vain05 commited on
Commit
380ea7b
1 Parent(s): 452994f

Model save

Browse files
README.md ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ base_model: stabilityai/stablelm-2-1_6b
4
+ tags:
5
+ - trl
6
+ - orpo
7
+ - generated_from_trainer
8
+ model-index:
9
+ - name: stablelm-2-1_6b-orpo-full-v3
10
+ results: []
11
+ ---
12
+
13
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
14
+ should probably proofread and complete it, then remove this comment. -->
15
+
16
+ # stablelm-2-1_6b-orpo-full-v3
17
+
18
+ This model is a fine-tuned version of [stabilityai/stablelm-2-1_6b](https://huggingface.co/stabilityai/stablelm-2-1_6b) on the None dataset.
19
+
20
+ ## Model description
21
+
22
+ More information needed
23
+
24
+ ## Intended uses & limitations
25
+
26
+ More information needed
27
+
28
+ ## Training and evaluation data
29
+
30
+ More information needed
31
+
32
+ ## Training procedure
33
+
34
+ ### Training hyperparameters
35
+
36
+ The following hyperparameters were used during training:
37
+ - learning_rate: 5e-07
38
+ - train_batch_size: 8
39
+ - eval_batch_size: 8
40
+ - seed: 42
41
+ - distributed_type: multi-GPU
42
+ - num_devices: 3
43
+ - gradient_accumulation_steps: 2
44
+ - total_train_batch_size: 48
45
+ - total_eval_batch_size: 24
46
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
47
+ - lr_scheduler_type: cosine
48
+ - lr_scheduler_warmup_ratio: 0.1
49
+ - num_epochs: 1
50
+
51
+ ### Training results
52
+
53
+
54
+
55
+ ### Framework versions
56
+
57
+ - Transformers 4.39.0.dev0
58
+ - Pytorch 2.2.1
59
+ - Datasets 2.18.0
60
+ - Tokenizers 0.15.2
all_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 1.2897131164005442,
4
+ "train_runtime": 9812.389,
5
+ "train_samples": 68479,
6
+ "train_samples_per_second": 6.979,
7
+ "train_steps_per_second": 0.145
8
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 100257,
4
+ "eos_token_id": 100257,
5
+ "transformers_version": "4.39.0.dev0"
6
+ }
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1b7192de297068fce4d243d305bc541a11bf70232e6074d6a4e5bb1d544a6e11
3
  size 3289069520
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9619a2550887591625f4e84dc1f7e1ea897637b427c1be4a1ea67717ce0b8770
3
  size 3289069520
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 1.2897131164005442,
4
+ "train_runtime": 9812.389,
5
+ "train_samples": 68479,
6
+ "train_samples_per_second": 6.979,
7
+ "train_steps_per_second": 0.145
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,2586 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 1.0,
5
+ "eval_steps": 500,
6
+ "global_step": 1427,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.01,
13
+ "grad_norm": 7.579595964430724,
14
+ "learning_rate": 3.4965034965034967e-08,
15
+ "log_odds_chosen": 0.4541015625,
16
+ "log_odds_ratio": -0.677050769329071,
17
+ "logits/chosen": -2.015625,
18
+ "logits/rejected": -2.015625,
19
+ "logps/chosen": -1.8671875,
20
+ "logps/rejected": -2.265625,
21
+ "loss": 1.5655,
22
+ "nll_loss": 1.5,
23
+ "rewards/accuracies": 0.6625000238418579,
24
+ "rewards/chosen": -0.1865234375,
25
+ "rewards/margins": 0.0400390625,
26
+ "rewards/rejected": -0.2265625,
27
+ "step": 10
28
+ },
29
+ {
30
+ "epoch": 0.01,
31
+ "grad_norm": 7.5249840340758425,
32
+ "learning_rate": 6.993006993006993e-08,
33
+ "log_odds_chosen": 0.22894287109375,
34
+ "log_odds_ratio": -0.7715820074081421,
35
+ "logits/chosen": -2.03125,
36
+ "logits/rejected": -2.0625,
37
+ "logps/chosen": -1.9921875,
38
+ "logps/rejected": -2.1875,
39
+ "loss": 1.5773,
40
+ "nll_loss": 1.4921875,
41
+ "rewards/accuracies": 0.5687500238418579,
42
+ "rewards/chosen": -0.19921875,
43
+ "rewards/margins": 0.01953125,
44
+ "rewards/rejected": -0.21875,
45
+ "step": 20
46
+ },
47
+ {
48
+ "epoch": 0.02,
49
+ "grad_norm": 5.9119698758492865,
50
+ "learning_rate": 1.0489510489510489e-07,
51
+ "log_odds_chosen": 0.28911131620407104,
52
+ "log_odds_ratio": -0.8036133050918579,
53
+ "logits/chosen": -2.015625,
54
+ "logits/rejected": -2.046875,
55
+ "logps/chosen": -2.109375,
56
+ "logps/rejected": -2.34375,
57
+ "loss": 1.5587,
58
+ "nll_loss": 1.515625,
59
+ "rewards/accuracies": 0.5687500238418579,
60
+ "rewards/chosen": -0.2109375,
61
+ "rewards/margins": 0.0242919921875,
62
+ "rewards/rejected": -0.2353515625,
63
+ "step": 30
64
+ },
65
+ {
66
+ "epoch": 0.03,
67
+ "grad_norm": 6.999607089742296,
68
+ "learning_rate": 1.3986013986013987e-07,
69
+ "log_odds_chosen": 0.25310057401657104,
70
+ "log_odds_ratio": -0.762402355670929,
71
+ "logits/chosen": -2.03125,
72
+ "logits/rejected": -2.015625,
73
+ "logps/chosen": -1.921875,
74
+ "logps/rejected": -2.125,
75
+ "loss": 1.5862,
76
+ "nll_loss": 1.4765625,
77
+ "rewards/accuracies": 0.5562499761581421,
78
+ "rewards/chosen": -0.1923828125,
79
+ "rewards/margins": 0.0208740234375,
80
+ "rewards/rejected": -0.212890625,
81
+ "step": 40
82
+ },
83
+ {
84
+ "epoch": 0.04,
85
+ "grad_norm": 5.309681737493818,
86
+ "learning_rate": 1.7482517482517481e-07,
87
+ "log_odds_chosen": 0.25770264863967896,
88
+ "log_odds_ratio": -0.697558581829071,
89
+ "logits/chosen": -2.109375,
90
+ "logits/rejected": -2.0625,
91
+ "logps/chosen": -1.921875,
92
+ "logps/rejected": -2.140625,
93
+ "loss": 1.5736,
94
+ "nll_loss": 1.546875,
95
+ "rewards/accuracies": 0.59375,
96
+ "rewards/chosen": -0.1923828125,
97
+ "rewards/margins": 0.0218505859375,
98
+ "rewards/rejected": -0.2138671875,
99
+ "step": 50
100
+ },
101
+ {
102
+ "epoch": 0.04,
103
+ "grad_norm": 5.569709247368732,
104
+ "learning_rate": 2.0979020979020979e-07,
105
+ "log_odds_chosen": 0.311614990234375,
106
+ "log_odds_ratio": -0.695605456829071,
107
+ "logits/chosen": -2.109375,
108
+ "logits/rejected": -2.15625,
109
+ "logps/chosen": -1.96875,
110
+ "logps/rejected": -2.25,
111
+ "loss": 1.5668,
112
+ "nll_loss": 1.484375,
113
+ "rewards/accuracies": 0.5562499761581421,
114
+ "rewards/chosen": -0.197265625,
115
+ "rewards/margins": 0.02734375,
116
+ "rewards/rejected": -0.224609375,
117
+ "step": 60
118
+ },
119
+ {
120
+ "epoch": 0.05,
121
+ "grad_norm": 5.773237000535201,
122
+ "learning_rate": 2.4475524475524473e-07,
123
+ "log_odds_chosen": 0.14488525688648224,
124
+ "log_odds_ratio": -0.786425769329071,
125
+ "logits/chosen": -2.171875,
126
+ "logits/rejected": -2.125,
127
+ "logps/chosen": -2.0,
128
+ "logps/rejected": -2.125,
129
+ "loss": 1.5723,
130
+ "nll_loss": 1.5234375,
131
+ "rewards/accuracies": 0.512499988079071,
132
+ "rewards/chosen": -0.2001953125,
133
+ "rewards/margins": 0.01287841796875,
134
+ "rewards/rejected": -0.212890625,
135
+ "step": 70
136
+ },
137
+ {
138
+ "epoch": 0.06,
139
+ "grad_norm": 5.5756538810115925,
140
+ "learning_rate": 2.7972027972027973e-07,
141
+ "log_odds_chosen": 0.2967529296875,
142
+ "log_odds_ratio": -0.7289062738418579,
143
+ "logits/chosen": -2.109375,
144
+ "logits/rejected": -2.140625,
145
+ "logps/chosen": -1.8125,
146
+ "logps/rejected": -2.046875,
147
+ "loss": 1.4668,
148
+ "nll_loss": 1.3515625,
149
+ "rewards/accuracies": 0.612500011920929,
150
+ "rewards/chosen": -0.181640625,
151
+ "rewards/margins": 0.023681640625,
152
+ "rewards/rejected": -0.205078125,
153
+ "step": 80
154
+ },
155
+ {
156
+ "epoch": 0.06,
157
+ "grad_norm": 4.2269586617524535,
158
+ "learning_rate": 3.146853146853147e-07,
159
+ "log_odds_chosen": 0.2103271484375,
160
+ "log_odds_ratio": -0.73095703125,
161
+ "logits/chosen": -2.171875,
162
+ "logits/rejected": -2.140625,
163
+ "logps/chosen": -1.8515625,
164
+ "logps/rejected": -2.03125,
165
+ "loss": 1.5357,
166
+ "nll_loss": 1.4921875,
167
+ "rewards/accuracies": 0.5874999761581421,
168
+ "rewards/chosen": -0.185546875,
169
+ "rewards/margins": 0.0186767578125,
170
+ "rewards/rejected": -0.2041015625,
171
+ "step": 90
172
+ },
173
+ {
174
+ "epoch": 0.07,
175
+ "grad_norm": 7.493002974943704,
176
+ "learning_rate": 3.4965034965034963e-07,
177
+ "log_odds_chosen": 0.14136353135108948,
178
+ "log_odds_ratio": -0.710644543170929,
179
+ "logits/chosen": -2.171875,
180
+ "logits/rejected": -2.1875,
181
+ "logps/chosen": -1.921875,
182
+ "logps/rejected": -2.03125,
183
+ "loss": 1.5193,
184
+ "nll_loss": 1.4375,
185
+ "rewards/accuracies": 0.581250011920929,
186
+ "rewards/chosen": -0.19140625,
187
+ "rewards/margins": 0.010498046875,
188
+ "rewards/rejected": -0.2021484375,
189
+ "step": 100
190
+ },
191
+ {
192
+ "epoch": 0.08,
193
+ "grad_norm": 6.173840413359453,
194
+ "learning_rate": 3.8461538461538463e-07,
195
+ "log_odds_chosen": 0.18465575575828552,
196
+ "log_odds_ratio": -0.7144531011581421,
197
+ "logits/chosen": -2.265625,
198
+ "logits/rejected": -2.296875,
199
+ "logps/chosen": -1.734375,
200
+ "logps/rejected": -1.890625,
201
+ "loss": 1.5043,
202
+ "nll_loss": 1.421875,
203
+ "rewards/accuracies": 0.612500011920929,
204
+ "rewards/chosen": -0.173828125,
205
+ "rewards/margins": 0.01556396484375,
206
+ "rewards/rejected": -0.189453125,
207
+ "step": 110
208
+ },
209
+ {
210
+ "epoch": 0.08,
211
+ "grad_norm": 3.9756802927191024,
212
+ "learning_rate": 4.1958041958041957e-07,
213
+ "log_odds_chosen": 0.22193603217601776,
214
+ "log_odds_ratio": -0.7044922113418579,
215
+ "logits/chosen": -2.296875,
216
+ "logits/rejected": -2.359375,
217
+ "logps/chosen": -1.4921875,
218
+ "logps/rejected": -1.703125,
219
+ "loss": 1.4443,
220
+ "nll_loss": 1.3125,
221
+ "rewards/accuracies": 0.543749988079071,
222
+ "rewards/chosen": -0.1494140625,
223
+ "rewards/margins": 0.0208740234375,
224
+ "rewards/rejected": -0.169921875,
225
+ "step": 120
226
+ },
227
+ {
228
+ "epoch": 0.09,
229
+ "grad_norm": 3.49115617326949,
230
+ "learning_rate": 4.545454545454545e-07,
231
+ "log_odds_chosen": 0.07476196438074112,
232
+ "log_odds_ratio": -0.741992175579071,
233
+ "logits/chosen": -2.28125,
234
+ "logits/rejected": -2.359375,
235
+ "logps/chosen": -1.5703125,
236
+ "logps/rejected": -1.640625,
237
+ "loss": 1.4535,
238
+ "nll_loss": 1.40625,
239
+ "rewards/accuracies": 0.53125,
240
+ "rewards/chosen": -0.1572265625,
241
+ "rewards/margins": 0.006622314453125,
242
+ "rewards/rejected": -0.1640625,
243
+ "step": 130
244
+ },
245
+ {
246
+ "epoch": 0.1,
247
+ "grad_norm": 3.2300418693285047,
248
+ "learning_rate": 4.895104895104895e-07,
249
+ "log_odds_chosen": 0.23361817002296448,
250
+ "log_odds_ratio": -0.694628894329071,
251
+ "logits/chosen": -2.265625,
252
+ "logits/rejected": -2.375,
253
+ "logps/chosen": -1.5,
254
+ "logps/rejected": -1.703125,
255
+ "loss": 1.4352,
256
+ "nll_loss": 1.3828125,
257
+ "rewards/accuracies": 0.581250011920929,
258
+ "rewards/chosen": -0.150390625,
259
+ "rewards/margins": 0.019775390625,
260
+ "rewards/rejected": -0.169921875,
261
+ "step": 140
262
+ },
263
+ {
264
+ "epoch": 0.11,
265
+ "grad_norm": 3.733581414979026,
266
+ "learning_rate": 4.999633338614865e-07,
267
+ "log_odds_chosen": 0.0267333984375,
268
+ "log_odds_ratio": -0.739453136920929,
269
+ "logits/chosen": -2.296875,
