BraylonDash commited on
Commit
2d19400
1 Parent(s): 6515a10

Model save

Browse files
README.md ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: peft
4
+ tags:
5
+ - trl
6
+ - dpo
7
+ - generated_from_trainer
8
+ base_model: DUAL-GPO/phi-2-gpo-new-i0
9
+ model-index:
10
+ - name: phi-2-gpo-newSFT-b0.001-v4-i1
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
+ # phi-2-gpo-newSFT-b0.001-v4-i1
18
+
19
+ This model is a fine-tuned version of [DUAL-GPO/phi-2-gpo-new-i0](https://huggingface.co/DUAL-GPO/phi-2-gpo-new-i0) on the None 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: 5e-06
39
+ - train_batch_size: 4
40
+ - eval_batch_size: 4
41
+ - seed: 42
42
+ - distributed_type: multi-GPU
43
+ - gradient_accumulation_steps: 4
44
+ - total_train_batch_size: 16
45
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
46
+ - lr_scheduler_type: cosine
47
+ - lr_scheduler_warmup_ratio: 0.1
48
+ - num_epochs: 1
49
+
50
+ ### Training results
51
+
52
+
53
+
54
+ ### Framework versions
55
+
56
+ - PEFT 0.7.1
57
+ - Transformers 4.36.2
58
+ - Pytorch 2.1.2
59
+ - Datasets 2.14.6
60
+ - Tokenizers 0.15.2
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a3d19c43e270266152ff30d1ec3c26628f29a38722eca3d1a9521c4c3dc9f0f1
3
  size 167807296
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:171359864dd186b9bec74b1c4a79ac39a10f9cbcb2cadce213cb9891943af920
3
  size 167807296
all_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 1.8189717344284058,
4
+ "train_runtime": 11934.9419,
5
+ "train_samples": 61135,
6
+ "train_samples_per_second": 1.676,
7
+ "train_steps_per_second": 0.105
8
+ }
runs/May13_17-58-53_gpu4-119-5/events.out.tfevents.1715587249.gpu4-119-5.388100.0 CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a468efdd7d0c6c7dc45a852cf27434ef0722bb802228d70c9ea2787caa075fe8
3
- size 81450
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:887dc53dc2702cebceb8c6ee1fe4373ff10ff59987c9b8eed4dccc677a7d7dd3
3
+ size 84974
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 1.8189717344284058,
4
+ "train_runtime": 11934.9419,
5
+ "train_samples": 61135,
6
+ "train_samples_per_second": 1.676,
7
+ "train_steps_per_second": 0.105
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,1794 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 1.0,
5
+ "eval_steps": 500,
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": 0.0,
13
+ "learning_rate": 4e-08,
14
+ "logits/chosen": 0.06502792239189148,
15
+ "logits/rejected": 0.2657509744167328,
16
+ "logps/chosen": -767.5213012695312,
17
+ "logps/rejected": -534.5206298828125,
18
+ "loss": 1.8815,
19
+ "rewards/accuracies": 0.0,
20
+ "rewards/chosen": 0.0,
21
+ "rewards/margins": 0.0,
22
+ "rewards/rejected": 0.0,
23
+ "step": 1
24
+ },
25
+ {
26
+ "epoch": 0.01,
27
+ "learning_rate": 4.0000000000000003e-07,
28
+ "logits/chosen": 0.1509028524160385,
29
+ "logits/rejected": 0.21932204067707062,
30
+ "logps/chosen": -511.6697998046875,
31
+ "logps/rejected": -560.0117797851562,
32
+ "loss": 2.1678,
33
+ "rewards/accuracies": 0.4236111044883728,
34
+ "rewards/chosen": 0.00018176721641793847,
35
+ "rewards/margins": 0.00014634148101322353,
36
+ "rewards/rejected": 3.5425720852799714e-05,
37
+ "step": 10
38
+ },
39
+ {
40
+ "epoch": 0.02,
41
+ "learning_rate": 8.000000000000001e-07,
42
+ "logits/chosen": 0.2015257328748703,
43
+ "logits/rejected": 0.23093406856060028,
44
+ "logps/chosen": -614.9334106445312,
45
+ "logps/rejected": -580.0980224609375,
46
+ "loss": 2.1481,
47
+ "rewards/accuracies": 0.512499988079071,
48
+ "rewards/chosen": 0.0002653285046108067,
49
+ "rewards/margins": 9.045841579791158e-05,
50
+ "rewards/rejected": 0.00017487008881289512,
51
+ "step": 20
52
+ },
53
+ {
54
+ "epoch": 0.02,
55
+ "learning_rate": 1.2000000000000002e-06,
56
+ "logits/chosen": 0.15271301567554474,
57
+ "logits/rejected": 0.20967456698417664,
58
+ "logps/chosen": -615.6820068359375,
59
+ "logps/rejected": -616.4979248046875,
60
+ "loss": 2.3544,
61
+ "rewards/accuracies": 0.44999998807907104,
62
+ "rewards/chosen": -0.0001579081581439823,
63
+ "rewards/margins": -0.00015658045595046133,
64
+ "rewards/rejected": -1.3277137895784108e-06,
65
+ "step": 30
66
+ },
67
+ {
68
+ "epoch": 0.03,
69
+ "learning_rate": 1.6000000000000001e-06,
70
+ "logits/chosen": 0.16454997658729553,
71
+ "logits/rejected": 0.15700042247772217,
72
+ "logps/chosen": -577.5484619140625,
73
+ "logps/rejected": -601.3185424804688,
74
+ "loss": 2.2042,
75
+ "rewards/accuracies": 0.48750001192092896,
76
+ "rewards/chosen": -0.0004847920499742031,
77
+ "rewards/margins": 7.897525210864842e-05,
78
+ "rewards/rejected": -0.0005637672729790211,
79
+ "step": 40
80
+ },
81
+ {
82
+ "epoch": 0.04,
83
+ "learning_rate": 2.0000000000000003e-06,
84
+ "logits/chosen": 0.15456734597682953,
85
+ "logits/rejected": 0.230211541056633,
86
+ "logps/chosen": -541.5357666015625,
87
+ "logps/rejected": -564.4284057617188,
88
+ "loss": 2.1489,
89
+ "rewards/accuracies": 0.4375,
90
+ "rewards/chosen": -0.0010174433700740337,
91
+ "rewards/margins": -0.00017645664047449827,
92
+ "rewards/rejected": -0.0008409869042225182,
93
+ "step": 50
94
+ },
95
+ {
96
+ "epoch": 0.05,
97
+ "learning_rate": 2.4000000000000003e-06,
98
+ "logits/chosen": 0.13179543614387512,
99
+ "logits/rejected": 0.23396913707256317,
100
+ "logps/chosen": -612.1072387695312,
101
+ "logps/rejected": -589.408203125,
102
+ "loss": 2.1442,
103
+ "rewards/accuracies": 0.5,
104
+ "rewards/chosen": -0.0011288017267361283,
105
+ "rewards/margins": 0.00029289646772667766,
106
+ "rewards/rejected": -0.001421698136255145,
107
+ "step": 60
108
+ },
109
+ {
110
+ "epoch": 0.06,
111
+ "learning_rate": 2.8000000000000003e-06,
112
+ "logits/chosen": 0.206586092710495,
113
+ "logits/rejected": 0.20121100544929504,
114
+ "logps/chosen": -562.5863037109375,
115
+ "logps/rejected": -549.765380859375,
116
+ "loss": 2.0063,
117
+ "rewards/accuracies": 0.5,
118
+ "rewards/chosen": -0.0016541840741410851,
119
+ "rewards/margins": 0.0002594069519545883,
120
+ "rewards/rejected": -0.001913591055199504,
121
+ "step": 70
122
+ },
123
+ {
124
+ "epoch": 0.06,
125
+ "learning_rate": 3.2000000000000003e-06,
126
+ "logits/chosen": 0.15584848821163177,
127
+ "logits/rejected": 0.22724422812461853,
128
+ "logps/chosen": -616.1813354492188,
129
+ "logps/rejected": -597.015625,
130
+ "loss": 2.263,
131
+ "rewards/accuracies": 0.53125,
132
+ "rewards/chosen": -0.0020384504459798336,
133
+ "rewards/margins": 0.0003365904267411679,
134
+ "rewards/rejected": -0.002375040901824832,
135
+ "step": 80
136
+ },
137
+ {
138
+ "epoch": 0.07,
139
+ "learning_rate": 3.6000000000000003e-06,
140
+ "logits/chosen": 0.13548582792282104,
141
+ "logits/rejected": 0.23147103190422058,
142
+ "logps/chosen": -588.42919921875,
143
+ "logps/rejected": -551.1621704101562,
144
+ "loss": 2.2401,
145
+ "rewards/accuracies": 0.612500011920929,
146
+ "rewards/chosen": -0.0012455533724278212,
147
+ "rewards/margins": 0.0012538704322651029,
148
+ "rewards/rejected": -0.002499423688277602,
149
+ "step": 90
150
+ },
151
+ {
152
+ "epoch": 0.08,
153
+ "learning_rate": 4.000000000000001e-06,
154
+ "logits/chosen": 0.15673957765102386,
155
+ "logits/rejected": 0.2734481692314148,
156
+ "logps/chosen": -574.0126342773438,
157
+ "logps/rejected": -630.3812255859375,
158
+ "loss": 2.0863,
159
+ "rewards/accuracies": 0.5874999761581421,
160
+ "rewards/chosen": -0.002293633297085762,
161
+ "rewards/margins": 0.002124248770996928,
162
+ "rewards/rejected": -0.004417881835252047,
163
+ "step": 100
164
+ },
165
+ {
166
+ "epoch": 0.09,
167
+ "learning_rate": 4.4e-06,
168
+ "logits/chosen": 0.15073809027671814,
169
+ "logits/rejected": 0.2886480689048767,
170
+ "logps/chosen": -578.6621704101562,
171
+ "logps/rejected": -523.6320190429688,
172
+ "loss": 2.106,
173
+ "rewards/accuracies": 0.612500011920929,
174
+ "rewards/chosen": -0.0048215631395578384,
175
+ "rewards/margins": 0.0025550012942403555,
176
+ "rewards/rejected": -0.007376563735306263,
177
+ "step": 110
178
+ },
179
+ {
180
+ "epoch": 0.1,
