QinLiuNLP commited on
Commit
10d8a09
1 Parent(s): 317b5ac

Model save

Browse files
Files changed (4) hide show
  1. README.md +62 -0
  2. all_results.json +9 -0
  3. train_results.json +9 -0
  4. trainer_state.json +1932 -0
README.md ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: llama3
3
+ library_name: peft
4
+ tags:
5
+ - trl
6
+ - dpo
7
+ - generated_from_trainer
8
+ base_model: Jackie999/llama3-sudo-5epochs-tofu_full_sft
9
+ model-index:
10
+ - name: llama3-sudo-dpo-10epochs-forget10mixall-1sft-2lora-1e-5
11
+ results: []
12
+ ---
13
+
14
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
15
+ should probably proofread and complete it, then remove this comment. -->
16
+
17
+ # llama3-sudo-dpo-10epochs-forget10mixall-1sft-2lora-1e-5
18
+
19
+ This model is a fine-tuned version of [Jackie999/llama3-sudo-5epochs-tofu_full_sft](https://huggingface.co/Jackie999/llama3-sudo-5epochs-tofu_full_sft) on an unknown dataset.
20
+
21
+ ## Model description
22
+
23
+ More information needed
24
+
25
+ ## Intended uses & limitations
26
+
27
+ More information needed
28
+
29
+ ## Training and evaluation data
30
+
31
+ More information needed
32
+
33
+ ## Training procedure
34
+
35
+ ### Training hyperparameters
36
+
37
+ The following hyperparameters were used during training:
38
+ - learning_rate: 1e-05
39
+ - train_batch_size: 4
40
+ - eval_batch_size: 8
41
+ - seed: 42
42
+ - distributed_type: multi-GPU
43
+ - num_devices: 4
44
+ - gradient_accumulation_steps: 4
45
+ - total_train_batch_size: 64
46
+ - total_eval_batch_size: 32
47
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
48
+ - lr_scheduler_type: cosine
49
+ - lr_scheduler_warmup_ratio: 0.1
50
+ - num_epochs: 10
51
+
52
+ ### Training results
53
+
54
+
55
+
56
+ ### Framework versions
57
+
58
+ - PEFT 0.12.0
59
+ - Transformers 4.44.0
60
+ - Pytorch 2.1.2
61
+ - Datasets 3.0.0
62
+ - Tokenizers 0.19.1
all_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 10.0,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.05206389323771,
5
+ "train_runtime": 14277.3883,
6
+ "train_samples": 8000,
7
+ "train_samples_per_second": 5.603,
8
+ "train_steps_per_second": 0.088
9
+ }
train_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 10.0,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.05206389323771,
5
+ "train_runtime": 14277.3883,
6
+ "train_samples": 8000,
7
+ "train_samples_per_second": 5.603,
8
+ "train_steps_per_second": 0.088
9
+ }
trainer_state.json ADDED
@@ -0,0 +1,1932 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 10.0,
5
+ "eval_steps": 1000,
6
+ "global_step": 1250,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.008,
13
+ "grad_norm": 0.5664508606614915,
14
+ "learning_rate": 8e-08,
15
+ "logits/chosen": -1.7671998739242554,
16
+ "logits/rejected": -2.2639822959899902,
17
+ "logps/chosen": -46.430763244628906,
18
+ "logps/rejected": -102.85381317138672,
19
+ "loss": 0.6931,
20
+ "rewards/accuracies": 0.0,
21
+ "rewards/chosen": 0.0,
22
+ "rewards/margins": 0.0,
23
+ "rewards/rejected": 0.0,
24
+ "step": 1
25
+ },
26
+ {
27
+ "epoch": 0.08,
28
+ "grad_norm": 0.5939624426575738,
29
+ "learning_rate": 8.000000000000001e-07,
30
+ "logits/chosen": -1.6349234580993652,
31
+ "logits/rejected": -2.0293447971343994,
32
+ "logps/chosen": -50.39720916748047,
33
+ "logps/rejected": -98.8401870727539,
34
+ "loss": 0.6929,
35
+ "rewards/accuracies": 0.5069444179534912,
36
+ "rewards/chosen": -0.00029719059239141643,
37
+ "rewards/margins": 0.0004301935259718448,
38
+ "rewards/rejected": -0.0007273841183632612,
39
+ "step": 10
40
+ },
41
+ {
42
+ "epoch": 0.16,
43
+ "grad_norm": 0.5700892501542819,
44
+ "learning_rate": 1.6000000000000001e-06,
45
+ "logits/chosen": -1.6655193567276,
46
+ "logits/rejected": -2.1240811347961426,
47
+ "logps/chosen": -47.0224609375,
48
+ "logps/rejected": -105.32938385009766,
49
+ "loss": 0.6875,
50
+ "rewards/accuracies": 0.981249988079071,
51
+ "rewards/chosen": 0.0002471635234542191,
52
+ "rewards/margins": 0.012360258027911186,
53
+ "rewards/rejected": -0.01211309339851141,
54
+ "step": 20
55
+ },
56
+ {
57
+ "epoch": 0.24,
58
+ "grad_norm": 0.7554778741427768,
59
+ "learning_rate": 2.4000000000000003e-06,
60
+ "logits/chosen": -1.7440729141235352,
61
+ "logits/rejected": -2.1188106536865234,
62
+ "logps/chosen": -50.84648513793945,
63
+ "logps/rejected": -106.61978912353516,
64
+ "loss": 0.6579,
65
+ "rewards/accuracies": 0.925000011920929,
66
+ "rewards/chosen": -0.0015375657239928842,
67
+ "rewards/margins": 0.07141353189945221,
68
+ "rewards/rejected": -0.07295109331607819,
69
+ "step": 30
70
+ },
71
+ {
72
+ "epoch": 0.32,
73
+ "grad_norm": 0.7532828556442333,
74
+ "learning_rate": 3.2000000000000003e-06,
75
+ "logits/chosen": -1.6423908472061157,
76
+ "logits/rejected": -2.07899808883667,
77
+ "logps/chosen": -48.13755416870117,
78
+ "logps/rejected": -131.5819854736328,
79
+ "loss": 0.5614,
80
+ "rewards/accuracies": 0.9750000238418579,
81
+ "rewards/chosen": 0.005950761027634144,
82
+ "rewards/margins": 0.3043977618217468,
83
+ "rewards/rejected": -0.29844698309898376,
84
+ "step": 40
85
+ },
86
+ {
87
+ "epoch": 0.4,
88
+ "grad_norm": 0.49124925197921077,
89
+ "learning_rate": 4.000000000000001e-06,
90
+ "logits/chosen": -1.577450156211853,
91
+ "logits/rejected": -2.0352962017059326,
92
+ "logps/chosen": -47.25825119018555,
93
+ "logps/rejected": -175.77737426757812,
94
+ "loss": 0.4101,
95
+ "rewards/accuracies": 0.9750000238418579,
96
+ "rewards/chosen": 0.014603497460484505,
97
+ "rewards/margins": 0.7516024708747864,
98
+ "rewards/rejected": -0.7369989156723022,
99
+ "step": 50
100
+ },
101
+ {
102
+ "epoch": 0.48,
103
+ "grad_norm": 0.3421632301388367,
104
+ "learning_rate": 4.800000000000001e-06,
105
+ "logits/chosen": -1.54917311668396,
106
+ "logits/rejected": -1.9336481094360352,
107
+ "logps/chosen": -54.58576583862305,
108
+ "logps/rejected": -217.2119598388672,
109
+ "loss": 0.2991,
110
+ "rewards/accuracies": 0.9312499761581421,
111
+ "rewards/chosen": -0.034352801740169525,
112
+ "rewards/margins": 1.1476414203643799,
113
+ "rewards/rejected": -1.1819943189620972,
114
+ "step": 60
115
+ },
116
+ {
117
+ "epoch": 0.56,
118
+ "grad_norm": 0.33380968035486847,
119
+ "learning_rate": 5.600000000000001e-06,
120
+ "logits/chosen": -1.4622433185577393,
121
+ "logits/rejected": -1.9104700088500977,
122
+ "logps/chosen": -53.6714973449707,
123
+ "logps/rejected": -297.1726379394531,
124
+ "loss": 0.2231,
125
+ "rewards/accuracies": 0.9437500238418579,
126
+ "rewards/chosen": -0.03121398389339447,
127
+ "rewards/margins": 1.926568627357483,
128
+ "rewards/rejected": -1.957782506942749,
129
+ "step": 70
130
+ },
131
+ {
132
+ "epoch": 0.64,
133
+ "grad_norm": 0.14294152601696442,
134
+ "learning_rate": 6.4000000000000006e-06,
135
+ "logits/chosen": -1.3913681507110596,
136
+ "logits/rejected": -1.818612813949585,
137
+ "logps/chosen": -54.882850646972656,
138
+ "logps/rejected": -364.594970703125,
139
+ "loss": 0.2104,
140
+ "rewards/accuracies": 0.949999988079071,
141
+ "rewards/chosen": -0.04058960825204849,
142
+ "rewards/margins": 2.5959861278533936,
143
+ "rewards/rejected": -2.636575937271118,
144
+ "step": 80
145
+ },
146
+ {
147
+ "epoch": 0.72,
148
+ "grad_norm": 0.2043412163104557,
149
+ "learning_rate": 7.2000000000000005e-06,
150
+ "logits/chosen": -1.3484817743301392,
151
+ "logits/rejected": -1.7438443899154663,
152
+ "logps/chosen": -57.729034423828125,
153
+ "logps/rejected": -391.08892822265625,
154
+ "loss": 0.194,
155
+ "rewards/accuracies": 0.9437500238418579,
156
+ "rewards/chosen": -0.0659557357430458,
157
+ "rewards/margins": 2.8331382274627686,
158
+ "rewards/rejected": -2.8990941047668457,
159
+ "step": 90
160
+ },
161
+ {
162
+ "epoch": 0.8,
163
+ "grad_norm": 0.45349092259459883,
164
+ "learning_rate": 8.000000000000001e-06,
165
+ "logits/chosen": -1.3855174779891968,
166
+ "logits/rejected": -1.819236159324646,
167
+ "logps/chosen": -49.61981964111328,
168
+ "logps/rejected": -360.9615173339844,
169
+ "loss": 0.1907,
170
+ "rewards/accuracies": 0.956250011920929,
171
+ "rewards/chosen": 0.004681958816945553,
172
+ "rewards/margins": 2.5954396724700928,
173
+ "rewards/rejected": -2.5907576084136963,
174
+ "step": 100
175
+ },
176
+ {
177
+ "epoch": 0.88,
178
+ "grad_norm": 0.2724569498424518,
179
+ "learning_rate": 8.8e-06,
180
+ "logits/chosen": -1.3040708303451538,
181
+ "logits/rejected": -1.7432258129119873,
182
+ "logps/chosen": -52.180030822753906,
183
+ "logps/rejected": -379.49664306640625,
184
+ "loss": 0.1868,
185
+ "rewards/accuracies": 0.949999988079071,
186
