just1nseo commited on
Commit
5a981aa
·
verified ·
1 Parent(s): 751a098

Model save

Browse files
README.md ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: peft
3
+ tags:
4
+ - trl
5
+ - dpo
6
+ - generated_from_trainer
7
+ base_model: allenai/tulu-2-7b
8
+ model-index:
9
+ - name: tulu2-7b-cost-UF-both-5e-7
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
+ # tulu2-7b-cost-UF-both-5e-7
17
+
18
+ This model is a fine-tuned version of [allenai/tulu-2-7b](https://huggingface.co/allenai/tulu-2-7b) on the None dataset.
19
+ It achieves the following results on the evaluation set:
20
+ - Loss: 0.6946
21
+ - Rewards/chosen: 0.0316
22
+ - Rewards/rejected: 0.0333
23
+ - Rewards/accuracies: 0.5195
24
+ - Rewards/margins: -0.0018
25
+ - Rewards/margins Max: 0.0952
26
+ - Rewards/margins Min: -0.1041
27
+ - Rewards/margins Std: 0.0646
28
+ - Logps/rejected: -316.1527
29
+ - Logps/chosen: -330.8240
30
+ - Logits/rejected: 0.8900
31
+ - Logits/chosen: 0.7447
32
+
33
+ ## Model description
34
+
35
+ More information needed
36
+
37
+ ## Intended uses & limitations
38
+
39
+ More information needed
40
+
41
+ ## Training and evaluation data
42
+
43
+ More information needed
44
+
45
+ ## Training procedure
46
+
47
+ ### Training hyperparameters
48
+
49
+ The following hyperparameters were used during training:
50
+ - learning_rate: 5e-07
51
+ - train_batch_size: 2
52
+ - eval_batch_size: 8
53
+ - seed: 42
54
+ - distributed_type: multi-GPU
55
+ - num_devices: 2
56
+ - gradient_accumulation_steps: 4
57
+ - total_train_batch_size: 16
58
+ - total_eval_batch_size: 16
59
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
60
+ - lr_scheduler_type: cosine
61
+ - lr_scheduler_warmup_ratio: 0.1
62
+ - num_epochs: 1
63
+
64
+ ### Training results
65
+
66
+ | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Rewards/margins Max | Rewards/margins Min | Rewards/margins Std | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
67
+ |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:-------------------:|:-------------------:|:-------------------:|:--------------:|:------------:|:---------------:|:-------------:|
68
+ | 0.6506 | 1.0 | 1359 | 0.6946 | 0.0316 | 0.0333 | 0.5195 | -0.0018 | 0.0952 | -0.1041 | 0.0646 | -316.1527 | -330.8240 | 0.8900 | 0.7447 |
69
+
70
+
71
+ ### Framework versions
72
+
73
+ - PEFT 0.7.1
74
+ - Transformers 4.39.0.dev0
75
+ - Pytorch 2.1.2+cu121
76
+ - Datasets 2.14.6
77
+ - Tokenizers 0.15.2
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7ecb99c73457cd701d126d54474431ac1820da65ed001e28ecfb809aab45c131
3
  size 639692768
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c28d759a4a3ea60879c44f369c367bcbe75dd001b8cc225b57e953604f92c8e
3
  size 639692768
all_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 0.6604200788417925,
4
+ "train_runtime": 10358.1715,
5
+ "train_samples": 21744,
6
+ "train_samples_per_second": 2.099,
7
+ "train_steps_per_second": 0.131
8
+ }
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 0.6604200788417925,
4
+ "train_runtime": 10358.1715,
5
+ "train_samples": 21744,
6
+ "train_samples_per_second": 2.099,
7
+ "train_steps_per_second": 0.131
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,2497 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 1.0,
5
+ "eval_steps": 100,
6
+ "global_step": 1359,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0,
13
+ "grad_norm": 0.416015625,
14
+ "learning_rate": 3.676470588235294e-09,
15
+ "logits/chosen": 0.3823161721229553,
16
+ "logits/rejected": 0.49768388271331787,
17
+ "logps/chosen": -143.86793518066406,
18
+ "logps/rejected": -156.11575317382812,
19
+ "loss": 0.6931,
20
+ "rewards/accuracies": 0.0,
21
+ "rewards/chosen": 0.0,
22
+ "rewards/margins": 0.0,
23
+ "rewards/margins_max": 0.0,
24
+ "rewards/margins_min": 0.0,
25
+ "rewards/margins_std": 0.0,
26
+ "rewards/rejected": 0.0,
27
+ "step": 1
28
+ },
29
+ {
30
+ "epoch": 0.01,
31
+ "grad_norm": 0.5,
32
+ "learning_rate": 3.676470588235294e-08,
33
+ "logits/chosen": -0.03673278167843819,
34
+ "logits/rejected": 0.7153948545455933,
35
+ "logps/chosen": -280.6231689453125,
36
+ "logps/rejected": -194.65892028808594,
37
+ "loss": 0.6933,
38
+ "rewards/accuracies": 0.3888888955116272,
39
+ "rewards/chosen": -0.0011863290565088391,
40
+ "rewards/margins": -0.00027060293359681964,
41
+ "rewards/margins_max": 0.0016658814856782556,
42
+ "rewards/margins_min": -0.002207087352871895,
43
+ "rewards/margins_std": 0.0027386024594306946,
44
+ "rewards/rejected": -0.0009157261229120195,
45
+ "step": 10
46
+ },
47
+ {
48
+ "epoch": 0.01,
49
+ "grad_norm": 0.37109375,
50
+ "learning_rate": 7.352941176470588e-08,
51
+ "logits/chosen": 0.007032014429569244,
52
+ "logits/rejected": 0.7839561104774475,
53
+ "logps/chosen": -332.548828125,
54
+ "logps/rejected": -255.7633819580078,
55
+ "loss": 0.6932,
56
+ "rewards/accuracies": 0.5249999761581421,
57
+ "rewards/chosen": -0.0009580751066096127,
58
+ "rewards/margins": 0.00016147561836987734,
59
+ "rewards/margins_max": 0.0031394611578434706,
60
+ "rewards/margins_min": -0.0028165101539343596,
61
+ "rewards/margins_std": 0.004211507737636566,
62
+ "rewards/rejected": -0.0011195506667718291,
63
+ "step": 20
64
+ },
65
+ {
66
+ "epoch": 0.02,
67
+ "grad_norm": 0.51171875,
68
+ "learning_rate": 1.1029411764705881e-07,
69
+ "logits/chosen": -0.04039617255330086,
70
+ "logits/rejected": 0.607696533203125,
71
+ "logps/chosen": -272.3553161621094,
72
+ "logps/rejected": -256.77227783203125,
73
+ "loss": 0.6932,
74
+ "rewards/accuracies": 0.4625000059604645,
75
+ "rewards/chosen": -0.000774255720898509,
76
+ "rewards/margins": 8.788631384959444e-05,
77
+ "rewards/margins_max": 0.001997451763600111,
78
+ "rewards/margins_min": -0.0018216792959719896,
79
+ "rewards/margins_std": 0.002700533252209425,
80
+ "rewards/rejected": -0.0008621420711278915,
81
+ "step": 30
82
+ },
83
+ {
84
+ "epoch": 0.03,
85
+ "grad_norm": 0.42578125,
86
+ "learning_rate": 1.4705882352941175e-07,
87
+ "logits/chosen": -0.06750641763210297,
88
+ "logits/rejected": 0.6604307889938354,
89
+ "logps/chosen": -295.84283447265625,
90
+ "logps/rejected": -260.4482421875,
91
+ "loss": 0.693,
92
+ "rewards/accuracies": 0.4625000059604645,
93
+ "rewards/chosen": -0.0007127780700102448,
94
+ "rewards/margins": 0.00020934769418090582,
95
+ "rewards/margins_max": 0.0021955964621156454,
96
+ "rewards/margins_min": -0.0017769013065844774,
97
+ "rewards/margins_std": 0.0028089801780879498,
98
+ "rewards/rejected": -0.0009221257641911507,
99
+ "step": 40
100
+ },
101
+ {
102
+ "epoch": 0.04,
103
+ "grad_norm": 0.470703125,
104
+ "learning_rate": 1.8382352941176472e-07,
105
+ "logits/chosen": -0.015060709789395332,
106
+ "logits/rejected": 0.6939523220062256,
107
+ "logps/chosen": -287.7695617675781,
108
+ "logps/rejected": -237.29794311523438,
109
+ "loss": 0.6927,
110
+ "rewards/accuracies": 0.5375000238418579,
111
+ "rewards/chosen": 0.0005769954295828938,
112
+ "rewards/margins": 0.0007454338483512402,
113
+ "rewards/margins_max": 0.0031318417750298977,
114
+ "rewards/margins_min": -0.0016409741947427392,
115
+ "rewards/margins_std": 0.0033748901914805174,
116
+ "rewards/rejected": -0.0001684384187683463,
117
+ "step": 50
118
+ },
119
+ {
120
+ "epoch": 0.04,
121
+ "grad_norm": 0.49609375,
122
+ "learning_rate": 2.2058823529411763e-07,
123
+ "logits/chosen": -0.0018195391166955233,
124
+ "logits/rejected": 0.7142434120178223,
125
+ "logps/chosen": -287.01068115234375,
126
+ "logps/rejected": -256.10614013671875,
127
+ "loss": 0.6923,
128
+ "rewards/accuracies": 0.625,
129
+ "rewards/chosen": 0.0012248576385900378,
130
+ "rewards/margins": 0.0020126227755099535,
131
+ "rewards/margins_max": 0.004514734260737896,
132
+ "rewards/margins_min": -0.0004894885933026671,
133
+ "rewards/margins_std": 0.0035385198425501585,
134
+ "rewards/rejected": -0.0007877650787122548,
135
+ "step": 60
136
+ },
137
+ {
138
+ "epoch": 0.05,
139
+ "grad_norm": 0.421875,
140
+ "learning_rate": 2.5735294117647057e-07,
141
+ "logits/chosen": 0.11708433926105499,
142
+ "logits/rejected": 0.5532101988792419,
143
+ "logps/chosen": -279.79791259765625,
144
+ "logps/rejected": -243.6463623046875,
145
+ "loss": 0.692,
146
+ "rewards/accuracies": 0.6625000238418579,
147
+ "rewards/chosen": 0.0009990496328100562,
148
+ "rewards/margins": 0.0019516434986144304,
149
+ "rewards/margins_max": 0.004653914365917444,
150
+ "rewards/margins_min": -0.0007506275433115661,
151
+ "rewards/margins_std": 0.0038215883541852236,
152
+ "rewards/rejected": -0.0009525938075967133,
153
+ "step": 70
154
+ },
155
+ {
156
+ "epoch": 0.06,
157
+ "grad_norm": 0.50390625,
158
+ "learning_rate": 2.941176470588235e-07,
159
+ "logits/chosen": -0.1581064760684967,
160
+ "logits/rejected": 0.7092851996421814,
161
+ "logps/chosen": -311.269775390625,
162
+ "logps/rejected": -251.24081420898438,
163
+ "loss": 0.6915,
164
+ "rewards/accuracies": 0.762499988079071,
165
+ "rewards/chosen": 0.002858319552615285,
166
+ "rewards/margins": 0.0039031144697219133,
167
+ "rewards/margins_max": 0.006772955413907766,
168
+ "rewards/margins_min": 0.001033273758366704,
169
+ "rewards/margins_std": 0.004058566875755787,
170
+ "rewards/rejected": -0.001044794567860663,
171
+ "step": 80
172
+ },
173
+ {
174
+ "epoch": 0.07,
175
+ "grad_norm": 0.390625,
176
+ "learning_rate": 3.3088235294117644e-07,
177
+ "logits/chosen": -0.045442212373018265,
178
+ "logits/rejected": 0.7158526182174683,
179
+ "logps/chosen": -265.78192138671875,
180
+ "logps/rejected": -218.0985870361328,
181
+ "loss": 0.691,
182
+ "rewards/accuracies": 0.75,
183
+ "rewards/chosen": 0.003086745273321867,
184
+ "rewards/margins": 0.0036944295279681683,
185
+ "rewards/margins_max": 0.007027274928987026,
186
+ "rewards/margins_min": 0.0003615838068071753,
187
+ "rewards/margins_std": 0.0047133550979197025,
188
+ "rewards/rejected": -0.0006076841382309794,
189
+ "step": 90
190
+ },
191
+ {
192
+ "epoch": 0.07,
193
+ "grad_norm": 0.458984375,
194
+ "learning_rate": 3.6764705882352943e-07,
195
+ "logits/chosen": 0.0736229196190834,
196
+ "logits/rejected": 0.7526043057441711,
197
+ "logps/chosen": -253.97119140625,
198
+ "logps/rejected": -216.62966918945312,
199
+ "loss": 0.6907,
200
+ "rewards/accuracies": 0.762499988079071,
201
+ "rewards/chosen": 0.004654365126043558,
202
+ "rewards/margins": 0.005358897615224123,
203
+ "rewards/margins_max": 0.008878113701939583,
204
+ "rewards/margins_min": 0.0018396815285086632,
205
+ "rewards/margins_std": 0.004976923577487469,
206
+ "rewards/rejected": -0.0007045331294648349,
207
+ "step": 100
208
+ },
209
+ {
210
+ "epoch": 0.08,
211
+ "grad_norm": 0.42578125,
212
+ "learning_rate": 4.044117647058823e-07,
213
+ "logits/chosen": -0.09733302891254425,
214
+ "logits/rejected": 0.6439273953437805,
215
+ "logps/chosen": -289.07769775390625,
216
+ "logps/rejected": -234.5022735595703,
217
+ "loss": 0.6899,
218
+ "rewards/accuracies": 0.800000011920929,
219
+ "rewards/chosen": 0.005760089494287968,
220
+ "rewards/margins": 0.006674544420093298,
221
+ "rewards/margins_max": 0.010509292595088482,
222
+ "rewards/margins_min": 0.002839796245098114,
223
+ "rewards/margins_std": 0.005423153750598431,
224
+ "rewards/rejected": -0.0009144555078819394,
225
+ "step": 110
226
+ },
227
+ {
228
+ "epoch": 0.09,
229
+ "grad_norm": 0.423828125,
230
+ "learning_rate": 4.4117647058823526e-07,
231
+ "logits/chosen": 0.018124500289559364,
