": 0.49383026361465454,
618
+ "rewards/accuracies": 0.925000011920929,
619
+ "rewards/chosen": -0.022399749606847763,
620
+ "rewards/margins": 0.038561251014471054,
621
+ "rewards/rejected": -0.060961008071899414,
622
+ "step": 330
623
+ },
624
+ {
625
+ "epoch": 1.61,
626
+ "grad_norm": 8.972176715125862,
627
+ "learning_rate": 2.888902908883069e-06,
628
+ "log_odds_chosen": 1.6073827743530273,
629
+ "log_odds_ratio": -0.25966399908065796,
630
+ "logits/chosen": -2.729976177215576,
631
+ "logits/rejected": -2.7247419357299805,
632
+ "logps/chosen": -0.3783184289932251,
633
+ "logps/rejected": -1.1157445907592773,
634
+ "loss": 0.503,
635
+ "nll_loss": 0.4228973388671875,
636
+ "rewards/accuracies": 0.9624999761581421,
637
+ "rewards/chosen": -0.018915923312306404,
638
+ "rewards/margins": 0.03687131032347679,
639
+ "rewards/rejected": -0.055787235498428345,
640
+ "step": 340
641
+ },
642
+ {
643
+ "epoch": 1.66,
644
+ "grad_norm": 8.93142291410328,
645
+ "learning_rate": 2.7427514237744e-06,
646
+ "log_odds_chosen": 1.4399652481079102,
647
+ "log_odds_ratio": -0.311657577753067,
648
+ "logits/chosen": -2.764589786529541,
649
+ "logits/rejected": -2.756528377532959,
650
+ "logps/chosen": -0.469074547290802,
651
+ "logps/rejected": -1.1970831155776978,
652
+ "loss": 0.4785,
653
+ "nll_loss": 0.4543870985507965,
654
+ "rewards/accuracies": 0.949999988079071,
655
+ "rewards/chosen": -0.0234537310898304,
656
+ "rewards/margins": 0.03640042990446091,
657
+ "rewards/rejected": -0.059854160994291306,
658
+ "step": 350
659
+ },
660
+ {
661
+ "epoch": 1.71,
662
+ "grad_norm": 10.098847690798312,
663
+ "learning_rate": 2.595756834225089e-06,
664
+ "log_odds_chosen": 1.34963059425354,
665
+ "log_odds_ratio": -0.3275851607322693,
666
+ "logits/chosen": -2.7097485065460205,
667
+ "logits/rejected": -2.73240327835083,
668
+ "logps/chosen": -0.4670106768608093,
669
+ "logps/rejected": -1.1147117614746094,
670
+ "loss": 0.5122,
671
+ "nll_loss": 0.5337762832641602,
672
+ "rewards/accuracies": 0.8999999761581421,
673
+ "rewards/chosen": -0.023350534960627556,
674
+ "rewards/margins": 0.03238505870103836,
675
+ "rewards/rejected": -0.055735595524311066,
676
+ "step": 360
677
+ },
678
+ {
679
+ "epoch": 1.75,
680
+ "grad_norm": 8.104028348470598,
681
+ "learning_rate": 2.448429669851285e-06,
682
+ "log_odds_chosen": 1.480303168296814,
683
+ "log_odds_ratio": -0.28436464071273804,
684
+ "logits/chosen": -2.731992483139038,
685
+ "logits/rejected": -2.7325727939605713,
686
+ "logps/chosen": -0.4118678569793701,
687
+ "logps/rejected": -1.0783556699752808,
688
+ "loss": 0.4646,
689
+ "nll_loss": 0.4363551139831543,
690
+ "rewards/accuracies": 0.9624999761581421,
691
+ "rewards/chosen": -0.020593393594026566,
692
+ "rewards/margins": 0.03332439437508583,
693
+ "rewards/rejected": -0.053917787969112396,
694
+ "step": 370
695
+ },
696
+ {
697
+ "epoch": 1.8,
698
+ "grad_norm": 9.21288938005839,
699
+ "learning_rate": 2.301281615340886e-06,
700
+ "log_odds_chosen": 1.5375401973724365,
701
+ "log_odds_ratio": -0.2999517321586609,
702
+ "logits/chosen": -2.7647042274475098,
703
+ "logits/rejected": -2.7518932819366455,