+ "rewards/chosen": -0.02268405631184578,
187
+ "rewards/margins": 2.7577483654022217,
188
+ "rewards/rejected": -2.78043270111084,
189
+ "step": 110
190
+ },
191
+ {
192
+ "epoch": 0.96,
193
+ "grad_norm": 0.17941572717033036,
194
+ "learning_rate": 9.600000000000001e-06,
195
+ "logits/chosen": -1.2530890703201294,
196
+ "logits/rejected": -1.7477638721466064,
197
+ "logps/chosen": -38.981895446777344,
198
+ "logps/rejected": -443.25347900390625,
199
+ "loss": 0.1556,
200
+ "rewards/accuracies": 0.9937499761581421,
201
+ "rewards/chosen": 0.07770782709121704,
202
+ "rewards/margins": 3.4688849449157715,
203
+ "rewards/rejected": -3.3911774158477783,
204
+ "step": 120
205
+ },
206
+ {
207
+ "epoch": 1.04,
208
+ "grad_norm": 0.3303028590674529,
209
+ "learning_rate": 9.999512620046523e-06,
210
+ "logits/chosen": -1.346740484237671,
211
+ "logits/rejected": -1.805371880531311,
212
+ "logps/chosen": -55.48865509033203,
213
+ "logps/rejected": -389.91607666015625,
214
+ "loss": 0.1713,
215
+ "rewards/accuracies": 0.9312499761581421,
216
+ "rewards/chosen": -0.04284561052918434,
217
+ "rewards/margins": 2.8560733795166016,
218
+ "rewards/rejected": -2.898919105529785,
219
+ "step": 130
220
+ },
221
+ {
222
+ "epoch": 1.12,
223
+ "grad_norm": 0.9056804720582001,
224
+ "learning_rate": 9.995614150494293e-06,
225
+ "logits/chosen": -1.234359622001648,
226
+ "logits/rejected": -1.6866785287857056,
227
+ "logps/chosen": -48.799949645996094,
228
+ "logps/rejected": -423.95452880859375,
229
+ "loss": 0.164,
230
+ "rewards/accuracies": 0.949999988079071,
231
+ "rewards/chosen": 0.011531209573149681,
232
+ "rewards/margins": 3.2391915321350098,
233
+ "rewards/rejected": -3.2276599407196045,
234
+ "step": 140
235
+ },
236
+ {
237
+ "epoch": 1.2,
238
+ "grad_norm": 0.5254416965749492,
239
+ "learning_rate": 9.987820251299121e-06,
240
+ "logits/chosen": -1.0958689451217651,
241
+ "logits/rejected": -1.4965614080429077,
242
+ "logps/chosen": -52.782798767089844,
243
+ "logps/rejected": -455.8980407714844,
244
+ "loss": 0.1648,
245
+ "rewards/accuracies": 0.9312499761581421,
246
+ "rewards/chosen": -0.012227678671479225,
247
+ "rewards/margins": 3.540536880493164,
248
+ "rewards/rejected": -3.552764415740967,
249
+ "step": 150
250
+ },
251
+ {
252
+ "epoch": 1.28,
253
+ "grad_norm": 0.5108498363771904,
254
+ "learning_rate": 9.976136999909156e-06,
255
+ "logits/chosen": -1.1107490062713623,
256
+ "logits/rejected": -1.5378646850585938,
257
+ "logps/chosen": -47.4257698059082,
258
+ "logps/rejected": -470.36163330078125,
259
+ "loss": 0.11,
260
+ "rewards/accuracies": 0.9437500238418579,
261
+ "rewards/chosen": 0.026664480566978455,
262
+ "rewards/margins": 3.716306209564209,
263
+ "rewards/rejected": -3.6896419525146484,
264
+ "step": 160
265
+ },
266
+ {
267
+ "epoch": 1.3599999999999999,
268
+ "grad_norm": 1.2383483480266742,
269
+ "learning_rate": 9.960573506572391e-06,
270
+ "logits/chosen": -1.028438687324524,
271
+ "logits/rejected": -1.5239307880401611,
272
+ "logps/chosen": -42.13895034790039,
273
+ "logps/rejected": -517.1678466796875,
274
+ "loss": 0.1026,
275
+ "rewards/accuracies": 0.96875,
276
+ "rewards/chosen": 0.05908266454935074,
277
+ "rewards/margins": 4.210160255432129,
278
+ "rewards/rejected": -4.151078224182129,
279
+ "step": 170
280
+ },
281
+ {
282
+ "epoch": 1.44,
283
+ "grad_norm": 0.9166087354544606,
284
+ "learning_rate": 9.941141907232766e-06,
285
+ "logits/chosen": -0.9610630869865417,
286
+ "logits/rejected": -1.4120290279388428,
287
+ "logps/chosen": -46.765838623046875,
288
+ "logps/rejected": -542.84716796875,
289
+ "loss": 0.109,
290
+ "rewards/accuracies": 0.9375,
291
+ "rewards/chosen": 0.03420247882604599,
292
+ "rewards/margins": 4.456036567687988,
293
+ "rewards/rejected": -4.421834468841553,
294
+ "step": 180
295
+ },
296
+ {
297
+ "epoch": 1.52,
298
+ "grad_norm": 1.6771630345432116,
299
+ "learning_rate": 9.91785735406693e-06,
300
+ "logits/chosen": -0.8817203640937805,
301
+ "logits/rejected": -1.3373805284500122,
302
+ "logps/chosen": -43.31850051879883,
303
+ "logps/rejected": -525.7540283203125,
304
+ "loss": 0.0982,
305
+ "rewards/accuracies": 0.9375,
306
+ "rewards/chosen": 0.06416326761245728,
307
+ "rewards/margins": 4.315359115600586,
308
+ "rewards/rejected": -4.251195907592773,
309
+ "step": 190
310
+ },
311
+ {
312
+ "epoch": 1.6,
313
+ "grad_norm": 1.2635622789324346,
314
+ "learning_rate": 9.890738003669029e-06,
315
+ "logits/chosen": -0.9383857846260071,
316
+ "logits/rejected": -1.3598968982696533,
317
+ "logps/chosen": -40.324180603027344,
318
+ "logps/rejected": -580.6539916992188,
319
+ "loss": 0.0748,
320
+ "rewards/accuracies": 0.956250011920929,
321
+ "rewards/chosen": 0.09767557680606842,
322
+ "rewards/margins": 4.886355876922607,
323
+ "rewards/rejected": -4.788680553436279,
324
+ "step": 200
325
+ },
326
+ {
327
+ "epoch": 1.6800000000000002,
328
+ "grad_norm": 2.051002862784939,
329
+ "learning_rate": 9.859805002892733e-06,
330
+ "logits/chosen": -0.747630774974823,
331
+ "logits/rejected": -1.189408302307129,
332
+ "logps/chosen": -38.997947692871094,
333
+ "logps/rejected": -600.232421875,
334
+ "loss": 0.0693,
335
+ "rewards/accuracies": 0.981249988079071,
336
+ "rewards/chosen": 0.10605404525995255,
337
+ "rewards/margins": 5.094006538391113,
338
+ "rewards/rejected": -4.98795223236084,
339
+ "step": 210
340
+ },
341
+ {
342
+ "epoch": 1.76,
343
+ "grad_norm": 0.5973861965703521,
344
+ "learning_rate": 9.825082472361558e-06,
345
+ "logits/chosen": -0.769231915473938,
346
+ "logits/rejected": -1.2444443702697754,
347
+ "logps/chosen": -37.80399703979492,
348
+ "logps/rejected": -661.2396850585938,
349
+ "loss": 0.0355,
350
+ "rewards/accuracies": 0.9937499761581421,
351
+ "rewards/chosen": 0.1041645035147667,
352
+ "rewards/margins": 5.686351299285889,
353
+ "rewards/rejected": -5.582187652587891,
354
+ "step": 220
355
+ },
356
+ {
357
+ "epoch": 1.8399999999999999,
358
+ "grad_norm": 0.36668201108212406,
359
+ "learning_rate": 9.786597487660336e-06,
360
+ "logits/chosen": -0.7400678396224976,
361
+ "logits/rejected": -1.185605764389038,
362
+ "logps/chosen": -38.35655975341797,
363
+ "logps/rejected": -675.3729248046875,
364
+ "loss": 0.0604,
365
+ "rewards/accuracies": 0.9937499761581421,
366
+ "rewards/chosen": 0.10024236142635345,
367
+ "rewards/margins": 5.824278831481934,
368
+ "rewards/rejected": -5.72403621673584,
369
+ "step": 230
370
+ },
371
+ {
372
+ "epoch": 1.92,
373
+ "grad_norm": 0.18097106498610982,
374
+ "learning_rate": 9.744380058222483e-06,
375
+ "logits/chosen": -0.8332279324531555,
376
+ "logits/rejected": -1.2189085483551025,
377
+ "logps/chosen": -39.80487060546875,
378
+ "logps/rejected": -665.7376708984375,
379
+ "loss": 0.0445,
380
+ "rewards/accuracies": 0.9750000238418579,
381
+ "rewards/chosen": 0.1086835116147995,
382
+ "rewards/margins": 5.7541303634643555,
383
+ "rewards/rejected": -5.64544677734375,
384
+ "step": 240
385
+ },
386
+ {
387
+ "epoch": 2.0,
388
+ "grad_norm": 0.5482510873677592,
389
+ "learning_rate": 9.698463103929542e-06,
390
+ "logits/chosen": -0.8593254089355469,
391
+ "logits/rejected": -1.219299077987671,
392
+ "logps/chosen": -40.509761810302734,
393
+ "logps/rejected": -678.4737548828125,
394
+ "loss": 0.0333,
395
+ "rewards/accuracies": 1.0,
396
+ "rewards/chosen": 0.10330041497945786,
397
+ "rewards/margins": 5.88486385345459,
398
+ "rewards/rejected": -5.781562805175781,
399
+ "step": 250
400
+ },
401
+ {
402
+ "epoch": 2.08,
403
+ "grad_norm": 0.06311428811879578,
404
+ "learning_rate": 9.648882429441258e-06,
405
+ "logits/chosen": -0.721664309501648,
406
+ "logits/rejected": -1.1366941928863525,
407
+ "logps/chosen": -43.40575408935547,
408
+ "logps/rejected": -718.126953125,
409
+ "loss": 0.043,
410
+ "rewards/accuracies": 0.9937499761581421,
411
+ "rewards/chosen": 0.06619155406951904,
412
+ "rewards/margins": 6.222557067871094,
413
+ "rewards/rejected": -6.156365394592285,
414
+ "step": 260
415
+ },
416
+ {
417
+ "epoch": 2.16,
418
+ "grad_norm": 0.10420712161857161,
419
+ "learning_rate": 9.595676696276173e-06,
420
+ "logits/chosen": -0.8397006988525391,
421
+ "logits/rejected": -1.1919224262237549,
422
+ "logps/chosen": -45.371559143066406,
423
+ "logps/rejected": -671.6702880859375,
424
+ "loss": 0.0399,
425
+ "rewards/accuracies": 1.0,
426
+ "rewards/chosen": 0.05775808170437813,
427
+ "rewards/margins": 5.775918006896973,
428
+ "rewards/rejected": -5.718161106109619,
429
+ "step": 270
430
+ },
431
+ {
432
+ "epoch": 2.24,
433
+ "grad_norm": 0.048219144314199394,
434
+ "learning_rate": 9.538887392664544e-06,