232
+ "logits/rejected": 0.7241690754890442,
233
+ "logps/chosen": -311.22100830078125,
234
+ "logps/rejected": -241.0907440185547,
235
+ "loss": 0.6888,
236
+ "rewards/accuracies": 0.8500000238418579,
237
+ "rewards/chosen": 0.008410803973674774,
238
+ "rewards/margins": 0.007946285419166088,
239
+ "rewards/margins_max": 0.013005959801375866,
240
+ "rewards/margins_min": 0.002886611968278885,
241
+ "rewards/margins_std": 0.0071554603055119514,
242
+ "rewards/rejected": 0.000464518612716347,
243
+ "step": 120
244
+ },
245
+ {
246
+ "epoch": 0.1,
247
+ "grad_norm": 0.494140625,
248
+ "learning_rate": 4.779411764705882e-07,
249
+ "logits/chosen": -0.1438770443201065,
250
+ "logits/rejected": 0.6687305569648743,
251
+ "logps/chosen": -265.34161376953125,
252
+ "logps/rejected": -225.00424194335938,
253
+ "loss": 0.6875,
254
+ "rewards/accuracies": 0.862500011920929,
255
+ "rewards/chosen": 0.011373260989785194,
256
+ "rewards/margins": 0.01142856478691101,
257
+ "rewards/margins_max": 0.018687166273593903,
258
+ "rewards/margins_min": 0.004169962834566832,
259
+ "rewards/margins_std": 0.010265212506055832,
260
+ "rewards/rejected": -5.530398993869312e-05,
261
+ "step": 130
262
+ },
263
+ {
264
+ "epoch": 0.1,
265
+ "grad_norm": 0.423828125,
266
+ "learning_rate": 4.999868030671756e-07,
267
+ "logits/chosen": 0.05402200296521187,
268
+ "logits/rejected": 0.6528539657592773,
269
+ "logps/chosen": -292.96112060546875,
270
+ "logps/rejected": -259.50225830078125,
271
+ "loss": 0.6862,
272
+ "rewards/accuracies": 0.9125000238418579,
273
+ "rewards/chosen": 0.013445700518786907,
274
+ "rewards/margins": 0.014345765113830566,
275
+ "rewards/margins_max": 0.022081127390265465,
276
+ "rewards/margins_min": 0.006610400043427944,
277
+ "rewards/margins_std": 0.010939454659819603,
278
+ "rewards/rejected": -0.0009000620921142399,
279
+ "step": 140
280
+ },
281
+ {
282
+ "epoch": 0.11,
283
+ "grad_norm": 0.462890625,
284
+ "learning_rate": 4.998383535732973e-07,
285
+ "logits/chosen": -0.16393637657165527,
286
+ "logits/rejected": 0.45549267530441284,
287
+ "logps/chosen": -290.1100158691406,
288
+ "logps/rejected": -285.8172302246094,
289
+ "loss": 0.6856,
290
+ "rewards/accuracies": 0.925000011920929,
291
+ "rewards/chosen": 0.016459070146083832,
292
+ "rewards/margins": 0.016152694821357727,
293
+ "rewards/margins_max": 0.02362522855401039,
294
+ "rewards/margins_min": 0.008680161088705063,
295
+ "rewards/margins_std": 0.010567759163677692,
296
+ "rewards/rejected": 0.00030637450981885195,
297
+ "step": 150
298
+ },
299
+ {
300
+ "epoch": 0.12,
301
+ "grad_norm": 0.494140625,
302
+ "learning_rate": 4.995250566954361e-07,
303
+ "logits/chosen": 0.018339941278100014,
304
+ "logits/rejected": 0.7964209914207458,
305
+ "logps/chosen": -283.98577880859375,
306
+ "logps/rejected": -241.65353393554688,
307
+ "loss": 0.6842,
308
+ "rewards/accuracies": 0.8999999761581421,
309
+ "rewards/chosen": 0.01847967877984047,
310
+ "rewards/margins": 0.01731337048113346,
311
+ "rewards/margins_max": 0.028514724224805832,
312
+ "rewards/margins_min": 0.006112017668783665,
313
+ "rewards/margins_std": 0.015841105952858925,
314
+ "rewards/rejected": 0.0011663079494610429,
315
+ "step": 160
316
+ },
317
+ {
318
+ "epoch": 0.13,
319
+ "grad_norm": 0.5546875,
320
+ "learning_rate": 4.990471191519357e-07,
321
+ "logits/chosen": 0.03216102719306946,
322
+ "logits/rejected": 0.7764664888381958,
323
+ "logps/chosen": -279.2666931152344,
324
+ "logps/rejected": -237.353515625,
325
+ "loss": 0.6827,
326
+ "rewards/accuracies": 0.925000011920929,
327
+ "rewards/chosen": 0.023826319724321365,
328
+ "rewards/margins": 0.02013733796775341,
329
+ "rewards/margins_max": 0.03089248575270176,
330
+ "rewards/margins_min": 0.009382191114127636,
331
+ "rewards/margins_std": 0.015210075303912163,
332
+ "rewards/rejected": 0.0036889780312776566,
333
+ "step": 170
334
+ },
335
+ {
336
+ "epoch": 0.13,
337
+ "grad_norm": 0.4453125,
338
+ "learning_rate": 4.984048562937129e-07,
339
+ "logits/chosen": -0.034670569002628326,
340
+ "logits/rejected": 0.5347191095352173,
341
+ "logps/chosen": -253.5879364013672,
342
+ "logps/rejected": -252.5656280517578,
343
+ "loss": 0.6827,
344
+ "rewards/accuracies": 0.9375,
345
+ "rewards/chosen": 0.021980373188853264,
346
+ "rewards/margins": 0.020695097744464874,
347
+ "rewards/margins_max": 0.03351113945245743,
348
+ "rewards/margins_min": 0.007879055105149746,
349
+ "rewards/margins_std": 0.018124621361494064,
350
+ "rewards/rejected": 0.0012852762592956424,
351
+ "step": 180
352
+ },
353
+ {
354
+ "epoch": 0.14,
355
+ "grad_norm": 0.37109375,
356
+ "learning_rate": 4.975986918961825e-07,
357
+ "logits/chosen": -0.0010436892043799162,
358
+ "logits/rejected": 0.5325745344161987,
359
+ "logps/chosen": -258.70465087890625,
360
+ "logps/rejected": -254.8517608642578,
361
+ "loss": 0.6811,
362
+ "rewards/accuracies": 0.925000011920929,
363
+ "rewards/chosen": 0.02575596049427986,
364
+ "rewards/margins": 0.0235120952129364,
365
+ "rewards/margins_max": 0.037295907735824585,
366
+ "rewards/margins_min": 0.009728283621370792,
367
+ "rewards/margins_std": 0.019493255764245987,
368
+ "rewards/rejected": 0.0022438622545450926,
369
+ "step": 190
370
+ },
371
+ {
372
+ "epoch": 0.15,
373
+ "grad_norm": 0.447265625,
374
+ "learning_rate": 4.966291578796448e-07,
375
+ "logits/chosen": -0.08659172803163528,
376
+ "logits/rejected": 0.8055790066719055,
377
+ "logps/chosen": -268.54559326171875,
378
+ "logps/rejected": -214.8706512451172,
379
+ "loss": 0.68,
380
+ "rewards/accuracies": 0.949999988079071,
381
+ "rewards/chosen": 0.02814793959259987,
382
+ "rewards/margins": 0.02799884043633938,
383
+ "rewards/margins_max": 0.042513541877269745,
384
+ "rewards/margins_min": 0.013484140858054161,
385
+ "rewards/margins_std": 0.020526882261037827,
386
+ "rewards/rejected": 0.00014909838500898331,
387
+ "step": 200
388
+ },
389
+ {
390
+ "epoch": 0.15,
391
+ "grad_norm": 0.3515625,
392
+ "learning_rate": 4.954968939583149e-07,
393
+ "logits/chosen": -0.15794704854488373,
394
+ "logits/rejected": 0.6090285181999207,
395
+ "logps/chosen": -289.7901306152344,
396
+ "logps/rejected": -238.60183715820312,
397
+ "loss": 0.6789,
398
+ "rewards/accuracies": 0.925000011920929,
399
+ "rewards/chosen": 0.03186321258544922,
400
+ "rewards/margins": 0.03159675747156143,
401
+ "rewards/margins_max": 0.04762989282608032,
402
+ "rewards/margins_min": 0.015563614666461945,
403
+ "rewards/margins_std": 0.022674284875392914,
404
+ "rewards/rejected": 0.0002664585190359503,
405
+ "step": 210
406
+ },
407
+ {
408
+ "epoch": 0.16,
409
+ "grad_norm": 0.44921875,
410
+ "learning_rate": 4.942026472182297e-07,
411
+ "logits/chosen": 0.012926039285957813,
412
+ "logits/rejected": 0.6630520820617676,
413
+ "logps/chosen": -269.59930419921875,
414
+ "logps/rejected": -248.7339324951172,
415
+ "loss": 0.6778,
416
+ "rewards/accuracies": 0.987500011920929,
417
+ "rewards/chosen": 0.0339738167822361,
418
+ "rewards/margins": 0.03334929794073105,
419
+ "rewards/margins_max": 0.04980397969484329,
420
+ "rewards/margins_min": 0.016894619911909103,
421
+ "rewards/margins_std": 0.023270433768630028,
422
+ "rewards/rejected": 0.0006245142430998385,
423
+ "step": 220
424
+ },
425
+ {
426
+ "epoch": 0.17,
427
+ "grad_norm": 0.43359375,
428
+ "learning_rate": 4.92747271624308e-07,
429
+ "logits/chosen": -0.10246638208627701,
430
+ "logits/rejected": 0.8088499903678894,
431
+ "logps/chosen": -293.9709167480469,
432
+ "logps/rejected": -229.0161895751953,
433
+ "loss": 0.6776,
434
+ "rewards/accuracies": 0.9624999761581421,
435
+ "rewards/chosen": 0.03623218461871147,
436
+ "rewards/margins": 0.03605912998318672,
437
+ "rewards/margins_max": 0.05394510179758072,
438
+ "rewards/margins_min": 0.018173154443502426,
439
+ "rewards/margins_std": 0.025294587016105652,
440
+ "rewards/rejected": 0.0001730593794491142,
441
+ "step": 230
442
+ },
443
+ {
444
+ "epoch": 0.18,
445
+ "grad_norm": 0.50390625,
446
+ "learning_rate": 4.911317274568909e-07,
447
+ "logits/chosen": -0.1043270006775856,
448
+ "logits/rejected": 0.7763813138008118,
449
+ "logps/chosen": -290.52032470703125,
450
+ "logps/rejected": -238.03591918945312,
451
+ "loss": 0.6748,
452
+ "rewards/accuracies": 0.9624999761581421,
453
+ "rewards/chosen": 0.04272182658314705,
454
+ "rewards/margins": 0.04140879958868027,
455
+ "rewards/margins_max": 0.0613585002720356,
456
+ "rewards/margins_min": 0.021459100767970085,
457
+ "rewards/margins_std": 0.0282131377607584,
458
+ "rewards/rejected": 0.0013130264123901725,
459
+ "step": 240
460
+ },
461
+ {
462
+ "epoch": 0.18,
463
+ "grad_norm": 0.4765625,
464
+ "learning_rate": 4.89357080678133e-07,
465
+ "logits/chosen": 0.042627763003110886,
466
+ "logits/rejected": 0.5050796270370483,
467
+ "logps/chosen": -225.12393188476562,
468
+ "logps/rejected": -208.74288940429688,
469
+ "loss": 0.6767,
470
+ "rewards/accuracies": 0.875,
471
+ "rewards/chosen": 0.02972862496972084,
472
+ "rewards/margins": 0.026549100875854492,
473
+ "rewards/margins_max": 0.04413127154111862,
474
+ "rewards/margins_min": 0.008966929279267788,
475
+ "rewards/margins_std": 0.02486494556069374,
476
+ "rewards/rejected": 0.00317952549085021,
477
+ "step": 250
478
+ },
479
+ {
480
+ "epoch": 0.19,
481
+ "grad_norm": 0.6015625,
482
+ "learning_rate": 4.874245022286637e-07,
483
+ "logits/chosen": 0.061648208647966385,
484
+ "logits/rejected": 0.6458357572555542,
485
+ "logps/chosen": -262.064208984375,
486
+ "logps/rejected": -268.61444091796875,
487
+ "loss": 0.677,
488
+ "rewards/accuracies": 0.9750000238418579,
489
+ "rewards/chosen": 0.03966163843870163,
490
+ "rewards/margins": 0.036633603274822235,
491
+ "rewards/margins_max": 0.05438825488090515,
492
+ "rewards/margins_min": 0.01887894608080387,
493
+ "rewards/margins_std": 0.02510887011885643,
494
+ "rewards/rejected": 0.0030280412174761295,
495
+ "step": 260
496
+ },
497
+ {
498
+ "epoch": 0.2,
499
+ "grad_norm": 0.3984375,
500
+ "learning_rate": 4.853352672549815e-07,
501
+ "logits/chosen": 0.05668522045016289,
502
+ "logits/rejected": 0.7193032503128052,
503
+ "logps/chosen": -261.01708984375,
504
+ "logps/rejected": -228.1970672607422,
505
+ "loss": 0.6732,
506
+ "rewards/accuracies": 0.9624999761581421,
507
+ "rewards/chosen": 0.038796816021203995,
508
+ "rewards/margins": 0.038500867784023285,
509
+ "rewards/margins_max": 0.05987783148884773,
510
+ "rewards/margins_min": 0.01712390035390854,
511
+ "rewards/margins_std": 0.030231595039367676,
512
+ "rewards/rejected": 0.0002959508856292814,
513
+ "step": 270
514
+ },
515
+ {
516
+ "epoch": 0.21,
517
+ "grad_norm": 0.3984375,
518
+ "learning_rate": 4.830907542680918e-07,
519
+ "logits/chosen": -0.022307882085442543,
520
+ "logits/rejected": 0.5763358473777771,
521
+ "logps/chosen": -259.81414794921875,
522
+ "logps/rejected": -243.997802734375,
523
+ "loss": 0.6746,
524
+ "rewards/accuracies": 0.9375,
525
+ "rewards/chosen": 0.04487748071551323,
526
+ "rewards/margins": 0.04191521182656288,
527
+ "rewards/margins_max": 0.06111626699566841,
528
+ "rewards/margins_min": 0.022714145481586456,
529
+ "rewards/margins_std": 0.027154404670000076,
530
+ "rewards/rejected": 0.0029622733127325773,
531
+ "step": 280
532
+ },
533
+ {
534
+ "epoch": 0.21,
535
+ "grad_norm": 0.337890625,
536
+ "learning_rate": 4.806924442339425e-07,
537
+ "logits/chosen": -0.14441627264022827,
538