270
+ "logits/rejected": -2.375,
271
+ "logps/chosen": -1.5234375,
272
+ "logps/rejected": -1.546875,
273
+ "loss": 1.4385,
274
+ "nll_loss": 1.3671875,
275
+ "rewards/accuracies": 0.5062500238418579,
276
+ "rewards/chosen": -0.15234375,
277
+ "rewards/margins": 0.0021209716796875,
278
+ "rewards/rejected": -0.154296875,
279
+ "step": 150
280
+ },
281
+ {
282
+ "epoch": 0.11,
283
+ "grad_norm": 4.063546439288504,
284
+ "learning_rate": 4.997837705025723e-07,
285
+ "log_odds_chosen": 0.14241942763328552,
286
+ "log_odds_ratio": -0.7054687738418579,
287
+ "logits/chosen": -2.203125,
288
+ "logits/rejected": -2.3125,
289
+ "logps/chosen": -1.3828125,
290
+ "logps/rejected": -1.4921875,
291
+ "loss": 1.3762,
292
+ "nll_loss": 1.2734375,
293
+ "rewards/accuracies": 0.5375000238418579,
294
+ "rewards/chosen": -0.138671875,
295
+ "rewards/margins": 0.0108642578125,
296
+ "rewards/rejected": -0.1494140625,
297
+ "step": 160
298
+ },
299
+ {
300
+ "epoch": 0.12,
301
+ "grad_norm": 2.8382728420037524,
302
+ "learning_rate": 4.994546826814266e-07,
303
+ "log_odds_chosen": 0.19875487685203552,
304
+ "log_odds_ratio": -0.658203125,
305
+ "logits/chosen": -2.375,
306
+ "logits/rejected": -2.46875,
307
+ "logps/chosen": -1.3359375,
308
+ "logps/rejected": -1.484375,
309
+ "loss": 1.3645,
310
+ "nll_loss": 1.3046875,
311
+ "rewards/accuracies": 0.581250011920929,
312
+ "rewards/chosen": -0.1337890625,
313
+ "rewards/margins": 0.0146484375,
314
+ "rewards/rejected": -0.1484375,
315
+ "step": 170
316
+ },
317
+ {
318
+ "epoch": 0.13,
319
+ "grad_norm": 2.5209932782786,
320
+ "learning_rate": 4.989762673951533e-07,
321
+ "log_odds_chosen": 0.16008301079273224,
322
+ "log_odds_ratio": -0.6830078363418579,
323
+ "logits/chosen": -2.359375,
324
+ "logits/rejected": -2.515625,
325
+ "logps/chosen": -1.2890625,
326
+ "logps/rejected": -1.40625,
327
+ "loss": 1.3044,
328
+ "nll_loss": 1.2265625,
329
+ "rewards/accuracies": 0.5687500238418579,
330
+ "rewards/chosen": -0.12890625,
331
+ "rewards/margins": 0.01129150390625,
332
+ "rewards/rejected": -0.140625,
333
+ "step": 180
334
+ },
335
+ {
336
+ "epoch": 0.13,
337
+ "grad_norm": 2.7624176664717623,
338
+ "learning_rate": 4.983488110306074e-07,
339
+ "log_odds_chosen": 0.08121337741613388,
340
+ "log_odds_ratio": -0.7232421636581421,
341
+ "logits/chosen": -2.375,
342
+ "logits/rejected": -2.4375,
343
+ "logps/chosen": -1.34375,
344
+ "logps/rejected": -1.3984375,
345
+ "loss": 1.3099,
346
+ "nll_loss": 1.265625,
347
+ "rewards/accuracies": 0.5062500238418579,
348
+ "rewards/chosen": -0.1337890625,
349
+ "rewards/margins": 0.005584716796875,
350
+ "rewards/rejected": -0.1396484375,
351
+ "step": 190
352
+ },
353
+ {
354
+ "epoch": 0.14,
355
+ "grad_norm": 2.5951453466247916,
356
+ "learning_rate": 4.975726891929584e-07,
357
+ "log_odds_chosen": 0.22237548232078552,
358
+ "log_odds_ratio": -0.66455078125,
359
+ "logits/chosen": -2.359375,
360
+ "logits/rejected": -2.46875,
361
+ "logps/chosen": -1.2421875,
362
+ "logps/rejected": -1.40625,
363
+ "loss": 1.3293,
364
+ "nll_loss": 1.1953125,
365
+ "rewards/accuracies": 0.5874999761581421,
366
+ "rewards/chosen": -0.1240234375,
367
+ "rewards/margins": 0.0169677734375,
368
+ "rewards/rejected": -0.140625,
369
+ "step": 200
370
+ },
371
+ {
372
+ "epoch": 0.15,
373
+ "grad_norm": 2.745043363887144,
374
+ "learning_rate": 4.966483664808476e-07,
375
+ "log_odds_chosen": 0.13164062798023224,
376
+ "log_odds_ratio": -0.7242187261581421,
377
+ "logits/chosen": -2.21875,
378
+ "logits/rejected": -2.21875,
379
+ "logps/chosen": -1.21875,
380
+ "logps/rejected": -1.3203125,
381
+ "loss": 1.3097,
382
+ "nll_loss": 1.1875,
383
+ "rewards/accuracies": 0.581250011920929,
384
+ "rewards/chosen": -0.12158203125,
385
+ "rewards/margins": 0.010498046875,
386
+ "rewards/rejected": -0.1318359375,
387
+ "step": 210
388
+ },
389
+ {
390
+ "epoch": 0.15,
391
+ "grad_norm": 2.544921558378996,
392
+ "learning_rate": 4.955763962082722e-07,
393
+ "log_odds_chosen": 0.2619995176792145,
394
+ "log_odds_ratio": -0.673046886920929,
395
+ "logits/chosen": -2.4375,
396
+ "logits/rejected": -2.453125,
397
+ "logps/chosen": -1.265625,
398
+ "logps/rejected": -1.46875,
399
+ "loss": 1.299,
400
+ "nll_loss": 1.2734375,
401
+ "rewards/accuracies": 0.59375,
402
+ "rewards/chosen": -0.126953125,
403
+ "rewards/margins": 0.0205078125,
404
+ "rewards/rejected": -0.1474609375,
405
+ "step": 220
406
+ },
407
+ {
408
+ "epoch": 0.16,
409
+ "grad_norm": 2.7422926363251467,
410
+ "learning_rate": 4.943574200733625e-07,
411
+ "log_odds_chosen": 0.197174072265625,
412
+ "log_odds_ratio": -0.6680663824081421,
413
+ "logits/chosen": -2.296875,
414
+ "logits/rejected": -2.390625,
415
+ "logps/chosen": -1.234375,
416
+ "logps/rejected": -1.375,
417
+ "loss": 1.3089,
418
+ "nll_loss": 1.234375,
419
+ "rewards/accuracies": 0.5249999761581421,
420
+ "rewards/chosen": -0.123046875,
421
+ "rewards/margins": 0.0142822265625,
422
+ "rewards/rejected": -0.1376953125,
423
+ "step": 230
424
+ },
425
+ {
426
+ "epoch": 0.17,
427
+ "grad_norm": 2.5904882528928845,
428
+ "learning_rate": 4.929921677742516e-07,
429
+ "log_odds_chosen": 0.24537964165210724,
430
+ "log_odds_ratio": -0.657031238079071,
431
+ "logits/chosen": -2.328125,
432
+ "logits/rejected": -2.421875,
433
+ "logps/chosen": -1.2734375,
434
+ "logps/rejected": -1.4453125,
435
+ "loss": 1.3289,
436
+ "nll_loss": 1.2421875,
437
+ "rewards/accuracies": 0.59375,
438
+ "rewards/chosen": -0.1279296875,
439
+ "rewards/margins": 0.0167236328125,
440
+ "rewards/rejected": -0.14453125,
441
+ "step": 240
442
+ },
443
+ {
444
+ "epoch": 0.18,
445
+ "grad_norm": 2.57374464764083,
446
+ "learning_rate": 4.91481456572267e-07,
447
+ "log_odds_chosen": 0.1439208984375,
448
+ "log_odds_ratio": -0.7132812738418579,
449
+ "logits/chosen": -2.296875,
450
+ "logits/rejected": -2.328125,
451
+ "logps/chosen": -1.21875,
452
+ "logps/rejected": -1.3203125,
453
+ "loss": 1.2807,
454
+ "nll_loss": 1.2109375,
455
+ "rewards/accuracies": 0.543749988079071,
456
+ "rewards/chosen": -0.1220703125,
457
+ "rewards/margins": 0.01019287109375,
458
+ "rewards/rejected": -0.1318359375,
459
+ "step": 250
460
+ },
461
+ {
462
+ "epoch": 0.18,
463
+ "grad_norm": 2.5866184628438873,
464
+ "learning_rate": 4.898261908027049e-07,
465
+ "log_odds_chosen": 0.20936889946460724,
466
+ "log_odds_ratio": -0.6578124761581421,
467
+ "logits/chosen": -2.28125,
468
+ "logits/rejected": -2.359375,
469
+ "logps/chosen": -1.2421875,
470
+ "logps/rejected": -1.40625,
471
+ "loss": 1.2753,
472
+ "nll_loss": 1.21875,
473
+ "rewards/accuracies": 0.581250011920929,
474
+ "rewards/chosen": -0.1240234375,
475
+ "rewards/margins": 0.01611328125,
476
+ "rewards/rejected": -0.140625,
477
+ "step": 260
478
+ },
479
+ {
480
+ "epoch": 0.19,
481
+ "grad_norm": 3.7405175636819643,
482
+ "learning_rate": 4.880273613334809e-07,
483
+ "log_odds_chosen": 0.22337035834789276,
484
+ "log_odds_ratio": -0.7001953125,
485
+ "logits/chosen": -2.328125,
486
+ "logits/rejected": -2.4375,
487
+ "logps/chosen": -1.2265625,
488
+ "logps/rejected": -1.40625,
489
+ "loss": 1.2719,
490
+ "nll_loss": 1.203125,
491
+ "rewards/accuracies": 0.550000011920929,
492
+ "rewards/chosen": -0.123046875,
493
+ "rewards/margins": 0.0179443359375,
494
+ "rewards/rejected": -0.140625,
495
+ "step": 270
496
+ },
497
+ {
498
+ "epoch": 0.2,
499
+ "grad_norm": 2.7574742318554923,
500
+ "learning_rate": 4.86086044971981e-07,
501
+ "log_odds_chosen": 0.15998534858226776,
502
+ "log_odds_ratio": -0.677539050579071,
503
+ "logits/chosen": -2.25,
504
+ "logits/rejected": -2.328125,
505
+ "logps/chosen": -1.2578125,
506
+ "logps/rejected": -1.3671875,
507
+ "loss": 1.3146,
508
+ "nll_loss": 1.28125,
509
+ "rewards/accuracies": 0.5562499761581421,
510
+ "rewards/chosen": -0.1259765625,
511
+ "rewards/margins": 0.0103759765625,
512
+ "rewards/rejected": -0.13671875,
513
+ "step": 280
514
+ },
515
+ {
516
+ "epoch": 0.2,
517
+ "grad_norm": 2.3017737036995194,
518
+ "learning_rate": 4.840034038204686e-07,
519
+ "log_odds_chosen": 0.244415283203125,
520
+ "log_odds_ratio": -0.649121105670929,
521
+ "logits/chosen": -2.328125,
522
+ "logits/rejected": -2.4375,
523
+ "logps/chosen": -1.1171875,
524
+ "logps/rejected": -1.2734375,
525
+ "loss": 1.2796,
526
+ "nll_loss": 1.1640625,
527
+ "rewards/accuracies": 0.581250011920929,
528
+ "rewards/chosen": -0.111328125,
529
+ "rewards/margins": 0.0159912109375,
530
+ "rewards/rejected": -0.1279296875,
531
+ "step": 290
532
+ },
533
+ {
534
+ "epoch": 0.21,
535
+ "grad_norm": 2.4440053287886823,
536
+ "learning_rate": 4.817806845804308e-07,
537
+ "log_odds_chosen": 0.03922118991613388,
538
+ "log_odds_ratio": -0.7237304449081421,
539
+ "logits/chosen": -2.28125,
540
+ "logits/rejected": -2.296875,
541
+ "logps/chosen": -1.25,
542
+ "logps/rejected": -1.2734375,
543
+ "loss": 1.3285,
544
+ "nll_loss": 1.25,
545
+ "rewards/accuracies": 0.4749999940395355,
546
+ "rewards/chosen": -0.125,
547
+ "rewards/margins": 0.002655029296875,
548
+ "rewards/rejected": -0.1279296875,
549
+ "step": 300
550
+ },
551
+ {
552
+ "epoch": 0.22,
553
+ "grad_norm": 2.404175576750603,
554
+ "learning_rate": 4.794192178062845e-07,
555
+ "log_odds_chosen": 0.07508544623851776,
556
+ "log_odds_ratio": -0.7403320074081421,
557
+ "logits/chosen": -2.265625,
558
+ "logits/rejected": -2.265625,
559
+ "logps/chosen": -1.2265625,
560
+ "logps/rejected": -1.2890625,
561
+ "loss": 1.2926,
562
+ "nll_loss": 1.2265625,
563
+ "rewards/accuracies": 0.512499988079071,
564
+ "rewards/chosen": -0.123046875,
565
+ "rewards/margins": 0.006195068359375,
566
+ "rewards/rejected": -0.12890625,
567
+ "step": 310
568
+ },
569
+ {
570
+ "epoch": 0.22,
571
+ "grad_norm": 2.4237084015053982,
572
+ "learning_rate": 4.769204171088849e-07,
573
+ "log_odds_chosen": 0.21917724609375,
574
+ "log_odds_ratio": -0.671875,
575
+ "logits/chosen": -2.359375,
576
+ "logits/rejected": -2.4375,
577
+ "logps/chosen": -1.203125,
578
+ "logps/rejected": -1.34375,
579
+ "loss": 1.3217,
580
+ "nll_loss": 1.2578125,
581
+ "rewards/accuracies": 0.5562499761581421,
582
+ "rewards/chosen": -0.1201171875,
583
+ "rewards/margins": 0.0147705078125,
584
+ "rewards/rejected": -0.134765625,
585
+ "step": 320
586
+ },
587
+ {
588
+ "epoch": 0.23,
589
+ "grad_norm": 2.684798684422195,
590
+ "learning_rate": 4.742857783093166e-07,
591
+ "log_odds_chosen": 0.278564453125,
592
+ "log_odds_ratio": -0.6395508050918579,
593
+ "logits/chosen": -2.203125,
594
+ "logits/rejected": -2.328125,
595
+ "logps/chosen": -1.140625,
596
+ "logps/rejected": -1.3359375,
597
+ "loss": 1.2549,
598
+ "nll_loss": 1.1640625,
599
+ "rewards/accuracies": 0.6187499761581421,
600
+ "rewards/chosen": -0.11376953125,
601
+ "rewards/margins": 0.0201416015625,
602
+ "rewards/rejected": -0.1337890625,
603
+ "step": 330
604
+ },
605
+ {
606
+ "epoch": 0.24,
607
+ "grad_norm": 3.6492535837021225,