181
+ "learning_rate": 4.800000000000001e-06,
182
+ "logits/chosen": 0.20540550351142883,
183
+ "logits/rejected": 0.22172732651233673,
184
+ "logps/chosen": -656.3192138671875,
185
+ "logps/rejected": -624.5650634765625,
186
+ "loss": 2.1297,
187
+ "rewards/accuracies": 0.5249999761581421,
188
+ "rewards/chosen": -0.010842840187251568,
189
+ "rewards/margins": 0.0032757048029452562,
190
+ "rewards/rejected": -0.01411854475736618,
191
+ "step": 120
192
+ },
193
+ {
194
+ "epoch": 0.1,
195
+ "learning_rate": 4.999756310023261e-06,
196
+ "logits/chosen": 0.13236024975776672,
197
+ "logits/rejected": 0.27400949597358704,
198
+ "logps/chosen": -594.3499145507812,
199
+ "logps/rejected": -589.4105834960938,
200
+ "loss": 2.0564,
201
+ "rewards/accuracies": 0.5562499761581421,
202
+ "rewards/chosen": -0.01253579556941986,
203
+ "rewards/margins": 0.005572236143052578,
204
+ "rewards/rejected": -0.018108030781149864,
205
+ "step": 130
206
+ },
207
+ {
208
+ "epoch": 0.11,
209
+ "learning_rate": 4.997807075247147e-06,
210
+ "logits/chosen": 0.207227423787117,
211
+ "logits/rejected": 0.17817452549934387,
212
+ "logps/chosen": -571.6937255859375,
213
+ "logps/rejected": -593.8927612304688,
214
+ "loss": 2.0839,
215
+ "rewards/accuracies": 0.6312500238418579,
216
+ "rewards/chosen": -0.013347337022423744,
217
+ "rewards/margins": 0.008908352814614773,
218
+ "rewards/rejected": -0.022255688905715942,
219
+ "step": 140
220
+ },
221
+ {
222
+ "epoch": 0.12,
223
+ "learning_rate": 4.993910125649561e-06,
224
+ "logits/chosen": 0.1696740686893463,
225
+ "logits/rejected": 0.18951769173145294,
226
+ "logps/chosen": -596.6219482421875,
227
+ "logps/rejected": -637.0972900390625,
228
+ "loss": 2.2635,
229
+ "rewards/accuracies": 0.6187499761581421,
230
+ "rewards/chosen": -0.022671109065413475,
231
+ "rewards/margins": 0.016697829589247704,
232
+ "rewards/rejected": -0.03936893865466118,
233
+ "step": 150
234
+ },
235
+ {
236
+ "epoch": 0.13,
237
+ "learning_rate": 4.988068499954578e-06,
238
+ "logits/chosen": 0.18265119194984436,
239
+ "logits/rejected": 0.27114540338516235,
240
+ "logps/chosen": -692.2119140625,
241
+ "logps/rejected": -658.11767578125,
242
+ "loss": 2.1075,
243
+ "rewards/accuracies": 0.518750011920929,
244
+ "rewards/chosen": -0.04613715782761574,
245
+ "rewards/margins": 0.007823893800377846,
246
+ "rewards/rejected": -0.053961049765348434,
247
+ "step": 160
248
+ },
249
+ {
250
+ "epoch": 0.14,
251
+ "learning_rate": 4.980286753286196e-06,
252
+ "logits/chosen": 0.23610782623291016,
253
+ "logits/rejected": 0.25577467679977417,
254
+ "logps/chosen": -607.8341674804688,
255
+ "logps/rejected": -622.9971923828125,
256
+ "loss": 2.1709,
257
+ "rewards/accuracies": 0.606249988079071,
258
+ "rewards/chosen": -0.04283341392874718,
259
+ "rewards/margins": 0.013730195350944996,
260
+ "rewards/rejected": -0.056563615798950195,
261
+ "step": 170
262
+ },
263
+ {
264
+ "epoch": 0.14,
265
+ "learning_rate": 4.970570953616383e-06,
266
+ "logits/chosen": 0.2259415090084076,
267
+ "logits/rejected": 0.2490004301071167,
268
+ "logps/chosen": -619.1112060546875,
269
+ "logps/rejected": -711.3251953125,
270
+ "loss": 2.1606,
271
+ "rewards/accuracies": 0.637499988079071,
272
+ "rewards/chosen": -0.04884644225239754,
273
+ "rewards/margins": 0.03266354650259018,
274
+ "rewards/rejected": -0.08150999248027802,
275
+ "step": 180
276
+ },
277
+ {
278
+ "epoch": 0.15,
279
+ "learning_rate": 4.958928677033465e-06,
280
+ "logits/chosen": 0.21490153670310974,
281
+ "logits/rejected": 0.20755355060100555,
282
+ "logps/chosen": -661.9138793945312,
283
+ "logps/rejected": -686.15869140625,
284
+ "loss": 2.1463,
285
+ "rewards/accuracies": 0.625,
286
+ "rewards/chosen": -0.0876893401145935,
287
+ "rewards/margins": 0.05007048696279526,
288
+ "rewards/rejected": -0.13775983452796936,
289
+ "step": 190
290
+ },
291
+ {
292
+ "epoch": 0.16,
293
+ "learning_rate": 4.9453690018345144e-06,
294
+ "logits/chosen": 0.22380857169628143,
295
+ "logits/rejected": 0.34608834981918335,
296
+ "logps/chosen": -859.7734375,
297
+ "logps/rejected": -910.9814453125,
298
+ "loss": 1.8887,
299
+ "rewards/accuracies": 0.5874999761581421,
300
+ "rewards/chosen": -0.2093479186296463,
301
+ "rewards/margins": 0.08500120043754578,
302
+ "rewards/rejected": -0.29434913396835327,
303
+ "step": 200
304
+ },
305
+ {
306
+ "epoch": 0.17,
307
+ "learning_rate": 4.9299025014463665e-06,
308
+ "logits/chosen": 0.2813585102558136,
309
+ "logits/rejected": 0.33269229531288147,
310
+ "logps/chosen": -944.7975463867188,
311
+ "logps/rejected": -1029.7379150390625,
312
+ "loss": 1.9249,
313
+ "rewards/accuracies": 0.574999988079071,
314
+ "rewards/chosen": -0.3255089223384857,
315
+ "rewards/margins": 0.09575793147087097,
316
+ "rewards/rejected": -0.4212668836116791,
317
+ "step": 210
318
+ },
319
+ {
320
+ "epoch": 0.18,
321
+ "learning_rate": 4.912541236180779e-06,
322
+ "logits/chosen": 0.2951585650444031,
323
+ "logits/rejected": 0.49975720047950745,
324
+ "logps/chosen": -904.1650390625,
325
+ "logps/rejected": -1103.083740234375,
326
+ "loss": 1.9609,
327
+ "rewards/accuracies": 0.59375,
328
+ "rewards/chosen": -0.35912585258483887,
329
+ "rewards/margins": 0.16983045637607574,
330
+ "rewards/rejected": -0.5289562940597534,
331
+ "step": 220
332
+ },
333
+ {
334
+ "epoch": 0.18,
335
+ "learning_rate": 4.893298743830168e-06,
336
+ "logits/chosen": 0.4890425205230713,
337
+ "logits/rejected": 0.5553490519523621,
338
+ "logps/chosen": -935.9508666992188,
339
+ "logps/rejected": -1199.338623046875,
340
+ "loss": 2.0485,
341
+ "rewards/accuracies": 0.6187499761581421,
342
+ "rewards/chosen": -0.3711962103843689,
343
+ "rewards/margins": 0.2363450974225998,
344
+ "rewards/rejected": -0.6075413823127747,
345
+ "step": 230
346
+ },
347
+ {
348
+ "epoch": 0.19,
349
+ "learning_rate": 4.8721900291112415e-06,
350
+ "logits/chosen": 0.3669169545173645,
351
+ "logits/rejected": 0.46804532408714294,
352
+ "logps/chosen": -862.8795166015625,
353
+ "logps/rejected": -937.4802856445312,
354
+ "loss": 2.1176,
355
+ "rewards/accuracies": 0.5562499761581421,
356
+ "rewards/chosen": -0.2683151364326477,
357
+ "rewards/margins": 0.10171107202768326,
358
+ "rewards/rejected": -0.37002623081207275,
359
+ "step": 240
360
+ },
361
+ {
362
+ "epoch": 0.2,
363
+ "learning_rate": 4.849231551964771e-06,
364
+ "logits/chosen": 0.28659749031066895,
365
+ "logits/rejected": 0.44282183051109314,
366
+ "logps/chosen": -833.5822143554688,
367
+ "logps/rejected": -910.1467895507812,
368
+ "loss": 1.9678,
369
+ "rewards/accuracies": 0.5249999761581421,
370
+ "rewards/chosen": -0.2521842122077942,
371
+ "rewards/margins": 0.07951636612415314,
372
+ "rewards/rejected": -0.33170050382614136,
373
+ "step": 250
374
+ },
375
+ {
376
+ "epoch": 0.21,
377
+ "learning_rate": 4.824441214720629e-06,
378
+ "logits/chosen": 0.40166163444519043,
379
+ "logits/rejected": 0.44364985823631287,
380
+ "logps/chosen": -788.6900634765625,
381
+ "logps/rejected": -991.3458862304688,
382
+ "loss": 1.7415,
383
+ "rewards/accuracies": 0.59375,
384
+ "rewards/chosen": -0.2276671826839447,
385
+ "rewards/margins": 0.16912813484668732,
386
+ "rewards/rejected": -0.3967953026294708,
387
+ "step": 260
388
+ },
389
+ {
390
+ "epoch": 0.22,
391
+ "learning_rate": 4.7978383481380865e-06,
392
+ "logits/chosen": 0.3444485664367676,
393
+ "logits/rejected": 0.5015489459037781,
394
+ "logps/chosen": -825.1337890625,
395
+ "logps/rejected": -1013.6962890625,
396
+ "loss": 1.8545,
397
+ "rewards/accuracies": 0.606249988079071,
398
+ "rewards/chosen": -0.2368713915348053,
399
+ "rewards/margins": 0.19953957200050354,
400
+ "rewards/rejected": -0.43641090393066406,
401
+ "step": 270
402
+ },
403
+ {
404
+ "epoch": 0.22,
405
+ "learning_rate": 4.769443696332272e-06,
406
+ "logits/chosen": 0.32490837574005127,
407
+ "logits/rejected": 0.4952085018157959,
408
+ "logps/chosen": -748.92578125,
409
+ "logps/rejected": -876.5540161132812,
410
+ "loss": 1.9791,
411
+ "rewards/accuracies": 0.5249999761581421,
412
