File size: 7,031 Bytes
f300aba
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
{
  "best_metric": 0.03174639865756035,
  "best_model_checkpoint": "./mixstral/31-03-24-Weni-WeniGPT-QA-Zephyr-7B-5.0.2-KTO_WeniGPT Experiment using KTO trainer with no collator, Mixstral model and amnesia system prompt.-2_max_steps-786_batch_32_2024-03-31_ppid_1264/checkpoint-200",
  "epoch": 1.5194681861348527,
  "eval_steps": 50,
  "global_step": 200,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.15,
      "grad_norm": 2.4299395084381104,
      "learning_rate": 0.00015000000000000001,
      "loss": 0.4387,
      "step": 20,
      "train/kl": 0.0385909266769886,
      "train/logps/chosen": -159.8199005126953,
      "train/logps/rejected": -160.9553680419922,
      "train/rewards/chosen": 0.08566487580537796,
      "train/rewards/margins": 0.623574860394001,
      "train/rewards/rejected": -0.537909984588623
    },
    {
      "epoch": 0.3,
      "grad_norm": 0.6239124536514282,
      "learning_rate": 0.0001963254593175853,
      "loss": 0.1567,
      "step": 40,
      "train/kl": 0.3582279086112976,
      "train/logps/chosen": -123.32263946533203,
      "train/logps/rejected": -219.208251953125,
      "train/rewards/chosen": 1.9985122680664062,
      "train/rewards/margins": 7.9115424156188965,
      "train/rewards/rejected": -5.91303014755249
    },
    {
      "epoch": 0.38,
      "eval/kl": 0.0,
      "eval/logps/chosen": -146.17608642578125,
      "eval/logps/rejected": -271.80560302734375,
      "eval/rewards/chosen": 3.3853886127471924,
      "eval/rewards/margins": 15.140896081924438,
      "eval/rewards/rejected": -11.755507469177246,
      "eval_loss": 0.057219523936510086,
      "eval_runtime": 1031.0752,
      "eval_samples_per_second": 0.467,
      "eval_steps_per_second": 0.117,
      "step": 50
    },
    {
      "epoch": 0.46,
      "grad_norm": 0.16106899082660675,
      "learning_rate": 0.00019107611548556432,
      "loss": 0.0596,
      "step": 60,
      "train/kl": 0.1449735164642334,
      "train/logps/chosen": -131.60433959960938,
      "train/logps/rejected": -292.1778869628906,
      "train/rewards/chosen": 3.4192304611206055,
      "train/rewards/margins": 16.215981483459473,
      "train/rewards/rejected": -12.796751022338867
    },
    {
      "epoch": 0.61,
      "grad_norm": 0.6279284358024597,
      "learning_rate": 0.00018582677165354333,
      "loss": 0.0528,
      "step": 80,
      "train/kl": 0.047291625291109085,
      "train/logps/chosen": -116.19053649902344,
      "train/logps/rejected": -327.73248291015625,
      "train/rewards/chosen": 3.602910041809082,
      "train/rewards/margins": 21.11134433746338,
      "train/rewards/rejected": -17.508434295654297
    },
    {
      "epoch": 0.76,
      "grad_norm": 1.3628041744232178,
      "learning_rate": 0.0001805774278215223,
      "loss": 0.0481,
      "step": 100,
      "train/kl": 0.0,
      "train/logps/chosen": -127.67686462402344,
      "train/logps/rejected": -339.4940185546875,
      "train/rewards/chosen": 3.723008155822754,
      "train/rewards/margins": 21.06360149383545,
      "train/rewards/rejected": -17.340593338012695
    },
    {
      "epoch": 0.76,
      "eval/kl": 0.0,
      "eval/logps/chosen": -140.9491424560547,
      "eval/logps/rejected": -331.2147216796875,
      "eval/rewards/chosen": 3.908082962036133,
      "eval/rewards/margins": 21.604503631591797,
      "eval/rewards/rejected": -17.696420669555664,
      "eval_loss": 0.04277317598462105,