608
+ "learning_rate": 4.7151687854347157e-07,
609
+ "log_odds_chosen": 0.1578369140625,
610
+ "log_odds_ratio": -0.69482421875,
611
+ "logits/chosen": -2.25,
612
+ "logits/rejected": -2.34375,
613
+ "logps/chosen": -1.1640625,
614
+ "logps/rejected": -1.2734375,
615
+ "loss": 1.2529,
616
+ "nll_loss": 1.1640625,
617
+ "rewards/accuracies": 0.5874999761581421,
618
+ "rewards/chosen": -0.11669921875,
619
+ "rewards/margins": 0.0106201171875,
620
+ "rewards/rejected": -0.126953125,
621
+ "step": 340
622
+ },
623
+ {
624
+ "epoch": 0.25,
625
+ "grad_norm": 2.8702543670932075,
626
+ "learning_rate": 4.6861537531795094e-07,
627
+ "log_odds_chosen": 0.20824584364891052,
628
+ "log_odds_ratio": -0.66015625,
629
+ "logits/chosen": -2.265625,
630
+ "logits/rejected": -2.359375,
631
+ "logps/chosen": -1.21875,
632
+ "logps/rejected": -1.3828125,
633
+ "loss": 1.2876,
634
+ "nll_loss": 1.25,
635
+ "rewards/accuracies": 0.59375,
636
+ "rewards/chosen": -0.1220703125,
637
+ "rewards/margins": 0.016357421875,
638
+ "rewards/rejected": -0.138671875,
639
+ "step": 350
640
+ },
641
+ {
642
+ "epoch": 0.25,
643
+ "grad_norm": 2.847118271789134,
644
+ "learning_rate": 4.655830055178557e-07,
645
+ "log_odds_chosen": 0.23935547471046448,
646
+ "log_odds_ratio": -0.6680663824081421,
647
+ "logits/chosen": -2.1875,
648
+ "logits/rejected": -2.328125,
649
+ "logps/chosen": -1.1328125,
650
+ "logps/rejected": -1.3046875,
651
+ "loss": 1.2766,
652
+ "nll_loss": 1.171875,
653
+ "rewards/accuracies": 0.5625,
654
+ "rewards/chosen": -0.11328125,
655
+ "rewards/margins": 0.016845703125,
656
+ "rewards/rejected": -0.130859375,
657
+ "step": 360
658
+ },
659
+ {
660
+ "epoch": 0.26,
661
+ "grad_norm": 2.510359096280682,
662
+ "learning_rate": 4.6242158436706015e-07,
663
+ "log_odds_chosen": 0.16656494140625,
664
+ "log_odds_ratio": -0.706347644329071,
665
+ "logits/chosen": -2.296875,
666
+ "logits/rejected": -2.359375,
667
+ "logps/chosen": -1.1875,
668
+ "logps/rejected": -1.3046875,
669
+ "loss": 1.2963,
670
+ "nll_loss": 1.2265625,
671
+ "rewards/accuracies": 0.5625,
672
+ "rewards/chosen": -0.119140625,
673
+ "rewards/margins": 0.011474609375,
674
+ "rewards/rejected": -0.130859375,
675
+ "step": 370
676
+ },
677
+ {
678
+ "epoch": 0.27,
679
+ "grad_norm": 2.2404018809724606,
680
+ "learning_rate": 4.5913300434159087e-07,
681
+ "log_odds_chosen": 0.2242431640625,
682
+ "log_odds_ratio": -0.6768554449081421,
683
+ "logits/chosen": -2.28125,
684
+ "logits/rejected": -2.359375,
685
+ "logps/chosen": -1.21875,
686
+ "logps/rejected": -1.390625,
687
+ "loss": 1.293,
688
+ "nll_loss": 1.21875,
689
+ "rewards/accuracies": 0.550000011920929,
690
+ "rewards/chosen": -0.1220703125,
691
+ "rewards/margins": 0.016845703125,
692
+ "rewards/rejected": -0.138671875,
693
+ "step": 380
694
+ },
695
+ {
696
+ "epoch": 0.27,
697
+ "grad_norm": 2.3810451499627434,
698
+ "learning_rate": 4.5571923403676047e-07,
699
+ "log_odds_chosen": 0.30363768339157104,
700
+ "log_odds_ratio": -0.6507812738418579,
701
+ "logits/chosen": -2.1875,
702
+ "logits/rejected": -2.265625,
703
+ "logps/chosen": -1.1328125,
704
+ "logps/rejected": -1.3359375,
705
+ "loss": 1.2838,
706
+ "nll_loss": 1.140625,
707
+ "rewards/accuracies": 0.59375,
708
+ "rewards/chosen": -0.11328125,
709
+ "rewards/margins": 0.0201416015625,
710
+ "rewards/rejected": -0.1328125,
711
+ "step": 390
712
+ },
713
+ {
714
+ "epoch": 0.28,
715
+ "grad_norm": 2.671734360059129,
716
+ "learning_rate": 4.5218231698873627e-07,
717
+ "log_odds_chosen": 0.0892333984375,
718
+ "log_odds_ratio": -0.735058605670929,
719
+ "logits/chosen": -2.296875,
720
+ "logits/rejected": -2.3125,
721
+ "logps/chosen": -1.234375,
722
+ "logps/rejected": -1.3046875,
723
+ "loss": 1.3283,
724
+ "nll_loss": 1.296875,
725
+ "rewards/accuracies": 0.5249999761581421,
726
+ "rewards/chosen": -0.12353515625,
727
+ "rewards/margins": 0.0067138671875,
728
+ "rewards/rejected": -0.1298828125,
729
+ "step": 400
730
+ },
731
+ {
732
+ "epoch": 0.29,
733
+ "grad_norm": 2.5118185799490043,
734
+ "learning_rate": 4.4852437045124735e-07,
735
+ "log_odds_chosen": 0.08201904594898224,
736
+ "log_odds_ratio": -0.7183593511581421,
737
+ "logits/chosen": -2.1875,
738
+ "logits/rejected": -2.3125,
739
+ "logps/chosen": -1.21875,
740
+ "logps/rejected": -1.265625,
741
+ "loss": 1.2777,
742
+ "nll_loss": 1.25,
743
+ "rewards/accuracies": 0.5062500238418579,
744
+ "rewards/chosen": -0.12158203125,
745
+ "rewards/margins": 0.0047607421875,
746
+ "rewards/rejected": -0.1259765625,
747
+ "step": 410
748
+ },
749
+ {
750
+ "epoch": 0.29,
751
+ "grad_norm": 2.5063550386397604,
752
+ "learning_rate": 4.447475841281635e-07,
753
+ "log_odds_chosen": 0.2621704041957855,
754
+ "log_odds_ratio": -0.6456054449081421,
755
+ "logits/chosen": -2.203125,
756
+ "logits/rejected": -2.296875,
757
+ "logps/chosen": -1.1796875,
758
+ "logps/rejected": -1.3515625,
759
+ "loss": 1.256,
760
+ "nll_loss": 1.25,
761
+ "rewards/accuracies": 0.581250011920929,
762
+ "rewards/chosen": -0.11767578125,
763
+ "rewards/margins": 0.01708984375,
764
+ "rewards/rejected": -0.134765625,
765
+ "step": 420
766
+ },
767
+ {
768
+ "epoch": 0.3,
769
+ "grad_norm": 2.3631407616001243,
770
+ "learning_rate": 4.408542188627044e-07,
771
+ "log_odds_chosen": 0.12116088718175888,
772
+ "log_odds_ratio": -0.7083984613418579,
773
+ "logits/chosen": -2.3125,
774
+ "logits/rejected": -2.328125,
775
+ "logps/chosen": -1.1796875,
776
+ "logps/rejected": -1.265625,
777
+ "loss": 1.2792,
778
+ "nll_loss": 1.1953125,
779
+ "rewards/accuracies": 0.5062500238418579,
780
+ "rewards/chosen": -0.1181640625,
781
+ "rewards/margins": 0.00860595703125,
782
+ "rewards/rejected": -0.126953125,
783
+ "step": 430
784
+ },
785
+ {
786
+ "epoch": 0.31,
787
+ "grad_norm": 2.6616497499686393,
788
+ "learning_rate": 4.368466052840636e-07,
789
+ "log_odds_chosen": 0.17690429091453552,
790
+ "log_odds_ratio": -0.6680663824081421,
791
+ "logits/chosen": -2.1875,
792
+ "logits/rejected": -2.3125,
793
+ "logps/chosen": -1.171875,
794
+ "logps/rejected": -1.296875,
795
+ "loss": 1.2943,
796
+ "nll_loss": 1.2109375,
797
+ "rewards/accuracies": 0.543749988079071,
798
+ "rewards/chosen": -0.11669921875,
799
+ "rewards/margins": 0.01300048828125,
800
+ "rewards/rejected": -0.1298828125,
801
+ "step": 440
802
+ },
803
+ {
804
+ "epoch": 0.32,
805
+ "grad_norm": 2.5349037092042406,
806
+ "learning_rate": 4.327271424122573e-07,
807
+ "log_odds_chosen": 0.2679077088832855,
808
+ "log_odds_ratio": -0.6576172113418579,
809
+ "logits/chosen": -2.1875,
810
+ "logits/rejected": -2.28125,
811
+ "logps/chosen": -1.1328125,
812
+ "logps/rejected": -1.3359375,
813
+ "loss": 1.2823,
814
+ "nll_loss": 1.1796875,
815
+ "rewards/accuracies": 0.5874999761581421,
816
+ "rewards/chosen": -0.11328125,
817
+ "rewards/margins": 0.019775390625,
818
+ "rewards/rejected": -0.1328125,
819
+ "step": 450
820
+ },
821
+ {
822
+ "epoch": 0.32,
823
+ "grad_norm": 2.485655487242489,
824
+ "learning_rate": 4.284982962220337e-07,
825
+ "log_odds_chosen": 0.212890625,
826
+ "log_odds_ratio": -0.680371105670929,
827
+ "logits/chosen": -2.15625,
828
+ "logits/rejected": -2.203125,
829
+ "logps/chosen": -1.15625,
830
+ "logps/rejected": -1.296875,
831
+ "loss": 1.2641,
832
+ "nll_loss": 1.171875,
833
+ "rewards/accuracies": 0.6000000238418579,
834
+ "rewards/chosen": -0.115234375,
835
+ "rewards/margins": 0.0140380859375,
836
+ "rewards/rejected": -0.12890625,
837
+ "step": 460
838
+ },
839
+ {
840
+ "epoch": 0.33,
841
+ "grad_norm": 2.569720946707249,
842
+ "learning_rate": 4.241625981667023e-07,
843
+ "log_odds_chosen": 0.20276489853858948,
844
+ "log_odds_ratio": -0.681933581829071,
845
+ "logits/chosen": -2.25,
846
+ "logits/rejected": -2.28125,
847
+ "logps/chosen": -1.1875,
848
+ "logps/rejected": -1.3125,
849
+ "loss": 1.2584,
850
+ "nll_loss": 1.21875,
851
+ "rewards/accuracies": 0.59375,
852
+ "rewards/chosen": -0.11865234375,
853
+ "rewards/margins": 0.01312255859375,
854
+ "rewards/rejected": -0.1318359375,
855
+ "step": 470
856
+ },
857
+ {
858
+ "epoch": 0.34,
859
+ "grad_norm": 2.6019962569531323,
860
+ "learning_rate": 4.19722643662766e-07,
861
+ "log_odds_chosen": 0.2719970643520355,
862
+ "log_odds_ratio": -0.64111328125,
863
+ "logits/chosen": -2.203125,
864
+ "logits/rejected": -2.28125,
865
+ "logps/chosen": -1.1328125,
866
+ "logps/rejected": -1.328125,
867
+ "loss": 1.2457,
868
+ "nll_loss": 1.15625,
869
+ "rewards/accuracies": 0.6499999761581421,
870
+ "rewards/chosen": -0.11279296875,
871
+ "rewards/margins": 0.01953125,
872
+ "rewards/rejected": -0.1328125,
873
+ "step": 480
874
+ },
875
+ {
876
+ "epoch": 0.34,
877
+ "grad_norm": 2.378793105209263,
878
+ "learning_rate": 4.151810905362643e-07,
879
+ "log_odds_chosen": 0.15797118842601776,
880
+ "log_odds_ratio": -0.710156261920929,
881
+ "logits/chosen": -2.109375,
882
+ "logits/rejected": -2.25,
883
+ "logps/chosen": -1.1484375,
884
+ "logps/rejected": -1.25,
885
+ "loss": 1.2691,
886
+ "nll_loss": 1.1640625,
887
+ "rewards/accuracies": 0.5375000238418579,
888
+ "rewards/chosen": -0.11474609375,
889
+ "rewards/margins": 0.01019287109375,
890
+ "rewards/rejected": -0.125,
891
+ "step": 490
892
+ },
893
+ {
894
+ "epoch": 0.35,
895
+ "grad_norm": 2.461627110160161,
896
+ "learning_rate": 4.105406574317578e-07,
897
+ "log_odds_chosen": 0.16668701171875,
898
+ "log_odds_ratio": -0.723925769329071,
899
+ "logits/chosen": -2.09375,
900
+ "logits/rejected": -2.171875,
901
+ "logps/chosen": -1.2265625,
902
+ "logps/rejected": -1.34375,
903
+ "loss": 1.2486,
904
+ "nll_loss": 1.25,
905
+ "rewards/accuracies": 0.4937500059604645,
906
+ "rewards/chosen": -0.1220703125,
907
+ "rewards/margins": 0.01226806640625,
908
+ "rewards/rejected": -0.134765625,
909
+ "step": 500
910
+ },
911
+ {
912
+ "epoch": 0.36,
913
+ "grad_norm": 2.3991709623693485,
914
+ "learning_rate": 4.0580412218490424e-07,
915
+ "log_odds_chosen": 0.3194335997104645,
916
+ "log_odds_ratio": -0.61669921875,
917
+ "logits/chosen": -2.15625,
918
+ "logits/rejected": -2.25,
919
+ "logps/chosen": -1.125,
920
+ "logps/rejected": -1.34375,
921
+ "loss": 1.261,
922
+ "nll_loss": 1.203125,
923
+ "rewards/accuracies": 0.625,
924
+ "rewards/chosen": -0.11279296875,
925
+ "rewards/margins": 0.0218505859375,
926
+ "rewards/rejected": -0.134765625,
927
+ "step": 510
928
+ },
929
+ {
930
+ "epoch": 0.36,
931
+ "grad_norm": 2.420136778893825,
932
+ "learning_rate": 4.009743201596034e-07,
933
+ "log_odds_chosen": 0.32293701171875,
934
+ "log_odds_ratio": -0.66357421875,
935
+ "logits/chosen": -2.203125,
936
+ "logits/rejected": -2.296875,
937
+ "logps/chosen": -1.140625,
938
+ "logps/rejected": -1.3671875,
939
+ "loss": 1.279,
940
+ "nll_loss": 1.203125,
941
+ "rewards/accuracies": 0.5625,
942
+ "rewards/chosen": -0.1142578125,
943
+ "rewards/margins": 0.0223388671875,
944
+ "rewards/rejected": -0.13671875,
945