+ "rewards/chosen": -0.20112459361553192,
413
+ "rewards/margins": 0.12988439202308655,
414
+ "rewards/rejected": -0.3310089707374573,
415
+ "step": 280
416
+ },
417
+ {
418
+ "epoch": 0.23,
419
+ "learning_rate": 4.7392794005985324e-06,
420
+ "logits/chosen": 0.41541141271591187,
421
+ "logits/rejected": 0.5006674528121948,
422
+ "logps/chosen": -831.7371215820312,
423
+ "logps/rejected": -1032.0704345703125,
424
+ "loss": 1.7353,
425
+ "rewards/accuracies": 0.606249988079071,
426
+ "rewards/chosen": -0.24899300932884216,
427
+ "rewards/margins": 0.20179975032806396,
428
+ "rewards/rejected": -0.45079272985458374,
429
+ "step": 290
430
+ },
431
+ {
432
+ "epoch": 0.24,
433
+ "learning_rate": 4.707368982147318e-06,
434
+ "logits/chosen": 0.36991387605667114,
435
+ "logits/rejected": 0.48434096574783325,
436
+ "logps/chosen": -898.0588989257812,
437
+ "logps/rejected": -1055.823486328125,
438
+ "loss": 1.9315,
439
+ "rewards/accuracies": 0.5874999761581421,
440
+ "rewards/chosen": -0.30798205733299255,
441
+ "rewards/margins": 0.15843333303928375,
442
+ "rewards/rejected": -0.4664154052734375,
443
+ "step": 300
444
+ },
445
+ {
446
+ "epoch": 0.25,
447
+ "learning_rate": 4.673737323763048e-06,
448
+ "logits/chosen": 0.47087937593460083,
449
+ "logits/rejected": 0.4668746590614319,
450
+ "logps/chosen": -921.783203125,
451
+ "logps/rejected": -1135.657470703125,
452
+ "loss": 1.9714,
453
+ "rewards/accuracies": 0.59375,
454
+ "rewards/chosen": -0.33973026275634766,
455
+ "rewards/margins": 0.1682974398136139,
456
+ "rewards/rejected": -0.5080276727676392,
457
+ "step": 310
458
+ },
459
+ {
460
+ "epoch": 0.26,
461
+ "learning_rate": 4.638410650401267e-06,
462
+ "logits/chosen": 0.32884734869003296,
463
+ "logits/rejected": 0.5667840242385864,
464
+ "logps/chosen": -793.4259643554688,
465
+ "logps/rejected": -1097.870361328125,
466
+ "loss": 1.6468,
467
+ "rewards/accuracies": 0.6499999761581421,
468
+ "rewards/chosen": -0.22830717265605927,
469
+ "rewards/margins": 0.2654227614402771,
470
+ "rewards/rejected": -0.49372991919517517,
471
+ "step": 320
472
+ },
473
+ {
474
+ "epoch": 0.26,
475
+ "learning_rate": 4.601416508739211e-06,
476
+ "logits/chosen": 0.3828834891319275,
477
+ "logits/rejected": 0.5297464728355408,
478
+ "logps/chosen": -948.02197265625,
479
+ "logps/rejected": -1136.756103515625,
480
+ "loss": 1.8833,
481
+ "rewards/accuracies": 0.606249988079071,
482
+ "rewards/chosen": -0.3334728181362152,
483
+ "rewards/margins": 0.1820012331008911,
484
+ "rewards/rejected": -0.5154740810394287,
485
+ "step": 330
486
+ },
487
+ {
488
+ "epoch": 0.27,
489
+ "learning_rate": 4.562783745695738e-06,
490
+ "logits/chosen": 0.41931548714637756,
491
+ "logits/rejected": 0.5291355848312378,
492
+ "logps/chosen": -862.7106323242188,
493
+ "logps/rejected": -1053.8846435546875,
494
+ "loss": 1.7946,
495
+ "rewards/accuracies": 0.6312500238418579,
496
+ "rewards/chosen": -0.2459833323955536,
497
+ "rewards/margins": 0.19779649376869202,
498
+ "rewards/rejected": -0.443779855966568,
499
+ "step": 340
500
+ },
501
+ {
502
+ "epoch": 0.28,
503
+ "learning_rate": 4.522542485937369e-06,
504
+ "logits/chosen": 0.3689717650413513,
505
+ "logits/rejected": 0.4881824553012848,
506
+ "logps/chosen": -875.2266845703125,
507
+ "logps/rejected": -1156.237060546875,
508
+ "loss": 1.6268,
509
+ "rewards/accuracies": 0.6875,
510
+ "rewards/chosen": -0.257233202457428,
511
+ "rewards/margins": 0.27138087153434753,
512
+ "rewards/rejected": -0.5286141037940979,
513
+ "step": 350
514
+ },
515
+ {
516
+ "epoch": 0.29,
517
+ "learning_rate": 4.4807241083879774e-06,
518
+ "logits/chosen": 0.4062356948852539,
519
+ "logits/rejected": 0.4961213171482086,
520
+ "logps/chosen": -890.9068603515625,
521
+ "logps/rejected": -1214.0831298828125,
522
+ "loss": 1.6478,
523
+ "rewards/accuracies": 0.643750011920929,
524
+ "rewards/chosen": -0.3261027932167053,
525
+ "rewards/margins": 0.30797919631004333,
526
+ "rewards/rejected": -0.6340819597244263,
527
+ "step": 360
528
+ },
529
+ {
530
+ "epoch": 0.3,
531
+ "learning_rate": 4.437361221760449e-06,
532
+ "logits/chosen": 0.3728734850883484,
533
+ "logits/rejected": 0.39377298951148987,
534
+ "logps/chosen": -995.19482421875,
535
+ "logps/rejected": -1160.993408203125,
536
+ "loss": 2.014,
537
+ "rewards/accuracies": 0.5625,
538
+ "rewards/chosen": -0.40234214067459106,
539
+ "rewards/margins": 0.1731390357017517,
540
+ "rewards/rejected": -0.5754811763763428,
541
+ "step": 370
542
+ },
543
+ {
544
+ "epoch": 0.3,
545
+ "learning_rate": 4.3924876391293915e-06,
546
+ "logits/chosen": 0.357779324054718,
547
+ "logits/rejected": 0.48311319947242737,
548
+ "logps/chosen": -854.5018310546875,
549
+ "logps/rejected": -1101.8800048828125,
550
+ "loss": 1.7634,
551
+ "rewards/accuracies": 0.643750011920929,
552
+ "rewards/chosen": -0.2472849190235138,
553
+ "rewards/margins": 0.23356270790100098,
554
+ "rewards/rejected": -0.48084765672683716,
555
+ "step": 380
556
+ },
557
+ {
558
+ "epoch": 0.31,
559
+ "learning_rate": 4.346138351564711e-06,
560
+ "logits/chosen": 0.32345646619796753,
561
+ "logits/rejected": 0.4409434199333191,
562
+ "logps/chosen": -821.7467651367188,
563
+ "logps/rejected": -1076.5284423828125,
564
+ "loss": 1.5989,
565
+ "rewards/accuracies": 0.706250011920929,
566
+ "rewards/chosen": -0.2140711098909378,
567
+ "rewards/margins": 0.2635394036769867,
568
+ "rewards/rejected": -0.4776105284690857,
569
+ "step": 390
570
+ },
571
+ {
572
+ "epoch": 0.32,
573
+ "learning_rate": 4.2983495008466285e-06,
574
+ "logits/chosen": 0.36422181129455566,
575
+ "logits/rejected": 0.45148295164108276,
576
+ "logps/chosen": -922.22705078125,
577
+ "logps/rejected": -1089.808837890625,
578
+ "loss": 1.7327,
579
+ "rewards/accuracies": 0.606249988079071,
580
+ "rewards/chosen": -0.3074890375137329,
581
+ "rewards/margins": 0.18984538316726685,
582
+ "rewards/rejected": -0.49733442068099976,
583
+ "step": 400
584
+ },
585
+ {
586
+ "epoch": 0.33,
587
+ "learning_rate": 4.249158351283414e-06,
588
+ "logits/chosen": 0.37963396310806274,
589
+ "logits/rejected": 0.43684640526771545,
590
+ "logps/chosen": -816.420166015625,
591
+ "logps/rejected": -1026.3609619140625,
592
+ "loss": 1.9326,
593
+ "rewards/accuracies": 0.5625,
594
+ "rewards/chosen": -0.26014500856399536,
595
+ "rewards/margins": 0.17839236557483673,
596
+ "rewards/rejected": -0.4385373592376709,
597
+ "step": 410
598
+ },
599
+ {
600
+ "epoch": 0.34,
601
+ "learning_rate": 4.198603260653792e-06,
602
+ "logits/chosen": 0.3479335308074951,
603
+ "logits/rejected": 0.5144117474555969,
604
+ "logps/chosen": -808.4892578125,
605
+ "logps/rejected": -1022.3094482421875,
606
+ "loss": 1.7198,
607
+ "rewards/accuracies": 0.59375,
608
+ "rewards/chosen": -0.24907036125659943,
609
+ "rewards/margins": 0.21405251324176788,
610
+ "rewards/rejected": -0.4631228446960449,
611
+ "step": 420
612
+ },
613
+ {
614
+ "epoch": 0.34,
615
+ "learning_rate": 4.146723650296701e-06,
616
+ "logits/chosen": 0.375163733959198,
617
+ "logits/rejected": 0.5701647996902466,
618
+ "logps/chosen": -894.6932373046875,
619
+ "logps/rejected": -1191.1158447265625,
620
+ "loss": 1.7984,
621
+ "rewards/accuracies": 0.625,
622
+ "rewards/chosen": -0.28425461053848267,
623
+ "rewards/margins": 0.2671293616294861,
624
+ "rewards/rejected": -0.5513839721679688,
625
+ "step": 430
626
+ },
627
+ {
628
+ "epoch": 0.35,
629
+ "learning_rate": 4.093559974371725e-06,
630
+ "logits/chosen": 0.39743003249168396,
631
+ "logits/rejected": 0.499051570892334,
632
+ "logps/chosen": -769.8292236328125,
633
+ "logps/rejected": -1028.2760009765625,
634
+ "loss": 1.8098,
635
+ "rewards/accuracies": 0.612500011920929,
636
+ "rewards/chosen": -0.2524164319038391,
637
+ "rewards/margins": 0.2162034809589386,
638
+ "rewards/rejected": -0.4686198830604553,
639
+ "step": 440
640
+ },
641
+ {
642
+ "epoch": 0.36,
643
+ "learning_rate": 4.039153688314146e-06,
644
+ "logits/chosen": 0.37791648507118225,
645
+ "logits/rejected": 0.46076732873916626,
646
+ "logps/chosen": -876.4410400390625,