704
+ "logps/chosen": -0.4663173258304596,
705
+ "logps/rejected": -1.2805049419403076,
706
+ "loss": 0.5002,
707
+ "nll_loss": 0.48935002088546753,
708
+ "rewards/accuracies": 0.9125000238418579,
709
+ "rewards/chosen": -0.02331586554646492,
710
+ "rewards/margins": 0.04070938378572464,
711
+ "rewards/rejected": -0.06402524560689926,
712
+ "step": 380
713
+ },
714
+ {
715
+ "epoch": 1.85,
716
+ "grad_norm": 8.485354651581948,
717
+ "learning_rate": 2.154823733312027e-06,
718
+ "log_odds_chosen": 1.4252521991729736,
719
+ "log_odds_ratio": -0.2968854308128357,
720
+ "logits/chosen": -2.723702907562256,
721
+ "logits/rejected": -2.7239809036254883,
722
+ "logps/chosen": -0.4829932153224945,
723
+ "logps/rejected": -1.1818339824676514,
724
+ "loss": 0.5363,
725
+ "nll_loss": 0.524863600730896,
726
+ "rewards/accuracies": 0.925000011920929,
727
+ "rewards/chosen": -0.024149660021066666,
728
+ "rewards/margins": 0.03494204580783844,
729
+ "rewards/rejected": -0.059091709554195404,
730
+ "step": 390
731
+ },
732
+ {
733
+ "epoch": 1.9,
734
+ "grad_norm": 9.4111350259119,
735
+ "learning_rate": 2.009564689332083e-06,
736
+ "log_odds_chosen": 1.6615612506866455,
737
+ "log_odds_ratio": -0.27174240350723267,
738
+ "logits/chosen": -2.7479844093322754,
739
+ "logits/rejected": -2.752495288848877,
740
+ "logps/chosen": -0.42523106932640076,
741
+ "logps/rejected": -1.2396804094314575,
742
+ "loss": 0.4901,
743
+ "nll_loss": 0.4686039984226227,
744
+ "rewards/accuracies": 0.9125000238418579,
745
+ "rewards/chosen": -0.02126155234873295,
746
+ "rewards/margins": 0.04072246700525284,
747
+ "rewards/rejected": -0.061984021216630936,
748
+ "step": 400
749
+ },
750
+ {
751
+ "epoch": 1.94,
752
+ "grad_norm": 8.90962135731115,
753
+ "learning_rate": 1.866008985261924e-06,
754
+ "log_odds_chosen": 1.4581377506256104,
755
+ "log_odds_ratio": -0.31245434284210205,
756
+ "logits/chosen": -2.7534964084625244,
757
+ "logits/rejected": -2.7579829692840576,
758
+ "logps/chosen": -0.4420735239982605,
759
+ "logps/rejected": -1.1250512599945068,
760
+ "loss": 0.4719,
761
+ "nll_loss": 0.4840390682220459,
762
+ "rewards/accuracies": 0.875,
763
+ "rewards/chosen": -0.022103676572442055,
764
+ "rewards/margins": 0.034148890525102615,
765
+ "rewards/rejected": -0.05625256896018982,
766
+ "step": 410
767
+ },
768
+ {
769
+ "epoch": 1.99,
770
+ "grad_norm": 10.982648615730804,
771
+ "learning_rate": 1.7246552070612087e-06,
772
+ "log_odds_chosen": 1.7540910243988037,
773
+ "log_odds_ratio": -0.25998422503471375,
774
+ "logits/chosen": -2.738429546356201,
775
+ "logits/rejected": -2.7384934425354004,
776
+ "logps/chosen": -0.36052900552749634,
777
+ "logps/rejected": -1.2116727828979492,
778
+ "loss": 0.472,
779
+ "nll_loss": 0.4113571047782898,
780
+ "rewards/accuracies": 0.925000011920929,
781
+ "rewards/chosen": -0.018026452511548996,
782
+ "rewards/margins": 0.042557187378406525,
783
+ "rewards/rejected": -0.060583632439374924,
784
+ "step": 420
785
+ },
786
+ {
787
+ "epoch": 2.0,
788
+ "eval_log_odds_chosen": 0.4637714624404907,
789
+ "eval_log_odds_ratio": -0.6059489846229553,