435
+ "logits/chosen": -0.932695746421814,
436
+ "logits/rejected": -1.2857800722122192,
437
+ "logps/chosen": -49.043556213378906,
438
+ "logps/rejected": -700.9938354492188,
439
+ "loss": 0.0302,
440
+ "rewards/accuracies": 1.0,
441
+ "rewards/chosen": 0.01686009392142296,
442
+ "rewards/margins": 6.021510601043701,
443
+ "rewards/rejected": -6.004650592803955,
444
+ "step": 280
445
+ },
446
+ {
447
+ "epoch": 2.32,
448
+ "grad_norm": 0.052699483559969486,
449
+ "learning_rate": 9.478558801197065e-06,
450
+ "logits/chosen": -1.0952008962631226,
451
+ "logits/rejected": -1.419945478439331,
452
+ "logps/chosen": -50.7900276184082,
453
+ "logps/rejected": -746.294921875,
454
+ "loss": 0.0287,
455
+ "rewards/accuracies": 1.0,
456
+ "rewards/chosen": -0.011282497085630894,
457
+ "rewards/margins": 6.445050239562988,
458
+ "rewards/rejected": -6.456332206726074,
459
+ "step": 290
460
+ },
461
+ {
462
+ "epoch": 2.4,
463
+ "grad_norm": 0.39282718133559014,
464
+ "learning_rate": 9.414737964294636e-06,
465
+ "logits/chosen": -1.276970386505127,
466
+ "logits/rejected": -1.5558645725250244,
467
+ "logps/chosen": -57.718162536621094,
468
+ "logps/rejected": -758.6500244140625,
469
+ "loss": 0.0284,
470
+ "rewards/accuracies": 0.9937499761581421,
471
+ "rewards/chosen": -0.07060392200946808,
472
+ "rewards/margins": 6.502586364746094,
473
+ "rewards/rejected": -6.573190212249756,
474
+ "step": 300
475
+ },
476
+ {
477
+ "epoch": 2.48,
478
+ "grad_norm": 0.12204254875967184,
479
+ "learning_rate": 9.347474647526095e-06,
480
+ "logits/chosen": -1.4385970830917358,
481
+ "logits/rejected": -1.6844545602798462,
482
+ "logps/chosen": -59.007118225097656,
483
+ "logps/rejected": -749.748046875,
484
+ "loss": 0.0353,
485
+ "rewards/accuracies": 0.9937499761581421,
486
+ "rewards/chosen": -0.08172997832298279,
487
+ "rewards/margins": 6.407279014587402,
488
+ "rewards/rejected": -6.489008903503418,
489
+ "step": 310
490
+ },
491
+ {
492
+ "epoch": 2.56,
493
+ "grad_norm": 0.05045051166529495,
494
+ "learning_rate": 9.276821300802535e-06,
495
+ "logits/chosen": -1.682716965675354,
496
+ "logits/rejected": -1.9395768642425537,
497
+ "logps/chosen": -69.35375213623047,
498
+ "logps/rejected": -762.8756103515625,
499
+ "loss": 0.0149,
500
+ "rewards/accuracies": 0.9937499761581421,
501
+ "rewards/chosen": -0.20327362418174744,
502
+ "rewards/margins": 6.410320281982422,
503
+ "rewards/rejected": -6.613594055175781,
504
+ "step": 320
505
+ },
506
+ {
507
+ "epoch": 2.64,
508
+ "grad_norm": 0.041644876319752826,
509
+ "learning_rate": 9.202833017478421e-06,
510
+ "logits/chosen": -1.9520126581192017,
511
+ "logits/rejected": -2.184046745300293,
512
+ "logps/chosen": -74.17578125,
513
+ "logps/rejected": -782.9901123046875,
514
+ "loss": 0.0139,
515
+ "rewards/accuracies": 1.0,
516
+ "rewards/chosen": -0.25414201617240906,
517
+ "rewards/margins": 6.548297882080078,
518
+ "rewards/rejected": -6.802439212799072,
519
+ "step": 330
520
+ },
521
+ {
522
+ "epoch": 2.7199999999999998,
523
+ "grad_norm": 0.10894239404290727,
524
+ "learning_rate": 9.125567491391476e-06,
525
+ "logits/chosen": -2.3633341789245605,
526
+ "logits/rejected": -2.5258421897888184,
527
+ "logps/chosen": -95.89537811279297,
528
+ "logps/rejected": -769.3350830078125,
529
+ "loss": 0.026,
530
+ "rewards/accuracies": 1.0,
531
+ "rewards/chosen": -0.4556370675563812,
532
+ "rewards/margins": 6.227226734161377,
533
+ "rewards/rejected": -6.682864189147949,
534
+ "step": 340
535
+ },
536
+ {
537
+ "epoch": 2.8,
538
+ "grad_norm": 0.09090196291657318,
539
+ "learning_rate": 9.045084971874738e-06,
540
+ "logits/chosen": -2.9310622215270996,
541
+ "logits/rejected": -3.0340473651885986,
542
+ "logps/chosen": -181.8387451171875,
543
+ "logps/rejected": -789.5302734375,
544
+ "loss": 0.0108,
545
+ "rewards/accuracies": 1.0,
546
+ "rewards/chosen": -1.3074408769607544,
547
+ "rewards/margins": 5.575064182281494,
548
+ "rewards/rejected": -6.882504463195801,
549
+ "step": 350
550
+ },
551
+ {
552
+ "epoch": 2.88,
553
+ "grad_norm": 0.12031593041735512,
554
+ "learning_rate": 8.961448216775955e-06,
555
+ "logits/chosen": -2.900047779083252,
556
+ "logits/rejected": -3.0111327171325684,
557
+ "logps/chosen": -186.22052001953125,
558
+ "logps/rejected": -826.2166137695312,
559
+ "loss": 0.0242,
560
+ "rewards/accuracies": 1.0,
561
+ "rewards/chosen": -1.3595960140228271,
562
+ "rewards/margins": 5.881885051727295,
563
+ "rewards/rejected": -7.241480350494385,
564
+ "step": 360
565
+ },
566
+ {
567
+ "epoch": 2.96,
568
+ "grad_norm": 0.055136260032898635,
569
+ "learning_rate": 8.874722443520898e-06,
570
+ "logits/chosen": -2.855480432510376,
571
+ "logits/rejected": -2.9430692195892334,
572
+ "logps/chosen": -199.16647338867188,
573
+ "logps/rejected": -827.880859375,
574
+ "loss": 0.0346,
575
+ "rewards/accuracies": 1.0,
576
+ "rewards/chosen": -1.4962244033813477,
577
+ "rewards/margins": 5.776437282562256,
578
+ "rewards/rejected": -7.272662162780762,
579
+ "step": 370
580
+ },
581
+ {
582
+ "epoch": 3.04,
583
+ "grad_norm": 0.05072076339049116,
584
+ "learning_rate": 8.784975278258783e-06,
585
+ "logits/chosen": -2.6045117378234863,
586
+ "logits/rejected": -2.729827404022217,
587
+ "logps/chosen": -133.8080596923828,
588
+ "logps/rejected": -809.306640625,
589
+ "loss": 0.0092,
590
+ "rewards/accuracies": 1.0,
591
+ "rewards/chosen": -0.8419109582901001,
592
+ "rewards/margins": 6.24552059173584,
593
+ "rewards/rejected": -7.08743143081665,
594
+ "step": 380
595
+ },
596
+ {
597
+ "epoch": 3.12,
598
+ "grad_norm": 0.7450627702438166,
599
+ "learning_rate": 8.692276703129421e-06,
600
+ "logits/chosen": -2.6937222480773926,
601
+ "logits/rejected": -2.7929458618164062,
602
+ "logps/chosen": -168.631103515625,
603
+ "logps/rejected": -842.021484375,
604
+ "loss": 0.0196,
605
+ "rewards/accuracies": 1.0,
606
+ "rewards/chosen": -1.18534255027771,
607
+ "rewards/margins": 6.214483737945557,
608
+ "rewards/rejected": -7.3998260498046875,
609
+ "step": 390
610
+ },
611
+ {
612
+ "epoch": 3.2,
613
+ "grad_norm": 0.05027116077254764,
614
+ "learning_rate": 8.596699001693257e-06,
615
+ "logits/chosen": -2.852487087249756,
616
+ "logits/rejected": -2.908648729324341,
617
+ "logps/chosen": -187.98727416992188,
618
+ "logps/rejected": -834.2393798828125,
619
+ "loss": 0.016,
620
+ "rewards/accuracies": 0.9937499761581421,
621
+ "rewards/chosen": -1.361481785774231,
622
+ "rewards/margins": 5.982842922210693,
623
+ "rewards/rejected": -7.344325065612793,
624
+ "step": 400
625
+ },
626
+ {
627
+ "epoch": 3.2800000000000002,
628
+ "grad_norm": 0.042611669944768574,
629
+ "learning_rate": 8.498316702566828e-06,
630
+ "logits/chosen": -2.6405489444732666,
631
+ "logits/rejected": -2.737513303756714,
632
+ "logps/chosen": -163.23654174804688,
633
+ "logps/rejected": -875.9517822265625,
634
+ "loss": 0.0099,
635
+ "rewards/accuracies": 0.9937499761581421,
636
+ "rewards/chosen": -1.1402196884155273,
637
+ "rewards/margins": 6.602635383605957,
638
+ "rewards/rejected": -7.742854118347168,
639
+ "step": 410
640
+ },
641
+ {
642
+ "epoch": 3.36,
643
+ "grad_norm": 0.08908259328663255,
644
+ "learning_rate": 8.397206521307584e-06,
645
+ "logits/chosen": -2.797304630279541,
646
+ "logits/rejected": -2.8749709129333496,
647
+ "logps/chosen": -200.6483917236328,
648
+ "logps/rejected": -878.53466796875,
649
+ "loss": 0.0056,
650
+ "rewards/accuracies": 1.0,
651
+ "rewards/chosen": -1.5122627019882202,
652
+ "rewards/margins": 6.271607398986816,
653
+ "rewards/rejected": -7.783869743347168,
654
+ "step": 420
655
+ },
656
+ {
657
+ "epoch": 3.44,
658
+ "grad_norm": 0.11190864225013811,
659
+ "learning_rate": 8.293447300593402e-06,
660
+ "logits/chosen": -2.712634563446045,
661
+ "logits/rejected": -2.790599822998047,
662
+ "logps/chosen": -156.58209228515625,
663
+ "logps/rejected": -865.5999755859375,
664
+ "loss": 0.0061,
665
+ "rewards/accuracies": 1.0,
666
+ "rewards/chosen": -1.0528513193130493,
667
+ "rewards/margins": 6.598712921142578,
668
+ "rewards/rejected": -7.651564598083496,
669
+ "step": 430
670
+ },
671
+ {
672
+ "epoch": 3.52,
673
+ "grad_norm": 0.06216733964642332,
674
+ "learning_rate": 8.18711994874345e-06,
675
+ "logits/chosen": -2.7393593788146973,
676
+ "logits/rejected": -2.810574531555176,
677
+ "logps/chosen": -159.99278259277344,
678
+ "logps/rejected": -852.8875732421875,
679
+ "loss": 0.0102,
680
+ "rewards/accuracies": 0.9937499761581421,
681
+ "rewards/chosen": -1.0680657625198364,
682
+ "rewards/margins": 6.475743770599365,
683
+ "rewards/rejected": -7.543809413909912,
684