647
+ "logps/rejected": -1076.7730712890625,
648
+ "loss": 1.824,
649
+ "rewards/accuracies": 0.6812499761581421,
650
+ "rewards/chosen": -0.2770997881889343,
651
+ "rewards/margins": 0.2473537176847458,
652
+ "rewards/rejected": -0.5244535207748413,
653
+ "step": 450
654
+ },
655
+ {
656
+ "epoch": 0.37,
657
+ "learning_rate": 3.983547216509254e-06,
658
+ "logits/chosen": 0.36932164430618286,
659
+ "logits/rejected": 0.5779580473899841,
660
+ "logps/chosen": -891.6867065429688,
661
+ "logps/rejected": -1252.7017822265625,
662
+ "loss": 1.5803,
663
+ "rewards/accuracies": 0.6812499761581421,
664
+ "rewards/chosen": -0.27173933386802673,
665
+ "rewards/margins": 0.33879798650741577,
666
+ "rewards/rejected": -0.6105372905731201,
667
+ "step": 460
668
+ },
669
+ {
670
+ "epoch": 0.38,
671
+ "learning_rate": 3.92678391921108e-06,
672
+ "logits/chosen": 0.376406729221344,
673
+ "logits/rejected": 0.46922606229782104,
674
+ "logps/chosen": -881.6818237304688,
675
+ "logps/rejected": -1075.5714111328125,
676
+ "loss": 1.9328,
677
+ "rewards/accuracies": 0.550000011920929,
678
+ "rewards/chosen": -0.3042740225791931,
679
+ "rewards/margins": 0.21284916996955872,
680
+ "rewards/rejected": -0.5171232223510742,
681
+ "step": 470
682
+ },
683
+ {
684
+ "epoch": 0.38,
685
+ "learning_rate": 3.868908058731376e-06,
686
+ "logits/chosen": 0.3659452497959137,
687
+ "logits/rejected": 0.47788533568382263,
688
+ "logps/chosen": -914.9581909179688,
689
+ "logps/rejected": -1192.8797607421875,
690
+ "loss": 1.8254,
691
+ "rewards/accuracies": 0.6000000238418579,
692
+ "rewards/chosen": -0.333695650100708,
693
+ "rewards/margins": 0.27644091844558716,
694
+ "rewards/rejected": -0.6101365685462952,
695
+ "step": 480
696
+ },
697
+ {
698
+ "epoch": 0.39,
699
+ "learning_rate": 3.8099647649251984e-06,
700
+ "logits/chosen": 0.3602338135242462,
701
+ "logits/rejected": 0.4133017063140869,
702
+ "logps/chosen": -875.7825927734375,
703
+ "logps/rejected": -1087.2510986328125,
704
+ "loss": 1.8227,
705
+ "rewards/accuracies": 0.59375,
706
+ "rewards/chosen": -0.3075736165046692,
707
+ "rewards/margins": 0.24024859070777893,
708
+ "rewards/rejected": -0.547822117805481,
709
+ "step": 490
710
+ },
711
+ {
712
+ "epoch": 0.4,
713
+ "learning_rate": 3.7500000000000005e-06,
714
+ "logits/chosen": 0.3213595449924469,
715
+ "logits/rejected": 0.4600197374820709,
716
+ "logps/chosen": -851.552734375,
717
+ "logps/rejected": -1143.1978759765625,
718
+ "loss": 1.7429,
719
+ "rewards/accuracies": 0.643750011920929,
720
+ "rewards/chosen": -0.2909749150276184,
721
+ "rewards/margins": 0.2939184010028839,
722
+ "rewards/rejected": -0.5848933458328247,
723
+ "step": 500
724
+ },
725
+ {
726
+ "epoch": 0.41,
727
+ "learning_rate": 3.689060522675689e-06,
728
+ "logits/chosen": 0.3785388469696045,
729
+ "logits/rejected": 0.37024298310279846,
730
+ "logps/chosen": -865.6300659179688,
731
+ "logps/rejected": -1103.9659423828125,
732
+ "loss": 1.7996,
733
+ "rewards/accuracies": 0.643750011920929,
734
+ "rewards/chosen": -0.29555124044418335,
735
+ "rewards/margins": 0.2511122524738312,
736
+ "rewards/rejected": -0.5466634631156921,
737
+ "step": 510
738
+ },
739
+ {
740
+ "epoch": 0.42,
741
+ "learning_rate": 3.627193851723577e-06,
742
+ "logits/chosen": 0.35357269644737244,
743
+ "logits/rejected": 0.44010990858078003,
744
+ "logps/chosen": -922.1881713867188,
745
+ "logps/rejected": -1222.6695556640625,
746
+ "loss": 1.8295,
747
+ "rewards/accuracies": 0.606249988079071,
748
+ "rewards/chosen": -0.34208646416664124,
749
+ "rewards/margins": 0.2784285545349121,
750
+ "rewards/rejected": -0.620514988899231,
751
+ "step": 520
752
+ },
753
+ {
754
+ "epoch": 0.42,
755
+ "learning_rate": 3.564448228912682e-06,
756
+ "logits/chosen": 0.3053657114505768,
757
+ "logits/rejected": 0.4411580562591553,
758
+ "logps/chosen": -964.7332763671875,
759
+ "logps/rejected": -1235.441650390625,
760
+ "loss": 1.7541,
761
+ "rewards/accuracies": 0.6625000238418579,
762
+ "rewards/chosen": -0.3641963601112366,
763
+ "rewards/margins": 0.26334619522094727,
764
+ "rewards/rejected": -0.6275425553321838,
765
+ "step": 530
766
+ },
767
+ {
768
+ "epoch": 0.43,
769
+ "learning_rate": 3.5008725813922383e-06,
770
+ "logits/chosen": 0.3650105595588684,
771
+ "logits/rejected": 0.35269036889076233,
772
+ "logps/chosen": -866.552734375,
773
+ "logps/rejected": -1229.647216796875,
774
+ "loss": 1.6454,
775
+ "rewards/accuracies": 0.6937500238418579,
776
+ "rewards/chosen": -0.29515185952186584,
777
+ "rewards/margins": 0.3350960910320282,
778
+ "rewards/rejected": -0.6302478909492493,
779
+ "step": 540
780
+ },
781
+ {
782
+ "epoch": 0.44,
783
+ "learning_rate": 3.436516483539781e-06,
784
+ "logits/chosen": 0.30932605266571045,
785
+ "logits/rejected": 0.382567435503006,
786
+ "logps/chosen": -870.6939697265625,
787
+ "logps/rejected": -1298.612060546875,
788
+ "loss": 1.616,
789
+ "rewards/accuracies": 0.668749988079071,
790
+ "rewards/chosen": -0.30489498376846313,
791
+ "rewards/margins": 0.3668510317802429,
792
+ "rewards/rejected": -0.6717459559440613,
793
+ "step": 550
794
+ },
795
+ {
796
+ "epoch": 0.45,
797
+ "learning_rate": 3.3714301183045382e-06,
798
+ "logits/chosen": 0.38849467039108276,
799
+ "logits/rejected": 0.4396263659000397,
800
+ "logps/chosen": -902.1296997070312,
801
+ "logps/rejected": -1177.967529296875,
802
+ "loss": 1.9176,
803
+ "rewards/accuracies": 0.59375,
804
+ "rewards/chosen": -0.3350718319416046,
805
+ "rewards/margins": 0.2506067156791687,
806
+ "rewards/rejected": -0.5856784582138062,
807
+ "step": 560
808
+ },
809
+ {
810
+ "epoch": 0.46,
811
+ "learning_rate": 3.3056642380762783e-06,
812
+ "logits/chosen": 0.29538553953170776,
813
+ "logits/rejected": 0.43000030517578125,
814
+ "logps/chosen": -952.9089965820312,
815
+ "logps/rejected": -1251.63525390625,
816
+ "loss": 1.6891,
817
+ "rewards/accuracies": 0.65625,
818
+ "rewards/chosen": -0.36573296785354614,
819
+ "rewards/margins": 0.29417911171913147,
820
+ "rewards/rejected": -0.6599120497703552,
821
+ "step": 570
822
+ },
823
+ {
824
+ "epoch": 0.46,
825
+ "learning_rate": 3.2392701251101172e-06,
826
+ "logits/chosen": 0.26259809732437134,
827
+ "logits/rejected": 0.4230431914329529,
828
+ "logps/chosen": -981.2227783203125,
829
+ "logps/rejected": -1164.7890625,
830
+ "loss": 1.8962,
831
+ "rewards/accuracies": 0.612500011920929,
832
+ "rewards/chosen": -0.3732236325740814,
833
+ "rewards/margins": 0.22464247047901154,
834
+ "rewards/rejected": -0.5978660583496094,
835
+ "step": 580
836
+ },
837
+ {
838
+ "epoch": 0.47,
839
+ "learning_rate": 3.1722995515381644e-06,
840
+ "logits/chosen": 0.3150332570075989,
841
+ "logits/rejected": 0.4163808226585388,
842
+ "logps/chosen": -904.3822021484375,
843
+ "logps/rejected": -1273.4002685546875,
844
+ "loss": 1.7318,
845
+ "rewards/accuracies": 0.6625000238418579,
846
+ "rewards/chosen": -0.3179934620857239,
847
+ "rewards/margins": 0.31932058930397034,
848
+ "rewards/rejected": -0.6373140215873718,
849
+ "step": 590
850
+ },
851
+ {
852
+ "epoch": 0.48,
853
+ "learning_rate": 3.1048047389991693e-06,
854
+ "logits/chosen": 0.2711288630962372,
855
+ "logits/rejected": 0.38649213314056396,
856
+ "logps/chosen": -961.5247802734375,
857
+ "logps/rejected": -1165.8568115234375,
858
+ "loss": 1.8103,
859
+ "rewards/accuracies": 0.606249988079071,
860
+ "rewards/chosen": -0.35747116804122925,
861
+ "rewards/margins": 0.2142975628376007,
862
+ "rewards/rejected": -0.5717687010765076,
863
+ "step": 600
864
+ },
865
+ {
866
+ "epoch": 0.49,
867
+ "learning_rate": 3.0368383179176584e-06,
868
+ "logits/chosen": 0.31565985083580017,
869
+ "logits/rejected": 0.3596263825893402,
870
+ "logps/chosen": -1067.418701171875,
871
+ "logps/rejected": -1284.5201416015625,
872
+ "loss": 1.7442,
873
+ "rewards/accuracies": 0.625,
874
+ "rewards/chosen": -0.42869243025779724,
875
+ "rewards/margins": 0.23898033797740936,
876
+ "rewards/rejected": -0.6676727533340454,
877
+ "step": 610
878
+ },
879
+ {
880
+ "epoch": 0.5,
881