+ "logits/rejected": 0.5065394639968872,
539
+ "logps/chosen": -243.6468505859375,
540
+ "logps/rejected": -224.4588165283203,
541
+ "loss": 0.6735,
542
+ "rewards/accuracies": 0.9125000238418579,
543
+ "rewards/chosen": 0.045001424849033356,
544
+ "rewards/margins": 0.04354133456945419,
545
+ "rewards/margins_max": 0.062143802642822266,
546
+ "rewards/margins_min": 0.02493886649608612,
547
+ "rewards/margins_std": 0.026307862251996994,
548
+ "rewards/rejected": 0.0014600924914702773,
549
+ "step": 290
550
+ },
551
+ {
552
+ "epoch": 0.22,
553
+ "grad_norm": 0.4140625,
554
+ "learning_rate": 4.781419195962598e-07,
555
+ "logits/chosen": -0.09291721135377884,
556
+ "logits/rejected": 0.8421966433525085,
557
+ "logps/chosen": -294.4910583496094,
558
+ "logps/rejected": -235.1195526123047,
559
+ "loss": 0.6669,
560
+ "rewards/accuracies": 0.9375,
561
+ "rewards/chosen": 0.05570490285754204,
562
+ "rewards/margins": 0.05652566999197006,
563
+ "rewards/margins_max": 0.0832170695066452,
564
+ "rewards/margins_min": 0.02983427606523037,
565
+ "rewards/margins_std": 0.0377473346889019,
566
+ "rewards/rejected": -0.0008207682403735816,
567
+ "step": 300
568
+ },
569
+ {
570
+ "epoch": 0.23,
571
+ "grad_norm": 0.427734375,
572
+ "learning_rate": 4.754408632324253e-07,
573
+ "logits/chosen": 0.08360351622104645,
574
+ "logits/rejected": 0.9180965423583984,
575
+ "logps/chosen": -274.46221923828125,
576
+ "logps/rejected": -218.58261108398438,
577
+ "loss": 0.6709,
578
+ "rewards/accuracies": 0.925000011920929,
579
+ "rewards/chosen": 0.04704751446843147,
580
+ "rewards/margins": 0.04507958143949509,
581
+ "rewards/margins_max": 0.07067537307739258,
582
+ "rewards/margins_min": 0.019483786076307297,
583
+ "rewards/margins_std": 0.03619792312383652,
584
+ "rewards/rejected": 0.0019679281394928694,
585
+ "step": 310
586
+ },
587
+ {
588
+ "epoch": 0.24,
589
+ "grad_norm": 0.392578125,
590
+ "learning_rate": 4.725910573430866e-07,
591
+ "logits/chosen": -0.03652850538492203,
592
+ "logits/rejected": 0.7102953195571899,
593
+ "logps/chosen": -259.0872497558594,
594
+ "logps/rejected": -220.06640625,
595
+ "loss": 0.6666,
596
+ "rewards/accuracies": 0.9375,
597
+ "rewards/chosen": 0.052086759358644485,
598
+ "rewards/margins": 0.05210639908909798,
599
+ "rewards/margins_max": 0.07583759725093842,
600
+ "rewards/margins_min": 0.02837519347667694,
601
+ "rewards/margins_std": 0.033560991287231445,
602
+ "rewards/rejected": -1.9636423530755565e-05,
603
+ "step": 320
604
+ },
605
+ {
606
+ "epoch": 0.24,
607
+ "grad_norm": 0.482421875,
608
+ "learning_rate": 4.6959438227623293e-07,
609
+ "logits/chosen": -0.07948180288076401,
610
+ "logits/rejected": 0.6266988515853882,
611
+ "logps/chosen": -291.2080078125,
612
+ "logps/rejected": -269.5301818847656,
613
+ "loss": 0.6677,
614
+ "rewards/accuracies": 0.987500011920929,
615
+ "rewards/chosen": 0.04915199428796768,
616
+ "rewards/margins": 0.050243012607097626,
617
+ "rewards/margins_max": 0.07468355447053909,
618
+ "rewards/margins_min": 0.025802483782172203,
619
+ "rewards/margins_std": 0.03456413000822067,
620
+ "rewards/rejected": -0.0010910223936662078,
621
+ "step": 330
622
+ },
623
+ {
624
+ "epoch": 0.25,
625
+ "grad_norm": 0.451171875,
626
+ "learning_rate": 4.664528152865105e-07,
627
+ "logits/chosen": 0.015779225155711174,
628
+ "logits/rejected": 0.6966221332550049,
629
+ "logps/chosen": -291.6715087890625,
630
+ "logps/rejected": -260.85784912109375,
631
+ "loss": 0.6643,
632
+ "rewards/accuracies": 0.9375,
633
+ "rewards/chosen": 0.06338638812303543,
634
+ "rewards/margins": 0.06281675398349762,
635
+ "rewards/margins_max": 0.09652182459831238,
636
+ "rewards/margins_min": 0.029111674055457115,
637
+ "rewards/margins_std": 0.04766617342829704,
638
+ "rewards/rejected": 0.0005696387379430234,
639
+ "step": 340
640
+ },
641
+ {
642
+ "epoch": 0.26,
643
+ "grad_norm": 0.56640625,
644
+ "learning_rate": 4.6316842923059816e-07,
645
+ "logits/chosen": -0.15278920531272888,
646
+ "logits/rejected": 0.6084717512130737,
647
+ "logps/chosen": -247.9635009765625,
648
+ "logps/rejected": -236.7021484375,
649
+ "loss": 0.6618,
650
+ "rewards/accuracies": 0.987500011920929,
651
+ "rewards/chosen": 0.07176478952169418,
652
+ "rewards/margins": 0.07197294384241104,
653
+ "rewards/margins_max": 0.09642891585826874,
654
+ "rewards/margins_min": 0.047516968101263046,
655
+ "rewards/margins_std": 0.03458596393465996,
656
+ "rewards/rejected": -0.0002081508719129488,
657
+ "step": 350
658
+ },
659
+ {
660
+ "epoch": 0.26,
661
+ "grad_norm": 0.45703125,
662
+ "learning_rate": 4.5974339119950334e-07,
663
+ "logits/chosen": -0.0013063043588772416,
664
+ "logits/rejected": 0.6344822645187378,
665
+ "logps/chosen": -263.31610107421875,
666
+ "logps/rejected": -204.03660583496094,
667
+ "loss": 0.6657,
668
+ "rewards/accuracies": 0.8999999761581421,
669
+ "rewards/chosen": 0.052458666265010834,
670
+ "rewards/margins": 0.04967236518859863,
671
+ "rewards/margins_max": 0.07377265393733978,
672
+ "rewards/margins_min": 0.025572070851922035,
673
+ "rewards/margins_std": 0.034082964062690735,
674
+ "rewards/rejected": 0.0027863027062267065,
675
+ "step": 360
676
+ },
677
+ {
678
+ "epoch": 0.27,
679
+ "grad_norm": 0.451171875,
680
+ "learning_rate": 4.5617996108867997e-07,
681
+ "logits/chosen": -0.09778765588998795,
682
+ "logits/rejected": 0.8527771234512329,
683
+ "logps/chosen": -272.1876525878906,
684
+ "logps/rejected": -224.5320281982422,
685
+ "loss": 0.6615,
686
+ "rewards/accuracies": 0.9624999761581421,
687
+ "rewards/chosen": 0.06909666955471039,
688
+ "rewards/margins": 0.06635682284832001,
689
+ "rewards/margins_max": 0.09376729279756546,
690
+ "rewards/margins_min": 0.03894636407494545,
691
+ "rewards/margins_std": 0.03876424580812454,
692
+ "rewards/rejected": 0.0027398415841162205,
693
+ "step": 370
694
+ },
695
+ {
696
+ "epoch": 0.28,
697
+ "grad_norm": 0.39453125,
698
+ "learning_rate": 4.5248049010691304e-07,
699
+ "logits/chosen": -0.04781482741236687,
700
+ "logits/rejected": 0.6669384241104126,
701
+ "logps/chosen": -270.28253173828125,
702
+ "logps/rejected": -231.1532440185547,
703
+ "loss": 0.6605,
704
+ "rewards/accuracies": 0.949999988079071,
705
+ "rewards/chosen": 0.06150456517934799,
706
+ "rewards/margins": 0.0636376217007637,
707
+ "rewards/margins_max": 0.09591880440711975,
708
+ "rewards/margins_min": 0.03135643154382706,
709
+ "rewards/margins_std": 0.04565249755978584,
710
+ "rewards/rejected": -0.0021330583840608597,
711
+ "step": 380
712
+ },
713
+ {
714
+ "epoch": 0.29,
715
+ "grad_norm": 0.5546875,
716
+ "learning_rate": 4.486474192249533e-07,
717
+ "logits/chosen": 0.034309376031160355,
718
+ "logits/rejected": 0.7420383095741272,
719
+ "logps/chosen": -246.19668579101562,
720
+ "logps/rejected": -215.16323852539062,
721
+ "loss": 0.6629,
722
+ "rewards/accuracies": 0.949999988079071,
723
+ "rewards/chosen": 0.056318044662475586,
724
+ "rewards/margins": 0.06057443469762802,
725
+ "rewards/margins_max": 0.08743326365947723,
726
+ "rewards/margins_min": 0.03371559828519821,
727
+ "rewards/margins_std": 0.037984125316143036,
728
+ "rewards/rejected": -0.0042563872411847115,
729
+ "step": 390
730
+ },
731
+ {
732
+ "epoch": 0.29,
733
+ "grad_norm": 0.5390625,
734
+ "learning_rate": 4.4468327756492504e-07,
735
+ "logits/chosen": 0.09597498923540115,
736
+ "logits/rejected": 0.7928622961044312,
737
+ "logps/chosen": -275.2891540527344,
738
+ "logps/rejected": -246.2664794921875,
739
+ "loss": 0.6631,
740
+ "rewards/accuracies": 0.949999988079071,
741
+ "rewards/chosen": 0.06533468514680862,
742
+ "rewards/margins": 0.06880095601081848,
743
+ "rewards/margins_max": 0.1037551760673523,
744
+ "rewards/margins_min": 0.03384672850370407,
745
+ "rewards/margins_std": 0.04943273216485977,
746
+ "rewards/rejected": -0.0034662713296711445,
747
+ "step": 400
748
+ },
749
+ {
750
+ "epoch": 0.3,
751
+ "grad_norm": 0.4765625,
752
+ "learning_rate": 4.405906807315705e-07,
753
+ "logits/chosen": -0.021347884088754654,
754
+ "logits/rejected": 0.6999279260635376,
755
+ "logps/chosen": -227.08447265625,
756
+ "logps/rejected": -222.6078643798828,
757
+ "loss": 0.6565,
758
+ "rewards/accuracies": 0.949999988079071,
759
+ "rewards/chosen": 0.0646035447716713,
760
+ "rewards/margins": 0.0676642507314682,
761
+ "rewards/margins_max": 0.09568590670824051,
762
+ "rewards/margins_min": 0.039642591029405594,
763
+ "rewards/margins_std": 0.039628610014915466,
764
+ "rewards/rejected": -0.003060706425458193,
765
+ "step": 410
766
+ },
767
+ {
768
+ "epoch": 0.31,
769
+ "grad_norm": 0.41796875,
770
+ "learning_rate": 4.363723290864314e-07,
771
+ "logits/chosen": 0.11118383705615997,
772
+ "logits/rejected": 0.9948932528495789,
773
+ "logps/chosen": -271.27496337890625,
774
+ "logps/rejected": -222.7627410888672,
775
+ "loss": 0.6589,
776
+ "rewards/accuracies": 0.949999988079071,
777
+ "rewards/chosen": 0.07119729369878769,
778
+ "rewards/margins": 0.07638658583164215,
779
+ "rewards/margins_max": 0.120518758893013,
780
+ "rewards/margins_min": 0.032254405319690704,
781
+ "rewards/margins_std": 0.0624123141169548,
782
+ "rewards/rejected": -0.005189282353967428,
783
+ "step": 420
784
+ },
785
+ {
786
+ "epoch": 0.32,
787
+ "grad_norm": 0.486328125,
788
+ "learning_rate": 4.3203100596610723e-07,
789
+ "logits/chosen": -0.016591979190707207,
790
+ "logits/rejected": 0.813875675201416,
791
+ "logps/chosen": -300.334716796875,
792
+ "logps/rejected": -259.6979064941406,
793
+ "loss": 0.656,
794
+ "rewards/accuracies": 0.9624999761581421,
795
+ "rewards/chosen": 0.07723168283700943,
796
+ "rewards/margins": 0.0816957950592041,
797
+ "rewards/margins_max": 0.1194806843996048,
798
+ "rewards/margins_min": 0.043910913169384,
799
+ "rewards/margins_std": 0.05343589186668396,
800
+ "rewards/rejected": -0.004464106634259224,
801
+ "step": 430
802
+ },
803
+ {
804
+ "epoch": 0.32,
805
+ "grad_norm": 0.384765625,
806
+ "learning_rate": 4.2756957584576436e-07,
807
+ "logits/chosen": 0.051903001964092255,
808
+ "logits/rejected": 0.7332680821418762,
809
+ "logps/chosen": -291.6619873046875,
810
+ "logps/rejected": -264.76763916015625,
811
+ "loss": 0.6551,
812
+ "rewards/accuracies": 0.949999988079071,
813
+ "rewards/chosen": 0.07283146679401398,
814
+ "rewards/margins": 0.07750260829925537,
815
+ "rewards/margins_max": 0.11894078552722931,
816
+ "rewards/margins_min": 0.03606444224715233,
817
+ "rewards/margins_std": 0.05860241502523422,
818
+ "rewards/rejected": -0.004671142902225256,
819
+ "step": 440
820
+ },
821
+ {
822
+ "epoch": 0.33,
823
+ "grad_norm": 0.38671875,
824
+ "learning_rate": 4.22990982449109e-07,
825
+ "logits/chosen": 0.00915027130395174,
826
+ "logits/rejected": 0.7316643595695496,
827
+ "logps/chosen": -236.2955780029297,
828
+ "logps/rejected": -220.3606414794922,
829
+ "loss": 0.6569,
830
+ "rewards/accuracies": 0.9375,
831
+ "rewards/chosen": 0.06984328478574753,
832
+ "rewards/margins": 0.07647494226694107,
833
+ "rewards/margins_max": 0.11472654342651367,
834
+ "rewards/margins_min": 0.038223348557949066,
835
+ "rewards/margins_std": 0.05409592390060425,
836
+ "rewards/rejected": -0.006631660275161266,
837
+ "step": 450
838
+ },
839
+ {
840
+ "epoch": 0.34,
841
+ "grad_norm": 0.412109375,
842
+ "learning_rate": 4.1829824680607104e-07,
843
+ "logits/chosen": -0.07657341659069061,
844
+ "logits/rejected": 0.6622528433799744,
845
+ "logps/chosen": -279.7924499511719,