+ "step": 440
685
+ },
686
+ {
687
+ "epoch": 3.6,
688
+ "grad_norm": 1.6020556191748092,
689
+ "learning_rate": 8.078307376628292e-06,
690
+ "logits/chosen": -2.8640763759613037,
691
+ "logits/rejected": -2.9100239276885986,
692
+ "logps/chosen": -194.4068603515625,
693
+ "logps/rejected": -889.5808715820312,
694
+ "loss": 0.0241,
695
+ "rewards/accuracies": 1.0,
696
+ "rewards/chosen": -1.4508755207061768,
697
+ "rewards/margins": 6.437768459320068,
698
+ "rewards/rejected": -7.888643741607666,
699
+ "step": 450
700
+ },
701
+ {
702
+ "epoch": 3.68,
703
+ "grad_norm": 0.08068344292732496,
704
+ "learning_rate": 7.967094433018508e-06,
705
+ "logits/chosen": -2.863485813140869,
706
+ "logits/rejected": -2.9134774208068848,
707
+ "logps/chosen": -192.51461791992188,
708
+ "logps/rejected": -875.9539184570312,
709
+ "loss": 0.0173,
710
+ "rewards/accuracies": 0.9937499761581421,
711
+ "rewards/chosen": -1.4307234287261963,
712
+ "rewards/margins": 6.317349433898926,
713
+ "rewards/rejected": -7.748072624206543,
714
+ "step": 460
715
+ },
716
+ {
717
+ "epoch": 3.76,
718
+ "grad_norm": 0.03890160957476439,
719
+ "learning_rate": 7.85356783842216e-06,
720
+ "logits/chosen": -2.780956983566284,
721
+ "logits/rejected": -2.8286776542663574,
722
+ "logps/chosen": -156.67994689941406,
723
+ "logps/rejected": -865.4143676757812,
724
+ "loss": 0.0076,
725
+ "rewards/accuracies": 1.0,
726
+ "rewards/chosen": -1.0601476430892944,
727
+ "rewards/margins": 6.580509185791016,
728
+ "rewards/rejected": -7.6406569480896,
729
+ "step": 470
730
+ },
731
+ {
732
+ "epoch": 3.84,
733
+ "grad_norm": 0.2797627859203357,
734
+ "learning_rate": 7.737816117462752e-06,
735
+ "logits/chosen": -2.872148036956787,
736
+ "logits/rejected": -2.9261298179626465,
737
+ "logps/chosen": -162.9639434814453,
738
+ "logps/rejected": -861.81201171875,
739
+ "loss": 0.0189,
740
+ "rewards/accuracies": 0.9937499761581421,
741
+ "rewards/chosen": -1.1223922967910767,
742
+ "rewards/margins": 6.500160217285156,
743
+ "rewards/rejected": -7.62255334854126,
744
+ "step": 480
745
+ },
746
+ {
747
+ "epoch": 3.92,
748
+ "grad_norm": 0.03202298420625548,
749
+ "learning_rate": 7.619929529850397e-06,
750
+ "logits/chosen": -3.1152751445770264,
751
+ "logits/rejected": -3.124176263809204,
752
+ "logps/chosen": -212.0701446533203,
753
+ "logps/rejected": -899.0244140625,
754
+ "loss": 0.0053,
755
+ "rewards/accuracies": 1.0,
756
+ "rewards/chosen": -1.6227067708969116,
757
+ "rewards/margins": 6.344549655914307,
758
+ "rewards/rejected": -7.96725606918335,
759
+ "step": 490
760
+ },
761
+ {
762
+ "epoch": 4.0,
763
+ "grad_norm": 0.056671500571648784,
764
+ "learning_rate": 7.500000000000001e-06,
765
+ "logits/chosen": -3.027770757675171,
766
+ "logits/rejected": -3.058871030807495,
767
+ "logps/chosen": -178.46084594726562,
768
+ "logps/rejected": -894.8692626953125,
769
+ "loss": 0.0051,
770
+ "rewards/accuracies": 1.0,
771
+ "rewards/chosen": -1.2816778421401978,
772
+ "rewards/margins": 6.661429405212402,
773
+ "rewards/rejected": -7.943107604980469,
774
+ "step": 500
775
+ },
776
+ {
777
+ "epoch": 4.08,
778
+ "grad_norm": 0.08007457084674505,
779
+ "learning_rate": 7.378121045351378e-06,
780
+ "logits/chosen": -3.0769410133361816,
781
+ "logits/rejected": -3.099307060241699,
782
+ "logps/chosen": -194.36056518554688,
783
+ "logps/rejected": -881.5306396484375,
784
+ "loss": 0.0097,
785
+ "rewards/accuracies": 1.0,
786
+ "rewards/chosen": -1.4257417917251587,
787
+ "rewards/margins": 6.408936977386475,
788
+ "rewards/rejected": -7.834680080413818,
789
+ "step": 510
790
+ },
791
+ {
792
+ "epoch": 4.16,
793
+ "grad_norm": 2.4134748485667603,
794
+ "learning_rate": 7.254387703447154e-06,
795
+ "logits/chosen": -3.0368213653564453,
796
+ "logits/rejected": -3.0519778728485107,
797
+ "logps/chosen": -203.78298950195312,
798
+ "logps/rejected": -914.9696044921875,
799
+ "loss": 0.0071,
800
+ "rewards/accuracies": 1.0,
801
+ "rewards/chosen": -1.5372952222824097,
802
+ "rewards/margins": 6.609500885009766,
803
+ "rewards/rejected": -8.146797180175781,
804
+ "step": 520
805
+ },
806
+ {
807
+ "epoch": 4.24,
808
+ "grad_norm": 0.029510070347368533,
809
+ "learning_rate": 7.128896457825364e-06,
810
+ "logits/chosen": -2.9700584411621094,
811
+ "logits/rejected": -3.0200304985046387,
812
+ "logps/chosen": -185.78460693359375,
813
+ "logps/rejected": -919.3395385742188,
814
+ "loss": 0.0036,
815
+ "rewards/accuracies": 1.0,
816
+ "rewards/chosen": -1.3613429069519043,
817
+ "rewards/margins": 6.823419094085693,
818
+ "rewards/rejected": -8.184762001037598,
819
+ "step": 530
820
+ },
821
+ {
822
+ "epoch": 4.32,
823
+ "grad_norm": 1.2493776168986983,
824
+ "learning_rate": 7.0017451627844765e-06,
825
+ "logits/chosen": -2.917212724685669,
826
+ "logits/rejected": -2.9564435482025146,
827
+ "logps/chosen": -197.5432891845703,
828
+ "logps/rejected": -924.7901611328125,
829
+ "loss": 0.0201,
830
+ "rewards/accuracies": 1.0,
831
+ "rewards/chosen": -1.476697325706482,
832
+ "rewards/margins": 6.7479658126831055,
833
+ "rewards/rejected": -8.224664688110352,
834
+ "step": 540
835
+ },
836
+ {
837
+ "epoch": 4.4,
838
+ "grad_norm": 0.2763373802778404,
839
+ "learning_rate": 6.873032967079562e-06,
840
+ "logits/chosen": -2.7449865341186523,
841
+ "logits/rejected": -2.7941806316375732,
842
+ "logps/chosen": -146.8518829345703,
843
+ "logps/rejected": -870.0069580078125,
844
+ "loss": 0.0121,
845
+ "rewards/accuracies": 1.0,
846
+ "rewards/chosen": -0.9301493763923645,
847
+ "rewards/margins": 6.802959442138672,
848
+ "rewards/rejected": -7.733109474182129,
849
+ "step": 550
850
+ },
851
+ {
852
+ "epoch": 4.48,
853
+ "grad_norm": 0.03334067080540203,
854
+ "learning_rate": 6.7428602366090764e-06,
855
+ "logits/chosen": -2.7536513805389404,
856
+ "logits/rejected": -2.8025503158569336,
857
+ "logps/chosen": -160.67921447753906,
858
+ "logps/rejected": -924.1072998046875,
859
+ "loss": 0.0208,
860
+ "rewards/accuracies": 1.0,
861
+ "rewards/chosen": -1.0948050022125244,
862
+ "rewards/margins": 7.13351583480835,
863
+ "rewards/rejected": -8.228321075439453,
864
+ "step": 560
865
+ },
866
+ {
867
+ "epoch": 4.5600000000000005,
868
+ "grad_norm": 0.10014923752683715,
869
+ "learning_rate": 6.611328476152557e-06,
870
+ "logits/chosen": -2.943023204803467,
871
+ "logits/rejected": -2.971611261367798,
872
+ "logps/chosen": -181.0494384765625,
873
+ "logps/rejected": -916.7213134765625,
874
+ "loss": 0.0064,
875
+ "rewards/accuracies": 1.0,
876
+ "rewards/chosen": -1.306639552116394,
877
+ "rewards/margins": 6.8469133377075195,
878
+ "rewards/rejected": -8.153553009033203,
879
+ "step": 570
880
+ },
881
+ {
882
+ "epoch": 4.64,
883
+ "grad_norm": 0.5303544380996533,
884
+ "learning_rate": 6.4785402502202345e-06,
885
+ "logits/chosen": -3.0545575618743896,
886
+ "logits/rejected": -3.05729079246521,
887
+ "logps/chosen": -207.95034790039062,
888
+ "logps/rejected": -939.8880615234375,
889
+ "loss": 0.0076,
890
+ "rewards/accuracies": 1.0,
891
+ "rewards/chosen": -1.5920554399490356,
892
+ "rewards/margins": 6.775687217712402,
893
+ "rewards/rejected": -8.367743492126465,
894
+ "step": 580
895
+ },
896
+ {
897
+ "epoch": 4.72,
898
+ "grad_norm": 0.04084615849116197,
899
+ "learning_rate": 6.344599103076329e-06,
900
+ "logits/chosen": -2.9924559593200684,
901
+ "logits/rejected": -2.992640972137451,
902
+ "logps/chosen": -198.8438720703125,
903
+ "logps/rejected": -943.1241455078125,
904
+ "loss": 0.0041,
905
+ "rewards/accuracies": 1.0,
906
+ "rewards/chosen": -1.5001078844070435,
907
+ "rewards/margins": 6.916898250579834,
908
+ "rewards/rejected": -8.417006492614746,
909
+ "step": 590
910
+ },
911
+ {
912
+ "epoch": 4.8,
913
+ "grad_norm": 1.0475565210531403,
914
+ "learning_rate": 6.209609477998339e-06,
915
+ "logits/chosen": -3.0877511501312256,
916
+ "logits/rejected": -3.0805959701538086,
917
+ "logps/chosen": -226.8428955078125,
918
+ "logps/rejected": -941.5935668945312,
919
+ "loss": 0.0178,
920
+ "rewards/accuracies": 0.987500011920929,
921
+ "rewards/chosen": -1.7735874652862549,
922
+ "rewards/margins": 6.6292877197265625,
923
+ "rewards/rejected": -8.402875900268555,
924
+ "step": 600
925
+ },
926
+ {
927
+ "epoch": 4.88,
928
+ "grad_norm": 0.0998487112139878,
929
+ "learning_rate": 6.073676635835317e-06,
930
+ "logits/chosen": -3.117692232131958,
931
+ "logits/rejected": -3.0967986583709717,
932
+ "logps/chosen": -220.5649871826172,
933
+ "logps/rejected": -916.0166015625,
934
+ "loss": 0.01,
935
+ "rewards/accuracies": 1.0,
936
+ "rewards/chosen": -1.7153180837631226,
937