790
+ "eval_logits/chosen": -2.7596328258514404,
791
+ "eval_logits/rejected": -2.7660648822784424,
792
+ "eval_logps/chosen": -0.8737161159515381,
793
+ "eval_logps/rejected": -1.1384042501449585,
794
+ "eval_loss": 0.9422717094421387,
795
+ "eval_nll_loss": 0.9339601397514343,
796
+ "eval_rewards/accuracies": 0.640625,
797
+ "eval_rewards/chosen": -0.04368580877780914,
798
+ "eval_rewards/margins": 0.013234402984380722,
799
+ "eval_rewards/rejected": -0.056920215487480164,
800
+ "eval_runtime": 57.3721,
801
+ "eval_samples_per_second": 13.073,
802
+ "eval_steps_per_second": 0.418,
803
+ "step": 422
804
+ },
805
+ {
806
+ "epoch": 2.04,
807
+ "grad_norm": 9.26767033665704,
808
+ "learning_rate": 1.585994293140307e-06,
809
+ "log_odds_chosen": 2.6479477882385254,
810
+ "log_odds_ratio": -0.14060619473457336,
811
+ "logits/chosen": -2.7426111698150635,
812
+ "logits/rejected": -2.7613494396209717,
813
+ "logps/chosen": -0.2339097261428833,
814
+ "logps/rejected": -1.2145100831985474,
815
+ "loss": 0.2354,
816
+ "nll_loss": 0.1956883668899536,
817
+ "rewards/accuracies": 0.9624999761581421,
818
+ "rewards/chosen": -0.011695485562086105,
819
+ "rewards/margins": 0.04903002083301544,
820
+ "rewards/rejected": -0.06072550266981125,
821
+ "step": 430
822
+ },
823
+ {
824
+ "epoch": 2.09,
825
+ "grad_norm": 7.3805660163482125,
826
+ "learning_rate": 1.4505078292730631e-06,
827
+ "log_odds_chosen": 3.2394683361053467,
828
+ "log_odds_ratio": -0.07881985604763031,
829
+ "logits/chosen": -2.725740909576416,
830
+ "logits/rejected": -2.7506771087646484,
831
+ "logps/chosen": -0.13437342643737793,
832
+ "logps/rejected": -1.3584119081497192,
833
+ "loss": 0.1704,
834
+ "nll_loss": 0.1273941844701767,
835
+ "rewards/accuracies": 0.987500011920929,
836
+ "rewards/chosen": -0.006718672811985016,
837
+ "rewards/margins": 0.06120193004608154,
838
+ "rewards/rejected": -0.06792060285806656,
839
+ "step": 440
840
+ },
841
+ {
842
+ "epoch": 2.13,
843
+ "grad_norm": 8.138860664723373,
844
+ "learning_rate": 1.3186663759923782e-06,
845
+ "log_odds_chosen": 3.6539645195007324,
846
+ "log_odds_ratio": -0.046128250658512115,
847
+ "logits/chosen": -2.75630521774292,
848
+ "logits/rejected": -2.7848801612854004,
849
+ "logps/chosen": -0.12164908647537231,
850
+ "logps/rejected": -1.645410180091858,
851
+ "loss": 0.1575,
852
+ "nll_loss": 0.13616196811199188,
853
+ "rewards/accuracies": 1.0,
854
+ "rewards/chosen": -0.006082454230636358,
855
+ "rewards/margins": 0.07618805766105652,
856
+ "rewards/rejected": -0.08227051794528961,
857
+ "step": 450
858
+ },
859
+ {
860
+ "epoch": 2.18,
861
+ "grad_norm": 8.550456232090282,
862
+ "learning_rate": 1.1909278342777513e-06,
863
+ "log_odds_chosen": 3.178800106048584,
864
+ "log_odds_ratio": -0.07669878005981445,
865
+ "logits/chosen": -2.731027841567993,
866
+ "logits/rejected": -2.7491018772125244,
867
+ "logps/chosen": -0.153743177652359,
868
+ "logps/rejected": -1.2796289920806885,
869
+ "loss": 0.1752,
870
+ "nll_loss": 0.19261114299297333,
871
+ "rewards/accuracies": 1.0,
872
+ "rewards/chosen": -0.0076871588826179504,
873
+ "rewards/margins": 0.05629429221153259,