+ "step": 520
946
+ },
947
+ {
948
+ "epoch": 0.37,
949
+ "grad_norm": 2.576899517162626,
950
+ "learning_rate": 3.960541425507039e-07,
951
+ "log_odds_chosen": 0.18942871689796448,
952
+ "log_odds_ratio": -0.6888672113418579,
953
+ "logits/chosen": -2.171875,
954
+ "logits/rejected": -2.296875,
955
+ "logps/chosen": -1.171875,
956
+ "logps/rejected": -1.3203125,
957
+ "loss": 1.3049,
958
+ "nll_loss": 1.2265625,
959
+ "rewards/accuracies": 0.512499988079071,
960
+ "rewards/chosen": -0.1171875,
961
+ "rewards/margins": 0.01458740234375,
962
+ "rewards/rejected": -0.1318359375,
963
+ "step": 530
964
+ },
965
+ {
966
+ "epoch": 0.38,
967
+ "grad_norm": 2.7625972509957144,
968
+ "learning_rate": 3.9104653465328853e-07,
969
+ "log_odds_chosen": 0.09678955376148224,
970
+ "log_odds_ratio": -0.721875011920929,
971
+ "logits/chosen": -2.109375,
972
+ "logits/rejected": -2.1875,
973
+ "logps/chosen": -1.203125,
974
+ "logps/rejected": -1.265625,
975
+ "loss": 1.2931,
976
+ "nll_loss": 1.234375,
977
+ "rewards/accuracies": 0.550000011920929,
978
+ "rewards/chosen": -0.1201171875,
979
+ "rewards/margins": 0.006103515625,
980
+ "rewards/rejected": -0.1259765625,
981
+ "step": 540
982
+ },
983
+ {
984
+ "epoch": 0.39,
985
+ "grad_norm": 2.45849449078066,
986
+ "learning_rate": 3.8595449409957516e-07,
987
+ "log_odds_chosen": 0.20249633491039276,
988
+ "log_odds_ratio": -0.668749988079071,
989
+ "logits/chosen": -1.9921875,
990
+ "logits/rejected": -2.109375,
991
+ "logps/chosen": -1.1015625,
992
+ "logps/rejected": -1.2421875,
993
+ "loss": 1.2424,
994
+ "nll_loss": 1.125,
995
+ "rewards/accuracies": 0.5625,
996
+ "rewards/chosen": -0.1103515625,
997
+ "rewards/margins": 0.01397705078125,
998
+ "rewards/rejected": -0.1240234375,
999
+ "step": 550
1000
+ },
1001
+ {
1002
+ "epoch": 0.39,
1003
+ "grad_norm": 2.336077999068494,
1004
+ "learning_rate": 3.807810690644868e-07,
1005
+ "log_odds_chosen": 0.15464477241039276,
1006
+ "log_odds_ratio": -0.6904296875,
1007
+ "logits/chosen": -2.0625,
1008
+ "logits/rejected": -2.1875,
1009
+ "logps/chosen": -1.1875,
1010
+ "logps/rejected": -1.3046875,
1011
+ "loss": 1.2639,
1012
+ "nll_loss": 1.2109375,
1013
+ "rewards/accuracies": 0.550000011920929,
1014
+ "rewards/chosen": -0.119140625,
1015
+ "rewards/margins": 0.011474609375,
1016
+ "rewards/rejected": -0.1298828125,
1017
+ "step": 560
1018
+ },
1019
+ {
1020
+ "epoch": 0.4,
1021
+ "grad_norm": 2.8019381749527845,
1022
+ "learning_rate": 3.75529356440967e-07,
1023
+ "log_odds_chosen": -0.002349853515625,
1024
+ "log_odds_ratio": -0.763867199420929,
1025
+ "logits/chosen": -2.140625,
1026
+ "logits/rejected": -2.1875,
1027
+ "logps/chosen": -1.1875,
1028
+ "logps/rejected": -1.203125,
1029
+ "loss": 1.2792,
1030
+ "nll_loss": 1.21875,
1031
+ "rewards/accuracies": 0.5,
1032
+ "rewards/chosen": -0.119140625,
1033
+ "rewards/margins": 0.0014190673828125,
1034
+ "rewards/rejected": -0.12060546875,
1035
+ "step": 570
1036
+ },
1037
+ {
1038
+ "epoch": 0.41,
1039
+ "grad_norm": 2.904460591649426,
1040
+ "learning_rate": 3.702024999861312e-07,
1041
+ "log_odds_chosen": 0.20991210639476776,
1042
+ "log_odds_ratio": -0.690234363079071,
1043
+ "logits/chosen": -2.140625,
1044
+ "logits/rejected": -2.203125,
1045
+ "logps/chosen": -1.1484375,
1046
+ "logps/rejected": -1.3125,
1047
+ "loss": 1.2689,
1048
+ "nll_loss": 1.21875,
1049
+ "rewards/accuracies": 0.5687500238418579,
1050
+ "rewards/chosen": -0.115234375,
1051
+ "rewards/margins": 0.0157470703125,
1052
+ "rewards/rejected": -0.130859375,
1053
+ "step": 580
1054
+ },
1055
+ {
1056
+ "epoch": 0.41,
1057
+ "grad_norm": 2.839022537148172,
1058
+ "learning_rate": 3.648036884393646e-07,
1059
+ "log_odds_chosen": 0.11424560844898224,
1060
+ "log_odds_ratio": -0.715136706829071,
1061
+ "logits/chosen": -2.078125,
1062
+ "logits/rejected": -2.234375,
1063
+ "logps/chosen": -1.1953125,
1064
+ "logps/rejected": -1.28125,
1065
+ "loss": 1.2823,
1066
+ "nll_loss": 1.234375,
1067
+ "rewards/accuracies": 0.4937500059604645,
1068
+ "rewards/chosen": -0.119140625,
1069
+ "rewards/margins": 0.008544921875,
1070
+ "rewards/rejected": -0.1279296875,
1071
+ "step": 590
1072
+ },
1073
+ {
1074
+ "epoch": 0.42,
1075
+ "grad_norm": 2.6089925230689737,
1076
+ "learning_rate": 3.593361536134931e-07,
1077
+ "log_odds_chosen": 0.2776428163051605,
1078
+ "log_odds_ratio": -0.6419922113418579,
1079
+ "logits/chosen": -1.9765625,
1080
+ "logits/rejected": -2.109375,
1081
+ "logps/chosen": -1.0859375,
1082
+ "logps/rejected": -1.2734375,
1083
+ "loss": 1.2798,
1084
+ "nll_loss": 1.1640625,
1085
+ "rewards/accuracies": 0.625,
1086
+ "rewards/chosen": -0.1083984375,
1087
+ "rewards/margins": 0.0189208984375,
1088
+ "rewards/rejected": -0.126953125,
1089
+ "step": 600
1090
+ },
1091
+ {
1092
+ "epoch": 0.43,
1093
+ "grad_norm": 2.603291469222202,
1094
+ "learning_rate": 3.538031684601698e-07,
1095
+ "log_odds_chosen": 0.06999512016773224,
1096
+ "log_odds_ratio": -0.740429699420929,
1097
+ "logits/chosen": -2.125,
1098
+ "logits/rejected": -2.234375,
1099
+ "logps/chosen": -1.171875,
1100
+ "logps/rejected": -1.234375,
1101
+ "loss": 1.2504,
1102
+ "nll_loss": 1.1796875,
1103
+ "rewards/accuracies": 0.48750001192092896,
1104
+ "rewards/chosen": -0.1171875,
1105
+ "rewards/margins": 0.006103515625,
1106
+ "rewards/rejected": -0.123046875,
1107
+ "step": 610
1108
+ },
1109
+ {
1110
+ "epoch": 0.43,
1111
+ "grad_norm": 2.524850145022778,
1112
+ "learning_rate": 3.4820804511063496e-07,
1113
+ "log_odds_chosen": 0.2641845643520355,
1114
+ "log_odds_ratio": -0.642285168170929,
1115
+ "logits/chosen": -2.015625,
1116
+ "logits/rejected": -2.09375,
1117
+ "logps/chosen": -1.09375,
1118
+ "logps/rejected": -1.25,
1119
+ "loss": 1.2574,
1120
+ "nll_loss": 1.140625,
1121
+ "rewards/accuracies": 0.5874999761581421,
1122
+ "rewards/chosen": -0.10888671875,
1123
+ "rewards/margins": 0.0157470703125,
1124
+ "rewards/rejected": -0.125,
1125
+ "step": 620
1126
+ },
1127
+ {
1128
+ "epoch": 0.44,
1129
+ "grad_norm": 2.821337451607177,
1130
+ "learning_rate": 3.425541328930226e-07,
1131
+ "log_odds_chosen": 0.26170653104782104,
1132
+ "log_odds_ratio": -0.6304687261581421,
1133
+ "logits/chosen": -2.046875,
1134
+ "logits/rejected": -2.15625,
1135
+ "logps/chosen": -1.125,
1136
+ "logps/rejected": -1.3125,
1137
+ "loss": 1.2199,
1138
+ "nll_loss": 1.15625,
1139
+ "rewards/accuracies": 0.625,
1140
+ "rewards/chosen": -0.11279296875,
1141
+ "rewards/margins": 0.0184326171875,
1142
+ "rewards/rejected": -0.130859375,
1143
+ "step": 630
1144
+ },
1145
+ {
1146
+ "epoch": 0.45,
1147
+ "grad_norm": 2.5621913998689734,
1148
+ "learning_rate": 3.3684481632740025e-07,
1149
+ "log_odds_chosen": 0.131927490234375,
1150
+ "log_odds_ratio": -0.692578136920929,
1151
+ "logits/chosen": -2.1875,
1152
+ "logits/rejected": -2.21875,
1153
+ "logps/chosen": -1.1875,
1154
+ "logps/rejected": -1.2734375,
1155
+ "loss": 1.2889,
1156
+ "nll_loss": 1.2421875,
1157
+ "rewards/accuracies": 0.53125,
1158
+ "rewards/chosen": -0.119140625,
1159
+ "rewards/margins": 0.0086669921875,
1160
+ "rewards/rejected": -0.1279296875,
1161
+ "step": 640
1162
+ },
1163
+ {
1164
+ "epoch": 0.46,
1165
+ "grad_norm": 2.556936502980912,
1166
+ "learning_rate": 3.310835130997428e-07,
1167
+ "log_odds_chosen": 0.18841552734375,
1168
+ "log_odds_ratio": -0.67431640625,
1169
+ "logits/chosen": -2.125,
1170
+ "logits/rejected": -2.234375,
1171
+ "logps/chosen": -1.1953125,
1172
+ "logps/rejected": -1.3359375,
1173
+ "loss": 1.3018,
1174
+ "nll_loss": 1.2578125,
1175
+ "rewards/accuracies": 0.5687500238418579,
1176
+ "rewards/chosen": -0.11962890625,
1177
+ "rewards/margins": 0.01348876953125,
1178
+ "rewards/rejected": -0.1337890625,
1179
+ "step": 650
1180
+ },
1181
+ {
1182
+ "epoch": 0.46,
1183
+ "grad_norm": 2.509648126252326,
1184
+ "learning_rate": 3.2527367201605215e-07,
1185
+ "log_odds_chosen": 0.28242188692092896,
1186
+ "log_odds_ratio": -0.6371093988418579,
1187
+ "logits/chosen": -2.0625,
1188
+ "logits/rejected": -2.15625,
1189
+ "logps/chosen": -1.15625,
1190
+ "logps/rejected": -1.3359375,
1191
+ "loss": 1.2715,
1192
+ "nll_loss": 1.2109375,
1193
+ "rewards/accuracies": 0.5562499761581421,
1194
+ "rewards/chosen": -0.115234375,
1195
+ "rewards/margins": 0.01806640625,
1196
+ "rewards/rejected": -0.1337890625,
1197
+ "step": 660
1198
+ },
1199
+ {
1200
+ "epoch": 0.47,
1201
+ "grad_norm": 2.572686228637093,
1202
+ "learning_rate": 3.194187709378479e-07,
1203
+ "log_odds_chosen": 0.10871581733226776,
1204
+ "log_odds_ratio": -0.723828136920929,
1205
+ "logits/chosen": -2.09375,
1206
+ "logits/rejected": -2.171875,
1207
+ "logps/chosen": -1.171875,
1208
+ "logps/rejected": -1.2421875,
1209
+ "loss": 1.2564,
1210
+ "nll_loss": 1.2421875,
1211
+ "rewards/accuracies": 0.5062500238418579,
1212
+ "rewards/chosen": -0.11669921875,
1213
+ "rewards/margins": 0.00714111328125,
1214
+ "rewards/rejected": -0.1240234375,
1215
+ "step": 670
1216
+ },
1217
+ {
1218
+ "epoch": 0.48,
1219
+ "grad_norm": 2.48714277601033,
1220
+ "learning_rate": 3.135223147002658e-07,
1221
+ "log_odds_chosen": 0.17313842475414276,
1222
+ "log_odds_ratio": -0.698437511920929,
1223
+ "logits/chosen": -2.15625,
1224
+ "logits/rejected": -2.21875,
1225
+ "logps/chosen": -1.1328125,
1226
+ "logps/rejected": -1.2578125,
1227
+ "loss": 1.2525,
1228
+ "nll_loss": 1.1875,
1229
+ "rewards/accuracies": 0.5249999761581421,
1230
+ "rewards/chosen": -0.11328125,
1231
+ "rewards/margins": 0.012451171875,
1232
+ "rewards/rejected": -0.1259765625,
1233
+ "step": 680
1234
+ },
1235
+ {
1236
+ "epoch": 0.48,
1237
+ "grad_norm": 2.5499221997702164,
1238
+ "learning_rate": 3.075878330140079e-07,
1239
+ "log_odds_chosen": 0.22934570908546448,
1240
+ "log_odds_ratio": -0.6927734613418579,
1241
+ "logits/chosen": -2.109375,
1242
+ "logits/rejected": -2.1875,
1243
+ "logps/chosen": -1.203125,
1244
+ "logps/rejected": -1.3515625,
1245
+ "loss": 1.3172,
1246
+ "nll_loss": 1.25,
1247
+ "rewards/accuracies": 0.5625,
1248
+ "rewards/chosen": -0.12060546875,
1249
+ "rewards/margins": 0.01470947265625,
1250
+ "rewards/rejected": -0.134765625,
1251
+ "step": 690
1252
+ },
1253
+ {
1254
+ "epoch": 0.49,
1255
+ "grad_norm": 2.4807391943530206,
1256
+ "learning_rate": 3.0161887835240353e-07,
1257
+ "log_odds_chosen": 0.16926269233226776,
1258
+ "log_odds_ratio": -0.692578136920929,
1259
+ "logits/chosen": -2.078125,
1260
+ "logits/rejected": -2.140625,
1261
+ "logps/chosen": -1.1640625,
1262
+ "logps/rejected": -1.28125,
1263
+ "loss": 1.2796,
1264
+ "nll_loss": 1.21875,
1265
+ "rewards/accuracies": 0.550000011920929,
1266
+ "rewards/chosen": -0.11669921875,
1267
+ "rewards/margins": 0.0113525390625,
1268
+ "rewards/rejected": -0.1279296875,
1269
+ "step": 700
1270
+ },
1271
+ {
1272
+ "epoch": 0.5,
1273
+ "grad_norm": 2.789334396425448,
1274
+ "learning_rate": 2.956190238248425e-07,
1275
+ "log_odds_chosen": 0.1761474609375,