+ "rewards/margins": 6.428615570068359,
938
+ "rewards/rejected": -8.143933296203613,
939
+ "step": 610
940
+ },
941
+ {
942
+ "epoch": 4.96,
943
+ "grad_norm": 0.04577054636911131,
944
+ "learning_rate": 5.936906572928625e-06,
945
+ "logits/chosen": -2.8556697368621826,
946
+ "logits/rejected": -2.8885579109191895,
947
+ "logps/chosen": -171.10992431640625,
948
+ "logps/rejected": -943.6531982421875,
949
+ "loss": 0.0063,
950
+ "rewards/accuracies": 1.0,
951
+ "rewards/chosen": -1.2215626239776611,
952
+ "rewards/margins": 7.193364143371582,
953
+ "rewards/rejected": -8.414926528930664,
954
+ "step": 620
955
+ },
956
+ {
957
+ "epoch": 5.04,
958
+ "grad_norm": 0.1261248214008726,
959
+ "learning_rate": 5.799405938459175e-06,
960
+ "logits/chosen": -2.886103868484497,
961
+ "logits/rejected": -2.930838108062744,
962
+ "logps/chosen": -162.7739715576172,
963
+ "logps/rejected": -950.7653198242188,
964
+ "loss": 0.0046,
965
+ "rewards/accuracies": 1.0,
966
+ "rewards/chosen": -1.1451432704925537,
967
+ "rewards/margins": 7.34460973739624,
968
+ "rewards/rejected": -8.489753723144531,
969
+ "step": 630
970
+ },
971
+ {
972
+ "epoch": 5.12,
973
+ "grad_norm": 0.1505621604680798,
974
+ "learning_rate": 5.661281951285613e-06,
975
+ "logits/chosen": -3.0620782375335693,
976
+ "logits/rejected": -3.0735926628112793,
977
+ "logps/chosen": -191.69161987304688,
978
+ "logps/rejected": -940.0349731445312,
979
+ "loss": 0.0038,
980
+ "rewards/accuracies": 1.0,
981
+ "rewards/chosen": -1.417407751083374,
982
+ "rewards/margins": 6.979997158050537,
983
+ "rewards/rejected": -8.397405624389648,
984
+ "step": 640
985
+ },
986
+ {
987
+ "epoch": 5.2,
988
+ "grad_norm": 0.019057548002213837,
989
+ "learning_rate": 5.522642316338268e-06,
990
+ "logits/chosen": -3.170454502105713,
991
+ "logits/rejected": -3.1733736991882324,
992
+ "logps/chosen": -206.1622314453125,
993
+ "logps/rejected": -938.7789306640625,
994
+ "loss": 0.0038,
995
+ "rewards/accuracies": 1.0,
996
+ "rewards/chosen": -1.5485303401947021,
997
+ "rewards/margins": 6.836629390716553,
998
+ "rewards/rejected": -8.385160446166992,
999
+ "step": 650
1000
+ },
1001
+ {
1002
+ "epoch": 5.28,
1003
+ "grad_norm": 0.04553800143055307,
1004
+ "learning_rate": 5.383595140634093e-06,
1005
+ "logits/chosen": -3.111215114593506,
1006
+ "logits/rejected": -3.113774061203003,
1007
+ "logps/chosen": -197.33230590820312,
1008
+ "logps/rejected": -965.2951049804688,
1009
+ "loss": 0.0032,
1010
+ "rewards/accuracies": 1.0,
1011
+ "rewards/chosen": -1.4725955724716187,
1012
+ "rewards/margins": 7.165165901184082,
1013
+ "rewards/rejected": -8.637761116027832,
1014
+ "step": 660
1015
+ },
1016
+ {
1017
+ "epoch": 5.36,
1018
+ "grad_norm": 0.023672933333312794,
1019
+ "learning_rate": 5.244248848978067e-06,
1020
+ "logits/chosen": -3.087512493133545,
1021
+ "logits/rejected": -3.0911624431610107,
1022
+ "logps/chosen": -215.00796508789062,
1023
+ "logps/rejected": -978.7565307617188,
1024
+ "loss": 0.0064,
1025
+ "rewards/accuracies": 1.0,
1026
+ "rewards/chosen": -1.669549584388733,
1027
+ "rewards/margins": 7.078073978424072,
1028
+ "rewards/rejected": -8.747623443603516,
1029
+ "step": 670
1030
+ },
1031
+ {
1032
+ "epoch": 5.44,
1033
+ "grad_norm": 0.027237164042707367,
1034
+ "learning_rate": 5.1047120994167855e-06,
1035
+ "logits/chosen": -3.069511890411377,
1036
+ "logits/rejected": -3.0692381858825684,
1037
+ "logps/chosen": -202.83555603027344,
1038
+ "logps/rejected": -972.17578125,
1039
+ "loss": 0.003,
1040
+ "rewards/accuracies": 1.0,
1041
+ "rewards/chosen": -1.5378749370574951,
1042
+ "rewards/margins": 7.15402889251709,
1043
+ "rewards/rejected": -8.691904067993164,
1044
+ "step": 680
1045
+ },
1046
+ {
1047
+ "epoch": 5.52,
1048
+ "grad_norm": 0.0238313232064652,
1049
+ "learning_rate": 4.965093698510192e-06,
1050
+ "logits/chosen": -3.045116424560547,
1051
+ "logits/rejected": -3.070358991622925,
1052
+ "logps/chosen": -189.43704223632812,
1053
+ "logps/rejected": -947.7220458984375,
1054
+ "loss": 0.0066,
1055
+ "rewards/accuracies": 1.0,
1056
+ "rewards/chosen": -1.3802012205123901,
1057
+ "rewards/margins": 7.0903000831604,
1058
+ "rewards/rejected": -8.470501899719238,
1059
+ "step": 690
1060
+ },
1061
+ {
1062
+ "epoch": 5.6,
1063
+ "grad_norm": 0.06266588262969701,
1064
+ "learning_rate": 4.825502516487497e-06,
1065
+ "logits/chosen": -2.9892053604125977,
1066
+ "logits/rejected": -3.0090491771698,
1067
+ "logps/chosen": -190.52420043945312,
1068
+ "logps/rejected": -957.5784301757812,
1069
+ "loss": 0.0041,
1070
+ "rewards/accuracies": 1.0,
1071
+ "rewards/chosen": -1.3977526426315308,
1072
+ "rewards/margins": 7.167298793792725,
1073
+ "rewards/rejected": -8.565052032470703,
1074
+ "step": 700
1075
+ },
1076
+ {
1077
+ "epoch": 5.68,
1078
+ "grad_norm": 0.21683933666142513,
1079
+ "learning_rate": 4.686047402353433e-06,
1080
+ "logits/chosen": -3.0948545932769775,
1081
+ "logits/rejected": -3.0949206352233887,
1082
+ "logps/chosen": -215.0717315673828,
1083
+ "logps/rejected": -974.2849731445312,
1084
+ "loss": 0.0048,
1085
+ "rewards/accuracies": 1.0,
1086
+ "rewards/chosen": -1.6559123992919922,
1087
+ "rewards/margins": 7.0673394203186035,
1088
+ "rewards/rejected": -8.723250389099121,
1089
+ "step": 710
1090
+ },
1091
+ {
1092
+ "epoch": 5.76,
1093
+ "grad_norm": 0.029076731867175645,
1094
+ "learning_rate": 4.546837099011101e-06,
1095
+ "logits/chosen": -3.138347864151001,
1096
+ "logits/rejected": -3.1147701740264893,
1097
+ "logps/chosen": -221.6696014404297,
1098
+ "logps/rejected": -976.2689208984375,
1099
+ "loss": 0.0034,
1100
+ "rewards/accuracies": 1.0,
1101
+ "rewards/chosen": -1.7190300226211548,
1102
+ "rewards/margins": 7.019604682922363,
1103
+ "rewards/rejected": -8.738635063171387,
1104
+ "step": 720
1105
+ },
1106
+ {
1107
+ "epoch": 5.84,
1108
+ "grad_norm": 0.036662851444810934,
1109
+ "learning_rate": 4.4079801584674955e-06,
1110
+ "logits/chosen": -3.0819013118743896,
1111
+ "logits/rejected": -3.0665931701660156,
1112
+ "logps/chosen": -216.7307891845703,
1113
+ "logps/rejected": -979.8884887695312,
1114
+ "loss": 0.0056,
1115
+ "rewards/accuracies": 0.9937499761581421,
1116
+ "rewards/chosen": -1.6721961498260498,
1117
+ "rewards/margins": 7.096380710601807,
1118
+ "rewards/rejected": -8.768576622009277,
1119
+ "step": 730
1120
+ },
1121
+ {
1122
+ "epoch": 5.92,
1123
+ "grad_norm": 0.03922484581534493,
1124
+ "learning_rate": 4.269584857187942e-06,
1125
+ "logits/chosen": -3.033648729324341,
1126
+ "logits/rejected": -3.03180193901062,
1127
+ "logps/chosen": -203.57614135742188,
1128
+ "logps/rejected": -953.6720581054688,
1129
+ "loss": 0.0031,
1130
+ "rewards/accuracies": 1.0,
1131
+ "rewards/chosen": -1.5226513147354126,
1132
+ "rewards/margins": 7.014796257019043,
1133
+ "rewards/rejected": -8.537446975708008,
1134
+ "step": 740
1135
+ },
1136
+ {
1137
+ "epoch": 6.0,
1138
+ "grad_norm": 0.031567842692926854,
1139
+ "learning_rate": 4.131759111665349e-06,
1140
+ "logits/chosen": -3.0094892978668213,
1141
+ "logits/rejected": -3.0100715160369873,
1142
+ "logps/chosen": -222.1414794921875,
1143
+ "logps/rejected": -981.8389892578125,
1144
+ "loss": 0.0031,
1145
+ "rewards/accuracies": 1.0,
1146
+ "rewards/chosen": -1.7311455011367798,
1147
+ "rewards/margins": 7.06912899017334,
1148
+ "rewards/rejected": -8.800275802612305,
1149
+ "step": 750
1150
+ },
1151
+ {
1152
+ "epoch": 6.08,
1153
+ "grad_norm": 0.02449878094582341,
1154
+ "learning_rate": 3.994610394270178e-06,
1155
+ "logits/chosen": -2.970534563064575,
1156
+ "logits/rejected": -2.9691994190216064,
1157
+ "logps/chosen": -207.33847045898438,
1158
+ "logps/rejected": -994.8795166015625,
1159
+ "loss": 0.0032,
1160
+ "rewards/accuracies": 1.0,
1161
+ "rewards/chosen": -1.5827735662460327,
1162
+ "rewards/margins": 7.3385796546936035,
1163
+ "rewards/rejected": -8.921354293823242,
1164
+ "step": 760
1165
+ },
1166
+ {
1167
+ "epoch": 6.16,
1168
+ "grad_norm": 0.02981087291288187,
1169
+ "learning_rate": 3.8582456494467214e-06,
1170
+ "logits/chosen": -2.9473607540130615,
1171
+ "logits/rejected": -2.956864833831787,
1172
+ "logps/chosen": -203.9072265625,
1173
+ "logps/rejected": -984.5435791015625,
1174
+ "loss": 0.0028,
1175
+ "rewards/accuracies": 1.0,
1176
+ "rewards/chosen": -1.5347377061843872,
1177
+ "rewards/margins": 7.284613132476807,
1178
+ "rewards/rejected": -8.819352149963379,
1179
+ "step": 770
1180
+ },
1181
+ {
1182
+ "epoch": 6.24,
1183
+ "grad_norm": 0.021043064932021155,
1184
+ "learning_rate": 3.7227712103210485e-06,
1185
+ "logits/chosen": -2.973010778427124,
1186