874
+ "rewards/rejected": -0.06398145854473114,
875
+ "step": 460
876
+ },
877
+ {
878
+ "epoch": 2.23,
879
+ "grad_norm": 8.542489688724254,
880
+ "learning_rate": 1.0677358552109617e-06,
881
+ "log_odds_chosen": 3.4661343097686768,
882
+ "log_odds_ratio": -0.0592988021671772,
883
+ "logits/chosen": -2.7476649284362793,
884
+ "logits/rejected": -2.74284291267395,
885
+ "logps/chosen": -0.14152076840400696,
886
+ "logps/rejected": -1.603075623512268,
887
+ "loss": 0.1517,
888
+ "nll_loss": 0.1476079523563385,
889
+ "rewards/accuracies": 1.0,
890
+ "rewards/chosen": -0.007076038978993893,
891
+ "rewards/margins": 0.0730777308344841,
892
+ "rewards/rejected": -0.08015377819538116,
893
+ "step": 470
894
+ },
895
+ {
896
+ "epoch": 2.27,
897
+ "grad_norm": 7.078338821762873,
898
+ "learning_rate": 9.495182991233237e-07,
899
+ "log_odds_chosen": 3.5714924335479736,
900
+ "log_odds_ratio": -0.05693133547902107,
901
+ "logits/chosen": -2.7550268173217773,
902
+ "logits/rejected": -2.76568341255188,
903
+ "logps/chosen": -0.1448669731616974,
904
+ "logps/rejected": -1.6731277704238892,
905
+ "loss": 0.1571,
906
+ "nll_loss": 0.1459461897611618,
907
+ "rewards/accuracies": 1.0,
908
+ "rewards/chosen": -0.007243348751217127,
909
+ "rewards/margins": 0.07641303539276123,
910
+ "rewards/rejected": -0.08365638554096222,
911
+ "step": 480
912
+ },
913
+ {
914
+ "epoch": 2.32,
915
+ "grad_norm": 7.953863925813501,
916
+ "learning_rate": 8.366857495860869e-07,
917
+ "log_odds_chosen": 3.53666353225708,
918
+ "log_odds_ratio": -0.06214044615626335,
919
+ "logits/chosen": -2.714313507080078,
920
+ "logits/rejected": -2.744480848312378,
921
+ "logps/chosen": -0.10873039066791534,
922
+ "logps/rejected": -1.4339990615844727,
923
+ "loss": 0.1535,
924
+ "nll_loss": 0.11224369704723358,
925
+ "rewards/accuracies": 0.987500011920929,
926
+ "rewards/chosen": -0.005436519160866737,
927
+ "rewards/margins": 0.06626342982053757,
928
+ "rewards/rejected": -0.07169995456933975,
929
+ "step": 490
930
+ },
931
+ {
932
+ "epoch": 2.37,
933
+ "grad_norm": 7.8747727873348765,
934
+ "learning_rate": 7.296300874050766e-07,
935
+ "log_odds_chosen": 3.4877212047576904,
936
+ "log_odds_ratio": -0.04788507893681526,
937
+ "logits/chosen": -2.7288925647735596,
938
+ "logits/rejected": -2.755156993865967,
939
+ "logps/chosen": -0.12493199110031128,
940
+ "logps/rejected": -1.555891513824463,
941
+ "loss": 0.1404,
942
+ "nll_loss": 0.13063213229179382,
943
+ "rewards/accuracies": 1.0,
944
+ "rewards/chosen": -0.006246599368751049,
945
+ "rewards/margins": 0.0715479776263237,
946
+ "rewards/rejected": -0.07779457420110703,
947
+ "step": 500
948
+ },
949
+ {
950
+ "epoch": 2.42,
951
+ "grad_norm": 8.163301368292721,
952
+ "learning_rate": 6.28723129572247e-07,
953
+ "log_odds_chosen": 3.2528719902038574,
954
+ "log_odds_ratio": -0.06903894245624542,
955
+ "logits/chosen": -2.723130941390991,
956
+ "logits/rejected": -2.731053113937378,
957
+ "logps/chosen": -0.1517888903617859,
958
+ "logps/rejected": -1.4444904327392578,
959
+ "loss": 0.1647,
960
+ "nll_loss": 0.1824251264333725,
961
+ "rewards/accuracies": 1.0,