+ "learning_rate": 2.9684532864643123e-06,
882
+ "logits/chosen": 0.2723945081233978,
883
+ "logits/rejected": 0.42879945039749146,
884
+ "logps/chosen": -907.2979736328125,
885
+ "logps/rejected": -1193.40625,
886
+ "loss": 1.7749,
887
+ "rewards/accuracies": 0.6312500238418579,
888
+ "rewards/chosen": -0.37917786836624146,
889
+ "rewards/margins": 0.25062739849090576,
890
+ "rewards/rejected": -0.6298052668571472,
891
+ "step": 620
892
+ },
893
+ {
894
+ "epoch": 0.5,
895
+ "learning_rate": 2.8997029692295875e-06,
896
+ "logits/chosen": 0.2836493253707886,
897
+ "logits/rejected": 0.36888745427131653,
898
+ "logps/chosen": -958.9281005859375,
899
+ "logps/rejected": -1170.929443359375,
900
+ "loss": 1.7337,
901
+ "rewards/accuracies": 0.6187499761581421,
902
+ "rewards/chosen": -0.36827540397644043,
903
+ "rewards/margins": 0.24554581940174103,
904
+ "rewards/rejected": -0.613821268081665,
905
+ "step": 630
906
+ },
907
+ {
908
+ "epoch": 0.51,
909
+ "learning_rate": 2.8306409756428067e-06,
910
+ "logits/chosen": 0.34693920612335205,
911
+ "logits/rejected": 0.33596283197402954,
912
+ "logps/chosen": -884.5350341796875,
913
+ "logps/rejected": -1140.420654296875,
914
+ "loss": 1.7815,
915
+ "rewards/accuracies": 0.643750011920929,
916
+ "rewards/chosen": -0.32642626762390137,
917
+ "rewards/margins": 0.22602030634880066,
918
+ "rewards/rejected": -0.5524466633796692,
919
+ "step": 640
920
+ },
921
+ {
922
+ "epoch": 0.52,
923
+ "learning_rate": 2.761321158169134e-06,
924
+ "logits/chosen": 0.29072973132133484,
925
+ "logits/rejected": 0.4021405279636383,
926
+ "logps/chosen": -920.5359497070312,
927
+ "logps/rejected": -1172.0166015625,
928
+ "loss": 1.7447,
929
+ "rewards/accuracies": 0.612500011920929,
930
+ "rewards/chosen": -0.31148773431777954,
931
+ "rewards/margins": 0.2528636157512665,
932
+ "rewards/rejected": -0.5643514394760132,
933
+ "step": 650
934
+ },
935
+ {
936
+ "epoch": 0.53,
937
+ "learning_rate": 2.6917975703170466e-06,
938
+ "logits/chosen": 0.30165717005729675,
939
+ "logits/rejected": 0.41581812500953674,
940
+ "logps/chosen": -998.71484375,
941
+ "logps/rejected": -1180.068115234375,
942
+ "loss": 1.7292,
943
+ "rewards/accuracies": 0.65625,
944
+ "rewards/chosen": -0.3652830719947815,
945
+ "rewards/margins": 0.22447367012500763,
946
+ "rewards/rejected": -0.5897567868232727,
947
+ "step": 660
948
+ },
949
+ {
950
+ "epoch": 0.54,
951
+ "learning_rate": 2.6221244244890336e-06,
952
+ "logits/chosen": 0.2630220949649811,
953
+ "logits/rejected": 0.38998615741729736,
954
+ "logps/chosen": -850.4860229492188,
955
+ "logps/rejected": -1299.91015625,
956
+ "loss": 1.4881,
957
+ "rewards/accuracies": 0.6875,
958
+ "rewards/chosen": -0.32001981139183044,
959
+ "rewards/margins": 0.3484157621860504,
960
+ "rewards/rejected": -0.6684354543685913,
961
+ "step": 670
962
+ },
963
+ {
964
+ "epoch": 0.54,
965
+ "learning_rate": 2.5523560497083927e-06,
966
+ "logits/chosen": 0.28038591146469116,
967
+ "logits/rejected": 0.4210067391395569,
968
+ "logps/chosen": -989.4406127929688,
969
+ "logps/rejected": -1252.375732421875,
970
+ "loss": 1.686,
971
+ "rewards/accuracies": 0.6625000238418579,
972
+ "rewards/chosen": -0.37420469522476196,
973
+ "rewards/margins": 0.259860098361969,
974
+ "rewards/rejected": -0.634064793586731,
975
+ "step": 680
976
+ },
977
+ {
978
+ "epoch": 0.55,
979
+ "learning_rate": 2.482546849255096e-06,
980
+ "logits/chosen": 0.3020229935646057,
981
+ "logits/rejected": 0.3828849792480469,
982
+ "logps/chosen": -956.5652465820312,
983
+ "logps/rejected": -1176.745849609375,
984
+ "loss": 1.6468,
985
+ "rewards/accuracies": 0.6187499761581421,
986
+ "rewards/chosen": -0.3711964190006256,
987
+ "rewards/margins": 0.2462393045425415,
988
+ "rewards/rejected": -0.6174356937408447,
989
+ "step": 690
990
+ },
991
+ {
992
+ "epoch": 0.56,
993
+ "learning_rate": 2.4127512582437486e-06,
994
+ "logits/chosen": 0.3271457552909851,
995
+ "logits/rejected": 0.3496400713920593,
996
+ "logps/chosen": -860.8416137695312,
997
+ "logps/rejected": -1315.4486083984375,
998
+ "loss": 1.4554,
999
+ "rewards/accuracies": 0.7124999761581421,
1000
+ "rewards/chosen": -0.3369373083114624,
1001
+ "rewards/margins": 0.36229121685028076,
1002
+ "rewards/rejected": -0.6992285251617432,
1003
+ "step": 700
1004
+ },
1005
+ {
1006
+ "epoch": 0.57,
1007
+ "learning_rate": 2.3430237011767166e-06,
1008
+ "logits/chosen": 0.32887858152389526,
1009
+ "logits/rejected": 0.38950568437576294,
1010
+ "logps/chosen": -962.7874755859375,
1011
+ "logps/rejected": -1309.929443359375,
1012
+ "loss": 1.6678,
1013
+ "rewards/accuracies": 0.6937500238418579,
1014
+ "rewards/chosen": -0.3625487685203552,
1015
+ "rewards/margins": 0.3222491443157196,
1016
+ "rewards/rejected": -0.6847978830337524,
1017
+ "step": 710
1018
+ },
1019
+ {
1020
+ "epoch": 0.58,
1021
+ "learning_rate": 2.2734185495055503e-06,
1022
+ "logits/chosen": 0.21860530972480774,
1023
+ "logits/rejected": 0.3752947151660919,
1024
+ "logps/chosen": -1050.0916748046875,
1025
+ "logps/rejected": -1280.526123046875,
1026
+ "loss": 1.9756,
1027
+ "rewards/accuracies": 0.6499999761581421,
1028
+ "rewards/chosen": -0.42955031991004944,
1029
+ "rewards/margins": 0.23833701014518738,
1030
+ "rewards/rejected": -0.6678873896598816,
1031
+ "step": 720
1032
+ },
1033
+ {
1034
+ "epoch": 0.58,
1035
+ "learning_rate": 2.2039900792337477e-06,
1036
+ "logits/chosen": 0.3153984248638153,
1037
+ "logits/rejected": 0.37083715200424194,
1038
+ "logps/chosen": -918.3059692382812,
1039
+ "logps/rejected": -1242.553466796875,
1040
+ "loss": 1.8289,
1041
+ "rewards/accuracies": 0.6625000238418579,
1042
+ "rewards/chosen": -0.34617164731025696,
1043
+ "rewards/margins": 0.2804465591907501,
1044
+ "rewards/rejected": -0.6266182065010071,
1045
+ "step": 730
1046
+ },
1047
+ {
1048
+ "epoch": 0.59,
1049
+ "learning_rate": 2.134792428593971e-06,
1050
+ "logits/chosen": 0.25939470529556274,
1051
+ "logits/rejected": 0.31975841522216797,
1052
+ "logps/chosen": -814.0809326171875,
1053
+ "logps/rejected": -1025.2806396484375,
1054
+ "loss": 1.8756,
1055
+ "rewards/accuracies": 0.6000000238418579,
1056
+ "rewards/chosen": -0.257251501083374,
1057
+ "rewards/margins": 0.21475358307361603,
1058
+ "rewards/rejected": -0.47200506925582886,
1059
+ "step": 740
1060
+ },
1061
+ {
1062
+ "epoch": 0.6,
1063
+ "learning_rate": 2.0658795558326745e-06,
1064
+ "logits/chosen": 0.24114593863487244,
1065
+ "logits/rejected": 0.3729786276817322,
1066
+ "logps/chosen": -844.0042724609375,
1067
+ "logps/rejected": -1063.400634765625,
1068
+ "loss": 1.701,
1069
+ "rewards/accuracies": 0.6625000238418579,
1070
+ "rewards/chosen": -0.2636208236217499,
1071
+ "rewards/margins": 0.22609353065490723,
1072
+ "rewards/rejected": -0.4897143840789795,
1073
+ "step": 750
1074
+ },
1075
+ {
1076
+ "epoch": 0.61,
1077
+ "learning_rate": 1.997305197135089e-06,
1078
+ "logits/chosen": 0.26230183243751526,
1079
+ "logits/rejected": 0.4085041582584381,
1080
+ "logps/chosen": -874.07568359375,
1081
+ "logps/rejected": -1172.9375,
1082
+ "loss": 1.4942,
1083
+ "rewards/accuracies": 0.6812499761581421,
1084
+ "rewards/chosen": -0.2855741083621979,
1085
+ "rewards/margins": 0.28826937079429626,
1086
+ "rewards/rejected": -0.5738434791564941,
1087
+ "step": 760
1088
+ },
1089
+ {
1090
+ "epoch": 0.62,
1091
+ "learning_rate": 1.9291228247233607e-06,
1092
+ "logits/chosen": 0.30484798550605774,
1093
+ "logits/rejected": 0.4110705256462097,
1094
+ "logps/chosen": -969.3850708007812,
1095
+ "logps/rejected": -1283.5828857421875,
1096
+ "loss": 1.8138,
1097
+ "rewards/accuracies": 0.625,
1098
+ "rewards/chosen": -0.34823209047317505,
1099
+ "rewards/margins": 0.2978925108909607,
1100
+ "rewards/rejected": -0.6461246013641357,
1101
+ "step": 770
1102
+ },
1103
+ {
1104
+ "epoch": 0.62,
1105
+ "learning_rate": 1.8613856051605242e-06,
1106
+ "logits/chosen": 0.26855385303497314,
1107
+ "logits/rejected": 0.42292389273643494,
1108
+ "logps/chosen": -959.6076049804688,
1109
+ "logps/rejected": -1120.3887939453125,