846
+ "logps/rejected": -269.6399230957031,
847
+ "loss": 0.6549,
848
+ "rewards/accuracies": 0.9750000238418579,
849
+ "rewards/chosen": 0.07170796394348145,
850
+ "rewards/margins": 0.0784129947423935,
851
+ "rewards/margins_max": 0.11032243072986603,
852
+ "rewards/margins_min": 0.046503566205501556,
853
+ "rewards/margins_std": 0.045126743614673615,
854
+ "rewards/rejected": -0.0067050280049443245,
855
+ "step": 460
856
+ },
857
+ {
858
+ "epoch": 0.35,
859
+ "grad_norm": 0.453125,
860
+ "learning_rate": 4.134944652594794e-07,
861
+ "logits/chosen": 0.0630350112915039,
862
+ "logits/rejected": 0.6474081873893738,
863
+ "logps/chosen": -261.22418212890625,
864
+ "logps/rejected": -239.87588500976562,
865
+ "loss": 0.6581,
866
+ "rewards/accuracies": 0.949999988079071,
867
+ "rewards/chosen": 0.05761648342013359,
868
+ "rewards/margins": 0.06358405947685242,
869
+ "rewards/margins_max": 0.09130324423313141,
870
+ "rewards/margins_min": 0.03586486726999283,
871
+ "rewards/margins_std": 0.03920084983110428,
872
+ "rewards/rejected": -0.0059675718657672405,
873
+ "step": 470
874
+ },
875
+ {
876
+ "epoch": 0.35,
877
+ "grad_norm": 0.392578125,
878
+ "learning_rate": 4.085828074220451e-07,
879
+ "logits/chosen": -0.19398275017738342,
880
+ "logits/rejected": 0.5935221910476685,
881
+ "logps/chosen": -270.58892822265625,
882
+ "logps/rejected": -245.28076171875,
883
+ "loss": 0.6561,
884
+ "rewards/accuracies": 0.9750000238418579,
885
+ "rewards/chosen": 0.06964448094367981,
886
+ "rewards/margins": 0.08161848783493042,
887
+ "rewards/margins_max": 0.12005958706140518,
888
+ "rewards/margins_min": 0.04317738488316536,
889
+ "rewards/margins_std": 0.0543639250099659,
890
+ "rewards/rejected": -0.011974002234637737,
891
+ "step": 480
892
+ },
893
+ {
894
+ "epoch": 0.36,
895
+ "grad_norm": 0.373046875,
896
+ "learning_rate": 4.035665140849994e-07,
897
+ "logits/chosen": -0.14574551582336426,
898
+ "logits/rejected": 0.5638002157211304,
899
+ "logps/chosen": -274.3948669433594,
900
+ "logps/rejected": -228.3678436279297,
901
+ "loss": 0.6602,
902
+ "rewards/accuracies": 0.949999988079071,
903
+ "rewards/chosen": 0.06451380997896194,
904
+ "rewards/margins": 0.07035050541162491,
905
+ "rewards/margins_max": 0.10534896701574326,
906
+ "rewards/margins_min": 0.035352036356925964,
907
+ "rewards/margins_std": 0.04949530214071274,
908
+ "rewards/rejected": -0.005836695898324251,
909
+ "step": 490
910
+ },
911
+ {
912
+ "epoch": 0.37,
913
+ "grad_norm": 0.470703125,
914
+ "learning_rate": 3.984488950797678e-07,
915
+ "logits/chosen": -0.040041010826826096,
916
+ "logits/rejected": 0.7338499426841736,
917
+ "logps/chosen": -265.7203063964844,
918
+ "logps/rejected": -230.03170776367188,
919
+ "loss": 0.6557,
920
+ "rewards/accuracies": 0.949999988079071,
921
+ "rewards/chosen": 0.07825618237257004,
922
+ "rewards/margins": 0.07796762883663177,
923
+ "rewards/margins_max": 0.11045694351196289,
924
+ "rewards/margins_min": 0.04547831788659096,
925
+ "rewards/margins_std": 0.04594682529568672,
926
+ "rewards/rejected": 0.00028855842538177967,
927
+ "step": 500
928
+ },
929
+ {
930
+ "epoch": 0.38,
931
+ "grad_norm": 0.46484375,
932
+ "learning_rate": 3.9323332709408904e-07,
933
+ "logits/chosen": 0.02443014457821846,
934
+ "logits/rejected": 0.604122519493103,
935
+ "logps/chosen": -254.78701782226562,
936
+ "logps/rejected": -233.02084350585938,
937
+ "loss": 0.6547,
938
+ "rewards/accuracies": 0.887499988079071,
939
+ "rewards/chosen": 0.060421328991651535,
940
+ "rewards/margins": 0.07197172939777374,
941
+ "rewards/margins_max": 0.11490993201732635,
942
+ "rewards/margins_min": 0.029033523052930832,
943
+ "rewards/margins_std": 0.060723792761564255,
944
+ "rewards/rejected": -0.01155039481818676,
945
+ "step": 510
946
+ },
947
+ {
948
+ "epoch": 0.38,
949
+ "grad_norm": 0.40234375,
950
+ "learning_rate": 3.879232514440227e-07,
951
+ "logits/chosen": 0.04711779206991196,
952
+ "logits/rejected": 0.9458179473876953,
953
+ "logps/chosen": -280.1850280761719,
954
+ "logps/rejected": -221.70785522460938,
955
+ "loss": 0.6533,
956
+ "rewards/accuracies": 0.9624999761581421,
957
+ "rewards/chosen": 0.07137273252010345,
958
+ "rewards/margins": 0.0817800909280777,
959
+ "rewards/margins_max": 0.12524864077568054,
960
+ "rewards/margins_min": 0.038311541080474854,
961
+ "rewards/margins_std": 0.06147381663322449,
962
+ "rewards/rejected": -0.01040735375136137,
963
+ "step": 520
964
+ },
965
+ {
966
+ "epoch": 0.39,
967
+ "grad_norm": 0.51171875,
968
+ "learning_rate": 3.825221718033129e-07,
969
+ "logits/chosen": 0.09109713137149811,
970
+ "logits/rejected": 0.7228142023086548,
971
+ "logps/chosen": -256.89202880859375,
972
+ "logps/rejected": -236.75460815429688,
973
+ "loss": 0.6552,
974
+ "rewards/accuracies": 0.9624999761581421,
975
+ "rewards/chosen": 0.06750547885894775,
976
+ "rewards/margins": 0.0867571011185646,
977
+ "rewards/margins_max": 0.12318144738674164,
978
+ "rewards/margins_min": 0.050332747399806976,
979
+ "rewards/margins_std": 0.05151181295514107,
980
+ "rewards/rejected": -0.019251624122262,
981
+ "step": 530
982
+ },
983
+ {
984
+ "epoch": 0.4,
985
+ "grad_norm": 0.375,
986
+ "learning_rate": 3.7703365189160746e-07,
987
+ "logits/chosen": -0.09472191333770752,
988
+ "logits/rejected": 0.6297842264175415,
989
+ "logps/chosen": -261.3214416503906,
990
+ "logps/rejected": -244.3570556640625,
991
+ "loss": 0.6562,
992
+ "rewards/accuracies": 0.925000011920929,
993
+ "rewards/chosen": 0.07096521556377411,
994
+ "rewards/margins": 0.07419043034315109,
995
+ "rewards/margins_max": 0.1066773533821106,
996
+ "rewards/margins_min": 0.041703514754772186,
997
+ "rewards/margins_std": 0.04594343155622482,
998
+ "rewards/rejected": -0.0032252087257802486,
999
+ "step": 540
1000
+ },
1001
+ {
1002
+ "epoch": 0.4,
1003
+ "grad_norm": 0.396484375,
1004
+ "learning_rate": 3.714613131230587e-07,
1005
+ "logits/chosen": -0.002119688782840967,
1006
+ "logits/rejected": 0.6269344091415405,
1007
+ "logps/chosen": -249.4778594970703,
1008
+ "logps/rejected": -223.02603149414062,
1009
+ "loss": 0.6572,
1010
+ "rewards/accuracies": 0.925000011920929,
1011
+ "rewards/chosen": 0.06649978458881378,
1012
+ "rewards/margins": 0.07572688907384872,
1013
+ "rewards/margins_max": 0.10936542600393295,
1014
+ "rewards/margins_min": 0.0420883484184742,
1015
+ "rewards/margins_std": 0.04757208377122879,
1016
+ "rewards/rejected": -0.009227104485034943,
1017
+ "step": 550
1018
+ },
1019
+ {
1020
+ "epoch": 0.41,
1021
+ "grad_norm": 0.427734375,
1022
+ "learning_rate": 3.6580883221685533e-07,
1023
+ "logits/chosen": 0.05928317829966545,
1024
+ "logits/rejected": 0.6542156338691711,
1025
+ "logps/chosen": -249.701171875,
1026
+ "logps/rejected": -235.28256225585938,
1027
+ "loss": 0.6546,
1028
+ "rewards/accuracies": 0.949999988079071,
1029
+ "rewards/chosen": 0.06236664578318596,
1030
+ "rewards/margins": 0.07323845475912094,
1031
+ "rewards/margins_max": 0.11050041764974594,
1032
+ "rewards/margins_min": 0.03597649186849594,
1033
+ "rewards/margins_std": 0.05269637703895569,
1034
+ "rewards/rejected": -0.010871809907257557,
1035
+ "step": 560
1036
+ },
1037
+ {
1038
+ "epoch": 0.42,
1039
+ "grad_norm": 0.39453125,
1040
+ "learning_rate": 3.6007993877126386e-07,
1041
+ "logits/chosen": -0.15580764412879944,
1042
+ "logits/rejected": 0.7158504724502563,
1043
+ "logps/chosen": -291.7669372558594,
1044
+ "logps/rejected": -232.95767211914062,
1045
+ "loss": 0.6561,
1046
+ "rewards/accuracies": 0.887499988079071,
1047
+ "rewards/chosen": 0.06417657434940338,
1048
+ "rewards/margins": 0.07455410063266754,
1049
+ "rewards/margins_max": 0.11633304506540298,
1050
+ "rewards/margins_min": 0.032775141298770905,
1051
+ "rewards/margins_std": 0.05908435583114624,
1052
+ "rewards/rejected": -0.010377511382102966,
1053
+ "step": 570
1054
+ },
1055
+ {
1056
+ "epoch": 0.43,
1057
+ "grad_norm": 0.490234375,
1058
+ "learning_rate": 3.5427841280277937e-07,
1059
+ "logits/chosen": 0.07802899181842804,
1060
+ "logits/rejected": 0.6800590753555298,
1061
+ "logps/chosen": -253.85107421875,
1062
+ "logps/rejected": -234.7869110107422,
1063
+ "loss": 0.6586,
1064
+ "rewards/accuracies": 0.9125000238418579,
1065
+ "rewards/chosen": 0.058535121381282806,
1066
+ "rewards/margins": 0.06843312084674835,
1067
+ "rewards/margins_max": 0.10552020370960236,
1068
+ "rewards/margins_min": 0.03134603053331375,
1069
+ "rewards/margins_std": 0.0524490587413311,
1070
+ "rewards/rejected": -0.0098979901522398,
1071
+ "step": 580
1072
+ },
1073
+ {
1074
+ "epoch": 0.43,
1075
+ "grad_norm": 0.482421875,
1076
+ "learning_rate": 3.484080822520096e-07,
1077
+ "logits/chosen": -0.049510225653648376,
1078
+ "logits/rejected": 0.7604032754898071,
1079
+ "logps/chosen": -268.8267822265625,
1080
+ "logps/rejected": -224.325439453125,
1081
+ "loss": 0.654,
1082
+ "rewards/accuracies": 0.9375,
1083
+ "rewards/chosen": 0.07341635227203369,
1084
+ "rewards/margins": 0.07777970284223557,
1085
+ "rewards/margins_max": 0.11911604553461075,
1086
+ "rewards/margins_min": 0.03644336014986038,
1087
+ "rewards/margins_std": 0.058458417654037476,
1088
+ "rewards/rejected": -0.004363355692476034,
1089
+ "step": 590
1090
+ },
1091
+ {
1092
+ "epoch": 0.44,
1093
+ "grad_norm": 0.400390625,
1094
+ "learning_rate": 3.4247282045793797e-07,
1095
+ "logits/chosen": -0.19325688481330872,
1096
+ "logits/rejected": 0.6270472407341003,
1097
+ "logps/chosen": -282.7298889160156,
1098
+ "logps/rejected": -252.6767578125,
1099
+ "loss": 0.6482,
1100
+ "rewards/accuracies": 0.987500011920929,
1101
+ "rewards/chosen": 0.07601656764745712,
1102
+ "rewards/margins": 0.09246112406253815,
1103
+ "rewards/margins_max": 0.13728852570056915,
1104
+ "rewards/margins_min": 0.04763372614979744,
1105
+ "rewards/margins_std": 0.06339551508426666,
1106
+ "rewards/rejected": -0.016444550827145576,
1107
+ "step": 600
1108
+ },
1109
+ {
1110
+ "epoch": 0.45,
1111
+ "grad_norm": 0.458984375,
1112
+ "learning_rate": 3.3647654360223144e-07,
1113
+ "logits/chosen": -0.05807774141430855,
1114
+ "logits/rejected": 0.6809174418449402,
1115
+ "logps/chosen": -299.19000244140625,
1116
+ "logps/rejected": -262.0146179199219,
1117
+ "loss": 0.6554,
1118
+ "rewards/accuracies": 0.9375,
1119
+ "rewards/chosen": 0.06890231370925903,
1120
+ "rewards/margins": 0.07707327604293823,
1121
+ "rewards/margins_max": 0.11320241540670395,
1122
+ "rewards/margins_min": 0.04094412922859192,
1123
+ "rewards/margins_std": 0.05109432339668274,
1124
+ "rewards/rejected": -0.0081709586083889,
1125
+ "step": 610
1126
+ },
1127
+ {
1128
+ "epoch": 0.46,
1129
+ "grad_norm": 0.4609375,
1130
+ "learning_rate": 3.30423208125281e-07,
1131
+ "logits/chosen": -0.15423426032066345,
1132
+ "logits/rejected": 0.6746962070465088,
1133
+ "logps/chosen": -361.3031921386719,
1134
+ "logps/rejected": -271.8617858886719,
1135
+ "loss": 0.6527,
1136
+ "rewards/accuracies": 0.987500011920929,
1137
+ "rewards/chosen": 0.08182433992624283,
1138
+ "rewards/margins": 0.08147110044956207,
1139
+ "rewards/margins_max": 0.1197957769036293,
1140
+ "rewards/margins_min": 0.04314642399549484,
1141
+ "rewards/margins_std": 0.05419927090406418,
1142
+ "rewards/rejected": 0.00035324436612427235,
1143
+ "step": 620
1144
+ },
1145
+ {
1146
+ "epoch": 0.46,
1147
+ "grad_norm": 0.53515625,
1148
+ "learning_rate": 3.2431680811567833e-07,
1149
+ "logits/chosen": -0.18638429045677185,