962
+ "rewards/chosen": -0.00758944358676672,
963
+ "rewards/margins": 0.06463508307933807,
964
+ "rewards/rejected": -0.07222452014684677,
965
+ "step": 510
966
+ },
967
+ {
968
+ "epoch": 2.46,
969
+ "grad_norm": 8.191096860807814,
970
+ "learning_rate": 5.343153379012445e-07,
971
+ "log_odds_chosen": 3.613795042037964,
972
+ "log_odds_ratio": -0.05437632277607918,
973
+ "logits/chosen": -2.7185778617858887,
974
+ "logits/rejected": -2.738831043243408,
975
+ "logps/chosen": -0.1386043280363083,
976
+ "logps/rejected": -1.6129226684570312,
977
+ "loss": 0.1547,
978
+ "nll_loss": 0.14940614998340607,
979
+ "rewards/accuracies": 1.0,
980
+ "rewards/chosen": -0.006930218078196049,
981
+ "rewards/margins": 0.07371591031551361,
982
+ "rewards/rejected": -0.08064613491296768,
983
+ "step": 520
984
+ },
985
+ {
986
+ "epoch": 2.51,
987
+ "grad_norm": 7.911676913184044,
988
+ "learning_rate": 4.467346018320198e-07,
989
+ "log_odds_chosen": 3.289904832839966,
990
+ "log_odds_ratio": -0.07905766367912292,
991
+ "logits/chosen": -2.738126754760742,
992
+ "logits/rejected": -2.765601634979248,
993
+ "logps/chosen": -0.15802282094955444,
994
+ "logps/rejected": -1.4789544343948364,
995
+ "loss": 0.164,
996
+ "nll_loss": 0.16227465867996216,
997
+ "rewards/accuracies": 0.987500011920929,
998
+ "rewards/chosen": -0.007901140488684177,
999
+ "rewards/margins": 0.0660465732216835,
1000
+ "rewards/rejected": -0.0739477127790451,
1001
+ "step": 530
1002
+ },
1003
+ {
1004
+ "epoch": 2.56,
1005
+ "grad_norm": 8.501165443462748,
1006
+ "learning_rate": 3.662850996319825e-07,
1007
+ "log_odds_chosen": 3.346526622772217,
1008
+ "log_odds_ratio": -0.0591273196041584,
1009
+ "logits/chosen": -2.6961731910705566,
1010
+ "logits/rejected": -2.705780506134033,
1011
+ "logps/chosen": -0.1303931623697281,
1012
+ "logps/rejected": -1.3868381977081299,
1013
+ "loss": 0.1629,
1014
+ "nll_loss": 0.1521378755569458,
1015
+ "rewards/accuracies": 1.0,
1016
+ "rewards/chosen": -0.006519658025354147,
1017
+ "rewards/margins": 0.06282225996255875,
1018
+ "rewards/rejected": -0.06934191286563873,
1019
+ "step": 540
1020
+ },
1021
+ {
1022
+ "epoch": 2.61,
1023
+ "grad_norm": 9.554188547629353,
1024
+ "learning_rate": 2.932462419488544e-07,
1025
+ "log_odds_chosen": 3.749110460281372,
1026
+ "log_odds_ratio": -0.04820709303021431,
1027
+ "logits/chosen": -2.7394113540649414,
1028
+ "logits/rejected": -2.752619981765747,
1029
+ "logps/chosen": -0.1146271601319313,
1030
+ "logps/rejected": -1.5809662342071533,
1031
+ "loss": 0.133,
1032
+ "nll_loss": 0.13163451850414276,
1033
+ "rewards/accuracies": 0.987500011920929,
1034
+ "rewards/chosen": -0.005731357727199793,
1035
+ "rewards/margins": 0.07331694662570953,
1036
+ "rewards/rejected": -0.07904830574989319,
1037
+ "step": 550
1038
+ },
1039
+ {
1040
+ "epoch": 2.65,
1041
+ "grad_norm": 8.43728434003558,
1042
+ "learning_rate": 2.278717013844059e-07,
1043
+ "log_odds_chosen": 3.417363405227661,
1044
+ "log_odds_ratio": -0.057405441999435425,
1045
+ "logits/chosen": -2.717993974685669,
1046
+ "logits/rejected": -2.7329213619232178,
1047