1110
+ "loss": 1.9167,
1111
+ "rewards/accuracies": 0.59375,
1112
+ "rewards/chosen": -0.35128024220466614,
1113
+ "rewards/margins": 0.20051094889640808,
1114
+ "rewards/rejected": -0.5517911911010742,
1115
+ "step": 780
1116
+ },
1117
+ {
1118
+ "epoch": 0.63,
1119
+ "learning_rate": 1.7941463578928088e-06,
1120
+ "logits/chosen": 0.3370608687400818,
1121
+ "logits/rejected": 0.38068529963493347,
1122
+ "logps/chosen": -885.8193359375,
1123
+ "logps/rejected": -1103.631591796875,
1124
+ "loss": 1.9313,
1125
+ "rewards/accuracies": 0.543749988079071,
1126
+ "rewards/chosen": -0.3268662095069885,
1127
+ "rewards/margins": 0.2425314486026764,
1128
+ "rewards/rejected": -0.5693976283073425,
1129
+ "step": 790
1130
+ },
1131
+ {
1132
+ "epoch": 0.64,
1133
+ "learning_rate": 1.7274575140626318e-06,
1134
+ "logits/chosen": 0.26554322242736816,
1135
+ "logits/rejected": 0.3851606249809265,
1136
+ "logps/chosen": -990.44140625,
1137
+ "logps/rejected": -1205.0343017578125,
1138
+ "loss": 1.7604,
1139
+ "rewards/accuracies": 0.612500011920929,
1140
+ "rewards/chosen": -0.35415148735046387,
1141
+ "rewards/margins": 0.21467110514640808,
1142
+ "rewards/rejected": -0.5688225626945496,
1143
+ "step": 800
1144
+ },
1145
+ {
1146
+ "epoch": 0.65,
1147
+ "learning_rate": 1.661371075624363e-06,
1148
+ "logits/chosen": 0.2818129062652588,
1149
+ "logits/rejected": 0.3556861877441406,
1150
+ "logps/chosen": -850.845703125,
1151
+ "logps/rejected": -1055.9957275390625,
1152
+ "loss": 1.7347,
1153
+ "rewards/accuracies": 0.625,
1154
+ "rewards/chosen": -0.30765876173973083,
1155
+ "rewards/margins": 0.22872476279735565,
1156
+ "rewards/rejected": -0.5363835096359253,
1157
+ "step": 810
1158
+ },
1159
+ {
1160
+ "epoch": 0.66,
1161
+ "learning_rate": 1.5959385747947697e-06,
1162
+ "logits/chosen": 0.2711453139781952,
1163
+ "logits/rejected": 0.38543039560317993,
1164
+ "logps/chosen": -874.6898193359375,
1165
+ "logps/rejected": -1049.7301025390625,
1166
+ "loss": 1.839,
1167
+ "rewards/accuracies": 0.612500011920929,
1168
+ "rewards/chosen": -0.2933613657951355,
1169
+ "rewards/margins": 0.20129230618476868,
1170
+ "rewards/rejected": -0.49465370178222656,
1171
+ "step": 820
1172
+ },
1173
+ {
1174
+ "epoch": 0.66,
1175
+ "learning_rate": 1.5312110338697427e-06,
1176
+ "logits/chosen": 0.3340443968772888,
1177
+ "logits/rejected": 0.3357875943183899,
1178
+ "logps/chosen": -926.2916870117188,
1179
+ "logps/rejected": -1192.5069580078125,
1180
+ "loss": 1.537,
1181
+ "rewards/accuracies": 0.6312500238418579,
1182
+ "rewards/chosen": -0.32535651326179504,
1183
+ "rewards/margins": 0.2853156626224518,
1184
+ "rewards/rejected": -0.6106721758842468,
1185
+ "step": 830
1186
+ },
1187
+ {
1188
+ "epoch": 0.67,
1189
+ "learning_rate": 1.467238925438646e-06,
1190
+ "logits/chosen": 0.25067275762557983,
1191
+ "logits/rejected": 0.36189913749694824,
1192
+ "logps/chosen": -882.9215698242188,
1193
+ "logps/rejected": -1149.0230712890625,
1194
+ "loss": 1.7236,
1195
+ "rewards/accuracies": 0.574999988079071,
1196
+ "rewards/chosen": -0.32856041193008423,
1197
+ "rewards/margins": 0.2509750723838806,
1198
+ "rewards/rejected": -0.5795354843139648,
1199
+ "step": 840
1200
+ },
1201
+ {
1202
+ "epoch": 0.68,
1203
+ "learning_rate": 1.4040721330273063e-06,
1204
+ "logits/chosen": 0.23793812096118927,
1205
+ "logits/rejected": 0.3621876537799835,
1206
+ "logps/chosen": -858.2410278320312,
1207
+ "logps/rejected": -1095.5906982421875,
1208
+ "loss": 1.6914,
1209
+ "rewards/accuracies": 0.643750011920929,
1210
+ "rewards/chosen": -0.2868640124797821,
1211
+ "rewards/margins": 0.2690005302429199,
1212
+ "rewards/rejected": -0.5558645129203796,
1213
+ "step": 850
1214
+ },
1215
+ {
1216
+ "epoch": 0.69,
1217
+ "learning_rate": 1.3417599122003464e-06,
1218
+ "logits/chosen": 0.34424689412117004,
1219
+ "logits/rejected": 0.3806430697441101,
1220
+ "logps/chosen": -927.8732299804688,
1221
+ "logps/rejected": -1146.036865234375,
1222
+ "loss": 1.7111,
1223
+ "rewards/accuracies": 0.6187499761581421,
1224
+ "rewards/chosen": -0.30680620670318604,
1225
+ "rewards/margins": 0.2561938464641571,
1226
+ "rewards/rejected": -0.5630000829696655,
1227
+ "step": 860
1228
+ },
1229
+ {
1230
+ "epoch": 0.7,
1231
+ "learning_rate": 1.280350852153168e-06,
1232
+ "logits/chosen": 0.3122938275337219,
1233
+ "logits/rejected": 0.2692076861858368,
1234
+ "logps/chosen": -810.717529296875,
1235
+ "logps/rejected": -1086.6156005859375,
1236
+ "loss": 1.7111,
1237
+ "rewards/accuracies": 0.643750011920929,
1238
+ "rewards/chosen": -0.28373444080352783,
1239
+ "rewards/margins": 0.24869227409362793,
1240
+ "rewards/rejected": -0.532426655292511,
1241
+ "step": 870
1242
+ },
1243
+ {
1244
+ "epoch": 0.7,
1245
+ "learning_rate": 1.2198928378235717e-06,
1246
+ "logits/chosen": 0.261820912361145,
1247
+ "logits/rejected": 0.49146905541419983,
1248
+ "logps/chosen": -936.8932495117188,
1249
+ "logps/rejected": -1308.829833984375,
1250
+ "loss": 1.4018,
1251
+ "rewards/accuracies": 0.6875,
1252
+ "rewards/chosen": -0.31694522500038147,
1253
+ "rewards/margins": 0.35484033823013306,
1254
+ "rewards/rejected": -0.6717855334281921,
1255
+ "step": 880
1256
+ },
1257
+ {
1258
+ "epoch": 0.71,
1259
+ "learning_rate": 1.160433012552508e-06,
1260
+ "logits/chosen": 0.29470381140708923,
1261
+ "logits/rejected": 0.3451955020427704,
1262
+ "logps/chosen": -989.2918090820312,
1263
+ "logps/rejected": -1229.162353515625,
1264
+ "loss": 1.6905,
1265
+ "rewards/accuracies": 0.65625,
1266
+ "rewards/chosen": -0.35454291105270386,
1267
+ "rewards/margins": 0.262814462184906,
1268
+ "rewards/rejected": -0.6173573732376099,
1269
+ "step": 890
1270
+ },
1271
+ {
1272
+ "epoch": 0.72,
1273
+ "learning_rate": 1.1020177413231334e-06,
1274
+ "logits/chosen": 0.3188125193119049,
1275
+ "logits/rejected": 0.38136720657348633,
1276
+ "logps/chosen": -893.96240234375,
1277
+ "logps/rejected": -1130.4246826171875,
1278
+ "loss": 1.6475,
1279
+ "rewards/accuracies": 0.6312500238418579,
1280
+ "rewards/chosen": -0.31285107135772705,
1281
+ "rewards/margins": 0.2777191996574402,
1282
+ "rewards/rejected": -0.5905702710151672,
1283
+ "step": 900
1284
+ },
1285
+ {
1286
+ "epoch": 0.73,
1287
+ "learning_rate": 1.0446925746067768e-06,
1288
+ "logits/chosen": 0.28160279989242554,
1289
+ "logits/rejected": 0.33634817600250244,
1290
+ "logps/chosen": -903.3307495117188,
1291
+ "logps/rejected": -1249.0274658203125,
1292
+ "loss": 1.6275,
1293
+ "rewards/accuracies": 0.6875,
1294
+ "rewards/chosen": -0.30794739723205566,
1295
+ "rewards/margins": 0.30408358573913574,
1296
+ "rewards/rejected": -0.6120310425758362,
1297
+ "step": 910
1298
+ },
1299
+ {
1300
+ "epoch": 0.74,
1301
+ "learning_rate": 9.88502212844063e-07,
1302
+ "logits/chosen": 0.2644272446632385,
1303
+ "logits/rejected": 0.4034818708896637,
1304
+ "logps/chosen": -837.537109375,
1305
+ "logps/rejected": -1180.740234375,
1306
+ "loss": 1.5707,
1307
+ "rewards/accuracies": 0.6937500238418579,
1308
+ "rewards/chosen": -0.2741805613040924,
1309
+ "rewards/margins": 0.31248125433921814,
1310
+ "rewards/rejected": -0.5866617560386658,
1311
+ "step": 920
1312
+ },
1313
+ {
1314
+ "epoch": 0.74,
1315
+ "learning_rate": 9.334904715888496e-07,
1316
+ "logits/chosen": 0.26665693521499634,
1317
+ "logits/rejected": 0.429934024810791,
1318
+ "logps/chosen": -959.6316528320312,
1319
+ "logps/rejected": -1225.3084716796875,
1320
+ "loss": 1.7093,
1321
+ "rewards/accuracies": 0.5874999761581421,
1322
+ "rewards/chosen": -0.3429884910583496,
1323
+ "rewards/margins": 0.2681938707828522,
1324
+ "rewards/rejected": -0.6111823320388794,
1325
+ "step": 930
1326
+ },
1327
+ {
1328
+ "epoch": 0.75,
1329
+ "learning_rate": 8.797002473421729e-07,
1330
+ "logits/chosen": 0.311528742313385,
1331
+ "logits/rejected": 0.3178673982620239,
1332
+ "logps/chosen": -891.2405395507812,
1333
+ "logps/rejected": -1275.465087890625,
1334
+ "loss": 1.4394,
1335
+ "rewards/accuracies": 0.7250000238418579,
1336
+ "rewards/chosen": -0.3114137053489685,