1276
+ "log_odds_ratio": -0.71728515625,
1277
+ "logits/chosen": -2.0625,
1278
+ "logits/rejected": -2.125,
1279
+ "logps/chosen": -1.15625,
1280
+ "logps/rejected": -1.28125,
1281
+ "loss": 1.2759,
1282
+ "nll_loss": 1.265625,
1283
+ "rewards/accuracies": 0.5249999761581421,
1284
+ "rewards/chosen": -0.11572265625,
1285
+ "rewards/margins": 0.0126953125,
1286
+ "rewards/rejected": -0.1279296875,
1287
+ "step": 710
1288
+ },
1289
+ {
1290
+ "epoch": 0.5,
1291
+ "grad_norm": 2.4358280165677817,
1292
+ "learning_rate": 2.8959186103785694e-07,
1293
+ "log_odds_chosen": 0.3479553163051605,
1294
+ "log_odds_ratio": -0.630664050579071,
1295
+ "logits/chosen": -2.0,
1296
+ "logits/rejected": -2.078125,
1297
+ "logps/chosen": -1.0625,
1298
+ "logps/rejected": -1.3046875,
1299
+ "loss": 1.2165,
1300
+ "nll_loss": 1.125,
1301
+ "rewards/accuracies": 0.612500011920929,
1302
+ "rewards/chosen": -0.10595703125,
1303
+ "rewards/margins": 0.0245361328125,
1304
+ "rewards/rejected": -0.130859375,
1305
+ "step": 720
1306
+ },
1307
+ {
1308
+ "epoch": 0.51,
1309
+ "grad_norm": 2.88074218743752,
1310
+ "learning_rate": 2.8354099794512876e-07,
1311
+ "log_odds_chosen": 0.14354248344898224,
1312
+ "log_odds_ratio": -0.6961914300918579,
1313
+ "logits/chosen": -2.078125,
1314
+ "logits/rejected": -2.21875,
1315
+ "logps/chosen": -1.109375,
1316
+ "logps/rejected": -1.1953125,
1317
+ "loss": 1.277,
1318
+ "nll_loss": 1.1640625,
1319
+ "rewards/accuracies": 0.5687500238418579,
1320
+ "rewards/chosen": -0.111328125,
1321
+ "rewards/margins": 0.00830078125,
1322
+ "rewards/rejected": -0.11962890625,
1323
+ "step": 730
1324
+ },
1325
+ {
1326
+ "epoch": 0.52,
1327
+ "grad_norm": 2.511351318161689,
1328
+ "learning_rate": 2.774700566877129e-07,
1329
+ "log_odds_chosen": 0.21488037705421448,
1330
+ "log_odds_ratio": -0.669921875,
1331
+ "logits/chosen": -2.046875,
1332
+ "logits/rejected": -2.109375,
1333
+ "logps/chosen": -1.140625,
1334
+ "logps/rejected": -1.3046875,
1335
+ "loss": 1.2247,
1336
+ "nll_loss": 1.1796875,
1337
+ "rewards/accuracies": 0.5687500238418579,
1338
+ "rewards/chosen": -0.1142578125,
1339
+ "rewards/margins": 0.0164794921875,
1340
+ "rewards/rejected": -0.130859375,
1341
+ "step": 740
1342
+ },
1343
+ {
1344
+ "epoch": 0.53,
1345
+ "grad_norm": 3.010290770764348,
1346
+ "learning_rate": 2.713826714257669e-07,
1347
+ "log_odds_chosen": 0.19861450791358948,
1348
+ "log_odds_ratio": -0.6786133050918579,
1349
+ "logits/chosen": -2.109375,
1350
+ "logits/rejected": -2.109375,
1351
+ "logps/chosen": -1.1796875,
1352
+ "logps/rejected": -1.3046875,
1353
+ "loss": 1.236,
1354
+ "nll_loss": 1.2109375,
1355
+ "rewards/accuracies": 0.5562499761581421,
1356
+ "rewards/chosen": -0.11767578125,
1357
+ "rewards/margins": 0.01239013671875,
1358
+ "rewards/rejected": -0.1298828125,
1359
+ "step": 750
1360
+ },
1361
+ {
1362
+ "epoch": 0.53,
1363
+ "grad_norm": 2.27093551536218,
1364
+ "learning_rate": 2.6528248616308595e-07,
1365
+ "log_odds_chosen": 0.31080931425094604,
1366
+ "log_odds_ratio": -0.6333984136581421,
1367
+ "logits/chosen": -2.0625,
1368
+ "logits/rejected": -2.140625,
1369
+ "logps/chosen": -1.1484375,
1370
+ "logps/rejected": -1.3671875,
1371
+ "loss": 1.2256,
1372
+ "nll_loss": 1.1953125,
1373
+ "rewards/accuracies": 0.612500011920929,
1374
+ "rewards/chosen": -0.115234375,
1375
+ "rewards/margins": 0.0216064453125,
1376
+ "rewards/rejected": -0.13671875,
1377
+ "step": 760
1378
+ },
1379
+ {
1380
+ "epoch": 0.54,
1381
+ "grad_norm": 2.5700766425464,
1382
+ "learning_rate": 2.591731525657454e-07,
1383
+ "log_odds_chosen": 0.2696166932582855,
1384
+ "log_odds_ratio": -0.6527343988418579,
1385
+ "logits/chosen": -2.046875,
1386
+ "logits/rejected": -2.140625,
1387
+ "logps/chosen": -1.1171875,
1388
+ "logps/rejected": -1.296875,
1389
+ "loss": 1.2227,
1390
+ "nll_loss": 1.1484375,
1391
+ "rewards/accuracies": 0.6312500238418579,
1392
+ "rewards/chosen": -0.11181640625,
1393
+ "rewards/margins": 0.0177001953125,
1394
+ "rewards/rejected": -0.1298828125,
1395
+ "step": 770
1396
+ },
1397
+ {
1398
+ "epoch": 0.55,
1399
+ "grad_norm": 3.3277253612496933,
1400
+ "learning_rate": 2.5305832777615644e-07,
1401
+ "log_odds_chosen": 0.19566650688648224,
1402
+ "log_odds_ratio": -0.688183605670929,
1403
+ "logits/chosen": -2.109375,
1404
+ "logits/rejected": -2.1875,
1405
+ "logps/chosen": -1.1640625,
1406
+ "logps/rejected": -1.296875,
1407
+ "loss": 1.2602,
1408
+ "nll_loss": 1.25,
1409
+ "rewards/accuracies": 0.550000011920929,
1410
+ "rewards/chosen": -0.1162109375,
1411
+ "rewards/margins": 0.01287841796875,
1412
+ "rewards/rejected": -0.12890625,
1413
+ "step": 780
1414
+ },
1415
+ {
1416
+ "epoch": 0.55,
1417
+ "grad_norm": 2.847511378794556,
1418
+ "learning_rate": 2.469416722238436e-07,
1419
+ "log_odds_chosen": 0.1658935546875,
1420
+ "log_odds_ratio": -0.706738293170929,
1421
+ "logits/chosen": -2.078125,
1422
+ "logits/rejected": -2.203125,
1423
+ "logps/chosen": -1.1875,
1424
+ "logps/rejected": -1.296875,
1425
+ "loss": 1.2591,
1426
+ "nll_loss": 1.21875,
1427
+ "rewards/accuracies": 0.5625,
1428
+ "rewards/chosen": -0.119140625,
1429
+ "rewards/margins": 0.01080322265625,
1430
+ "rewards/rejected": -0.1298828125,
1431
+ "step": 790
1432
+ },
1433
+ {
1434
+ "epoch": 0.56,
1435
+ "grad_norm": 2.6549295176096672,
1436
+ "learning_rate": 2.4082684743425454e-07,
1437
+ "log_odds_chosen": 0.08632812649011612,
1438
+ "log_odds_ratio": -0.722851574420929,
1439
+ "logits/chosen": -2.0625,
1440
+ "logits/rejected": -2.15625,
1441
+ "logps/chosen": -1.140625,
1442
+ "logps/rejected": -1.203125,
1443
+ "loss": 1.2585,
1444
+ "nll_loss": 1.1875,
1445
+ "rewards/accuracies": 0.53125,
1446
+ "rewards/chosen": -0.11376953125,
1447
+ "rewards/margins": 0.0067138671875,
1448
+ "rewards/rejected": -0.12060546875,
1449
+ "step": 800
1450
+ },
1451
+ {
1452
+ "epoch": 0.57,
1453
+ "grad_norm": 2.743059460411326,
1454
+ "learning_rate": 2.3471751383691403e-07,
1455
+ "log_odds_chosen": 0.22829589247703552,
1456
+ "log_odds_ratio": -0.6942383050918579,
1457
+ "logits/chosen": -2.046875,
1458
+ "logits/rejected": -2.125,
1459
+ "logps/chosen": -1.140625,
1460
+ "logps/rejected": -1.2890625,
1461
+ "loss": 1.2288,
1462
+ "nll_loss": 1.1875,
1463
+ "rewards/accuracies": 0.5874999761581421,
1464
+ "rewards/chosen": -0.1142578125,
1465
+ "rewards/margins": 0.014892578125,
1466
+ "rewards/rejected": -0.12890625,
1467
+ "step": 810
1468
+ },
1469
+ {
1470
+ "epoch": 0.57,
1471
+ "grad_norm": 2.5772075525222253,
1472
+ "learning_rate": 2.2861732857423306e-07,
1473
+ "log_odds_chosen": 0.27777099609375,
1474
+ "log_odds_ratio": -0.6568359136581421,
1475
+ "logits/chosen": -1.9609375,
1476
+ "logits/rejected": -2.015625,
1477
+ "logps/chosen": -1.0859375,
1478
+ "logps/rejected": -1.28125,
1479
+ "loss": 1.2301,
1480
+ "nll_loss": 1.140625,
1481
+ "rewards/accuracies": 0.6312500238418579,
1482
+ "rewards/chosen": -0.1083984375,
1483
+ "rewards/margins": 0.02001953125,
1484
+ "rewards/rejected": -0.12890625,
1485
+ "step": 820
1486
+ },
1487
+ {
1488
+ "epoch": 0.58,
1489
+ "grad_norm": 2.9548400351876465,
1490
+ "learning_rate": 2.225299433122871e-07,
1491
+ "log_odds_chosen": 0.19624023139476776,
1492
+ "log_odds_ratio": -0.6937500238418579,
1493
+ "logits/chosen": -2.046875,
1494
+ "logits/rejected": -2.109375,
1495
+ "logps/chosen": -1.21875,
1496
+ "logps/rejected": -1.3828125,
1497
+ "loss": 1.2425,
1498
+ "nll_loss": 1.21875,
1499
+ "rewards/accuracies": 0.543749988079071,
1500
+ "rewards/chosen": -0.1220703125,
1501
+ "rewards/margins": 0.015869140625,
1502
+ "rewards/rejected": -0.1376953125,
1503
+ "step": 830
1504
+ },
1505
+ {
1506
+ "epoch": 0.59,
1507
+ "grad_norm": 3.0054061279771362,
1508
+ "learning_rate": 2.1645900205487122e-07,
1509
+ "log_odds_chosen": 0.21121826767921448,
1510
+ "log_odds_ratio": -0.654492199420929,
1511
+ "logits/chosen": -2.03125,
1512
+ "logits/rejected": -2.0625,
1513
+ "logps/chosen": -1.09375,
1514
+ "logps/rejected": -1.234375,
1515
+ "loss": 1.2646,
1516
+ "nll_loss": 1.140625,
1517
+ "rewards/accuracies": 0.612500011920929,
1518
+ "rewards/chosen": -0.10888671875,
1519
+ "rewards/margins": 0.014892578125,
1520
+ "rewards/rejected": -0.12353515625,
1521
+ "step": 840
1522
+ },
1523
+ {
1524
+ "epoch": 0.6,
1525
+ "grad_norm": 2.6164133362622484,
1526
+ "learning_rate": 2.1040813896214307e-07,
1527
+ "log_odds_chosen": 0.150909423828125,
1528
+ "log_odds_ratio": -0.6998046636581421,
1529
+ "logits/chosen": -2.0,
1530
+ "logits/rejected": -2.078125,
1531
+ "logps/chosen": -1.1171875,
1532
+ "logps/rejected": -1.234375,
1533
+ "loss": 1.263,
1534
+ "nll_loss": 1.15625,
1535
+ "rewards/accuracies": 0.59375,
1536
+ "rewards/chosen": -0.11181640625,
1537
+ "rewards/margins": 0.0118408203125,
1538
+ "rewards/rejected": -0.12353515625,
1539
+ "step": 850
1540
+ },
1541
+ {
1542
+ "epoch": 0.6,
1543
+ "grad_norm": 2.507875262223547,
1544
+ "learning_rate": 2.0438097617515741e-07,
1545
+ "log_odds_chosen": 0.241455078125,
1546
+ "log_odds_ratio": -0.6556640863418579,
1547
+ "logits/chosen": -1.890625,
1548
+ "logits/rejected": -2.015625,
1549
+ "logps/chosen": -1.0859375,
1550
+ "logps/rejected": -1.2265625,
1551
+ "loss": 1.2638,
1552
+ "nll_loss": 1.1484375,
1553
+ "rewards/accuracies": 0.543749988079071,
1554
+ "rewards/chosen": -0.10888671875,
1555
+ "rewards/margins": 0.01409912109375,
1556
+ "rewards/rejected": -0.123046875,
1557
+ "step": 860
1558
+ },
1559
+ {
1560
+ "epoch": 0.61,
1561
+ "grad_norm": 2.507656226813357,
1562
+ "learning_rate": 1.9838112164759652e-07,
1563
+ "log_odds_chosen": 0.12969970703125,
1564
+ "log_odds_ratio": -0.7134765386581421,
1565
+ "logits/chosen": -2.046875,
1566
+ "logits/rejected": -2.15625,
1567
+ "logps/chosen": -1.1328125,
1568
+ "logps/rejected": -1.234375,
1569
+ "loss": 1.2807,
1570
+ "nll_loss": 1.15625,
1571
+ "rewards/accuracies": 0.5625,
1572
+ "rewards/chosen": -0.11279296875,
1573
+ "rewards/margins": 0.0103759765625,
1574
+ "rewards/rejected": -0.12353515625,
1575
+ "step": 870
1576
+ },
1577
+ {
1578
+ "epoch": 0.62,
1579
+ "grad_norm": 2.4386366953549508,
1580
+ "learning_rate": 1.9241216698599202e-07,
1581
+ "log_odds_chosen": 0.16312256455421448,
1582
+ "log_odds_ratio": -0.6949218511581421,
1583
+ "logits/chosen": -2.125,
1584
+ "logits/rejected": -2.171875,
1585
+ "logps/chosen": -1.171875,
1586
+ "logps/rejected": -1.2890625,
1587
+ "loss": 1.2391,
1588
+ "nll_loss": 1.234375,
1589
+ "rewards/accuracies": 0.5874999761581421,
1590
+ "rewards/chosen": -0.11669921875,
1591
+ "rewards/margins": 0.01214599609375,
1592
+ "rewards/rejected": -0.12890625,
1593
+ "step": 880
1594
+ },
1595
+ {
1596
+ "epoch": 0.62,
1597
+ "grad_norm": 2.4804086256409157,
1598
+ "learning_rate": 1.8647768529973423e-07,
1599
+ "log_odds_chosen": 0.17501220107078552,
1600
+ "log_odds_ratio": -0.6958984136581421,
1601