+ "logps/chosen": -0.12498118728399277,
1048
+ "logps/rejected": -1.3357540369033813,
1049
+ "loss": 0.1578,
1050
+ "nll_loss": 0.13818371295928955,
1051
+ "rewards/accuracies": 1.0,
1052
+ "rewards/chosen": -0.0062490589916706085,
1053
+ "rewards/margins": 0.0605386421084404,
1054
+ "rewards/rejected": -0.06678769737482071,
1055
+ "step": 560
1056
+ },
1057
+ {
1058
+ "epoch": 2.7,
1059
+ "grad_norm": 7.63488898142434,
1060
+ "learning_rate": 1.7038853145946804e-07,
1061
+ "log_odds_chosen": 3.4726767539978027,
1062
+ "log_odds_ratio": -0.05518152192234993,
1063
+ "logits/chosen": -2.723026752471924,
1064
+ "logits/rejected": -2.7421092987060547,
1065
+ "logps/chosen": -0.1284557282924652,
1066
+ "logps/rejected": -1.4232304096221924,
1067
+ "loss": 0.1488,
1068
+ "nll_loss": 0.1334700882434845,
1069
+ "rewards/accuracies": 1.0,
1070
+ "rewards/chosen": -0.006422785576432943,
1071
+ "rewards/margins": 0.06473874300718307,
1072
+ "rewards/rejected": -0.07116152346134186,
1073
+ "step": 570
1074
+ },
1075
+ {
1076
+ "epoch": 2.75,
1077
+ "grad_norm": 7.573438983056155,
1078
+ "learning_rate": 1.2099637803016983e-07,
1079
+ "log_odds_chosen": 3.2544848918914795,
1080
+ "log_odds_ratio": -0.07813612371683121,
1081
+ "logits/chosen": -2.7465689182281494,
1082
+ "logits/rejected": -2.760995388031006,
1083
+ "logps/chosen": -0.2028094083070755,
1084
+ "logps/rejected": -1.4702503681182861,
1085
+ "loss": 0.1724,
1086
+ "nll_loss": 0.21864549815654755,
1087
+ "rewards/accuracies": 0.987500011920929,
1088
+ "rewards/chosen": -0.01014047022908926,
1089
+ "rewards/margins": 0.06337204575538635,
1090
+ "rewards/rejected": -0.07351251691579819,
1091
+ "step": 580
1092
+ },
1093
+ {
1094
+ "epoch": 2.8,
1095
+ "grad_norm": 7.42006307957704,
1096
+ "learning_rate": 7.986678589423757e-08,
1097
+ "log_odds_chosen": 3.658898115158081,
1098
+ "log_odds_ratio": -0.043824274092912674,
1099
+ "logits/chosen": -2.735332489013672,
1100
+ "logits/rejected": -2.7602334022521973,
1101
+ "logps/chosen": -0.13853900134563446,
1102
+ "logps/rejected": -1.616575837135315,
1103
+ "loss": 0.1584,
1104
+ "nll_loss": 0.1556694209575653,
1105
+ "rewards/accuracies": 1.0,
1106
+ "rewards/chosen": -0.006926950998604298,
1107
+ "rewards/margins": 0.07390183210372925,
1108
+ "rewards/rejected": -0.08082877844572067,
1109
+ "step": 590
1110
+ },
1111
+ {
1112
+ "epoch": 2.84,
1113
+ "grad_norm": 7.043177228838798,
1114
+ "learning_rate": 4.7142602995598754e-08,
1115
+ "log_odds_chosen": 3.6085071563720703,
1116
+ "log_odds_ratio": -0.05147845670580864,
1117
+ "logits/chosen": -2.7238612174987793,
1118
+ "logits/rejected": -2.7488350868225098,
1119
+ "logps/chosen": -0.12226458638906479,
1120
+ "logps/rejected": -1.4737613201141357,
1121
+ "loss": 0.147,
1122
+ "nll_loss": 0.13915380835533142,
1123
+ "rewards/accuracies": 1.0,
1124
+ "rewards/chosen": -0.006113228853791952,
1125
+ "rewards/margins": 0.06757483631372452,
1126
+ "rewards/rejected": -0.0736880749464035,
1127
+ "step": 600
1128
+ },
1129
+ {
1130
+ "epoch": 2.89,
1131
+ "grad_norm": 8.244444250935933,
1132
+ "learning_rate": 2.2937484296556566e-08,
1133