1150
+ "logits/rejected": 0.5596809387207031,
1151
+ "logps/chosen": -265.38189697265625,
1152
+ "logps/rejected": -216.4676513671875,
1153
+ "loss": 0.6498,
1154
+ "rewards/accuracies": 0.9375,
1155
+ "rewards/chosen": 0.07413068413734436,
1156
+ "rewards/margins": 0.08364250510931015,
1157
+ "rewards/margins_max": 0.11661893129348755,
1158
+ "rewards/margins_min": 0.05066608637571335,
1159
+ "rewards/margins_std": 0.046635713428258896,
1160
+ "rewards/rejected": -0.009511834010481834,
1161
+ "step": 630
1162
+ },
1163
+ {
1164
+ "epoch": 0.47,
1165
+ "grad_norm": 0.4765625,
1166
+ "learning_rate": 3.1816137267485136e-07,
1167
+ "logits/chosen": -0.01266663521528244,
1168
+ "logits/rejected": 0.7700913548469543,
1169
+ "logps/chosen": -270.8904113769531,
1170
+ "logps/rejected": -227.817138671875,
1171
+ "loss": 0.6499,
1172
+ "rewards/accuracies": 0.9624999761581421,
1173
+ "rewards/chosen": 0.08034569025039673,
1174
+ "rewards/margins": 0.09530878067016602,
1175
+ "rewards/margins_max": 0.13598743081092834,
1176
+ "rewards/margins_min": 0.05463013797998428,
1177
+ "rewards/margins_std": 0.0575282946228981,
1178
+ "rewards/rejected": -0.014963088557124138,
1179
+ "step": 640
1180
+ },
1181
+ {
1182
+ "epoch": 0.48,
1183
+ "grad_norm": 0.458984375,
1184
+ "learning_rate": 3.1196096325859815e-07,
1185
+ "logits/chosen": -0.09016970545053482,
1186
+ "logits/rejected": 0.5758002400398254,
1187
+ "logps/chosen": -248.17721557617188,
1188
+ "logps/rejected": -231.0363311767578,
1189
+ "loss": 0.6539,
1190
+ "rewards/accuracies": 0.925000011920929,
1191
+ "rewards/chosen": 0.06288131326436996,
1192
+ "rewards/margins": 0.08216816931962967,
1193
+ "rewards/margins_max": 0.12130717188119888,
1194
+ "rewards/margins_min": 0.04302916303277016,
1195
+ "rewards/margins_std": 0.05535091087222099,
1196
+ "rewards/rejected": -0.019286850467324257,
1197
+ "step": 650
1198
+ },
1199
+ {
1200
+ "epoch": 0.49,
1201
+ "grad_norm": 0.357421875,
1202
+ "learning_rate": 3.057196709972727e-07,
1203
+ "logits/chosen": -0.06700615584850311,
1204
+ "logits/rejected": 0.7238128781318665,
1205
+ "logps/chosen": -273.21044921875,
1206
+ "logps/rejected": -222.5702667236328,
1207
+ "loss": 0.6541,
1208
+ "rewards/accuracies": 0.925000011920929,
1209
+ "rewards/chosen": 0.07693549245595932,
1210
+ "rewards/margins": 0.08857451379299164,
1211
+ "rewards/margins_max": 0.13141193985939026,
1212
+ "rewards/margins_min": 0.045737091451883316,
1213
+ "rewards/margins_std": 0.0605812594294548,
1214
+ "rewards/rejected": -0.011639016680419445,
1215
+ "step": 660
1216
+ },
1217
+ {
1218
+ "epoch": 0.49,
1219
+ "grad_norm": 0.384765625,
1220
+ "learning_rate": 2.9944161399639086e-07,
1221
+ "logits/chosen": -0.17069509625434875,
1222
+ "logits/rejected": 0.5526743531227112,
1223
+ "logps/chosen": -274.2879943847656,
1224
+ "logps/rejected": -238.9512176513672,
1225
+ "loss": 0.6554,
1226
+ "rewards/accuracies": 0.9624999761581421,
1227
+ "rewards/chosen": 0.07316674292087555,
1228
+ "rewards/margins": 0.0783548504114151,
1229
+ "rewards/margins_max": 0.12053277343511581,
1230
+ "rewards/margins_min": 0.03617694973945618,
1231
+ "rewards/margins_std": 0.059648580849170685,
1232
+ "rewards/rejected": -0.0051881177350878716,
1233
+ "step": 670
1234
+ },
1235
+ {
1236
+ "epoch": 0.5,
1237
+ "grad_norm": 0.4453125,
1238
+ "learning_rate": 2.9313093461943824e-07,
1239
+ "logits/chosen": -0.16692019999027252,
1240
+ "logits/rejected": 0.6132463216781616,
1241
+ "logps/chosen": -266.98297119140625,
1242
+ "logps/rejected": -240.8970489501953,
1243
+ "loss": 0.6514,
1244
+ "rewards/accuracies": 0.949999988079071,
1245
+ "rewards/chosen": 0.0738702192902565,
1246
+ "rewards/margins": 0.08596795052289963,
1247
+ "rewards/margins_max": 0.12484346330165863,
1248
+ "rewards/margins_min": 0.04709243029356003,
1249
+ "rewards/margins_std": 0.05497828125953674,
1250
+ "rewards/rejected": -0.012097733095288277,
1251
+ "step": 680
1252
+ },
1253
+ {
1254
+ "epoch": 0.51,
1255
+ "grad_norm": 0.404296875,
1256
+ "learning_rate": 2.8679179675467104e-07,
1257
+ "logits/chosen": -0.028652098029851913,
1258
+ "logits/rejected": 0.7112724781036377,
1259
+ "logps/chosen": -263.5270080566406,
1260
+ "logps/rejected": -232.5438232421875,
1261
+ "loss": 0.6505,
1262
+ "rewards/accuracies": 0.9375,
1263
+ "rewards/chosen": 0.07920118421316147,
1264
+ "rewards/margins": 0.08954565227031708,
1265
+ "rewards/margins_max": 0.13445377349853516,
1266
+ "rewards/margins_min": 0.04463755339384079,
1267
+ "rewards/margins_std": 0.06350965052843094,
1268
+ "rewards/rejected": -0.010344477370381355,
1269
+ "step": 690
1270
+ },
1271
+ {
1272
+ "epoch": 0.52,
1273
+ "grad_norm": 0.44140625,
1274
+ "learning_rate": 2.80428383067716e-07,
1275
+ "logits/chosen": 0.04036666825413704,
1276
+ "logits/rejected": 0.5802733302116394,
1277
+ "logps/chosen": -276.1773681640625,
1278
+ "logps/rejected": -284.2413024902344,
1279
+ "loss": 0.6492,
1280
+ "rewards/accuracies": 0.949999988079071,
1281
+ "rewards/chosen": 0.07053406536579132,
1282
+ "rewards/margins": 0.09340489655733109,
1283
+ "rewards/margins_max": 0.13880935311317444,
1284
+ "rewards/margins_min": 0.048000432550907135,
1285
+ "rewards/margins_std": 0.06421160697937012,
1286
+ "rewards/rejected": -0.022870829328894615,
1287
+ "step": 700
1288
+ },
1289
+ {
1290
+ "epoch": 0.52,
1291
+ "grad_norm": 0.484375,
1292
+ "learning_rate": 2.7404489224177973e-07,
1293
+ "logits/chosen": -0.10626095533370972,
1294
+ "logits/rejected": 0.6656783819198608,
1295
+ "logps/chosen": -273.10369873046875,
1296
+ "logps/rejected": -240.2863006591797,
1297
+ "loss": 0.6555,
1298
+ "rewards/accuracies": 0.925000011920929,
1299
+ "rewards/chosen": 0.06227077171206474,
1300
+ "rewards/margins": 0.07790277898311615,
1301
+ "rewards/margins_max": 0.12013592571020126,
1302
+ "rewards/margins_min": 0.03566960617899895,
1303
+ "rewards/margins_std": 0.05972672253847122,
1304
+ "rewards/rejected": -0.015631992369890213,
1305
+ "step": 710
1306
+ },
1307
+ {
1308
+ "epoch": 0.53,
1309
+ "grad_norm": 0.392578125,
1310
+ "learning_rate": 2.676455362072894e-07,
1311
+ "logits/chosen": -0.09553918987512589,
1312
+ "logits/rejected": 0.7383521795272827,
1313
+ "logps/chosen": -302.948974609375,
1314
+ "logps/rejected": -269.911376953125,
1315
+ "loss": 0.6506,
1316
+ "rewards/accuracies": 0.9750000238418579,
1317
+ "rewards/chosen": 0.08486850559711456,
1318
+ "rewards/margins": 0.10140830278396606,
1319
+ "rewards/margins_max": 0.14273163676261902,
1320
+ "rewards/margins_min": 0.06008495017886162,
1321
+ "rewards/margins_std": 0.05844002962112427,
1322
+ "rewards/rejected": -0.016539786010980606,
1323
+ "step": 720
1324
+ },
1325
+ {
1326
+ "epoch": 0.54,
1327
+ "grad_norm": 0.37890625,
1328
+ "learning_rate": 2.612345373627937e-07,
1329
+ "logits/chosen": 0.11825668811798096,
1330
+ "logits/rejected": 0.7459693551063538,
1331
+ "logps/chosen": -275.9749450683594,
1332
+ "logps/rejected": -238.9975128173828,
1333
+ "loss": 0.6551,
1334
+ "rewards/accuracies": 0.9375,
1335
+ "rewards/chosen": 0.06423775851726532,
1336
+ "rewards/margins": 0.08155209571123123,
1337
+ "rewards/margins_max": 0.12171584367752075,
1338
+ "rewards/margins_min": 0.04138834401965141,
1339
+ "rewards/margins_std": 0.05680011585354805,
1340
+ "rewards/rejected": -0.017314324155449867,
1341
+ "step": 730
1342
+ },
1343
+ {
1344
+ "epoch": 0.54,
1345
+ "grad_norm": 0.40234375,
1346
+ "learning_rate": 2.54816125788955e-07,
1347
+ "logits/chosen": -0.04539821669459343,
1348
+ "logits/rejected": 0.8194332122802734,
1349
+ "logps/chosen": -290.15557861328125,
1350
+ "logps/rejected": -230.5435333251953,
1351
+ "loss": 0.6515,
1352
+ "rewards/accuracies": 0.9750000238418579,
1353
+ "rewards/chosen": 0.074457548558712,
1354
+ "rewards/margins": 0.08972884714603424,
1355
+ "rewards/margins_max": 0.13167297840118408,
1356
+ "rewards/margins_min": 0.04778473079204559,
1357
+ "rewards/margins_std": 0.059317946434020996,
1358
+ "rewards/rejected": -0.015271296724677086,
1359
+ "step": 740
1360
+ },
1361
+ {
1362
+ "epoch": 0.55,
1363
+ "grad_norm": 0.38671875,
1364
+ "learning_rate": 2.4839453645747467e-07,
1365
+ "logits/chosen": 0.04359906539320946,
1366
+ "logits/rejected": 0.622580349445343,
1367
+ "logps/chosen": -276.761474609375,
1368
+ "logps/rejected": -252.15261840820312,
1369
+ "loss": 0.6543,
1370
+ "rewards/accuracies": 0.987500011920929,
1371
+ "rewards/chosen": 0.07819774001836777,
1372
+ "rewards/margins": 0.0873616486787796,
1373
+ "rewards/margins_max": 0.14198613166809082,
1374
+ "rewards/margins_min": 0.03273715078830719,
1375
+ "rewards/margins_std": 0.07725068926811218,
1376
+ "rewards/rejected": -0.009163893759250641,
1377
+ "step": 750
1378
+ },
1379
+ {
1380
+ "epoch": 0.56,
1381
+ "grad_norm": 0.33984375,
1382
+ "learning_rate": 2.4197400643678987e-07,
1383
+ "logits/chosen": 0.03581676632165909,
1384
+ "logits/rejected": 0.8077651858329773,
1385
+ "logps/chosen": -280.2861328125,
1386
+ "logps/rejected": -233.9840545654297,
1387
+ "loss": 0.6528,
1388
+ "rewards/accuracies": 0.987500011920929,
1389
+ "rewards/chosen": 0.07635350525379181,
1390
+ "rewards/margins": 0.09026426076889038,
1391
+ "rewards/margins_max": 0.1272718459367752,
1392
+ "rewards/margins_min": 0.05325668305158615,
1393
+ "rewards/margins_std": 0.05233662202954292,
1394
+ "rewards/rejected": -0.013910764828324318,
1395
+ "step": 760
1396
+ },
1397
+ {
1398
+ "epoch": 0.57,
1399
+ "grad_norm": 0.435546875,
1400
+ "learning_rate": 2.3555877209638726e-07,
1401
+ "logits/chosen": -0.12947851419448853,
1402
+ "logits/rejected": 0.7489079833030701,
1403
+ "logps/chosen": -290.2663879394531,
1404
+ "logps/rejected": -254.322021484375,
1405
+ "loss": 0.6506,
1406
+ "rewards/accuracies": 0.9375,
1407
+ "rewards/chosen": 0.0825168713927269,
1408
+ "rewards/margins": 0.08864180743694305,
1409
+ "rewards/margins_max": 0.12674041092395782,
1410
+ "rewards/margins_min": 0.050543200224637985,
1411
+ "rewards/margins_std": 0.05387955904006958,
1412
+ "rewards/rejected": -0.006124935112893581,
1413
+ "step": 770
1414
+ },
1415
+ {
1416
+ "epoch": 0.57,
1417
+ "grad_norm": 0.390625,
1418
+ "learning_rate": 2.2915306631157817e-07,
1419
+ "logits/chosen": -0.04638870805501938,
1420
+ "logits/rejected": 0.6960999965667725,
1421
+ "logps/chosen": -238.04745483398438,
1422
+ "logps/rejected": -215.9326629638672,
1423
+ "loss": 0.6497,
1424
+ "rewards/accuracies": 0.949999988079071,
1425
+ "rewards/chosen": 0.06742905080318451,
1426
+ "rewards/margins": 0.08723022788763046,
1427
+ "rewards/margins_max": 0.13323740661144257,
1428
+ "rewards/margins_min": 0.04122306779026985,
1429
+ "rewards/margins_std": 0.0650639608502388,
1430
+ "rewards/rejected": -0.019801190122961998,
1431
+ "step": 780
1432
+ },
1433
+ {
1434
+ "epoch": 0.58,
1435
+ "grad_norm": 0.451171875,
1436
+ "learning_rate": 2.2276111567057887e-07,
1437
+ "logits/chosen": 0.19736722111701965,
1438
+ "logits/rejected": 0.8112510442733765,
1439
+ "logps/chosen": -263.0642395019531,
1440
+ "logps/rejected": -247.5417022705078,
1441
+ "loss": 0.6481,
1442
+ "rewards/accuracies": 0.9375,
1443
+ "rewards/chosen": 0.07030580937862396,
1444
+ "rewards/margins": 0.08302759379148483,
1445
+ "rewards/margins_max": 0.12399895489215851,
1446
+ "rewards/margins_min": 0.042056240141391754,
1447
+ "rewards/margins_std": 0.05794224888086319,
1448