1337
+ "rewards/margins": 0.3879539370536804,
1338
+ "rewards/rejected": -0.6993676424026489,
1339
+ "step": 940
1340
+ },
1341
+ {
1342
+ "epoch": 0.76,
1343
+ "learning_rate": 8.271734841028553e-07,
1344
+ "logits/chosen": 0.2898734211921692,
1345
+ "logits/rejected": 0.3742389678955078,
1346
+ "logps/chosen": -850.2423095703125,
1347
+ "logps/rejected": -1203.037841796875,
1348
+ "loss": 1.6261,
1349
+ "rewards/accuracies": 0.6625000238418579,
1350
+ "rewards/chosen": -0.31859657168388367,
1351
+ "rewards/margins": 0.3100232481956482,
1352
+ "rewards/rejected": -0.6286198496818542,
1353
+ "step": 950
1354
+ },
1355
+ {
1356
+ "epoch": 0.77,
1357
+ "learning_rate": 7.759511406608255e-07,
1358
+ "logits/chosen": 0.30733975768089294,
1359
+ "logits/rejected": 0.44421592354774475,
1360
+ "logps/chosen": -934.5499267578125,
1361
+ "logps/rejected": -1207.3531494140625,
1362
+ "loss": 1.9113,
1363
+ "rewards/accuracies": 0.6312500238418579,
1364
+ "rewards/chosen": -0.3752412497997284,
1365
+ "rewards/margins": 0.25361892580986023,
1366
+ "rewards/rejected": -0.6288601756095886,
1367
+ "step": 960
1368
+ },
1369
+ {
1370
+ "epoch": 0.78,
1371
+ "learning_rate": 7.260731586586983e-07,
1372
+ "logits/chosen": 0.3483877182006836,
1373
+ "logits/rejected": 0.3553027808666229,
1374
+ "logps/chosen": -893.8357543945312,
1375
+ "logps/rejected": -1164.4613037109375,
1376
+ "loss": 1.776,
1377
+ "rewards/accuracies": 0.606249988079071,
1378
+ "rewards/chosen": -0.35817593336105347,
1379
+ "rewards/margins": 0.2500895857810974,
1380
+ "rewards/rejected": -0.6082655191421509,
1381
+ "step": 970
1382
+ },
1383
+ {
1384
+ "epoch": 0.78,
1385
+ "learning_rate": 6.775784314464717e-07,
1386
+ "logits/chosen": 0.353769451379776,
1387
+ "logits/rejected": 0.4078153967857361,
1388
+ "logps/chosen": -968.4454956054688,
1389
+ "logps/rejected": -1228.4012451171875,
1390
+ "loss": 1.716,
1391
+ "rewards/accuracies": 0.612500011920929,
1392
+ "rewards/chosen": -0.36273428797721863,
1393
+ "rewards/margins": 0.2496825009584427,
1394
+ "rewards/rejected": -0.6124168634414673,
1395
+ "step": 980
1396
+ },
1397
+ {
1398
+ "epoch": 0.79,
1399
+ "learning_rate": 6.305047737536707e-07,
1400
+ "logits/chosen": 0.30879533290863037,
1401
+ "logits/rejected": 0.452120304107666,
1402
+ "logps/chosen": -877.1207275390625,
1403
+ "logps/rejected": -1282.6204833984375,
1404
+ "loss": 1.5403,
1405
+ "rewards/accuracies": 0.6875,
1406
+ "rewards/chosen": -0.30206242203712463,
1407
+ "rewards/margins": 0.3925813138484955,
1408
+ "rewards/rejected": -0.6946438550949097,
1409
+ "step": 990
1410
+ },
1411
+ {
1412
+ "epoch": 0.8,
1413
+ "learning_rate": 5.848888922025553e-07,
1414
+ "logits/chosen": 0.27114662528038025,
1415
+ "logits/rejected": 0.45444902777671814,
1416
+ "logps/chosen": -969.3070068359375,
1417
+ "logps/rejected": -1175.6612548828125,
1418
+ "loss": 1.8379,
1419
+ "rewards/accuracies": 0.6000000238418579,
1420
+ "rewards/chosen": -0.3690924048423767,
1421
+ "rewards/margins": 0.2309758961200714,
1422
+ "rewards/rejected": -0.6000683903694153,
1423
+ "step": 1000
1424
+ },
1425
+ {
1426
+ "epoch": 0.81,
1427
+ "learning_rate": 5.407663566854008e-07,
1428
+ "logits/chosen": 0.2744256854057312,
1429
+ "logits/rejected": 0.41505903005599976,
1430
+ "logps/chosen": -845.2996826171875,
1431
+ "logps/rejected": -1316.6790771484375,
1432
+ "loss": 1.4883,
1433
+ "rewards/accuracies": 0.7437499761581421,
1434
+ "rewards/chosen": -0.27676206827163696,
1435
+ "rewards/margins": 0.4259292483329773,
1436
+ "rewards/rejected": -0.702691376209259,
1437
+ "step": 1010
1438
+ },
1439
+ {
1440
+ "epoch": 0.82,
1441
+ "learning_rate": 4.981715726281666e-07,
1442
+ "logits/chosen": 0.2770433723926544,
1443
+ "logits/rejected": 0.4135493338108063,
1444
+ "logps/chosen": -950.1988525390625,
1445
+ "logps/rejected": -1196.9830322265625,
1446
+ "loss": 2.0066,
1447
+ "rewards/accuracies": 0.574999988079071,
1448
+ "rewards/chosen": -0.38483771681785583,
1449
+ "rewards/margins": 0.2553882300853729,
1450
+ "rewards/rejected": -0.6402260065078735,
1451
+ "step": 1020
1452
+ },
1453
+ {
1454
+ "epoch": 0.82,
1455
+ "learning_rate": 4.5713775416217884e-07,
1456
+ "logits/chosen": 0.3525824546813965,
1457
+ "logits/rejected": 0.35387757420539856,
1458
+ "logps/chosen": -931.20068359375,
1459
+ "logps/rejected": -1218.2432861328125,
1460
+ "loss": 1.7446,
1461
+ "rewards/accuracies": 0.612500011920929,
1462
+ "rewards/chosen": -0.3306300938129425,
1463
+ "rewards/margins": 0.3051949739456177,
1464
+ "rewards/rejected": -0.6358250379562378,
1465
+ "step": 1030
1466
+ },
1467
+ {
1468
+ "epoch": 0.83,
1469
+ "learning_rate": 4.1769689822475147e-07,
1470
+ "logits/chosen": 0.30737197399139404,
1471
+ "logits/rejected": 0.4157963693141937,
1472
+ "logps/chosen": -1038.634033203125,
1473
+ "logps/rejected": -1335.1549072265625,
1474
+ "loss": 1.6302,
1475
+ "rewards/accuracies": 0.643750011920929,
1476
+ "rewards/chosen": -0.37811559438705444,
1477
+ "rewards/margins": 0.3240962624549866,
1478
+ "rewards/rejected": -0.7022119760513306,
1479
+ "step": 1040
1480
+ },
1481
+ {
1482
+ "epoch": 0.84,
1483
+ "learning_rate": 3.798797596089351e-07,
1484
+ "logits/chosen": 0.3381016254425049,
1485
+ "logits/rejected": 0.4121452867984772,
1486
+ "logps/chosen": -985.3694458007812,
1487
+ "logps/rejected": -1181.4552001953125,
1488
+ "loss": 1.8847,
1489
+ "rewards/accuracies": 0.643750011920929,
1490
+ "rewards/chosen": -0.37644538283348083,
1491
+ "rewards/margins": 0.2359016388654709,
1492
+ "rewards/rejected": -0.6123470067977905,
1493
+ "step": 1050
1494
+ },
1495
+ {
1496
+ "epoch": 0.85,
1497
+ "learning_rate": 3.4371582698185636e-07,
1498
+ "logits/chosen": 0.2851170003414154,
1499
+ "logits/rejected": 0.3804434835910797,
1500
+ "logps/chosen": -922.8662109375,
1501
+ "logps/rejected": -1263.484375,
1502
+ "loss": 1.729,
1503
+ "rewards/accuracies": 0.706250011920929,
1504
+ "rewards/chosen": -0.3330380320549011,
1505
+ "rewards/margins": 0.33658528327941895,
1506
+ "rewards/rejected": -0.6696232557296753,
1507
+ "step": 1060
1508
+ },
1509
+ {
1510
+ "epoch": 0.86,
1511
+ "learning_rate": 3.092332998903416e-07,
1512
+ "logits/chosen": 0.2671842873096466,
1513
+ "logits/rejected": 0.34879574179649353,
1514
+ "logps/chosen": -977.2224731445312,
1515
+ "logps/rejected": -1249.781982421875,
1516
+ "loss": 1.7384,
1517
+ "rewards/accuracies": 0.637499988079071,
1518
+ "rewards/chosen": -0.3930615484714508,
1519
+ "rewards/margins": 0.2609555125236511,
1520
+ "rewards/rejected": -0.6540170907974243,
1521
+ "step": 1070
1522
+ },
1523
+ {
1524
+ "epoch": 0.86,
1525
+ "learning_rate": 2.764590667717562e-07,
1526
+ "logits/chosen": 0.36944884061813354,
1527
+ "logits/rejected": 0.3970021903514862,
1528
+ "logps/chosen": -830.1124877929688,
1529
+ "logps/rejected": -1250.266357421875,
1530
+ "loss": 1.5305,
1531
+ "rewards/accuracies": 0.6625000238418579,
1532
+ "rewards/chosen": -0.3065842092037201,
1533
+ "rewards/margins": 0.34969645738601685,
1534
+ "rewards/rejected": -0.6562806367874146,
1535
+ "step": 1080
1536
+ },
1537
+ {
1538
+ "epoch": 0.87,
1539
+ "learning_rate": 2.454186839872158e-07,
1540
+ "logits/chosen": 0.30845582485198975,
1541
+ "logits/rejected": 0.3318074643611908,
1542
+ "logps/chosen": -933.2980346679688,
1543
+ "logps/rejected": -1170.7999267578125,
1544
+ "loss": 1.8682,
1545
+ "rewards/accuracies": 0.6187499761581421,
1546
+ "rewards/chosen": -0.3466985821723938,
1547
+ "rewards/margins": 0.2223629206418991,
1548
+ "rewards/rejected": -0.5690615177154541,
1549
+ "step": 1090
1550
+ },
1551
+ {
1552
+ "epoch": 0.88,
1553
+ "learning_rate": 2.1613635589349756e-07,
1554
+ "logits/chosen": 0.27249905467033386,
1555
+ "logits/rejected": 0.4322986602783203,
1556
+ "logps/chosen": -986.0386962890625,
1557
+ "logps/rejected": -1213.860107421875,
1558
+ "loss": 1.792,
1559
+ "rewards/accuracies": 0.643750011920929,
1560
+ "rewards/chosen": -0.3734912574291229,
1561
+ "rewards/margins": 0.2489509880542755,
1562