+ "logits/rejected": -2.9938042163848877,
1187
+ "logps/chosen": -190.59471130371094,
1188
+ "logps/rejected": -966.3521728515625,
1189
+ "loss": 0.0033,
1190
+ "rewards/accuracies": 1.0,
1191
+ "rewards/chosen": -1.3837454319000244,
1192
+ "rewards/margins": 7.294589996337891,
1193
+ "rewards/rejected": -8.678335189819336,
1194
+ "step": 780
1195
+ },
1196
+ {
1197
+ "epoch": 6.32,
1198
+ "grad_norm": 0.039673872651556516,
1199
+ "learning_rate": 3.5882927157856175e-06,
1200
+ "logits/chosen": -2.9964919090270996,
1201
+ "logits/rejected": -3.003882884979248,
1202
+ "logps/chosen": -211.8938446044922,
1203
+ "logps/rejected": -996.83544921875,
1204
+ "loss": 0.0025,
1205
+ "rewards/accuracies": 1.0,
1206
+ "rewards/chosen": -1.6143487691879272,
1207
+ "rewards/margins": 7.334100246429443,
1208
+ "rewards/rejected": -8.948450088500977,
1209
+ "step": 790
1210
+ },
1211
+ {
1212
+ "epoch": 6.4,
1213
+ "grad_norm": 0.04339817998984244,
1214
+ "learning_rate": 3.4549150281252635e-06,
1215
+ "logits/chosen": -2.9923062324523926,
1216
+ "logits/rejected": -3.008997917175293,
1217
+ "logps/chosen": -199.03875732421875,
1218
+ "logps/rejected": -962.5731201171875,
1219
+ "loss": 0.0033,
1220
+ "rewards/accuracies": 1.0,
1221
+ "rewards/chosen": -1.4706881046295166,
1222
+ "rewards/margins": 7.157661437988281,
1223
+ "rewards/rejected": -8.628351211547852,
1224
+ "step": 800
1225
+ },
1226
+ {
1227
+ "epoch": 6.48,
1228
+ "grad_norm": 0.03256715178755916,
1229
+ "learning_rate": 3.322742151248726e-06,
1230
+ "logits/chosen": -2.947535514831543,
1231
+ "logits/rejected": -2.9602136611938477,
1232
+ "logps/chosen": -194.71202087402344,
1233
+ "logps/rejected": -992.3277587890625,
1234
+ "loss": 0.0033,
1235
+ "rewards/accuracies": 1.0,
1236
+ "rewards/chosen": -1.4545636177062988,
1237
+ "rewards/margins": 7.46028995513916,
1238
+ "rewards/rejected": -8.914854049682617,
1239
+ "step": 810
1240
+ },
1241
+ {
1242
+ "epoch": 6.5600000000000005,
1243
+ "grad_norm": 0.03103799595540118,
1244
+ "learning_rate": 3.1918771495895395e-06,
1245
+ "logits/chosen": -2.986626625061035,
1246
+ "logits/rejected": -3.0086371898651123,
1247
+ "logps/chosen": -199.30032348632812,
1248
+ "logps/rejected": -983.8270263671875,
1249
+ "loss": 0.0051,
1250
+ "rewards/accuracies": 0.9937499761581421,
1251
+ "rewards/chosen": -1.479447603225708,
1252
+ "rewards/margins": 7.346575736999512,
1253
+ "rewards/rejected": -8.826024055480957,
1254
+ "step": 820
1255
+ },
1256
+ {
1257
+ "epoch": 6.64,
1258
+ "grad_norm": 0.012923263616360448,
1259
+ "learning_rate": 3.0624220677394854e-06,
1260
+ "logits/chosen": -2.980841636657715,
1261
+ "logits/rejected": -2.9940218925476074,
1262
+ "logps/chosen": -201.14207458496094,
1263
+ "logps/rejected": -989.4503784179688,
1264
+ "loss": 0.002,
1265
+ "rewards/accuracies": 1.0,
1266
+ "rewards/chosen": -1.5033818483352661,
1267
+ "rewards/margins": 7.376742362976074,
1268
+ "rewards/rejected": -8.880125045776367,
1269
+ "step": 830
1270
+ },
1271
+ {
1272
+ "epoch": 6.72,
1273
+ "grad_norm": 0.4008776832999778,
1274
+ "learning_rate": 2.934477850877292e-06,
1275
+ "logits/chosen": -2.8989338874816895,
1276
+ "logits/rejected": -2.904329299926758,
1277
+ "logps/chosen": -175.6615447998047,
1278
+ "logps/rejected": -993.4110107421875,
1279
+ "loss": 0.0059,
1280
+ "rewards/accuracies": 1.0,
1281
+ "rewards/chosen": -1.2583926916122437,
1282
+ "rewards/margins": 7.667746543884277,
1283
+ "rewards/rejected": -8.926138877868652,
1284
+ "step": 840
1285
+ },
1286
+ {
1287
+ "epoch": 6.8,
1288
+ "grad_norm": 0.03242501786191666,
1289
+ "learning_rate": 2.8081442660546126e-06,
1290
+ "logits/chosen": -2.9362475872039795,
1291
+ "logits/rejected": -2.942796230316162,
1292
+ "logps/chosen": -200.99786376953125,
1293
+ "logps/rejected": -999.9691162109375,
1294
+ "loss": 0.0026,
1295
+ "rewards/accuracies": 1.0,
1296
+ "rewards/chosen": -1.530400037765503,
1297
+ "rewards/margins": 7.44219970703125,
1298
+ "rewards/rejected": -8.9725980758667,
1299
+ "step": 850
1300
+ },
1301
+ {
1302
+ "epoch": 6.88,
1303
+ "grad_norm": 0.04701366652822717,
1304
+ "learning_rate": 2.683519824400693e-06,
1305
+ "logits/chosen": -2.9499258995056152,
1306
+ "logits/rejected": -2.9498021602630615,
1307
+ "logps/chosen": -200.02708435058594,
1308
+ "logps/rejected": -1003.5309448242188,
1309
+ "loss": 0.0022,
1310
+ "rewards/accuracies": 1.0,
1311
+ "rewards/chosen": -1.5131416320800781,
1312
+ "rewards/margins": 7.508401393890381,
1313
+ "rewards/rejected": -9.021543502807617,
1314
+ "step": 860
1315
+ },
1316
+ {
1317
+ "epoch": 6.96,
1318
+ "grad_norm": 0.06965442317131018,
1319
+ "learning_rate": 2.560701704306336e-06,
1320
+ "logits/chosen": -2.992658853530884,
1321
+ "logits/rejected": -3.0063395500183105,
1322
+ "logps/chosen": -196.2028045654297,
1323
+ "logps/rejected": -988.9591064453125,
1324
+ "loss": 0.0028,
1325
+ "rewards/accuracies": 1.0,
1326
+ "rewards/chosen": -1.4468988180160522,
1327
+ "rewards/margins": 7.434958457946777,
1328
+ "rewards/rejected": -8.881857872009277,
1329
+ "step": 870
1330
+ },
1331
+ {
1332
+ "epoch": 7.04,
1333
+ "grad_norm": 0.04943704962243662,
1334
+ "learning_rate": 2.4397856756471435e-06,
1335
+ "logits/chosen": -2.967582941055298,
1336
+ "logits/rejected": -2.969144344329834,
1337
+ "logps/chosen": -201.17965698242188,
1338
+ "logps/rejected": -1002.7322387695312,
1339
+ "loss": 0.0023,
1340
+ "rewards/accuracies": 1.0,
1341
+ "rewards/chosen": -1.5157372951507568,
1342
+ "rewards/margins": 7.4925055503845215,
1343
+ "rewards/rejected": -9.0082426071167,
1344
+ "step": 880
1345
+ },
1346
+ {
1347
+ "epoch": 7.12,
1348
+ "grad_norm": 0.026898832512229635,
1349
+ "learning_rate": 2.320866025105016e-06,
1350
+ "logits/chosen": -3.0048301219940186,
1351
+ "logits/rejected": -3.0222504138946533,
1352
+ "logps/chosen": -212.1355438232422,
1353
+ "logps/rejected": -974.49072265625,
1354
+ "loss": 0.0028,
1355
+ "rewards/accuracies": 1.0,
1356
+ "rewards/chosen": -1.6027580499649048,
1357
+ "rewards/margins": 7.149576663970947,
1358
+ "rewards/rejected": -8.752334594726562,
1359
+ "step": 890
1360
+ },
1361
+ {
1362
+ "epoch": 7.2,
1363
+ "grad_norm": 0.018773487381868556,
1364
+ "learning_rate": 2.204035482646267e-06,
1365
+ "logits/chosen": -3.004668712615967,
1366
+ "logits/rejected": -3.02616024017334,
1367
+ "logps/chosen": -204.20449829101562,
1368
+ "logps/rejected": -988.2781372070312,
1369
+ "loss": 0.0023,
1370
+ "rewards/accuracies": 1.0,
1371
+ "rewards/chosen": -1.5249879360198975,
1372
+ "rewards/margins": 7.354971408843994,
1373
+ "rewards/rejected": -8.879959106445312,
1374
+ "step": 900
1375
+ },
1376
+ {
1377
+ "epoch": 7.28,
1378
+ "grad_norm": 0.010977198093830244,
1379
+ "learning_rate": 2.0893851492135536e-06,
1380
+ "logits/chosen": -3.026869773864746,
1381
+ "logits/rejected": -3.0388169288635254,
1382
+ "logps/chosen": -211.68359375,
1383
+ "logps/rejected": -987.3611450195312,
1384
+ "loss": 0.0021,
1385
+ "rewards/accuracies": 1.0,
1386
+ "rewards/chosen": -1.6020605564117432,
1387
+ "rewards/margins": 7.269850254058838,
1388
+ "rewards/rejected": -8.871912002563477,
1389
+ "step": 910
1390
+ },
1391
+ {
1392
+ "epoch": 7.36,
1393
+ "grad_norm": 0.29173762576971574,
1394
+ "learning_rate": 1.977004425688126e-06,
1395
+ "logits/chosen": -3.0339274406433105,
1396
+ "logits/rejected": -3.066455602645874,
1397
+ "logps/chosen": -212.5125732421875,
1398
+ "logps/rejected": -1007.0792846679688,
1399
+ "loss": 0.005,
1400
+ "rewards/accuracies": 1.0,
1401
+ "rewards/chosen": -1.6347990036010742,
1402
+ "rewards/margins": 7.410192966461182,
1403
+ "rewards/rejected": -9.044992446899414,
1404
+ "step": 920
1405
+ },
1406
+ {
1407
+ "epoch": 7.44,
1408
+ "grad_norm": 0.021055806854853823,
1409
+ "learning_rate": 1.8669809431776991e-06,
1410
+ "logits/chosen": -2.9736626148223877,
1411
+ "logits/rejected": -2.9851651191711426,
1412
+ "logps/chosen": -196.3205108642578,
1413
+ "logps/rejected": -990.7384643554688,
1414
+ "loss": 0.0022,
1415
+ "rewards/accuracies": 1.0,
1416
+ "rewards/chosen": -1.459997296333313,
1417
+ "rewards/margins": 7.45266580581665,
1418
+ "rewards/rejected": -8.912662506103516,
1419
+ "step": 930
1420
+ },
1421
+ {
1422
+ "epoch": 7.52,
1423
+ "grad_norm": 0.18348617533249506,
1424
+ "learning_rate": 1.7594004946843458e-06,
1425
+ "logits/chosen": -2.942028522491455,
1426
+ "logits/rejected": -2.9542624950408936,
1427
+ "logps/chosen": -199.4943389892578,
1428
+ "logps/rejected": -1008.6310424804688,
1429
+ "loss": 0.003,
1430
+ "rewards/accuracies": 1.0,
1431
+ "rewards/chosen": -1.5059641599655151,