+ "rewards/rejected": -0.01272178627550602,
1449
+ "step": 790
1450
+ },
1451
+ {
1452
+ "epoch": 0.59,
1453
+ "grad_norm": 0.474609375,
1454
+ "learning_rate": 2.1638713768573936e-07,
1455
+ "logits/chosen": 0.0460067018866539,
1456
+ "logits/rejected": 0.6669069528579712,
1457
+ "logps/chosen": -227.3536834716797,
1458
+ "logps/rejected": -220.83984375,
1459
+ "loss": 0.6476,
1460
+ "rewards/accuracies": 0.925000011920929,
1461
+ "rewards/chosen": 0.0669163167476654,
1462
+ "rewards/margins": 0.07873805612325668,
1463
+ "rewards/margins_max": 0.12541484832763672,
1464
+ "rewards/margins_min": 0.03206125646829605,
1465
+ "rewards/margins_std": 0.0660109594464302,
1466
+ "rewards/rejected": -0.011821739375591278,
1467
+ "step": 800
1468
+ },
1469
+ {
1470
+ "epoch": 0.6,
1471
+ "grad_norm": 0.484375,
1472
+ "learning_rate": 2.100353380107609e-07,
1473
+ "logits/chosen": -0.007617829833179712,
1474
+ "logits/rejected": 0.7070167660713196,
1475
+ "logps/chosen": -271.02691650390625,
1476
+ "logps/rejected": -219.64944458007812,
1477
+ "loss": 0.655,
1478
+ "rewards/accuracies": 0.925000011920929,
1479
+ "rewards/chosen": 0.05975126102566719,
1480
+ "rewards/margins": 0.07805275171995163,
1481
+ "rewards/margins_max": 0.11311063915491104,
1482
+ "rewards/margins_min": 0.04299486428499222,
1483
+ "rewards/margins_std": 0.04957933351397514,
1484
+ "rewards/rejected": -0.01830149069428444,
1485
+ "step": 810
1486
+ },
1487
+ {
1488
+ "epoch": 0.6,
1489
+ "grad_norm": 0.400390625,
1490
+ "learning_rate": 2.0370990766573698e-07,
1491
+ "logits/chosen": 0.11678284406661987,
1492
+ "logits/rejected": 0.8694367408752441,
1493
+ "logps/chosen": -319.4522705078125,
1494
+ "logps/rejected": -231.9202423095703,
1495
+ "loss": 0.653,
1496
+ "rewards/accuracies": 0.925000011920929,
1497
+ "rewards/chosen": 0.060270898044109344,
1498
+ "rewards/margins": 0.08039253205060959,
1499
+ "rewards/margins_max": 0.11875177919864655,
1500
+ "rewards/margins_min": 0.042033277451992035,
1501
+ "rewards/margins_std": 0.054248176515102386,
1502
+ "rewards/rejected": -0.020121637731790543,
1503
+ "step": 820
1504
+ },
1505
+ {
1506
+ "epoch": 0.61,
1507
+ "grad_norm": 0.357421875,
1508
+ "learning_rate": 1.974150202718513e-07,
1509
+ "logits/chosen": 0.037704307585954666,
1510
+ "logits/rejected": 0.6765593886375427,
1511
+ "logps/chosen": -252.72128295898438,
1512
+ "logps/rejected": -251.72232055664062,
1513
+ "loss": 0.6551,
1514
+ "rewards/accuracies": 0.9750000238418579,
1515
+ "rewards/chosen": 0.06962188333272934,
1516
+ "rewards/margins": 0.07926660031080246,
1517
+ "rewards/margins_max": 0.11337349563837051,
1518
+ "rewards/margins_min": 0.0451597236096859,
1519
+ "rewards/margins_std": 0.04823441430926323,
1520
+ "rewards/rejected": -0.009644721634685993,
1521
+ "step": 830
1522
+ },
1523
+ {
1524
+ "epoch": 0.62,
1525
+ "grad_norm": 0.4296875,
1526
+ "learning_rate": 1.9115482929755445e-07,
1527
+ "logits/chosen": 0.21195778250694275,
1528
+ "logits/rejected": 0.8628422021865845,
1529
+ "logps/chosen": -255.17141723632812,
1530
+ "logps/rejected": -230.54837036132812,
1531
+ "loss": 0.6542,
1532
+ "rewards/accuracies": 0.9624999761581421,
1533
+ "rewards/chosen": 0.05505312606692314,
1534
+ "rewards/margins": 0.08028857409954071,
1535
+ "rewards/margins_max": 0.12079598009586334,
1536
+ "rewards/margins_min": 0.03978116437792778,
1537
+ "rewards/margins_std": 0.05728612467646599,
1538
+ "rewards/rejected": -0.025235449895262718,
1539
+ "step": 840
1540
+ },
1541
+ {
1542
+ "epoch": 0.63,
1543
+ "grad_norm": 0.451171875,
1544
+ "learning_rate": 1.8493346531803887e-07,
1545
+ "logits/chosen": -0.0924936980009079,
1546
+ "logits/rejected": 0.6660225987434387,
1547
+ "logps/chosen": -253.78762817382812,
1548
+ "logps/rejected": -230.6340789794922,
1549
+ "loss": 0.654,
1550
+ "rewards/accuracies": 0.925000011920929,
1551
+ "rewards/chosen": 0.07504816353321075,
1552
+ "rewards/margins": 0.08388366550207138,
1553
+ "rewards/margins_max": 0.13147205114364624,
1554
+ "rewards/margins_min": 0.03629531338810921,
1555
+ "rewards/margins_std": 0.06730011850595474,
1556
+ "rewards/rejected": -0.00883551873266697,
1557
+ "step": 850
1558
+ },
1559
+ {
1560
+ "epoch": 0.63,
1561
+ "grad_norm": 0.62890625,
1562
+ "learning_rate": 1.7875503328981807e-07,
1563
+ "logits/chosen": 0.03313580900430679,
1564
+ "logits/rejected": 0.7846869826316833,
1565
+ "logps/chosen": -289.10919189453125,
1566
+ "logps/rejected": -260.46209716796875,
1567
+ "loss": 0.651,
1568
+ "rewards/accuracies": 0.9750000238418579,
1569
+ "rewards/chosen": 0.07691006362438202,
1570
+ "rewards/margins": 0.09373887628316879,
1571
+ "rewards/margins_max": 0.14286434650421143,
1572
+ "rewards/margins_min": 0.04461339861154556,
1573
+ "rewards/margins_std": 0.06947391480207443,
1574
+ "rewards/rejected": -0.01682879962027073,
1575
+ "step": 860
1576
+ },
1577
+ {
1578
+ "epoch": 0.64,
1579
+ "grad_norm": 0.4765625,
1580
+ "learning_rate": 1.7262360984221006e-07,
1581
+ "logits/chosen": 0.07170330733060837,
1582
+ "logits/rejected": 0.585077166557312,
1583
+ "logps/chosen": -232.71224975585938,
1584
+ "logps/rejected": -241.5399932861328,
1585
+ "loss": 0.649,
1586
+ "rewards/accuracies": 0.949999988079071,
1587
+ "rewards/chosen": 0.06494490802288055,
1588
+ "rewards/margins": 0.082070492208004,
1589
+ "rewards/margins_max": 0.11704879999160767,
1590
+ "rewards/margins_min": 0.04709218442440033,
1591
+ "rewards/margins_std": 0.049466803669929504,
1592
+ "rewards/rejected": -0.017125584185123444,
1593
+ "step": 870
1594
+ },
1595
+ {
1596
+ "epoch": 0.65,
1597
+ "grad_norm": 0.46875,
1598
+ "learning_rate": 1.6654324058751175e-07,
1599
+ "logits/chosen": -0.03041740693151951,
1600
+ "logits/rejected": 0.655995786190033,
1601
+ "logps/chosen": -307.1095275878906,
1602
+ "logps/rejected": -223.6023712158203,
1603
+ "loss": 0.6511,
1604
+ "rewards/accuracies": 0.9624999761581421,
1605
+ "rewards/chosen": 0.05577283352613449,
1606
+ "rewards/margins": 0.0676901713013649,
1607
+ "rewards/margins_max": 0.10265400260686874,
1608
+ "rewards/margins_min": 0.03272632881999016,
1609
+ "rewards/margins_std": 0.04944632947444916,
1610
+ "rewards/rejected": -0.011917328462004662,
1611
+ "step": 880
1612
+ },
1613
+ {
1614
+ "epoch": 0.65,
1615
+ "grad_norm": 0.439453125,
1616
+ "learning_rate": 1.6051793745163812e-07,
1617
+ "logits/chosen": -0.03592555597424507,
1618
+ "logits/rejected": 0.5853763222694397,
1619
+ "logps/chosen": -272.7641296386719,
1620
+ "logps/rejected": -243.679443359375,
1621
+ "loss": 0.6543,
1622
+ "rewards/accuracies": 0.949999988079071,
1623
+ "rewards/chosen": 0.07209287583827972,
1624
+ "rewards/margins": 0.0827474445104599,
1625
+ "rewards/margins_max": 0.11978437751531601,
1626
+ "rewards/margins_min": 0.045710500329732895,
1627
+ "rewards/margins_std": 0.05237815901637077,
1628
+ "rewards/rejected": -0.010654577054083347,
1629
+ "step": 890
1630
+ },
1631
+ {
1632
+ "epoch": 0.66,
1633
+ "grad_norm": 0.435546875,
1634
+ "learning_rate": 1.5455167602698915e-07,
1635
+ "logits/chosen": 0.10257561504840851,
1636
+ "logits/rejected": 0.7078714370727539,
1637
+ "logps/chosen": -239.56201171875,
1638
+ "logps/rejected": -240.7050323486328,
1639
+ "loss": 0.6508,
1640
+ "rewards/accuracies": 0.949999988079071,
1641
+ "rewards/chosen": 0.06524372100830078,
1642
+ "rewards/margins": 0.08618183434009552,
1643
+ "rewards/margins_max": 0.12894438207149506,
1644
+ "rewards/margins_min": 0.043419282883405685,
1645
+ "rewards/margins_std": 0.060475386679172516,
1646
+ "rewards/rejected": -0.020938115194439888,
1647
+ "step": 900
1648
+ },
1649
+ {
1650
+ "epoch": 0.67,
1651
+ "grad_norm": 0.5,
1652
+ "learning_rate": 1.4864839294928924e-07,
1653
+ "logits/chosen": 0.03240091726183891,
1654
+ "logits/rejected": 0.7821232080459595,
1655
+ "logps/chosen": -283.28338623046875,
1656
+ "logps/rejected": -228.65771484375,
1657
+ "loss": 0.6537,
1658
+ "rewards/accuracies": 0.9624999761581421,
1659
+ "rewards/chosen": 0.06496749818325043,
1660
+ "rewards/margins": 0.08208582550287247,
1661
+ "rewards/margins_max": 0.12210579216480255,
1662
+ "rewards/margins_min": 0.04206586629152298,
1663
+ "rewards/margins_std": 0.056596774607896805,
1664
+ "rewards/rejected": -0.01711832545697689,
1665
+ "step": 910
1666
+ },
1667
+ {
1668
+ "epoch": 0.68,
1669
+ "grad_norm": 0.515625,
1670
+ "learning_rate": 1.428119833001315e-07,
1671
+ "logits/chosen": -0.23715606331825256,
1672
+ "logits/rejected": 0.6829497218132019,
1673
+ "logps/chosen": -352.04071044921875,
1674
+ "logps/rejected": -288.7782287597656,
1675
+ "loss": 0.6512,
1676
+ "rewards/accuracies": 0.987500011920929,
1677
+ "rewards/chosen": 0.07856696844100952,
1678
+ "rewards/margins": 0.08786121755838394,
1679
+ "rewards/margins_max": 0.1261076182126999,
1680
+ "rewards/margins_min": 0.049614839255809784,
1681
+ "rewards/margins_std": 0.05408855527639389,
1682
+ "rewards/rejected": -0.009294259361922741,
1683
+ "step": 920
1684
+ },
1685
+ {
1686
+ "epoch": 0.68,
1687
+ "grad_norm": 0.41796875,
1688
+ "learning_rate": 1.370462980369401e-07,
1689
+ "logits/chosen": 0.029560232535004616,
1690
+ "logits/rejected": 0.729602038860321,
1691
+ "logps/chosen": -257.4678955078125,
1692
+ "logps/rejected": -246.0548858642578,
1693
+ "loss": 0.6487,
1694
+ "rewards/accuracies": 0.9624999761581421,
1695
+ "rewards/chosen": 0.0831000953912735,
1696
+ "rewards/margins": 0.09744415432214737,
1697
+ "rewards/margins_max": 0.15836606919765472,
1698
+ "rewards/margins_min": 0.03652223199605942,
1699
+ "rewards/margins_std": 0.08615661412477493,
1700
+ "rewards/rejected": -0.014344056136906147,
1701
+ "step": 930
1702
+ },
1703
+ {
1704
+ "epoch": 0.69,
1705
+ "grad_norm": 0.4453125,
1706
+ "learning_rate": 1.3135514145204606e-07,
1707
+ "logits/chosen": 0.046013571321964264,
1708
+ "logits/rejected": 0.5954084396362305,
1709
+ "logps/chosen": -269.13629150390625,
1710
+ "logps/rejected": -249.99008178710938,
1711
+ "loss": 0.6511,
1712
+ "rewards/accuracies": 0.9624999761581421,
1713
+ "rewards/chosen": 0.06792394816875458,
1714
+ "rewards/margins": 0.08472023159265518,
1715
+ "rewards/margins_max": 0.12758344411849976,
1716
+ "rewards/margins_min": 0.0418570414185524,
1717
+ "rewards/margins_std": 0.06061771512031555,
1718
+ "rewards/rejected": -0.016796285286545753,
1719
+ "step": 940
1720
+ },
1721
+ {
1722
+ "epoch": 0.7,
1723
+ "grad_norm": 0.515625,
1724
+ "learning_rate": 1.257422686625539e-07,
1725
+ "logits/chosen": -0.12281881272792816,
1726
+ "logits/rejected": 0.7430060505867004,
1727
+ "logps/chosen": -328.2705383300781,
1728
+ "logps/rejected": -239.69070434570312,
1729
+ "loss": 0.6552,
1730
+ "rewards/accuracies": 0.9624999761581421,
1731
+ "rewards/chosen": 0.06796546280384064,
1732
+ "rewards/margins": 0.08045408874750137,
1733
+ "rewards/margins_max": 0.1251390427350998,
1734
+ "rewards/margins_min": 0.03576914221048355,
1735
+ "rewards/margins_std": 0.06319407373666763,
1736
+ "rewards/rejected": -0.01248864270746708,
1737
+ "step": 950
1738
+ },
1739
+ {
1740
+ "epoch": 0.71,
1741
+ "grad_norm": 0.49609375,
1742
+ "learning_rate": 1.2021138313265444e-07,
1743
+ "logits/chosen": -0.05693725496530533,
1744
+ "logits/rejected": 0.5191583633422852,
1745
+ "logps/chosen": -267.0467224121094,
1746
+ "logps/rejected": -253.3563995361328,
1747
+ "loss": 0.6547,
1748
+ "rewards/accuracies": 0.9750000238418579,