+ "rewards/rejected": -0.6224422454833984,
1563
+ "step": 1100
1564
+ },
1565
+ {
1566
+ "epoch": 0.89,
1567
+ "learning_rate": 1.8863491596921745e-07,
1568
+ "logits/chosen": 0.31418293714523315,
1569
+ "logits/rejected": 0.39671963453292847,
1570
+ "logps/chosen": -923.6838989257812,
1571
+ "logps/rejected": -1139.2310791015625,
1572
+ "loss": 1.8913,
1573
+ "rewards/accuracies": 0.5874999761581421,
1574
+ "rewards/chosen": -0.3423645496368408,
1575
+ "rewards/margins": 0.2353101521730423,
1576
+ "rewards/rejected": -0.5776747465133667,
1577
+ "step": 1110
1578
+ },
1579
+ {
1580
+ "epoch": 0.9,
1581
+ "learning_rate": 1.629358090099639e-07,
1582
+ "logits/chosen": 0.31794118881225586,
1583
+ "logits/rejected": 0.4100438058376312,
1584
+ "logps/chosen": -999.4279174804688,
1585
+ "logps/rejected": -1165.24169921875,
1586
+ "loss": 1.7528,
1587
+ "rewards/accuracies": 0.5562499761581421,
1588
+ "rewards/chosen": -0.361406147480011,
1589
+ "rewards/margins": 0.24150419235229492,
1590
+ "rewards/rejected": -0.6029103994369507,
1591
+ "step": 1120
1592
+ },
1593
+ {
1594
+ "epoch": 0.9,
1595
+ "learning_rate": 1.3905907440629752e-07,
1596
+ "logits/chosen": 0.28145748376846313,
1597
+ "logits/rejected": 0.3526201844215393,
1598
+ "logps/chosen": -896.6658935546875,
1599
+ "logps/rejected": -1153.307861328125,
1600
+ "loss": 1.7645,
1601
+ "rewards/accuracies": 0.6187499761581421,
1602
+ "rewards/chosen": -0.33654385805130005,
1603
+ "rewards/margins": 0.23922619223594666,
1604
+ "rewards/rejected": -0.5757700204849243,
1605
+ "step": 1130
1606
+ },
1607
+ {
1608
+ "epoch": 0.91,
1609
+ "learning_rate": 1.1702333051763271e-07,
1610
+ "logits/chosen": 0.30934053659439087,
1611
+ "logits/rejected": 0.3852478861808777,
1612
+ "logps/chosen": -949.3416137695312,
1613
+ "logps/rejected": -1139.066650390625,
1614
+ "loss": 1.7438,
1615
+ "rewards/accuracies": 0.6499999761581421,
1616
+ "rewards/chosen": -0.33037489652633667,
1617
+ "rewards/margins": 0.2449624091386795,
1618
+ "rewards/rejected": -0.575337290763855,
1619
+ "step": 1140
1620
+ },
1621
+ {
1622
+ "epoch": 0.92,
1623
+ "learning_rate": 9.684576015420277e-08,
1624
+ "logits/chosen": 0.27784961462020874,
1625
+ "logits/rejected": 0.40725311636924744,
1626
+ "logps/chosen": -865.6135864257812,
1627
+ "logps/rejected": -1142.362060546875,
1628
+ "loss": 1.8127,
1629
+ "rewards/accuracies": 0.6187499761581421,
1630
+ "rewards/chosen": -0.30999353528022766,
1631
+ "rewards/margins": 0.2895258069038391,
1632
+ "rewards/rejected": -0.5995193123817444,
1633
+ "step": 1150
1634
+ },
1635
+ {
1636
+ "epoch": 0.93,
1637
+ "learning_rate": 7.854209717842231e-08,
1638
+ "logits/chosen": 0.28656622767448425,
1639
+ "logits/rejected": 0.394423246383667,
1640
+ "logps/chosen": -983.4553833007812,
1641
+ "logps/rejected": -1191.291748046875,
1642
+ "loss": 1.897,
1643
+ "rewards/accuracies": 0.6187499761581421,
1644
+ "rewards/chosen": -0.3752153813838959,
1645
+ "rewards/margins": 0.2264389544725418,
1646
+ "rewards/rejected": -0.6016542911529541,
1647
+ "step": 1160
1648
+ },
1649
+ {
1650
+ "epoch": 0.94,
1651
+ "learning_rate": 6.212661423609184e-08,
1652
+ "logits/chosen": 0.33070552349090576,
1653
+ "logits/rejected": 0.4134686589241028,
1654
+ "logps/chosen": -823.2513427734375,
1655
+ "logps/rejected": -1224.76318359375,
1656
+ "loss": 1.4644,
1657
+ "rewards/accuracies": 0.6937500238418579,
1658
+ "rewards/chosen": -0.2675178647041321,
1659
+ "rewards/margins": 0.37183278799057007,
1660
+ "rewards/rejected": -0.6393505930900574,
1661
+ "step": 1170
1662
+ },
1663
+ {
1664
+ "epoch": 0.94,
1665
+ "learning_rate": 4.761211162702117e-08,
1666
+ "logits/chosen": 0.30609947443008423,
1667
+ "logits/rejected": 0.3751484751701355,
1668
+ "logps/chosen": -889.3785400390625,
1669
+ "logps/rejected": -1164.7288818359375,
1670
+ "loss": 1.9301,
1671
+ "rewards/accuracies": 0.574999988079071,
1672
+ "rewards/chosen": -0.33601176738739014,
1673
+ "rewards/margins": 0.24511654675006866,
1674
+ "rewards/rejected": -0.5811282396316528,
1675
+ "step": 1180
1676
+ },
1677
+ {
1678
+ "epoch": 0.95,
1679
+ "learning_rate": 3.5009907323737826e-08,
1680
+ "logits/chosen": 0.3077315390110016,
1681
+ "logits/rejected": 0.32115936279296875,
1682
+ "logps/chosen": -940.2276611328125,
1683
+ "logps/rejected": -1252.099853515625,
1684
+ "loss": 1.6158,
1685
+ "rewards/accuracies": 0.6812499761581421,
1686
+ "rewards/chosen": -0.3381388485431671,
1687
+ "rewards/margins": 0.2894209623336792,
1688
+ "rewards/rejected": -0.6275597810745239,
1689
+ "step": 1190
1690
+ },
1691
+ {
1692
+ "epoch": 0.96,
1693
+ "learning_rate": 2.4329828146074096e-08,
1694
+ "logits/chosen": 0.3073035478591919,
1695
+ "logits/rejected": 0.4198598265647888,
1696
+ "logps/chosen": -872.7296752929688,
1697
+ "logps/rejected": -1125.4498291015625,
1698
+ "loss": 1.7422,
1699
+ "rewards/accuracies": 0.637499988079071,
1700
+ "rewards/chosen": -0.30542269349098206,
1701
+ "rewards/margins": 0.29255709052085876,
1702
+ "rewards/rejected": -0.5979797840118408,
1703
+ "step": 1200
1704
+ },
1705
+ {
1706
+ "epoch": 0.97,
1707
+ "learning_rate": 1.5580202098509078e-08,
1708
+ "logits/chosen": 0.31831032037734985,
1709
+ "logits/rejected": 0.4222412109375,
1710
+ "logps/chosen": -889.1136474609375,
1711
+ "logps/rejected": -1244.5892333984375,
1712
+ "loss": 1.7276,
1713
+ "rewards/accuracies": 0.6312500238418579,
1714
+ "rewards/chosen": -0.31571874022483826,
1715
+ "rewards/margins": 0.3139411509037018,
1716
+ "rewards/rejected": -0.62965989112854,
1717
+ "step": 1210
1718
+ },
1719
+ {
1720
+ "epoch": 0.98,
1721
+ "learning_rate": 8.767851876239075e-09,
1722
+ "logits/chosen": 0.2873593270778656,
1723
+ "logits/rejected": 0.42444127798080444,
1724
+ "logps/chosen": -910.4513549804688,
1725
+ "logps/rejected": -1105.3863525390625,
1726
+ "loss": 1.6767,
1727
+ "rewards/accuracies": 0.65625,
1728
+ "rewards/chosen": -0.29710885882377625,
1729
+ "rewards/margins": 0.2506183981895447,
1730
+ "rewards/rejected": -0.5477272272109985,
1731
+ "step": 1220
1732
+ },
1733
+ {
1734
+ "epoch": 0.98,
1735
+ "learning_rate": 3.8980895450474455e-09,
1736
+ "logits/chosen": 0.32995566725730896,
1737
+ "logits/rejected": 0.36712411046028137,
1738
+ "logps/chosen": -900.4547729492188,
1739
+ "logps/rejected": -1215.38134765625,
1740
+ "loss": 1.764,
1741
+ "rewards/accuracies": 0.6499999761581421,
1742
+ "rewards/chosen": -0.30327844619750977,
1743
+ "rewards/margins": 0.2899893820285797,
1744
+ "rewards/rejected": -0.5932677984237671,
1745
+ "step": 1230
1746
+ },
1747
+ {
1748
+ "epoch": 0.99,
1749
+ "learning_rate": 9.747123991141193e-10,
1750
+ "logits/chosen": 0.30442702770233154,
1751
+ "logits/rejected": 0.3750511407852173,
1752
+ "logps/chosen": -829.1936645507812,
1753
+ "logps/rejected": -1163.27783203125,
1754
+ "loss": 1.8085,
1755
+ "rewards/accuracies": 0.6187499761581421,
1756
+ "rewards/chosen": -0.3106636106967926,
1757
+ "rewards/margins": 0.2759193778038025,
1758
+ "rewards/rejected": -0.5865829586982727,
1759
+ "step": 1240
1760
+ },
1761
+ {
1762
+ "epoch": 1.0,
1763
+ "learning_rate": 0.0,
1764
+ "logits/chosen": 0.3020361661911011,
1765
+ "logits/rejected": 0.362225204706192,
1766
+ "logps/chosen": -867.2174682617188,
1767
+ "logps/rejected": -1196.6907958984375,
1768
+ "loss": 1.7083,
1769
+ "rewards/accuracies": 0.65625,
1770
+ "rewards/chosen": -0.3051598370075226,
1771
+ "rewards/margins": 0.2913387715816498,
1772
+ "rewards/rejected": -0.5964986085891724,
1773
+ "step": 1250
1774
+ },
1775
+ {
1776
+ "epoch": 1.0,
1777
+ "step": 1250,
1778
+ "total_flos": 0.0,
1779
+ "train_loss": 1.8189717344284058,
1780
+ "train_runtime": 11934.9419,
1781
+ "train_samples_per_second": 1.676,
1782
+ "train_steps_per_second": 0.105
1783
+ }
1784
+ ],
1785
+ "logging_steps": 10,
1786
+ "max_steps": 1250,
1787
+ "num_input_tokens_seen": 0,
1788
+ "num_train_epochs": 1,
1789
+ "save_steps": 100,
1790
+ "total_flos": 0.0,
1791
+ "train_batch_size": 4,
1792
+ "trial_name": null,
1793
+ "trial_params": null
1794
+ }