      "eval_runtime": 1009.2383,
      "eval_samples_per_second": 0.478,
      "eval_steps_per_second": 0.12,
      "step": 100
    },
    {
      "epoch": 0.91,
      "grad_norm": 0.2859511077404022,
      "learning_rate": 0.00017532808398950132,
      "loss": 0.0438,
      "step": 120,
      "train/kl": 0.0,
      "train/logps/chosen": -107.13507080078125,
      "train/logps/rejected": -362.26336669921875,
      "train/rewards/chosen": 3.8003525733947754,
      "train/rewards/margins": 23.768267154693604,
      "train/rewards/rejected": -19.967914581298828
    },
    {
      "epoch": 1.06,
      "grad_norm": 0.27395951747894287,
      "learning_rate": 0.00017007874015748033,
      "loss": 0.0391,
      "step": 140,
      "train/kl": 0.0,
      "train/logps/chosen": -116.65599822998047,
      "train/logps/rejected": -404.44793701171875,
      "train/rewards/chosen": 3.8443548679351807,
      "train/rewards/margins": 28.49311852455139,
      "train/rewards/rejected": -24.64876365661621
    },
    {
      "epoch": 1.14,
      "eval/kl": 0.0,
      "eval/logps/chosen": -137.66822814941406,
      "eval/logps/rejected": -436.40203857421875,
      "eval/rewards/chosen": 4.236174583435059,
      "eval/rewards/margins": 32.45132541656494,
      "eval/rewards/rejected": -28.215150833129883,
      "eval_loss": 0.03409990668296814,
      "eval_runtime": 1008.6969,
      "eval_samples_per_second": 0.478,
      "eval_steps_per_second": 0.12,
      "step": 150
    },
    {
      "epoch": 1.22,
      "grad_norm": 0.5016167163848877,
      "learning_rate": 0.00016482939632545934,
      "loss": 0.0283,
      "step": 160,
      "train/kl": 0.11301638185977936,
      "train/logps/chosen": -113.57328796386719,
      "train/logps/rejected": -447.1547546386719,
      "train/rewards/chosen": 4.402313709259033,
      "train/rewards/margins": 33.74688768386841,
      "train/rewards/rejected": -29.344573974609375
    },
    {
      "epoch": 1.37,
      "grad_norm": 0.051358386874198914,
      "learning_rate": 0.00015958005249343832,
      "loss": 0.0242,
      "step": 180,
      "train/kl": 0.05546154826879501,
      "train/logps/chosen": -101.12237548828125,
      "train/logps/rejected": -464.6999206542969,
      "train/rewards/chosen": 4.362555027008057,
      "train/rewards/margins": 34.47586107254028,
      "train/rewards/rejected": -30.113306045532227
    },
    {
      "epoch": 1.52,
      "grad_norm": 0.044250741600990295,
      "learning_rate": 0.00015433070866141733,
      "loss": 0.0209,
      "step": 200,
      "train/kl": 0.17102402448654175,
      "train/logps/chosen": -117.83503723144531,
      "train/logps/rejected": -454.4959411621094,
      "train/rewards/chosen": 4.654594898223877,
      "train/rewards/margins": 34.24535036087036,
      "train/rewards/rejected": -29.590755462646484
    },
    {
      "epoch": 1.52,
      "eval/kl": 0.0,
      "eval/logps/chosen": -136.6546630859375,
      "eval/logps/rejected": -475.3577575683594,
      "eval/rewards/chosen": 4.337530612945557,
      "eval/rewards/margins": 36.44825220108032,
      "eval/rewards/rejected": -32.110721588134766,
      "eval_loss": 0.03174639865756035,
      "eval_runtime": 1008.2427,
      "eval_samples_per_second": 0.478,
      "eval_steps_per_second": 0.12,
      "step": 200
    }
  ],
  "logging_steps": 20,
  "max_steps": 786,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 6,
  "save_steps": 100,
  "total_flos": 0.0,
  "train_batch_size": 4,
  "trial_name": null,
  "trial_params": null
}