1749
+ "rewards/chosen": 0.06574182212352753,
1750
+ "rewards/margins": 0.0813496932387352,
1751
+ "rewards/margins_max": 0.11537833511829376,
1752
+ "rewards/margins_min": 0.04732104018330574,
1753
+ "rewards/margins_std": 0.048123784363269806,
1754
+ "rewards/rejected": -0.015607871115207672,
1755
+ "step": 960
1756
+ },
1757
+ {
1758
+ "epoch": 0.71,
1759
+ "grad_norm": 0.421875,
1760
+ "learning_rate": 1.1476613423001974e-07,
1761
+ "logits/chosen": 0.01074630580842495,
1762
+ "logits/rejected": 0.7058780789375305,
1763
+ "logps/chosen": -268.10162353515625,
1764
+ "logps/rejected": -246.88095092773438,
1765
+ "loss": 0.6503,
1766
+ "rewards/accuracies": 0.925000011920929,
1767
+ "rewards/chosen": 0.06480704247951508,
1768
+ "rewards/margins": 0.07971958816051483,
1769
+ "rewards/margins_max": 0.12575621902942657,
1770
+ "rewards/margins_min": 0.03368296101689339,
1771
+ "rewards/margins_std": 0.0651056244969368,
1772
+ "rewards/rejected": -0.014912550337612629,
1773
+ "step": 970
1774
+ },
1775
+ {
1776
+ "epoch": 0.72,
1777
+ "grad_norm": 0.447265625,
1778
+ "learning_rate": 1.0941011481789042e-07,
1779
+ "logits/chosen": 0.16233515739440918,
1780
+ "logits/rejected": 0.6619356274604797,
1781
+ "logps/chosen": -268.57537841796875,
1782
+ "logps/rejected": -248.046142578125,
1783
+ "loss": 0.6547,
1784
+ "rewards/accuracies": 0.9750000238418579,
1785
+ "rewards/chosen": 0.061273593455553055,
1786
+ "rewards/margins": 0.07566945254802704,
1787
+ "rewards/margins_max": 0.1124287024140358,
1788
+ "rewards/margins_min": 0.03891021013259888,
1789
+ "rewards/margins_std": 0.051985424011945724,
1790
+ "rewards/rejected": -0.014395867474377155,
1791
+ "step": 980
1792
+ },
1793
+ {
1794
+ "epoch": 0.73,
1795
+ "grad_norm": 0.4296875,
1796
+ "learning_rate": 1.041468588844476e-07,
1797
+ "logits/chosen": 0.021765699610114098,
1798
+ "logits/rejected": 0.5723878145217896,
1799
+ "logps/chosen": -252.99365234375,
1800
+ "logps/rejected": -237.05233764648438,
1801
+ "loss": 0.6506,
1802
+ "rewards/accuracies": 0.925000011920929,
1803
+ "rewards/chosen": 0.06569187343120575,
1804
+ "rewards/margins": 0.08357974141836166,
1805
+ "rewards/margins_max": 0.12318392097949982,
1806
+ "rewards/margins_min": 0.043975573033094406,
1807
+ "rewards/margins_std": 0.056008756160736084,
1808
+ "rewards/rejected": -0.017887871712446213,
1809
+ "step": 990
1810
+ },
1811
+ {
1812
+ "epoch": 0.74,
1813
+ "grad_norm": 0.474609375,
1814
+ "learning_rate": 9.897983921102954e-08,
1815
+ "logits/chosen": -0.08432263880968094,
1816
+ "logits/rejected": 0.5865111351013184,
1817
+ "logps/chosen": -265.72735595703125,
1818
+ "logps/rejected": -234.95028686523438,
1819
+ "loss": 0.6572,
1820
+ "rewards/accuracies": 0.949999988079071,
1821
+ "rewards/chosen": 0.060374390333890915,
1822
+ "rewards/margins": 0.070238396525383,
1823
+ "rewards/margins_max": 0.10305947065353394,
1824
+ "rewards/margins_min": 0.037417326122522354,
1825
+ "rewards/margins_std": 0.04641600325703621,
1826
+ "rewards/rejected": -0.009864002466201782,
1827
+ "step": 1000
1828
+ },
1829
+ {
1830
+ "epoch": 0.74,
1831
+ "grad_norm": 0.42578125,
1832
+ "learning_rate": 9.391246508073433e-08,
1833
+ "logits/chosen": -0.17238160967826843,
1834
+ "logits/rejected": 0.6868064999580383,
1835
+ "logps/chosen": -321.6485595703125,
1836
+ "logps/rejected": -263.62188720703125,
1837
+ "loss": 0.6505,
1838
+ "rewards/accuracies": 0.9375,
1839
+ "rewards/chosen": 0.07812955230474472,
1840
+ "rewards/margins": 0.09409363567829132,
1841
+ "rewards/margins_max": 0.13336913287639618,
1842
+ "rewards/margins_min": 0.054818131029605865,
1843
+ "rewards/margins_std": 0.05554394796490669,
1844
+ "rewards/rejected": -0.015964074060320854,
1845
+ "step": 1010
1846
+ },
1847
+ {
1848
+ "epoch": 0.75,
1849
+ "grad_norm": 0.408203125,
1850
+ "learning_rate": 8.894808002892037e-08,
1851
+ "logits/chosen": -0.048636335879564285,
1852
+ "logits/rejected": 0.6718664765357971,
1853
+ "logps/chosen": -286.724365234375,
1854
+ "logps/rejected": -256.17767333984375,
1855
+ "loss": 0.6521,
1856
+ "rewards/accuracies": 0.9750000238418579,
1857
+ "rewards/chosen": 0.07918758690357208,
1858
+ "rewards/margins": 0.09503775835037231,
1859
+ "rewards/margins_max": 0.1390329748392105,
1860
+ "rewards/margins_min": 0.05104254558682442,
1861
+ "rewards/margins_std": 0.06221862882375717,
1862
+ "rewards/rejected": -0.015850163996219635,
1863
+ "step": 1020
1864
+ },
1865
+ {
1866
+ "epoch": 0.76,
1867
+ "grad_norm": 0.423828125,
1868
+ "learning_rate": 8.408995963708756e-08,
1869
+ "logits/chosen": -0.03906983509659767,
1870
+ "logits/rejected": 0.7512571215629578,
1871
+ "logps/chosen": -289.33660888671875,
1872
+ "logps/rejected": -260.1453857421875,
1873
+ "loss": 0.6546,
1874
+ "rewards/accuracies": 0.9624999761581421,
1875
+ "rewards/chosen": 0.07737907022237778,
1876
+ "rewards/margins": 0.08742411434650421,
1877
+ "rewards/margins_max": 0.12180934101343155,
1878
+ "rewards/margins_min": 0.053038887679576874,
1879
+ "rewards/margins_std": 0.048628050833940506,
1880
+ "rewards/rejected": -0.010045038536190987,
1881
+ "step": 1030
1882
+ },
1883
+ {
1884
+ "epoch": 0.77,
1885
+ "grad_norm": 0.484375,
1886
+ "learning_rate": 7.934130937159508e-08,
1887
+ "logits/chosen": 0.0004897325998172164,
1888
+ "logits/rejected": 0.7238715887069702,
1889
+ "logps/chosen": -279.21624755859375,
1890
+ "logps/rejected": -248.09957885742188,
1891
+ "loss": 0.6519,
1892
+ "rewards/accuracies": 0.987500011920929,
1893
+ "rewards/chosen": 0.07424648851156235,
1894
+ "rewards/margins": 0.08449728786945343,
1895
+ "rewards/margins_max": 0.1293594241142273,
1896
+ "rewards/margins_min": 0.039635173976421356,
1897
+ "rewards/margins_std": 0.06344461441040039,
1898
+ "rewards/rejected": -0.010250799357891083,
1899
+ "step": 1040
1900
+ },
1901
+ {
1902
+ "epoch": 0.77,
1903
+ "grad_norm": 0.455078125,
1904
+ "learning_rate": 7.470526246864364e-08,
1905
+ "logits/chosen": -0.028243955224752426,
1906
+ "logits/rejected": 0.6930115818977356,
1907
+ "logps/chosen": -301.51458740234375,
1908
+ "logps/rejected": -251.8220672607422,
1909
+ "loss": 0.6559,
1910
+ "rewards/accuracies": 0.9750000238418579,
1911
+ "rewards/chosen": 0.06157411262392998,
1912
+ "rewards/margins": 0.07462558895349503,
1913
+ "rewards/margins_max": 0.10729455947875977,
1914
+ "rewards/margins_min": 0.04195661470293999,
1915
+ "rewards/margins_std": 0.04620090872049332,
1916
+ "rewards/rejected": -0.01305147260427475,
1917
+ "step": 1050
1918
+ },
1919
+ {
1920
+ "epoch": 0.78,
1921
+ "grad_norm": 0.494140625,
1922
+ "learning_rate": 7.018487786691512e-08,
1923
+ "logits/chosen": -0.18514272570610046,
1924
+ "logits/rejected": 0.8259115219116211,
1925
+ "logps/chosen": -284.2102355957031,
1926
+ "logps/rejected": -236.52206420898438,
1927
+ "loss": 0.6474,
1928
+ "rewards/accuracies": 0.987500011920929,
1929
+ "rewards/chosen": 0.09142851829528809,
1930
+ "rewards/margins": 0.10867790132761002,
1931
+ "rewards/margins_max": 0.16026386618614197,
1932
+ "rewards/margins_min": 0.05709192156791687,
1933
+ "rewards/margins_std": 0.07295358926057816,
1934
+ "rewards/rejected": -0.01724938116967678,
1935
+ "step": 1060
1936
+ },
1937
+ {
1938
+ "epoch": 0.79,
1939
+ "grad_norm": 0.4296875,
1940
+ "learning_rate": 6.578313818923559e-08,
1941
+ "logits/chosen": 0.12218723446130753,
1942
+ "logits/rejected": 0.6379785537719727,
1943
+ "logps/chosen": -238.366943359375,
1944
+ "logps/rejected": -233.61026000976562,
1945
+ "loss": 0.6535,
1946
+ "rewards/accuracies": 0.9624999761581421,
1947
+ "rewards/chosen": 0.055001966655254364,
1948
+ "rewards/margins": 0.06915679574012756,
1949
+ "rewards/margins_max": 0.09845878183841705,
1950
+ "rewards/margins_min": 0.039854805916547775,
1951
+ "rewards/margins_std": 0.04143926501274109,
1952
+ "rewards/rejected": -0.014154826290905476,
1953
+ "step": 1070
1954
+ },
1955
+ {
1956
+ "epoch": 0.79,
1957
+ "grad_norm": 0.40234375,
1958
+ "learning_rate": 6.15029477745925e-08,
1959
+ "logits/chosen": -0.1947169005870819,
1960
+ "logits/rejected": 0.7179273366928101,
1961
+ "logps/chosen": -295.55096435546875,
1962
+ "logps/rejected": -222.1858367919922,
1963
+ "loss": 0.6512,
1964
+ "rewards/accuracies": 0.9375,
1965
+ "rewards/chosen": 0.07841251790523529,
1966
+ "rewards/margins": 0.08169189840555191,
1967
+ "rewards/margins_max": 0.11758895218372345,
1968
+ "rewards/margins_min": 0.045794855803251266,
1969
+ "rewards/margins_std": 0.05076608806848526,
1970
+ "rewards/rejected": -0.003279387950897217,
1971
+ "step": 1080
1972
+ },
1973
+ {
1974
+ "epoch": 0.8,
1975
+ "grad_norm": 0.44921875,
1976
+ "learning_rate": 5.734713076180486e-08,
1977
+ "logits/chosen": 0.006981179118156433,
1978
+ "logits/rejected": 0.7452491521835327,
1979
+ "logps/chosen": -260.1736145019531,
1980
+ "logps/rejected": -224.355224609375,
1981
+ "loss": 0.6557,
1982
+ "rewards/accuracies": 0.987500011920929,
1983
+ "rewards/chosen": 0.06287099421024323,
1984
+ "rewards/margins": 0.07665695995092392,
1985
+ "rewards/margins_max": 0.11319824308156967,
1986
+ "rewards/margins_min": 0.04011568799614906,
1987
+ "rewards/margins_std": 0.05167717486619949,
1988
+ "rewards/rejected": -0.013785977847874165,
1989
+ "step": 1090
1990
+ },
1991
+ {
1992
+ "epoch": 0.81,
1993
+ "grad_norm": 0.48046875,
1994
+ "learning_rate": 5.3318429226110875e-08,
1995
+ "logits/chosen": 0.015549438074231148,
1996
+ "logits/rejected": 0.6923533082008362,
1997
+ "logps/chosen": -269.09747314453125,
1998
+ "logps/rejected": -236.7939910888672,
1999
+ "loss": 0.6512,
2000
+ "rewards/accuracies": 0.9624999761581421,
2001
+ "rewards/chosen": 0.06426068395376205,
2002
+ "rewards/margins": 0.08864103257656097,
2003
+ "rewards/margins_max": 0.12798473238945007,
2004
+ "rewards/margins_min": 0.04929735139012337,
2005
+ "rewards/margins_std": 0.05564037710428238,
2006
+ "rewards/rejected": -0.024380357936024666,
2007
+ "step": 1100
2008
+ },
2009
+ {
2010
+ "epoch": 0.82,
2011
+ "grad_norm": 0.419921875,
2012
+ "learning_rate": 4.9419501369902026e-08,
2013
+ "logits/chosen": -0.029879633337259293,
2014
+ "logits/rejected": 0.7326146960258484,
2015
+ "logps/chosen": -260.98968505859375,
2016
+ "logps/rejected": -220.9407958984375,
2017
+ "loss": 0.6499,
2018
+ "rewards/accuracies": 0.949999988079071,
2019
+ "rewards/chosen": 0.07826592028141022,
2020
+ "rewards/margins": 0.09226210415363312,
2021
+ "rewards/margins_max": 0.1359224021434784,
2022
+ "rewards/margins_min": 0.04860180616378784,
2023
+ "rewards/margins_std": 0.06174498051404953,
2024
+ "rewards/rejected": -0.0139961838722229,
2025
+ "step": 1110
2026
+ },
2027
+ {
2028
+ "epoch": 0.82,
2029
+ "grad_norm": 0.42578125,
2030
+ "learning_rate": 4.5652919768798896e-08,
2031
+ "logits/chosen": -0.1553410142660141,
2032
+ "logits/rejected": 0.6449904441833496,
2033
+ "logps/chosen": -280.3263244628906,
2034
+ "logps/rejected": -242.4776153564453,
2035
+ "loss": 0.6512,
2036
+ "rewards/accuracies": 0.987500011920929,
2037
+ "rewards/chosen": 0.074976846575737,
2038
+ "rewards/margins": 0.08818279206752777,
2039
+ "rewards/margins_max": 0.12042693048715591,
2040
+ "rewards/margins_min": 0.05593867227435112,
2041
+ "rewards/margins_std": 0.04560008645057678,
2042
+ "rewards/rejected": -0.013205950148403645,
2043
+ "step": 1120
2044
+ },
2045
+ {
2046
+ "epoch": 0.83,
2047
+ "grad_norm": 0.474609375,
2048