1432
+ "rewards/margins": 7.565948486328125,
1433
+ "rewards/rejected": -9.07191276550293,
1434
+ "step": 940
1435
+ },
1436
+ {
1437
+ "epoch": 7.6,
1438
+ "grad_norm": 0.02961600216806713,
1439
+ "learning_rate": 1.6543469682057105e-06,
1440
+ "logits/chosen": -2.9790632724761963,
1441
+ "logits/rejected": -2.981333017349243,
1442
+ "logps/chosen": -225.8614959716797,
1443
+ "logps/rejected": -1018.3231201171875,
1444
+ "loss": 0.0025,
1445
+ "rewards/accuracies": 1.0,
1446
+ "rewards/chosen": -1.773397445678711,
1447
+ "rewards/margins": 7.389155387878418,
1448
+ "rewards/rejected": -9.162553787231445,
1449
+ "step": 950
1450
+ },
1451
+ {
1452
+ "epoch": 7.68,
1453
+ "grad_norm": 0.033303351274835385,
1454
+ "learning_rate": 1.551902281321651e-06,
1455
+ "logits/chosen": -3.0063486099243164,
1456
+ "logits/rejected": -3.0057759284973145,
1457
+ "logps/chosen": -210.657470703125,
1458
+ "logps/rejected": -1013.4367065429688,
1459
+ "loss": 0.0023,
1460
+ "rewards/accuracies": 1.0,
1461
+ "rewards/chosen": -1.6198650598526,
1462
+ "rewards/margins": 7.4931464195251465,
1463
+ "rewards/rejected": -9.113012313842773,
1464
+ "step": 960
1465
+ },
1466
+ {
1467
+ "epoch": 7.76,
1468
+ "grad_norm": 0.02991133657233578,
1469
+ "learning_rate": 1.4521463173173966e-06,
1470
+ "logits/chosen": -2.9552433490753174,
1471
+ "logits/rejected": -2.986539363861084,
1472
+ "logps/chosen": -185.34751892089844,
1473
+ "logps/rejected": -995.6087646484375,
1474
+ "loss": 0.0022,
1475
+ "rewards/accuracies": 1.0,
1476
+ "rewards/chosen": -1.3463085889816284,
1477
+ "rewards/margins": 7.610987186431885,
1478
+ "rewards/rejected": -8.957294464111328,
1479
+ "step": 970
1480
+ },
1481
+ {
1482
+ "epoch": 7.84,
1483
+ "grad_norm": 0.03956036447686216,
1484
+ "learning_rate": 1.3551568628929434e-06,
1485
+ "logits/chosen": -2.987436532974243,
1486
+ "logits/rejected": -2.99739146232605,
1487
+ "logps/chosen": -202.7963409423828,
1488
+ "logps/rejected": -1010.6976318359375,
1489
+ "loss": 0.0028,
1490
+ "rewards/accuracies": 1.0,
1491
+ "rewards/chosen": -1.5336072444915771,
1492
+ "rewards/margins": 7.547433376312256,
1493
+ "rewards/rejected": -9.081039428710938,
1494
+ "step": 980
1495
+ },
1496
+ {
1497
+ "epoch": 7.92,
1498
+ "grad_norm": 0.0380908854419757,
1499
+ "learning_rate": 1.2610095475073415e-06,
1500
+ "logits/chosen": -2.966024160385132,
1501
+ "logits/rejected": -2.9880573749542236,
1502
+ "logps/chosen": -188.83627319335938,
1503
+ "logps/rejected": -1017.0061645507812,
1504
+ "loss": 0.0022,
1505
+ "rewards/accuracies": 1.0,
1506
+ "rewards/chosen": -1.397072196006775,
1507
+ "rewards/margins": 7.754624366760254,
1508
+ "rewards/rejected": -9.15169620513916,
1509
+ "step": 990
1510
+ },
1511
+ {
1512
+ "epoch": 8.0,
1513
+ "grad_norm": 0.056813053850928295,
1514
+ "learning_rate": 1.1697777844051105e-06,
1515
+ "logits/chosen": -3.0522654056549072,
1516
+ "logits/rejected": -3.037459135055542,
1517
+ "logps/chosen": -224.0611114501953,
1518
+ "logps/rejected": -1007.8203125,
1519
+ "loss": 0.0028,
1520
+ "rewards/accuracies": 1.0,
1521
+ "rewards/chosen": -1.7456802129745483,
1522
+ "rewards/margins": 7.323793888092041,
1523
+ "rewards/rejected": -9.069474220275879,
1524
+ "step": 1000
1525
+ },
1526
+ {
1527
+ "epoch": 8.08,
1528
+ "grad_norm": 0.04428932232427895,
1529
+ "learning_rate": 1.0815327133708015e-06,
1530
+ "logits/chosen": -2.974698305130005,
1531
+ "logits/rejected": -2.989356756210327,
1532
+ "logps/chosen": -199.12106323242188,
1533
+ "logps/rejected": -999.2364501953125,
1534
+ "loss": 0.0048,
1535
+ "rewards/accuracies": 0.9937499761581421,
1536
+ "rewards/chosen": -1.4827978610992432,
1537
+ "rewards/margins": 7.491432189941406,
1538
+ "rewards/rejected": -8.97422981262207,
1539
+ "step": 1010
1540
+ },
1541
+ {
1542
+ "epoch": 8.16,
1543
+ "grad_norm": 0.03094089083601851,
1544
+ "learning_rate": 9.963431452563331e-07,
1545
+ "logits/chosen": -2.9914603233337402,
1546
+ "logits/rejected": -2.9979825019836426,
1547
+ "logps/chosen": -202.41212463378906,
1548
+ "logps/rejected": -983.4681396484375,
1549
+ "loss": 0.0022,
1550
+ "rewards/accuracies": 1.0,
1551
+ "rewards/chosen": -1.5089391469955444,
1552
+ "rewards/margins": 7.3303656578063965,
1553
+ "rewards/rejected": -8.839305877685547,
1554
+ "step": 1020
1555
+ },
1556
+ {
1557
+ "epoch": 8.24,
1558
+ "grad_norm": 0.027522596865270775,
1559
+ "learning_rate": 9.142755083243577e-07,
1560
+ "logits/chosen": -3.0142979621887207,
1561
+ "logits/rejected": -3.0246782302856445,
1562
+ "logps/chosen": -206.26199340820312,
1563
+ "logps/rejected": -1013.3017578125,
1564
+ "loss": 0.0028,
1565
+ "rewards/accuracies": 1.0,
1566
+ "rewards/chosen": -1.5663325786590576,
1567
+ "rewards/margins": 7.545605659484863,
1568
+ "rewards/rejected": -9.111937522888184,
1569
+ "step": 1030
1570
+ },
1571
+ {
1572
+ "epoch": 8.32,
1573
+ "grad_norm": 0.02636838723753889,
1574
+ "learning_rate": 8.353937964495029e-07,
1575
+ "logits/chosen": -2.9712257385253906,
1576
+ "logits/rejected": -2.994713544845581,
1577
+ "logps/chosen": -199.3070526123047,
1578
+ "logps/rejected": -992.4100341796875,
1579
+ "loss": 0.0021,
1580
+ "rewards/accuracies": 1.0,
1581
+ "rewards/chosen": -1.476993441581726,
1582
+ "rewards/margins": 7.447300910949707,
1583
+ "rewards/rejected": -8.924293518066406,
1584
+ "step": 1040
1585
+ },
1586
+ {
1587
+ "epoch": 8.4,
1588
+ "grad_norm": 0.021646075748453193,
1589
+ "learning_rate": 7.597595192178702e-07,
1590
+ "logits/chosen": -2.9999868869781494,
1591
+ "logits/rejected": -3.0006492137908936,
1592
+ "logps/chosen": -215.9517364501953,
1593
+ "logps/rejected": -997.6002197265625,
1594
+ "loss": 0.0026,
1595
+ "rewards/accuracies": 1.0,
1596
+ "rewards/chosen": -1.6561391353607178,
1597
+ "rewards/margins": 7.31753396987915,
1598
+ "rewards/rejected": -8.973673820495605,
1599
+ "step": 1050
1600
+ },
1601
+ {
1602
+ "epoch": 8.48,
1603
+ "grad_norm": 0.030961558657159674,
1604
+ "learning_rate": 6.874316539637127e-07,
1605
+ "logits/chosen": -2.9921913146972656,
1606
+ "logits/rejected": -2.9953560829162598,
1607
+ "logps/chosen": -215.5575408935547,
1608
+ "logps/rejected": -1014.0217895507812,
1609
+ "loss": 0.0021,
1610
+ "rewards/accuracies": 1.0,
1611
+ "rewards/chosen": -1.659396767616272,
1612
+ "rewards/margins": 7.469141960144043,
1613
+ "rewards/rejected": -9.128539085388184,
1614
+ "step": 1060
1615
+ },
1616
+ {
1617
+ "epoch": 8.56,
1618
+ "grad_norm": 0.021967484279154154,
1619
+ "learning_rate": 6.184665997806832e-07,
1620
+ "logits/chosen": -3.006669282913208,
1621
+ "logits/rejected": -3.005312442779541,
1622
+ "logps/chosen": -213.9212188720703,
1623
+ "logps/rejected": -990.9972534179688,
1624
+ "loss": 0.0026,
1625
+ "rewards/accuracies": 1.0,
1626
+ "rewards/chosen": -1.6195297241210938,
1627
+ "rewards/margins": 7.294030666351318,
1628
+ "rewards/rejected": -8.91356086730957,
1629
+ "step": 1070
1630
+ },
1631
+ {
1632
+ "epoch": 8.64,
1633
+ "grad_norm": 0.042081431226620426,
1634
+ "learning_rate": 5.529181335435124e-07,
1635
+ "logits/chosen": -3.008005142211914,
1636
+ "logits/rejected": -3.0087175369262695,
1637
+ "logps/chosen": -218.51760864257812,
1638
+ "logps/rejected": -1024.052978515625,
1639
+ "loss": 0.002,
1640
+ "rewards/accuracies": 1.0,
1641
+ "rewards/chosen": -1.690387487411499,
1642
+ "rewards/margins": 7.524999141693115,
1643
+ "rewards/rejected": -9.215387344360352,
1644
+ "step": 1080
1645
+ },
1646
+ {
1647
+ "epoch": 8.72,
1648
+ "grad_norm": 0.043812953916419525,
1649
+ "learning_rate": 4.908373679744316e-07,
1650
+ "logits/chosen": -3.0348081588745117,
1651
+ "logits/rejected": -3.0385255813598633,
1652
+ "logps/chosen": -226.70779418945312,
1653
+ "logps/rejected": -1010.8932495117188,
1654
+ "loss": 0.0023,
1655
+ "rewards/accuracies": 1.0,
1656
+ "rewards/chosen": -1.7693846225738525,
1657
+ "rewards/margins": 7.316775321960449,
1658
+ "rewards/rejected": -9.086159706115723,
1659
+ "step": 1090
1660
+ },
1661
+ {
1662
+ "epoch": 8.8,
1663
+ "grad_norm": 0.02641646477682858,
1664
+ "learning_rate": 4.322727117869951e-07,
1665
+ "logits/chosen": -3.014754056930542,
1666
+ "logits/rejected": -3.0084919929504395,
1667
+ "logps/chosen": -217.29788208007812,
1668
+ "logps/rejected": -1013.4778442382812,
1669
+ "loss": 0.0021,
1670
+ "rewards/accuracies": 1.0,
1671
+ "rewards/chosen": -1.679574728012085,
1672
+ "rewards/margins": 7.428310394287109,
1673
+ "rewards/rejected": -9.107885360717773,
1674
+ "step": 1100
1675
+ },
1676
+ {
1677
+ "epoch": 8.88,
1678
+ "grad_norm": 0.027581407278890403,