+ "logits/chosen": -2.015625,
1602
+ "logits/rejected": -2.109375,
1603
+ "logps/chosen": -1.0625,
1604
+ "logps/rejected": -1.203125,
1605
+ "loss": 1.227,
1606
+ "nll_loss": 1.125,
1607
+ "rewards/accuracies": 0.59375,
1608
+ "rewards/chosen": -0.1064453125,
1609
+ "rewards/margins": 0.0135498046875,
1610
+ "rewards/rejected": -0.1201171875,
1611
+ "step": 890
1612
+ },
1613
+ {
1614
+ "epoch": 0.63,
1615
+ "grad_norm": 2.4623419233165986,
1616
+ "learning_rate": 1.80581229062152e-07,
1617
+ "log_odds_chosen": 0.17165526747703552,
1618
+ "log_odds_ratio": -0.6900390386581421,
1619
+ "logits/chosen": -2.03125,
1620
+ "logits/rejected": -2.078125,
1621
+ "logps/chosen": -1.1640625,
1622
+ "logps/rejected": -1.2734375,
1623
+ "loss": 1.2806,
1624
+ "nll_loss": 1.234375,
1625
+ "rewards/accuracies": 0.543749988079071,
1626
+ "rewards/chosen": -0.1162109375,
1627
+ "rewards/margins": 0.0111083984375,
1628
+ "rewards/rejected": -0.126953125,
1629
+ "step": 900
1630
+ },
1631
+ {
1632
+ "epoch": 0.64,
1633
+ "grad_norm": 2.797106319585485,
1634
+ "learning_rate": 1.7472632798394788e-07,
1635
+ "log_odds_chosen": 0.102783203125,
1636
+ "log_odds_ratio": -0.738574206829071,
1637
+ "logits/chosen": -2.03125,
1638
+ "logits/rejected": -2.125,
1639
+ "logps/chosen": -1.125,
1640
+ "logps/rejected": -1.1796875,
1641
+ "loss": 1.2623,
1642
+ "nll_loss": 1.203125,
1643
+ "rewards/accuracies": 0.48124998807907104,
1644
+ "rewards/chosen": -0.1123046875,
1645
+ "rewards/margins": 0.0057373046875,
1646
+ "rewards/rejected": -0.1181640625,
1647
+ "step": 910
1648
+ },
1649
+ {
1650
+ "epoch": 0.64,
1651
+ "grad_norm": 2.502768284944301,
1652
+ "learning_rate": 1.6891648690025718e-07,
1653
+ "log_odds_chosen": 0.20386962592601776,
1654
+ "log_odds_ratio": -0.687695324420929,
1655
+ "logits/chosen": -2.046875,
1656
+ "logits/rejected": -2.09375,
1657
+ "logps/chosen": -1.09375,
1658
+ "logps/rejected": -1.2421875,
1659
+ "loss": 1.2348,
1660
+ "nll_loss": 1.1640625,
1661
+ "rewards/accuracies": 0.53125,
1662
+ "rewards/chosen": -0.109375,
1663
+ "rewards/margins": 0.01507568359375,
1664
+ "rewards/rejected": -0.12451171875,
1665
+ "step": 920
1666
+ },
1667
+ {
1668
+ "epoch": 0.65,
1669
+ "grad_norm": 2.656181511294093,
1670
+ "learning_rate": 1.6315518367259978e-07,
1671
+ "log_odds_chosen": 0.27000731229782104,
1672
+ "log_odds_ratio": -0.647265613079071,
1673
+ "logits/chosen": -1.9765625,
1674
+ "logits/rejected": -2.0625,
1675
+ "logps/chosen": -1.1015625,
1676
+ "logps/rejected": -1.2890625,
1677
+ "loss": 1.2463,
1678
+ "nll_loss": 1.140625,
1679
+ "rewards/accuracies": 0.6000000238418579,
1680
+ "rewards/chosen": -0.1103515625,
1681
+ "rewards/margins": 0.0185546875,
1682
+ "rewards/rejected": -0.12890625,
1683
+ "step": 930
1684
+ },
1685
+ {
1686
+ "epoch": 0.66,
1687
+ "grad_norm": 3.0809025740755227,
1688
+ "learning_rate": 1.574458671069774e-07,
1689
+ "log_odds_chosen": 0.10646972805261612,
1690
+ "log_odds_ratio": -0.731249988079071,
1691
+ "logits/chosen": -1.9609375,
1692
+ "logits/rejected": -2.09375,
1693
+ "logps/chosen": -1.1328125,
1694
+ "logps/rejected": -1.2109375,
1695
+ "loss": 1.243,
1696
+ "nll_loss": 1.171875,
1697
+ "rewards/accuracies": 0.46875,
1698
+ "rewards/chosen": -0.11279296875,
1699
+ "rewards/margins": 0.0078125,
1700
+ "rewards/rejected": -0.12060546875,
1701
+ "step": 940
1702
+ },
1703
+ {
1704
+ "epoch": 0.67,
1705
+ "grad_norm": 2.5307625663876787,
1706
+ "learning_rate": 1.5179195488936504e-07,
1707
+ "log_odds_chosen": 0.22603759169578552,
1708
+ "log_odds_ratio": -0.6683593988418579,
1709
+ "logits/chosen": -2.046875,
1710
+ "logits/rejected": -2.109375,
1711
+ "logps/chosen": -1.109375,
1712
+ "logps/rejected": -1.265625,
1713
+ "loss": 1.224,
1714
+ "nll_loss": 1.1796875,
1715
+ "rewards/accuracies": 0.5687500238418579,
1716
+ "rewards/chosen": -0.1103515625,
1717
+ "rewards/margins": 0.0162353515625,
1718
+ "rewards/rejected": -0.126953125,
1719
+ "step": 950
1720
+ },
1721
+ {
1722
+ "epoch": 0.67,
1723
+ "grad_norm": 2.587513457015136,
1724
+ "learning_rate": 1.4619683153983016e-07,
1725
+ "log_odds_chosen": 0.15230712294578552,
1726
+ "log_odds_ratio": -0.70263671875,
1727
+ "logits/chosen": -2.0625,
1728
+ "logits/rejected": -2.171875,
1729
+ "logps/chosen": -1.1484375,
1730
+ "logps/rejected": -1.2421875,
1731
+ "loss": 1.2417,
1732
+ "nll_loss": 1.1953125,
1733
+ "rewards/accuracies": 0.53125,
1734
+ "rewards/chosen": -0.11474609375,
1735
+ "rewards/margins": 0.00909423828125,
1736
+ "rewards/rejected": -0.1240234375,
1737
+ "step": 960
1738
+ },
1739
+ {
1740
+ "epoch": 0.68,
1741
+ "grad_norm": 2.468276881587768,
1742
+ "learning_rate": 1.4066384638650687e-07,
1743
+ "log_odds_chosen": 0.26385498046875,
1744
+ "log_odds_ratio": -0.650195300579071,
1745
+ "logits/chosen": -2.015625,
1746
+ "logits/rejected": -2.046875,
1747
+ "logps/chosen": -1.15625,
1748
+ "logps/rejected": -1.3359375,
1749
+ "loss": 1.2174,
1750
+ "nll_loss": 1.1796875,
1751
+ "rewards/accuracies": 0.625,
1752
+ "rewards/chosen": -0.11572265625,
1753
+ "rewards/margins": 0.0177001953125,
1754
+ "rewards/rejected": -0.1337890625,
1755
+ "step": 970
1756
+ },
1757
+ {
1758
+ "epoch": 0.69,
1759
+ "grad_norm": 2.5705148701270537,
1760
+ "learning_rate": 1.351963115606354e-07,
1761
+ "log_odds_chosen": 0.18297119438648224,
1762
+ "log_odds_ratio": -0.689746081829071,
1763
+ "logits/chosen": -2.03125,
1764
+ "logits/rejected": -2.171875,
1765
+ "logps/chosen": -1.109375,
1766
+ "logps/rejected": -1.2265625,
1767
+ "loss": 1.2253,
1768
+ "nll_loss": 1.1640625,
1769
+ "rewards/accuracies": 0.550000011920929,
1770
+ "rewards/chosen": -0.11083984375,
1771
+ "rewards/margins": 0.011962890625,
1772
+ "rewards/rejected": -0.123046875,
1773
+ "step": 980
1774
+ },
1775
+ {
1776
+ "epoch": 0.69,
1777
+ "grad_norm": 2.5100478432366558,
1778
+ "learning_rate": 1.297975000138688e-07,
1779
+ "log_odds_chosen": 0.10567627102136612,
1780
+ "log_odds_ratio": -0.7080078125,
1781
+ "logits/chosen": -2.03125,
1782
+ "logits/rejected": -2.109375,
1783
+ "logps/chosen": -1.15625,
1784
+ "logps/rejected": -1.234375,
1785
+ "loss": 1.257,
1786
+ "nll_loss": 1.2109375,
1787
+ "rewards/accuracies": 0.53125,
1788
+ "rewards/chosen": -0.11572265625,
1789
+ "rewards/margins": 0.00787353515625,
1790
+ "rewards/rejected": -0.12353515625,
1791
+ "step": 990
1792
+ },
1793
+ {
1794
+ "epoch": 0.7,
1795
+ "grad_norm": 2.6847113433850924,
1796
+ "learning_rate": 1.24470643559033e-07,
1797
+ "log_odds_chosen": 0.27998048067092896,
1798
+ "log_odds_ratio": -0.650585949420929,
1799
+ "logits/chosen": -2.0,
1800
+ "logits/rejected": -2.109375,
1801
+ "logps/chosen": -1.125,
1802
+ "logps/rejected": -1.3125,
1803
+ "loss": 1.2519,
1804
+ "nll_loss": 1.1875,
1805
+ "rewards/accuracies": 0.6499999761581421,
1806
+ "rewards/chosen": -0.11279296875,
1807
+ "rewards/margins": 0.018798828125,
1808
+ "rewards/rejected": -0.1318359375,
1809
+ "step": 1000
1810
+ },
1811
+ {
1812
+ "epoch": 0.71,
1813
+ "grad_norm": 2.695150048132808,
1814
+ "learning_rate": 1.1921893093551323e-07,
1815
+ "log_odds_chosen": 0.21813353896141052,
1816
+ "log_odds_ratio": -0.673144519329071,
1817
+ "logits/chosen": -1.9921875,
1818
+ "logits/rejected": -2.0625,
1819
+ "logps/chosen": -1.1171875,
1820
+ "logps/rejected": -1.265625,
1821
+ "loss": 1.2468,
1822
+ "nll_loss": 1.1171875,
1823
+ "rewards/accuracies": 0.59375,
1824
+ "rewards/chosen": -0.11181640625,
1825
+ "rewards/margins": 0.0147705078125,
1826
+ "rewards/rejected": -0.1259765625,
1827
+ "step": 1010
1828
+ },
1829
+ {
1830
+ "epoch": 0.71,
1831
+ "grad_norm": 2.575139032183112,
1832
+ "learning_rate": 1.140455059004248e-07,
1833
+ "log_odds_chosen": 0.33759766817092896,
1834
+ "log_odds_ratio": -0.63525390625,
1835
+ "logits/chosen": -2.078125,
1836
+ "logits/rejected": -2.203125,
1837
+ "logps/chosen": -1.1171875,
1838
+ "logps/rejected": -1.359375,
1839
+ "loss": 1.2473,
1840
+ "nll_loss": 1.1796875,
1841
+ "rewards/accuracies": 0.6187499761581421,
1842
+ "rewards/chosen": -0.11181640625,
1843
+ "rewards/margins": 0.0244140625,
1844
+ "rewards/rejected": -0.13671875,
1845
+ "step": 1020
1846
+ },
1847
+ {
1848
+ "epoch": 0.72,
1849
+ "grad_norm": 2.3412172556068858,
1850
+ "learning_rate": 1.0895346534671146e-07,
1851
+ "log_odds_chosen": 0.28547364473342896,
1852
+ "log_odds_ratio": -0.6714843511581421,
1853
+ "logits/chosen": -2.0625,
1854
+ "logits/rejected": -2.171875,
1855
+ "logps/chosen": -1.109375,
1856
+ "logps/rejected": -1.296875,
1857
+ "loss": 1.223,
1858
+ "nll_loss": 1.1484375,
1859
+ "rewards/accuracies": 0.612500011920929,
1860
+ "rewards/chosen": -0.111328125,
1861
+ "rewards/margins": 0.018798828125,
1862
+ "rewards/rejected": -0.1298828125,
1863
+ "step": 1030
1864
+ },
1865
+ {
1866
+ "epoch": 0.73,
1867
+ "grad_norm": 2.440451621766773,
1868
+ "learning_rate": 1.0394585744929605e-07,
1869
+ "log_odds_chosen": 0.09172363579273224,
1870
+ "log_odds_ratio": -0.706250011920929,
1871
+ "logits/chosen": -2.03125,
1872
+ "logits/rejected": -2.078125,
1873
+ "logps/chosen": -1.1796875,
1874
+ "logps/rejected": -1.25,
1875
+ "loss": 1.2244,
1876
+ "nll_loss": 1.203125,
1877
+ "rewards/accuracies": 0.550000011920929,
1878
+ "rewards/chosen": -0.1181640625,
1879
+ "rewards/margins": 0.0067138671875,
1880
+ "rewards/rejected": -0.125,
1881
+ "step": 1040
1882
+ },
1883
+ {
1884
+ "epoch": 0.74,
1885
+ "grad_norm": 2.7872635135771486,
1886
+ "learning_rate": 9.902567984039659e-08,
1887
+ "log_odds_chosen": 0.10517577826976776,
1888
+ "log_odds_ratio": -0.718554675579071,
1889
+ "logits/chosen": -2.015625,
1890
+ "logits/rejected": -2.109375,
1891
+ "logps/chosen": -1.1796875,
1892
+ "logps/rejected": -1.25,
1893
+ "loss": 1.2433,
1894
+ "nll_loss": 1.234375,
1895
+ "rewards/accuracies": 0.543749988079071,
1896
+ "rewards/chosen": -0.1181640625,
1897
+ "rewards/margins": 0.0067138671875,
1898
+ "rewards/rejected": -0.125,
1899
+ "step": 1050
1900
+ },
1901
+ {
1902
+ "epoch": 0.74,
1903
+ "grad_norm": 2.6062059656472667,
1904
+ "learning_rate": 9.419587781509572e-08,
1905
+ "log_odds_chosen": 0.21335449814796448,
1906
+ "log_odds_ratio": -0.6875,
1907
+ "logits/chosen": -1.9296875,
1908
+ "logits/rejected": -2.046875,
1909
+ "logps/chosen": -1.0859375,
1910
+ "logps/rejected": -1.2421875,
1911
+ "loss": 1.2339,
1912
+ "nll_loss": 1.109375,
1913
+ "rewards/accuracies": 0.550000011920929,
1914
+ "rewards/chosen": -0.1083984375,
1915
+ "rewards/margins": 0.016357421875,
1916
+ "rewards/rejected": -0.12451171875,
1917
+ "step": 1060
1918
+ },
1919
+ {
1920
+ "epoch": 0.75,
1921
+ "grad_norm": 2.387458644437053,
1922
+ "learning_rate": 8.945934256824217e-08,
1923
+ "log_odds_chosen": 0.19075927138328552,
1924
+ "log_odds_ratio": -0.67236328125,
1925
+ "logits/chosen": -2.0,
1926
+ "logits/rejected": -2.0625,
1927
+ "logps/chosen": -1.125,
1928
+ "logps/rejected": -1.2578125,