+ "log_odds_chosen": 3.4587013721466064,
1134
+ "log_odds_ratio": -0.05050401762127876,
1135
+ "logits/chosen": -2.7173829078674316,
1136
+ "logits/rejected": -2.74656343460083,
1137
+ "logps/chosen": -0.11210664361715317,
1138
+ "logps/rejected": -1.383599042892456,
1139
+ "loss": 0.145,
1140
+ "nll_loss": 0.12569257616996765,
1141
+ "rewards/accuracies": 1.0,
1142
+ "rewards/chosen": -0.005605332553386688,
1143
+ "rewards/margins": 0.06357462704181671,
1144
+ "rewards/rejected": -0.0691799595952034,
1145
+ "step": 610
1146
+ },
1147
+ {
1148
+ "epoch": 2.94,
1149
+ "grad_norm": 8.912901656210648,
1150
+ "learning_rate": 7.335497040648898e-09,
1151
+ "log_odds_chosen": 3.397965908050537,
1152
+ "log_odds_ratio": -0.05181251838803291,
1153
+ "logits/chosen": -2.70883846282959,
1154
+ "logits/rejected": -2.7197322845458984,
1155
+ "logps/chosen": -0.1189185157418251,
1156
+ "logps/rejected": -1.3503382205963135,
1157
+ "loss": 0.1465,
1158
+ "nll_loss": 0.13937658071517944,
1159
+ "rewards/accuracies": 1.0,
1160
+ "rewards/chosen": -0.00594592560082674,
1161
+ "rewards/margins": 0.06157098338007927,
1162
+ "rewards/rejected": -0.06751690804958344,
1163
+ "step": 620
1164
+ },
1165
+ {
1166
+ "epoch": 2.99,
1167
+ "grad_norm": 8.441114994591647,
1168
+ "learning_rate": 3.908287771542396e-10,
1169
+ "log_odds_chosen": 3.680515766143799,
1170
+ "log_odds_ratio": -0.04670769348740578,
1171
+ "logits/chosen": -2.716243028640747,
1172
+ "logits/rejected": -2.7199196815490723,
1173
+ "logps/chosen": -0.14760281145572662,
1174
+ "logps/rejected": -1.711219072341919,
1175
+ "loss": 0.1537,
1176
+ "nll_loss": 0.14854022860527039,
1177
+ "rewards/accuracies": 1.0,
1178
+ "rewards/chosen": -0.007380140013992786,
1179
+ "rewards/margins": 0.07818081974983215,
1180
+ "rewards/rejected": -0.08556096255779266,
1181
+ "step": 630
1182
+ },
1183
+ {
1184
+ "epoch": 3.0,
1185
+ "eval_log_odds_chosen": 0.5296839475631714,
1186
+ "eval_log_odds_ratio": -0.6105475425720215,
1187
+ "eval_logits/chosen": -2.7270965576171875,
1188
+ "eval_logits/rejected": -2.7365875244140625,
1189
+ "eval_logps/chosen": -1.0283128023147583,
1190
+ "eval_logps/rejected": -1.333586573600769,
1191
+ "eval_loss": 1.1074548959732056,
1192
+ "eval_nll_loss": 1.0961836576461792,
1193
+ "eval_rewards/accuracies": 0.6354166865348816,
1194
+ "eval_rewards/chosen": -0.05141563341021538,
1195
+ "eval_rewards/margins": 0.015263698063790798,
1196
+ "eval_rewards/rejected": -0.06667933613061905,
1197
+ "eval_runtime": 57.2818,
1198
+ "eval_samples_per_second": 13.093,
1199
+ "eval_steps_per_second": 0.419,
1200
+ "step": 633
1201
+ },
1202
+ {
1203
+ "epoch": 3.0,
1204
+ "step": 633,
1205
+ "total_flos": 0.0,
1206
+ "train_loss": 0.5264878700506442,
1207
+ "train_runtime": 6083.3362,
1208
+ "train_samples_per_second": 3.329,
1209
+ "train_steps_per_second": 0.104
1210
+ }
1211
+ ],
1212
+ "logging_steps": 10,
1213
+ "max_steps": 633,
1214
+ "num_input_tokens_seen": 0,
1215
+ "num_train_epochs": 3,
1216
+ "save_steps": 500,
1217
+ "total_flos": 0.0,
1218
+ "train_batch_size": 8,
1219
+ "trial_name": null,
1220
+ "trial_params": null
1221
+ }