1679
+ "learning_rate": 3.772698319384349e-07,
1680
+ "logits/chosen": -2.9870715141296387,
1681
+ "logits/rejected": -2.974972724914551,
1682
+ "logps/chosen": -213.857177734375,
1683
+ "logps/rejected": -1029.269287109375,
1684
+ "loss": 0.0023,
1685
+ "rewards/accuracies": 1.0,
1686
+ "rewards/chosen": -1.6536537408828735,
1687
+ "rewards/margins": 7.60448694229126,
1688
+ "rewards/rejected": -9.258139610290527,
1689
+ "step": 1110
1690
+ },
1691
+ {
1692
+ "epoch": 8.96,
1693
+ "grad_norm": 0.018398802557307123,
1694
+ "learning_rate": 3.258716180199278e-07,
1695
+ "logits/chosen": -2.969782590866089,
1696
+ "logits/rejected": -2.9666781425476074,
1697
+ "logps/chosen": -201.89608764648438,
1698
+ "logps/rejected": -1012.4627075195312,
1699
+ "loss": 0.0021,
1700
+ "rewards/accuracies": 1.0,
1701
+ "rewards/chosen": -1.521287441253662,
1702
+ "rewards/margins": 7.587810516357422,
1703
+ "rewards/rejected": -9.109098434448242,
1704
+ "step": 1120
1705
+ },
1706
+ {
1707
+ "epoch": 9.04,
1708
+ "grad_norm": 0.02530896697144264,
1709
+ "learning_rate": 2.7811814881259503e-07,
1710
+ "logits/chosen": -2.967005491256714,
1711
+ "logits/rejected": -2.9650299549102783,
1712
+ "logps/chosen": -211.93923950195312,
1713
+ "logps/rejected": -1015.2527465820312,
1714
+ "loss": 0.0022,
1715
+ "rewards/accuracies": 1.0,
1716
+ "rewards/chosen": -1.6247079372406006,
1717
+ "rewards/margins": 7.509281158447266,
1718
+ "rewards/rejected": -9.133989334106445,
1719
+ "step": 1130
1720
+ },
1721
+ {
1722
+ "epoch": 9.12,
1723
+ "grad_norm": 0.021443293328022403,
1724
+ "learning_rate": 2.3404666103526542e-07,
1725
+ "logits/chosen": -2.951347827911377,
1726
+ "logits/rejected": -2.967103958129883,
1727
+ "logps/chosen": -192.74740600585938,
1728
+ "logps/rejected": -1004.8844604492188,
1729
+ "loss": 0.0023,
1730
+ "rewards/accuracies": 1.0,
1731
+ "rewards/chosen": -1.4152947664260864,
1732
+ "rewards/margins": 7.6158270835876465,
1733
+ "rewards/rejected": -9.031121253967285,
1734
+ "step": 1140
1735
+ },
1736
+ {
1737
+ "epoch": 9.2,
1738
+ "grad_norm": 0.028269129055998828,
1739
+ "learning_rate": 1.9369152030840553e-07,
1740
+ "logits/chosen": -2.968433141708374,
1741
+ "logits/rejected": -2.975369930267334,
1742
+ "logps/chosen": -208.6031036376953,
1743
+ "logps/rejected": -1004.6807861328125,
1744
+ "loss": 0.002,
1745
+ "rewards/accuracies": 1.0,
1746
+ "rewards/chosen": -1.5802977085113525,
1747
+ "rewards/margins": 7.448237419128418,
1748
+ "rewards/rejected": -9.028534889221191,
1749
+ "step": 1150
1750
+ },
1751
+ {
1752
+ "epoch": 9.28,
1753
+ "grad_norm": 0.012741391504073269,
1754
+ "learning_rate": 1.5708419435684463e-07,
1755
+ "logits/chosen": -2.979626417160034,
1756
+ "logits/rejected": -3.0008533000946045,
1757
+ "logps/chosen": -203.78268432617188,
1758
+ "logps/rejected": -998.7214965820312,
1759
+ "loss": 0.0022,
1760
+ "rewards/accuracies": 1.0,
1761
+ "rewards/chosen": -1.53437077999115,
1762
+ "rewards/margins": 7.443058967590332,
1763
+ "rewards/rejected": -8.97743034362793,
1764
+ "step": 1160
1765
+ },
1766
+ {
1767
+ "epoch": 9.36,
1768
+ "grad_norm": 0.018928993356126772,
1769
+ "learning_rate": 1.2425322847218368e-07,
1770
+ "logits/chosen": -2.9456331729888916,
1771
+ "logits/rejected": -2.947418689727783,
1772
+ "logps/chosen": -208.3197784423828,
1773
+ "logps/rejected": -1002.4910888671875,
1774
+ "loss": 0.0024,
1775
+ "rewards/accuracies": 1.0,
1776
+ "rewards/chosen": -1.5662791728973389,
1777
+ "rewards/margins": 7.459691524505615,
1778
+ "rewards/rejected": -9.025972366333008,
1779
+ "step": 1170
1780
+ },
1781
+ {
1782
+ "epoch": 9.44,
1783
+ "grad_norm": 0.11171748274756513,
1784
+ "learning_rate": 9.522422325404234e-08,
1785
+ "logits/chosen": -2.972970485687256,
1786
+ "logits/rejected": -2.9851505756378174,
1787
+ "logps/chosen": -196.48709106445312,
1788
+ "logps/rejected": -1000.6891479492188,
1789
+ "loss": 0.0025,
1790
+ "rewards/accuracies": 1.0,
1791
+ "rewards/chosen": -1.4484045505523682,
1792
+ "rewards/margins": 7.5574445724487305,
1793
+ "rewards/rejected": -9.005849838256836,
1794
+ "step": 1180
1795
+ },
1796
+ {
1797
+ "epoch": 9.52,
1798
+ "grad_norm": 0.014643254346465179,
1799
+ "learning_rate": 7.001981464747565e-08,
1800
+ "logits/chosen": -3.006410598754883,
1801
+ "logits/rejected": -3.001279354095459,
1802
+ "logps/chosen": -239.8344268798828,
1803
+ "logps/rejected": -1013.0524291992188,
1804
+ "loss": 0.0044,
1805
+ "rewards/accuracies": 1.0,
1806
+ "rewards/chosen": -1.9079118967056274,
1807
+ "rewards/margins": 7.20407247543335,
1808
+ "rewards/rejected": -9.111984252929688,
1809
+ "step": 1190
1810
+ },
1811
+ {
1812
+ "epoch": 9.6,
1813
+ "grad_norm": 0.023445989539047585,
1814
+ "learning_rate": 4.865965629214819e-08,
1815
+ "logits/chosen": -2.9561855792999268,
1816
+ "logits/rejected": -2.9647486209869385,
1817
+ "logps/chosen": -201.7306671142578,
1818
+ "logps/rejected": -982.6613159179688,
1819
+ "loss": 0.0026,
1820
+ "rewards/accuracies": 1.0,
1821
+ "rewards/chosen": -1.502721905708313,
1822
+ "rewards/margins": 7.332386016845703,
1823
+ "rewards/rejected": -8.835107803344727,
1824
+ "step": 1200
1825
+ },
1826
+ {
1827
+ "epoch": 9.68,
1828
+ "grad_norm": 0.0197958747361798,
1829
+ "learning_rate": 3.1160404197018155e-08,
1830
+ "logits/chosen": -2.9367284774780273,
1831
+ "logits/rejected": -2.94865083694458,
1832
+ "logps/chosen": -199.8970947265625,
1833
+ "logps/rejected": -1011.3294677734375,
1834
+ "loss": 0.0027,
1835
+ "rewards/accuracies": 1.0,
1836
+ "rewards/chosen": -1.5013561248779297,
1837
+ "rewards/margins": 7.6023712158203125,
1838
+ "rewards/rejected": -9.103727340698242,
1839
+ "step": 1210
1840
+ },
1841
+ {
1842
+ "epoch": 9.76,
1843
+ "grad_norm": 0.07635646097593517,
1844
+ "learning_rate": 1.753570375247815e-08,
1845
+ "logits/chosen": -3.0195131301879883,
1846
+ "logits/rejected": -3.0240790843963623,
1847
+ "logps/chosen": -220.675048828125,
1848
+ "logps/rejected": -1000.5875854492188,
1849
+ "loss": 0.0024,
1850
+ "rewards/accuracies": 1.0,
1851
+ "rewards/chosen": -1.7074737548828125,
1852
+ "rewards/margins": 7.296438694000244,
1853
+ "rewards/rejected": -9.003911972045898,
1854
+ "step": 1220
1855
+ },
1856
+ {
1857
+ "epoch": 9.84,
1858
+ "grad_norm": 0.011307093391428234,
1859
+ "learning_rate": 7.796179090094891e-09,
1860
+ "logits/chosen": -2.937976837158203,
1861
+ "logits/rejected": -2.9619266986846924,
1862
+ "logps/chosen": -190.8544921875,
1863
+ "logps/rejected": -1021.5460815429688,
1864
+ "loss": 0.0018,
1865
+ "rewards/accuracies": 1.0,
1866
+ "rewards/chosen": -1.419277548789978,
1867
+ "rewards/margins": 7.7647881507873535,
1868
+ "rewards/rejected": -9.184064865112305,
1869
+ "step": 1230
1870
+ },
1871
+ {
1872
+ "epoch": 9.92,
1873
+ "grad_norm": 0.017541852350439674,
1874
+ "learning_rate": 1.9494247982282386e-09,
1875
+ "logits/chosen": -2.976008653640747,
1876
+ "logits/rejected": -2.9958183765411377,
1877
+ "logps/chosen": -201.06101989746094,
1878
+ "logps/rejected": -1008.154296875,
1879
+ "loss": 0.0021,
1880
+ "rewards/accuracies": 1.0,
1881
+ "rewards/chosen": -1.5034410953521729,
1882
+ "rewards/margins": 7.560837745666504,
1883
+ "rewards/rejected": -9.064278602600098,
1884
+ "step": 1240
1885
+ },
1886
+ {
1887
+ "epoch": 10.0,
1888
+ "grad_norm": 0.01886261936533329,
1889
+ "learning_rate": 0.0,
1890
+ "logits/chosen": -2.998544454574585,
1891
+ "logits/rejected": -3.019519329071045,
1892
+ "logps/chosen": -217.5911102294922,
1893
+ "logps/rejected": -1009.5062255859375,
1894
+ "loss": 0.002,
1895
+ "rewards/accuracies": 1.0,
1896
+ "rewards/chosen": -1.6805696487426758,
1897
+ "rewards/margins": 7.409621238708496,
1898
+ "rewards/rejected": -9.090190887451172,
1899
+ "step": 1250
1900
+ },
1901
+ {
1902
+ "epoch": 10.0,
1903
+ "step": 1250,
1904
+ "total_flos": 0.0,
1905
+ "train_loss": 0.05206389323771,
1906
+ "train_runtime": 14277.3883,
1907
+ "train_samples_per_second": 5.603,
1908
+ "train_steps_per_second": 0.088
1909
+ }
1910
+ ],
1911
+ "logging_steps": 10,
1912
+ "max_steps": 1250,
1913
+ "num_input_tokens_seen": 0,
1914
+ "num_train_epochs": 10,
1915
+ "save_steps": 1000,
1916
+ "stateful_callbacks": {
1917
+ "TrainerControl": {
1918
+ "args": {
1919
+ "should_epoch_stop": false,
1920
+ "should_evaluate": false,
1921
+ "should_log": false,
1922
+ "should_save": true,
1923
+ "should_training_stop": true
1924
+ },
1925
+ "attributes": {}
1926
+ }
1927
+ },
1928
+ "total_flos": 0.0,
1929
+ "train_batch_size": 4,
1930
+ "trial_name": null,
1931
+ "trial_params": null
1932
+ }