1929
+ "loss": 1.2428,
1930
+ "nll_loss": 1.171875,
1931
+ "rewards/accuracies": 0.5625,
1932
+ "rewards/chosen": -0.1123046875,
1933
+ "rewards/margins": 0.0137939453125,
1934
+ "rewards/rejected": -0.1259765625,
1935
+ "step": 1070
1936
+ },
1937
+ {
1938
+ "epoch": 0.76,
1939
+ "grad_norm": 2.7020254464929536,
1940
+ "learning_rate": 8.481890946373562e-08,
1941
+ "log_odds_chosen": 0.16676025092601776,
1942
+ "log_odds_ratio": -0.6944335699081421,
1943
+ "logits/chosen": -1.9921875,
1944
+ "logits/rejected": -2.09375,
1945
+ "logps/chosen": -1.1484375,
1946
+ "logps/rejected": -1.265625,
1947
+ "loss": 1.2652,
1948
+ "nll_loss": 1.1953125,
1949
+ "rewards/accuracies": 0.5874999761581421,
1950
+ "rewards/chosen": -0.115234375,
1951
+ "rewards/margins": 0.01165771484375,
1952
+ "rewards/rejected": -0.126953125,
1953
+ "step": 1080
1954
+ },
1955
+ {
1956
+ "epoch": 0.76,
1957
+ "grad_norm": 2.5506342978080365,
1958
+ "learning_rate": 8.027735633723401e-08,
1959
+ "log_odds_chosen": 0.19404907524585724,
1960
+ "log_odds_ratio": -0.678906261920929,
1961
+ "logits/chosen": -2.015625,
1962
+ "logits/rejected": -2.0625,
1963
+ "logps/chosen": -1.15625,
1964
+ "logps/rejected": -1.2890625,
1965
+ "loss": 1.2371,
1966
+ "nll_loss": 1.1640625,
1967
+ "rewards/accuracies": 0.5874999761581421,
1968
+ "rewards/chosen": -0.11572265625,
1969
+ "rewards/margins": 0.01336669921875,
1970
+ "rewards/rejected": -0.12890625,
1971
+ "step": 1090
1972
+ },
1973
+ {
1974
+ "epoch": 0.77,
1975
+ "grad_norm": 2.9131209890812215,
1976
+ "learning_rate": 7.583740183329768e-08,
1977
+ "log_odds_chosen": 0.25482177734375,
1978
+ "log_odds_ratio": -0.654589831829071,
1979
+ "logits/chosen": -2.078125,
1980
+ "logits/rejected": -2.1875,
1981
+ "logps/chosen": -1.1328125,
1982
+ "logps/rejected": -1.3125,
1983
+ "loss": 1.2099,
1984
+ "nll_loss": 1.2109375,
1985
+ "rewards/accuracies": 0.59375,
1986
+ "rewards/chosen": -0.11376953125,
1987
+ "rewards/margins": 0.017578125,
1988
+ "rewards/rejected": -0.130859375,
1989
+ "step": 1100
1990
+ },
1991
+ {
1992
+ "epoch": 0.78,
1993
+ "grad_norm": 2.4151795293745417,
1994
+ "learning_rate": 7.150170377796627e-08,
1995
+ "log_odds_chosen": 0.09471435844898224,
1996
+ "log_odds_ratio": -0.719921886920929,
1997
+ "logits/chosen": -2.015625,
1998
+ "logits/rejected": -2.125,
1999
+ "logps/chosen": -1.15625,
2000
+ "logps/rejected": -1.2109375,
2001
+ "loss": 1.2681,
2002
+ "nll_loss": 1.1875,
2003
+ "rewards/accuracies": 0.543749988079071,
2004
+ "rewards/chosen": -0.11572265625,
2005
+ "rewards/margins": 0.0057373046875,
2006
+ "rewards/rejected": -0.12109375,
2007
+ "step": 1110
2008
+ },
2009
+ {
2010
+ "epoch": 0.78,
2011
+ "grad_norm": 2.9894142206703074,
2012
+ "learning_rate": 6.727285758774276e-08,
2013
+ "log_odds_chosen": 0.20228271186351776,
2014
+ "log_odds_ratio": -0.686718761920929,
2015
+ "logits/chosen": -1.9296875,
2016
+ "logits/rejected": -2.0625,
2017
+ "logps/chosen": -1.078125,
2018
+ "logps/rejected": -1.2265625,
2019
+ "loss": 1.2048,
2020
+ "nll_loss": 1.1171875,
2021
+ "rewards/accuracies": 0.5562499761581421,
2022
+ "rewards/chosen": -0.107421875,
2023
+ "rewards/margins": 0.01513671875,
2024
+ "rewards/rejected": -0.12255859375,
2025
+ "step": 1120
2026
+ },
2027
+ {
2028
+ "epoch": 0.79,
2029
+ "grad_norm": 2.292724337045773,
2030
+ "learning_rate": 6.315339471593645e-08,
2031
+ "log_odds_chosen": 0.2957824766635895,
2032
+ "log_odds_ratio": -0.6419922113418579,
2033
+ "logits/chosen": -1.9921875,
2034
+ "logits/rejected": -2.171875,
2035
+ "logps/chosen": -1.125,
2036
+ "logps/rejected": -1.328125,
2037
+ "loss": 1.2562,
2038
+ "nll_loss": 1.1640625,
2039
+ "rewards/accuracies": 0.606249988079071,
2040
+ "rewards/chosen": -0.11279296875,
2041
+ "rewards/margins": 0.0201416015625,
2042
+ "rewards/rejected": -0.1328125,
2043
+ "step": 1130
2044
+ },
2045
+ {
2046
+ "epoch": 0.8,
2047
+ "grad_norm": 2.4858379206596783,
2048
+ "learning_rate": 5.9145781137295596e-08,
2049
+ "log_odds_chosen": 0.11273193359375,
2050
+ "log_odds_ratio": -0.712109386920929,
2051
+ "logits/chosen": -1.9765625,
2052
+ "logits/rejected": -2.109375,
2053
+ "logps/chosen": -1.1640625,
2054
+ "logps/rejected": -1.25,
2055
+ "loss": 1.2469,
2056
+ "nll_loss": 1.1796875,
2057
+ "rewards/accuracies": 0.5687500238418579,
2058
+ "rewards/chosen": -0.11669921875,
2059
+ "rewards/margins": 0.00823974609375,
2060
+ "rewards/rejected": -0.125,
2061
+ "step": 1140
2062
+ },
2063
+ {
2064
+ "epoch": 0.81,
2065
+ "grad_norm": 2.9554773310288303,
2066
+ "learning_rate": 5.5252415871836514e-08,
2067
+ "log_odds_chosen": 0.19149169325828552,
2068
+ "log_odds_ratio": -0.6700195074081421,
2069
+ "logits/chosen": -2.078125,
2070
+ "logits/rejected": -2.125,
2071
+ "logps/chosen": -1.140625,
2072
+ "logps/rejected": -1.265625,
2073
+ "loss": 1.2691,
2074
+ "nll_loss": 1.1796875,
2075
+ "rewards/accuracies": 0.574999988079071,
2076
+ "rewards/chosen": -0.11376953125,
2077
+ "rewards/margins": 0.01287841796875,
2078
+ "rewards/rejected": -0.126953125,
2079
+ "step": 1150
2080
+ },
2081
+ {
2082
+ "epoch": 0.81,
2083
+ "grad_norm": 2.455900615151912,
2084
+ "learning_rate": 5.147562954875267e-08,
2085
+ "log_odds_chosen": 0.25535887479782104,
2086
+ "log_odds_ratio": -0.662304699420929,
2087
+ "logits/chosen": -1.9765625,
2088
+ "logits/rejected": -2.0625,
2089
+ "logps/chosen": -1.0625,
2090
+ "logps/rejected": -1.2265625,
2091
+ "loss": 1.2103,
2092
+ "nll_loss": 1.078125,
2093
+ "rewards/accuracies": 0.574999988079071,
2094
+ "rewards/chosen": -0.1064453125,
2095
+ "rewards/margins": 0.01611328125,
2096
+ "rewards/rejected": -0.12255859375,
2097
+ "step": 1160
2098
+ },
2099
+ {
2100
+ "epoch": 0.82,
2101
+ "grad_norm": 2.6012002451079055,
2102
+ "learning_rate": 4.781768301126374e-08,
2103
+ "log_odds_chosen": 0.10560913383960724,
2104
+ "log_odds_ratio": -0.7105468511581421,
2105
+ "logits/chosen": -2.0625,
2106
+ "logits/rejected": -2.09375,
2107
+ "logps/chosen": -1.171875,
2108
+ "logps/rejected": -1.2578125,
2109
+ "loss": 1.2439,
2110
+ "nll_loss": 1.1875,
2111
+ "rewards/accuracies": 0.5375000238418579,
2112
+ "rewards/chosen": -0.11767578125,
2113
+ "rewards/margins": 0.00823974609375,
2114
+ "rewards/rejected": -0.1259765625,
2115
+ "step": 1170
2116
+ },
2117
+ {
2118
+ "epoch": 0.83,
2119
+ "grad_norm": 2.2529151127172526,
2120
+ "learning_rate": 4.428076596323954e-08,
2121
+ "log_odds_chosen": 0.21270751953125,
2122
+ "log_odds_ratio": -0.6869140863418579,
2123
+ "logits/chosen": -1.953125,
2124
+ "logits/rejected": -2.078125,
2125
+ "logps/chosen": -1.125,
2126
+ "logps/rejected": -1.28125,
2127
+ "loss": 1.2745,
2128
+ "nll_loss": 1.2265625,
2129
+ "rewards/accuracies": 0.53125,
2130
+ "rewards/chosen": -0.11279296875,
2131
+ "rewards/margins": 0.01513671875,
2132
+ "rewards/rejected": -0.1279296875,
2133
+ "step": 1180
2134
+ },
2135
+ {
2136
+ "epoch": 0.83,
2137
+ "grad_norm": 2.4154203157415512,
2138
+ "learning_rate": 4.0866995658409145e-08,
2139
+ "log_odds_chosen": 0.25830078125,
2140
+ "log_odds_ratio": -0.65966796875,
2141
+ "logits/chosen": -1.9921875,
2142
+ "logits/rejected": -2.125,
2143
+ "logps/chosen": -1.09375,
2144
+ "logps/rejected": -1.265625,
2145
+ "loss": 1.2827,
2146
+ "nll_loss": 1.1484375,
2147
+ "rewards/accuracies": 0.6187499761581421,
2148
+ "rewards/chosen": -0.109375,
2149
+ "rewards/margins": 0.0174560546875,
2150
+ "rewards/rejected": -0.126953125,
2151
+ "step": 1190
2152
+ },
2153
+ {
2154
+ "epoch": 0.84,
2155
+ "grad_norm": 2.447937914506964,
2156
+ "learning_rate": 3.757841563293987e-08,
2157
+ "log_odds_chosen": 0.24925537407398224,
2158
+ "log_odds_ratio": -0.6611328125,
2159
+ "logits/chosen": -2.03125,
2160
+ "logits/rejected": -2.09375,
2161
+ "logps/chosen": -1.125,
2162
+ "logps/rejected": -1.296875,
2163
+ "loss": 1.2345,
2164
+ "nll_loss": 1.1640625,
2165
+ "rewards/accuracies": 0.6000000238418579,
2166
+ "rewards/chosen": -0.1123046875,
2167
+ "rewards/margins": 0.0167236328125,
2168
+ "rewards/rejected": -0.12890625,
2169
+ "step": 1200
2170
+ },
2171
+ {
2172
+ "epoch": 0.85,
2173
+ "grad_norm": 2.5349608625933375,
2174
+ "learning_rate": 3.441699448214433e-08,
2175
+ "log_odds_chosen": 0.20963135361671448,
2176
+ "log_odds_ratio": -0.689257800579071,
2177
+ "logits/chosen": -2.015625,
2178
+ "logits/rejected": -2.109375,
2179
+ "logps/chosen": -1.140625,
2180
+ "logps/rejected": -1.28125,
2181
+ "loss": 1.2633,
2182
+ "nll_loss": 1.1875,
2183
+ "rewards/accuracies": 0.581250011920929,
2184
+ "rewards/chosen": -0.11376953125,
2185
+ "rewards/margins": 0.01409912109375,
2186
+ "rewards/rejected": -0.1279296875,
2187
+ "step": 1210
2188
+ },
2189
+ {
2190
+ "epoch": 0.85,
2191
+ "grad_norm": 2.6135424725635987,
2192
+ "learning_rate": 3.138462468204914e-08,
2193
+ "log_odds_chosen": 0.25098878145217896,
2194
+ "log_odds_ratio": -0.660449206829071,
2195
+ "logits/chosen": -2.015625,
2196
+ "logits/rejected": -2.03125,
2197
+ "logps/chosen": -1.078125,
2198
+ "logps/rejected": -1.25,
2199
+ "loss": 1.2263,
2200
+ "nll_loss": 1.15625,
2201
+ "rewards/accuracies": 0.606249988079071,
2202
+ "rewards/chosen": -0.10791015625,
2203
+ "rewards/margins": 0.01708984375,
2204
+ "rewards/rejected": -0.125,
2205
+ "step": 1220
2206
+ },
2207
+ {
2208
+ "epoch": 0.86,
2209
+ "grad_norm": 2.450199821395702,
2210
+ "learning_rate": 2.8483121456528454e-08,
2211
+ "log_odds_chosen": 0.24144287407398224,
2212
+ "log_odds_ratio": -0.661816418170929,
2213
+ "logits/chosen": -2.046875,
2214
+ "logits/rejected": -2.125,
2215
+ "logps/chosen": -1.171875,
2216
+ "logps/rejected": -1.34375,
2217
+ "loss": 1.2541,
2218
+ "nll_loss": 1.2421875,
2219
+ "rewards/accuracies": 0.6000000238418579,
2220
+ "rewards/chosen": -0.1171875,
2221
+ "rewards/margins": 0.016845703125,
2222
+ "rewards/rejected": -0.1337890625,
2223
+ "step": 1230
2224
+ },
2225
+ {
2226
+ "epoch": 0.87,
2227
+ "grad_norm": 2.4363743869256904,
2228
+ "learning_rate": 2.5714221690683462e-08,
2229
+ "log_odds_chosen": 0.122314453125,
2230
+ "log_odds_ratio": -0.7119140625,
2231
+ "logits/chosen": -1.9609375,
2232
+ "logits/rejected": -2.046875,
2233
+ "logps/chosen": -1.1484375,
2234
+ "logps/rejected": -1.2265625,
2235
+ "loss": 1.3106,
2236
+ "nll_loss": 1.1953125,
2237
+ "rewards/accuracies": 0.518750011920929,
2238
+ "rewards/chosen": -0.115234375,
2239
+ "rewards/margins": 0.0079345703125,
2240
+ "rewards/rejected": -0.123046875,
2241
+ "step": 1240
2242
+ },
2243
+ {
2244
+ "epoch": 0.88,
2245
+ "grad_norm": 2.5756303193348327,
2246
+ "learning_rate": 2.307958289111514e-08,
2247
+ "log_odds_chosen": 0.19765624403953552,
2248
+ "log_odds_ratio": -0.6796875,
2249
+ "logits/chosen": -2.0,
2250
+ "logits/rejected": -2.03125,
2251
+ "logps/chosen": -1.15625,
2252
+ "logps/rejected": -1.296875,
2253
+ "loss": 1.2333,
2254
+ "nll_loss": 1.1875,
2255
+ "rewards/accuracies": 0.5625,
2256
+ "rewards/chosen": -0.11572265625,
2257
+ "rewards/margins": 0.01397705078125,
2258
+ "rewards/rejected": -0.1298828125,
2259
+ "step": 1250
2260
+ },
2261
+ {
2262
+ "epoch": 0.88,
2263
+ "grad_norm": 2.553964198420633,
2264
+ "learning_rate": 2.0580782193715506e-08,
2265
+ "log_odds_chosen": 0.15519408881664276,
2266
+ "log_odds_ratio": -0.704882800579071,
2267
+ "logits/chosen": -1.9296875,
2268
+ "logits/rejected": -2.0,
2269
+ "logps/chosen": -1.15625,
2270
+ "logps/rejected": -1.2578125,
2271
+ "loss": 1.2388,
2272
+ "nll_loss": 1.1640625,
2273
+ "rewards/accuracies": 0.5375000238418579,
2274
+ "rewards/chosen": -0.11572265625,
2275
+ "rewards/margins": 0.01031494140625,
2276
+ "rewards/rejected": -0.1259765625,
2277
+ "step": 1260
2278
+ },
2279
+ {
2280
+ "epoch": 0.89,
2281
+ "grad_norm": 2.469702033641708,
2282
+ "learning_rate": 1.821931541956914e-08,
2283
+ "log_odds_chosen": 0.4207519590854645,
2284
+ "log_odds_ratio": -0.582226574420929,
2285
+ "logits/chosen": -2.03125,
2286
+ "logits/rejected": -2.125,
2287
+ "logps/chosen": -1.09375,
2288
+ "logps/rejected": -1.359375,
2289
+ "loss": 1.2354,
2290
+ "nll_loss": 1.1484375,
2291
+ "rewards/accuracies": 0.6937500238418579,
2292
+ "rewards/chosen": -0.109375,
2293
+ "rewards/margins": 0.027099609375,
2294
+ "rewards/rejected": -0.1357421875,
2295
+ "step": 1270
2296
+ },
2297
+ {
2298
+ "epoch": 0.9,
2299
+ "grad_norm": 2.4889196799977786,
2300
+ "learning_rate": 1.5996596179531364e-08,
2301
+ "log_odds_chosen": 0.12117920070886612,
2302
+ "log_odds_ratio": -0.710644543170929,
2303
+ "logits/chosen": -2.015625,
2304
+ "logits/rejected": -2.046875,
2305
+ "logps/chosen": -1.1484375,
2306
+ "logps/rejected": -1.234375,
2307
+ "loss": 1.2454,
2308
+ "nll_loss": 1.21875,
2309
+ "rewards/accuracies": 0.5062500238418579,
2310
+ "rewards/chosen": -0.115234375,
2311
+ "rewards/margins": 0.0084228515625,
2312
+ "rewards/rejected": -0.12353515625,
2313
+ "step": 1280
2314
+ },
2315
+ {
2316
+ "epoch": 0.9,
2317
+ "grad_norm": 2.6921341795857088,
2318
+ "learning_rate": 1.3913955028018925e-08,
2319
+ "log_odds_chosen": 0.0946044921875,
2320
+ "log_odds_ratio": -0.734375,
2321
+ "logits/chosen": -1.9921875,
2322
+ "logits/rejected": -2.046875,
2323
+ "logps/chosen": -1.15625,
2324
+ "logps/rejected": -1.21875,
2325
+ "loss": 1.2338,
2326
+ "nll_loss": 1.203125,
2327
+ "rewards/accuracies": 0.5,
2328
+ "rewards/chosen": -0.115234375,
2329
+ "rewards/margins": 0.00634765625,
2330
+ "rewards/rejected": -0.12158203125,
2331
+ "step": 1290
2332
+ },
2333
+ {
2334
+ "epoch": 0.91,
2335
+ "grad_norm": 2.9234303938288444,
2336
+ "learning_rate": 1.1972638666519153e-08,
2337
+ "log_odds_chosen": 0.30915528535842896,
2338
+ "log_odds_ratio": -0.6298828125,
2339
+ "logits/chosen": -1.9453125,
2340
+ "logits/rejected": -2.046875,
2341
+ "logps/chosen": -1.09375,
2342
+ "logps/rejected": -1.328125,
2343
+ "loss": 1.2258,
2344
+ "nll_loss": 1.140625,
2345
+ "rewards/accuracies": 0.637499988079071,
2346
+ "rewards/chosen": -0.109375,
2347
+ "rewards/margins": 0.0230712890625,
2348
+ "rewards/rejected": -0.1328125,
2349
+ "step": 1300
2350
+ },
2351
+ {
2352
+ "epoch": 0.92,
2353
+ "grad_norm": 2.488742998417297,
2354
+ "learning_rate": 1.0173809197295074e-08,
2355
+ "log_odds_chosen": 0.1483154296875,
2356
+ "log_odds_ratio": -0.6932617425918579,
2357
+ "logits/chosen": -2.015625,
2358
+ "logits/rejected": -2.078125,
2359
+ "logps/chosen": -1.171875,
2360
+ "logps/rejected": -1.28125,
2361
+ "loss": 1.2868,
2362
+ "nll_loss": 1.234375,
2363
+ "rewards/accuracies": 0.5625,
2364
+ "rewards/chosen": -0.11767578125,
2365
+ "rewards/margins": 0.01080322265625,
2366
+ "rewards/rejected": -0.12890625,
2367
+ "step": 1310
2368
+ },
2369
+ {
2370
+ "epoch": 0.93,
2371
+ "grad_norm": 2.930123126278221,
2372
+ "learning_rate": 8.518543427732949e-09,
2373
+ "log_odds_chosen": 0.2968383729457855,
2374
+ "log_odds_ratio": -0.6597656011581421,
2375
+ "logits/chosen": -1.9609375,
2376
+ "logits/rejected": -2.0,
2377
+ "logps/chosen": -1.1015625,
2378
+ "logps/rejected": -1.3359375,
2379
+ "loss": 1.2515,
2380
+ "nll_loss": 1.1484375,
2381
+ "rewards/accuracies": 0.6187499761581421,
2382
+ "rewards/chosen": -0.1103515625,
2383
+ "rewards/margins": 0.0230712890625,
2384
+ "rewards/rejected": -0.1337890625,
2385
+ "step": 1320
2386
+ },
2387
+ {
2388
+ "epoch": 0.93,
2389
+ "grad_norm": 2.4910759658116306,
2390
+ "learning_rate": 7.007832225748356e-09,
2391
+ "log_odds_chosen": 0.3276123106479645,
2392
+ "log_odds_ratio": -0.623339831829071,
2393
+ "logits/chosen": -1.8359375,
2394
+ "logits/rejected": -1.953125,
2395
+ "logps/chosen": -1.03125,
2396
+ "logps/rejected": -1.25,
2397
+ "loss": 1.2122,
2398
+ "nll_loss": 1.1015625,
2399
+ "rewards/accuracies": 0.6312500238418579,
2400
+ "rewards/chosen": -0.1025390625,
2401
+ "rewards/margins": 0.0220947265625,
2402
+ "rewards/rejected": -0.125,
2403
+ "step": 1330
2404
+ },
2405
+ {
2406
+ "epoch": 0.94,
2407
+ "grad_norm": 2.6952772741810462,
2408
+ "learning_rate": 5.642579926637553e-09,
2409
+ "log_odds_chosen": 0.102996826171875,
2410
+ "log_odds_ratio": -0.718554675579071,
2411
+ "logits/chosen": -1.953125,
2412
+ "logits/rejected": -1.984375,
2413
+ "logps/chosen": -1.1875,
2414
+ "logps/rejected": -1.265625,
2415
+ "loss": 1.2679,
2416
+ "nll_loss": 1.2421875,
2417
+ "rewards/accuracies": 0.4937500059604645,
2418
+ "rewards/chosen": -0.11865234375,
2419
+ "rewards/margins": 0.007659912109375,
2420
+ "rewards/rejected": -0.1259765625,
2421
+ "step": 1340
2422
+ },
2423
+ {
2424
+ "epoch": 0.95,
2425
+ "grad_norm": 2.564277781118033,
2426
+ "learning_rate": 4.4236037917277644e-09,
2427
+ "log_odds_chosen": 0.17406006157398224,
2428
+ "log_odds_ratio": -0.697949230670929,
2429
+ "logits/chosen": -2.078125,
2430
+ "logits/rejected": -2.171875,
2431
+ "logps/chosen": -1.1875,
2432
+ "logps/rejected": -1.3125,
2433
+ "loss": 1.2736,
2434
+ "nll_loss": 1.2265625,
2435
+ "rewards/accuracies": 0.5625,
2436
+ "rewards/chosen": -0.11865234375,
2437
+ "rewards/margins": 0.012939453125,
2438
+ "rewards/rejected": -0.1318359375,
2439
+ "step": 1350
2440
+ },
2441
+ {
2442
+ "epoch": 0.95,
2443
+ "grad_norm": 2.6044066137402764,
2444
+ "learning_rate": 3.3516335191523528e-09,
2445
+ "log_odds_chosen": 0.22431640326976776,
2446
+ "log_odds_ratio": -0.6830078363418579,
2447
+ "logits/chosen": -1.9765625,
2448
+ "logits/rejected": -2.078125,
2449
+ "logps/chosen": -1.15625,
2450
+ "logps/rejected": -1.3046875,
2451
+ "loss": 1.221,
2452
+ "nll_loss": 1.1640625,
2453
+ "rewards/accuracies": 0.643750011920929,
2454
+ "rewards/chosen": -0.115234375,
2455
+ "rewards/margins": 0.01483154296875,
2456
+ "rewards/rejected": -0.1298828125,
2457
+ "step": 1360
2458
+ },
2459
+ {
2460
+ "epoch": 0.96,
2461
+ "grad_norm": 2.569982717697751,
2462
+ "learning_rate": 2.4273108070415605e-09,
2463
+ "log_odds_chosen": 0.24323730170726776,
2464
+ "log_odds_ratio": -0.675000011920929,
2465
+ "logits/chosen": -1.984375,
2466
+ "logits/rejected": -2.078125,
2467
+ "logps/chosen": -1.125,
2468
+ "logps/rejected": -1.3046875,
2469
+ "loss": 1.2342,
2470
+ "nll_loss": 1.1953125,
2471
+ "rewards/accuracies": 0.6000000238418579,
2472
+ "rewards/chosen": -0.1123046875,
2473
+ "rewards/margins": 0.017822265625,
2474
+ "rewards/rejected": -0.130859375,
2475
+ "step": 1370
2476
+ },
2477
+ {
2478
+ "epoch": 0.97,
2479
+ "grad_norm": 2.9474176384005117,
2480
+ "learning_rate": 1.651188969392564e-09,
2481
+ "log_odds_chosen": 0.18778076767921448,
2482
+ "log_odds_ratio": -0.675000011920929,
2483
+ "logits/chosen": -2.03125,
2484
+ "logits/rejected": -2.171875,
2485
+ "logps/chosen": -1.1484375,
2486
+ "logps/rejected": -1.2734375,
2487
+ "loss": 1.2187,
2488
+ "nll_loss": 1.1796875,
2489
+ "rewards/accuracies": 0.5562499761581421,
2490
+ "rewards/chosen": -0.11474609375,
2491
+ "rewards/margins": 0.012939453125,
2492
+ "rewards/rejected": -0.1279296875,
2493
+ "step": 1380
2494
+ },
2495
+ {
2496
+ "epoch": 0.97,
2497
+ "grad_norm": 2.5311989884303747,
2498
+ "learning_rate": 1.0237326048466443e-09,
2499
+ "log_odds_chosen": 0.2604003846645355,
2500
+ "log_odds_ratio": -0.6591796875,
2501
+ "logits/chosen": -1.9296875,
2502
+ "logits/rejected": -1.96875,
2503
+ "logps/chosen": -1.1328125,
2504
+ "logps/rejected": -1.3125,
2505
+ "loss": 1.2465,
2506
+ "nll_loss": 1.1875,
2507
+ "rewards/accuracies": 0.5874999761581421,
2508
+ "rewards/chosen": -0.11328125,
2509
+ "rewards/margins": 0.0185546875,
2510
+ "rewards/rejected": -0.1318359375,
2511
+ "step": 1390
2512
+ },
2513
+ {
2514
+ "epoch": 0.98,
2515
+ "grad_norm": 2.4286187387911005,
2516
+ "learning_rate": 5.453173185734073e-10,
2517
+ "log_odds_chosen": 0.158447265625,
2518
+ "log_odds_ratio": -0.71044921875,
2519
+ "logits/chosen": -2.0,
2520
+ "logits/rejected": -2.09375,
2521
+ "logps/chosen": -1.15625,
2522
+ "logps/rejected": -1.2734375,
2523
+ "loss": 1.2469,
2524
+ "nll_loss": 1.2265625,
2525
+ "rewards/accuracies": 0.550000011920929,
2526
+ "rewards/chosen": -0.1162109375,
2527
+ "rewards/margins": 0.01123046875,
2528
+ "rewards/rejected": -0.126953125,
2529
+ "step": 1400
2530
+ },
2531
+ {
2532
+ "epoch": 0.99,
2533
+ "grad_norm": 2.7786302717518256,
2534
+ "learning_rate": 2.1622949742761133e-10,
2535
+ "log_odds_chosen": 0.21599121391773224,
2536
+ "log_odds_ratio": -0.6719726324081421,
2537
+ "logits/chosen": -1.9609375,
2538
+ "logits/rejected": -2.0625,
2539
+ "logps/chosen": -1.09375,
2540
+ "logps/rejected": -1.2265625,
2541
+ "loss": 1.2478,
2542
+ "nll_loss": 1.1796875,
2543
+ "rewards/accuracies": 0.5874999761581421,
2544
+ "rewards/chosen": -0.109375,
2545
+ "rewards/margins": 0.0133056640625,
2546
+ "rewards/rejected": -0.123046875,
2547
+ "step": 1410
2548
+ },
2549
+ {
2550
+ "epoch": 1.0,
2551
+ "grad_norm": 2.2980939714984134,
2552
+ "learning_rate": 3.666613851355249e-11,
2553
+ "log_odds_chosen": 0.22458496689796448,
2554
+ "log_odds_ratio": -0.666210949420929,
2555
+ "logits/chosen": -2.078125,
2556
+ "logits/rejected": -2.203125,
2557
+ "logps/chosen": -1.1640625,
2558
+ "logps/rejected": -1.34375,
2559
+ "loss": 1.2281,
2560
+ "nll_loss": 1.2109375,
2561
+ "rewards/accuracies": 0.5687500238418579,
2562
+ "rewards/chosen": -0.1162109375,
2563
+ "rewards/margins": 0.017822265625,
2564
+ "rewards/rejected": -0.1337890625,
2565
+ "step": 1420
2566
+ },
2567
+ {
2568
+ "epoch": 1.0,
2569
+ "step": 1427,
2570
+ "total_flos": 0.0,
2571
+ "train_loss": 1.2897131164005442,
2572
+ "train_runtime": 9812.389,
2573
+ "train_samples_per_second": 6.979,
2574
+ "train_steps_per_second": 0.145
2575
+ }
2576
+ ],
2577
+ "logging_steps": 10,
2578
+ "max_steps": 1427,
2579
+ "num_input_tokens_seen": 0,
2580
+ "num_train_epochs": 1,
2581
+ "save_steps": 100,
2582
+ "total_flos": 0.0,
2583
+ "train_batch_size": 8,
2584
+ "trial_name": null,
2585
+ "trial_params": null
2586
+ }