+ "learning_rate": 4.2021169674223536e-08,
2049
+ "logits/chosen": -0.12581007182598114,
2050
+ "logits/rejected": 0.6627474427223206,
2051
+ "logps/chosen": -274.0786437988281,
2052
+ "logps/rejected": -250.4473114013672,
2053
+ "loss": 0.6483,
2054
+ "rewards/accuracies": 0.987500011920929,
2055
+ "rewards/chosen": 0.0676179900765419,
2056
+ "rewards/margins": 0.08669795840978622,
2057
+ "rewards/margins_max": 0.12913702428340912,
2058
+ "rewards/margins_min": 0.04425888508558273,
2059
+ "rewards/margins_std": 0.060017913579940796,
2060
+ "rewards/rejected": -0.01907997578382492,
2061
+ "step": 1130
2062
+ },
2063
+ {
2064
+ "epoch": 0.84,
2065
+ "grad_norm": 0.369140625,
2066
+ "learning_rate": 3.852664737359046e-08,
2067
+ "logits/chosen": 0.08468042314052582,
2068
+ "logits/rejected": 0.7667611241340637,
2069
+ "logps/chosen": -249.0133819580078,
2070
+ "logps/rejected": -232.16561889648438,
2071
+ "loss": 0.6497,
2072
+ "rewards/accuracies": 0.9750000238418579,
2073
+ "rewards/chosen": 0.07480134069919586,
2074
+ "rewards/margins": 0.09523622691631317,
2075
+ "rewards/margins_max": 0.15331092476844788,
2076
+ "rewards/margins_min": 0.037161558866500854,
2077
+ "rewards/margins_std": 0.08212999999523163,
2078
+ "rewards/rejected": -0.020434897392988205,
2079
+ "step": 1140
2080
+ },
2081
+ {
2082
+ "epoch": 0.85,
2083
+ "grad_norm": 0.373046875,
2084
+ "learning_rate": 3.5171658609197824e-08,
2085
+ "logits/chosen": -0.03485478088259697,
2086
+ "logits/rejected": 0.5982181429862976,
2087
+ "logps/chosen": -252.3005828857422,
2088
+ "logps/rejected": -240.3730926513672,
2089
+ "loss": 0.6515,
2090
+ "rewards/accuracies": 0.9750000238418579,
2091
+ "rewards/chosen": 0.06789231300354004,
2092
+ "rewards/margins": 0.0819520652294159,
2093
+ "rewards/margins_max": 0.11909864842891693,
2094
+ "rewards/margins_min": 0.04480547830462456,
2095
+ "rewards/margins_std": 0.05253320187330246,
2096
+ "rewards/rejected": -0.014059755019843578,
2097
+ "step": 1150
2098
+ },
2099
+ {
2100
+ "epoch": 0.85,
2101
+ "grad_norm": 0.435546875,
2102
+ "learning_rate": 3.195841705686139e-08,
2103
+ "logits/chosen": 0.001758894301019609,
2104
+ "logits/rejected": 0.6828774213790894,
2105
+ "logps/chosen": -259.75640869140625,
2106
+ "logps/rejected": -243.09970092773438,
2107
+ "loss": 0.6517,
2108
+ "rewards/accuracies": 0.987500011920929,
2109
+ "rewards/chosen": 0.06515143811702728,
2110
+ "rewards/margins": 0.07973488420248032,
2111
+ "rewards/margins_max": 0.11704728752374649,
2112
+ "rewards/margins_min": 0.042422495782375336,
2113
+ "rewards/margins_std": 0.05276770144701004,
2114
+ "rewards/rejected": -0.014583444222807884,
2115
+ "step": 1160
2116
+ },
2117
+ {
2118
+ "epoch": 0.86,
2119
+ "grad_norm": 0.4375,
2120
+ "learning_rate": 2.8889042865294837e-08,
2121
+ "logits/chosen": 0.0009852380026131868,
2122
+ "logits/rejected": 0.7770395278930664,
2123
+ "logps/chosen": -252.7977752685547,
2124
+ "logps/rejected": -211.03494262695312,
2125
+ "loss": 0.652,
2126
+ "rewards/accuracies": 0.949999988079071,
2127
+ "rewards/chosen": 0.06879086047410965,
2128
+ "rewards/margins": 0.0823785737156868,
2129
+ "rewards/margins_max": 0.126399964094162,
2130
+ "rewards/margins_min": 0.038357168436050415,
2131
+ "rewards/margins_std": 0.06225566938519478,
2132
+ "rewards/rejected": -0.013587711378932,
2133
+ "step": 1170
2134
+ },
2135
+ {
2136
+ "epoch": 0.87,
2137
+ "grad_norm": 0.5234375,
2138
+ "learning_rate": 2.5965561257202036e-08,
2139
+ "logits/chosen": -0.0783575028181076,
2140
+ "logits/rejected": 0.7545520067214966,
2141
+ "logps/chosen": -272.24041748046875,
2142
+ "logps/rejected": -247.2455291748047,
2143
+ "loss": 0.6473,
2144
+ "rewards/accuracies": 0.949999988079071,
2145
+ "rewards/chosen": 0.09078168123960495,
2146
+ "rewards/margins": 0.10268386453390121,
2147
+ "rewards/margins_max": 0.14412382245063782,
2148
+ "rewards/margins_min": 0.06124391406774521,
2149
+ "rewards/margins_std": 0.05860493704676628,
2150
+ "rewards/rejected": -0.011902175843715668,
2151
+ "step": 1180
2152
+ },
2153
+ {
2154
+ "epoch": 0.88,
2155
+ "grad_norm": 0.80078125,
2156
+ "learning_rate": 2.318990119300218e-08,
2157
+ "logits/chosen": 0.05255628377199173,
2158
+ "logits/rejected": 0.6947780847549438,
2159
+ "logps/chosen": -243.50509643554688,
2160
+ "logps/rejected": -212.8872528076172,
2161
+ "loss": 0.6546,
2162
+ "rewards/accuracies": 0.9375,
2163
+ "rewards/chosen": 0.05941227823495865,
2164
+ "rewards/margins": 0.07317686825990677,
2165
+ "rewards/margins_max": 0.11058169603347778,
2166
+ "rewards/margins_min": 0.03577204793691635,
2167
+ "rewards/margins_std": 0.052898406982421875,
2168
+ "rewards/rejected": -0.013764592818915844,
2169
+ "step": 1190
2170
+ },
2171
+ {
2172
+ "epoch": 0.88,
2173
+ "grad_norm": 0.4296875,
2174
+ "learning_rate": 2.0563894098070216e-08,
2175
+ "logits/chosen": -0.0692586898803711,
2176
+ "logits/rejected": 0.7870053052902222,
2177
+ "logps/chosen": -309.2232666015625,
2178
+ "logps/rejected": -266.5531921386719,
2179
+ "loss": 0.6476,
2180
+ "rewards/accuracies": 0.925000011920929,
2181
+ "rewards/chosen": 0.08217885345220566,
2182
+ "rewards/margins": 0.09460541605949402,
2183
+ "rewards/margins_max": 0.13726098835468292,
2184
+ "rewards/margins_min": 0.05194983631372452,
2185
+ "rewards/margins_std": 0.06032410264015198,
2186
+ "rewards/rejected": -0.012426558881998062,
2187
+ "step": 1200
2188
+ },
2189
+ {
2190
+ "epoch": 0.89,
2191
+ "grad_norm": 0.34375,
2192
+ "learning_rate": 1.8089272654333353e-08,
2193
+ "logits/chosen": -0.046720363199710846,
2194
+ "logits/rejected": 0.6504586338996887,
2195
+ "logps/chosen": -261.49298095703125,
2196
+ "logps/rejected": -223.7759552001953,
2197
+ "loss": 0.6526,
2198
+ "rewards/accuracies": 0.9375,
2199
+ "rewards/chosen": 0.07228533923625946,
2200
+ "rewards/margins": 0.07905267179012299,
2201
+ "rewards/margins_max": 0.12283986806869507,
2202
+ "rewards/margins_min": 0.035265471786260605,
2203
+ "rewards/margins_std": 0.06192445755004883,
2204
+ "rewards/rejected": -0.006767333950847387,
2205
+ "step": 1210
2206
+ },
2207
+ {
2208
+ "epoch": 0.9,
2209
+ "grad_norm": 0.4140625,
2210
+ "learning_rate": 1.5767669657019005e-08,
2211
+ "logits/chosen": -0.1003817468881607,
2212
+ "logits/rejected": 0.7918148636817932,
2213
+ "logps/chosen": -284.4376525878906,
2214
+ "logps/rejected": -226.739990234375,
2215
+ "loss": 0.6514,
2216
+ "rewards/accuracies": 0.949999988079071,
2217
+ "rewards/chosen": 0.07430224120616913,
2218
+ "rewards/margins": 0.08967499434947968,
2219
+ "rewards/margins_max": 0.13941441476345062,
2220
+ "rewards/margins_min": 0.03993559628725052,
2221
+ "rewards/margins_std": 0.07034213840961456,
2222
+ "rewards/rejected": -0.015372763387858868,
2223
+ "step": 1220
2224
+ },
2225
+ {
2226
+ "epoch": 0.91,
2227
+ "grad_norm": 0.42578125,
2228
+ "learning_rate": 1.3600616937310267e-08,
2229
+ "logits/chosen": -0.13235566020011902,
2230
+ "logits/rejected": 0.602156400680542,
2231
+ "logps/chosen": -261.45050048828125,
2232
+ "logps/rejected": -231.56277465820312,
2233
+ "loss": 0.647,
2234
+ "rewards/accuracies": 0.9624999761581421,
2235
+ "rewards/chosen": 0.0806741714477539,
2236
+ "rewards/margins": 0.10508724302053452,
2237
+ "rewards/margins_max": 0.1610412895679474,
2238
+ "rewards/margins_min": 0.049133218824863434,
2239
+ "rewards/margins_std": 0.07913094758987427,
2240
+ "rewards/rejected": -0.024413082748651505,
2241
+ "step": 1230
2242
+ },
2243
+ {
2244
+ "epoch": 0.91,
2245
+ "grad_norm": 0.478515625,
2246
+ "learning_rate": 1.1589544351619047e-08,
2247
+ "logits/chosen": 0.02703939750790596,
2248
+ "logits/rejected": 0.8025895357131958,
2249
+ "logps/chosen": -334.37176513671875,
2250
+ "logps/rejected": -263.4154968261719,
2251
+ "loss": 0.6488,
2252
+ "rewards/accuracies": 0.949999988079071,
2253
+ "rewards/chosen": 0.07323239743709564,
2254
+ "rewards/margins": 0.08194652944803238,
2255
+ "rewards/margins_max": 0.12296142429113388,
2256
+ "rewards/margins_min": 0.04093163460493088,
2257
+ "rewards/margins_std": 0.05800382047891617,
2258
+ "rewards/rejected": -0.008714130148291588,
2259
+ "step": 1240
2260
+ },
2261
+ {
2262
+ "epoch": 0.92,
2263
+ "grad_norm": 0.359375,
2264
+ "learning_rate": 9.735778838143749e-09,
2265
+ "logits/chosen": -0.0286453515291214,
2266
+ "logits/rejected": 0.7510533332824707,
2267
+ "logps/chosen": -311.01959228515625,
2268
+ "logps/rejected": -240.82870483398438,
2269
+ "loss": 0.6492,
2270
+ "rewards/accuracies": 0.987500011920929,
2271
+ "rewards/chosen": 0.07665716856718063,
2272
+ "rewards/margins": 0.09343422949314117,
2273
+ "rewards/margins_max": 0.13625505566596985,
2274
+ "rewards/margins_min": 0.0506133958697319,
2275
+ "rewards/margins_std": 0.06055780500173569,
2276
+ "rewards/rejected": -0.01677705906331539,
2277
+ "step": 1250
2278
+ },
2279
+ {
2280
+ "epoch": 0.93,
2281
+ "grad_norm": 0.44140625,
2282
+ "learning_rate": 8.040543541333655e-09,
2283
+ "logits/chosen": 0.05244514346122742,
2284
+ "logits/rejected": 0.7751520276069641,
2285
+ "logps/chosen": -310.7460021972656,
2286
+ "logps/rejected": -286.03826904296875,
2287
+ "loss": 0.6489,
2288
+ "rewards/accuracies": 0.987500011920929,
2289
+ "rewards/chosen": 0.06985687464475632,
2290
+ "rewards/margins": 0.0900772362947464,
2291
+ "rewards/margins_max": 0.12786877155303955,
2292
+ "rewards/margins_min": 0.052285682410001755,
2293
+ "rewards/margins_std": 0.053445328027009964,
2294
+ "rewards/rejected": -0.020220357924699783,
2295
+ "step": 1260
2296
+ },
2297
+ {
2298
+ "epoch": 0.93,
2299
+ "grad_norm": 0.5,
2300
+ "learning_rate": 6.504957004838746e-09,
2301
+ "logits/chosen": -0.03165416046977043,
2302
+ "logits/rejected": 0.822761058807373,
2303
+ "logps/chosen": -310.04400634765625,
2304
+ "logps/rejected": -237.48782348632812,
2305
+ "loss": 0.6523,
2306
+ "rewards/accuracies": 0.9750000238418579,
2307
+ "rewards/chosen": 0.06924634426832199,
2308
+ "rewards/margins": 0.08929922431707382,
2309
+ "rewards/margins_max": 0.14400045573711395,
2310
+ "rewards/margins_min": 0.034597981721162796,
2311
+ "rewards/margins_std": 0.07735923677682877,
2312
+ "rewards/rejected": -0.020052876323461533,
2313
+ "step": 1270
2314
+ },
2315
+ {
2316
+ "epoch": 0.94,
2317
+ "grad_norm": 0.4765625,
2318
+ "learning_rate": 5.130032433476483e-09,
2319
+ "logits/chosen": -0.011189639568328857,
2320
+ "logits/rejected": 0.691806972026825,
2321
+ "logps/chosen": -271.3976135253906,
2322
+ "logps/rejected": -269.19390869140625,
2323
+ "loss": 0.6545,
2324
+ "rewards/accuracies": 0.9624999761581421,
2325
+ "rewards/chosen": 0.08424172550439835,
2326
+ "rewards/margins": 0.09692950546741486,
2327
+ "rewards/margins_max": 0.14971934258937836,
2328
+ "rewards/margins_min": 0.044139645993709564,
2329
+ "rewards/margins_std": 0.07465612888336182,
2330
+ "rewards/rejected": -0.012687774375081062,
2331
+ "step": 1280
2332
+ },
2333
+ {
2334
+ "epoch": 0.95,
2335
+ "grad_norm": 0.43359375,
2336
+ "learning_rate": 3.916677024702858e-09,
2337
+ "logits/chosen": 0.009526084177196026,
2338
+ "logits/rejected": 0.7200338244438171,
2339
+ "logps/chosen": -255.1676788330078,
2340
+ "logps/rejected": -234.27432250976562,
2341
+ "loss": 0.6534,
2342
+ "rewards/accuracies": 0.9750000238418579,
2343
+ "rewards/chosen": 0.06782687455415726,
2344
+ "rewards/margins": 0.08498277515172958,
2345
+ "rewards/margins_max": 0.1279430389404297,
2346
+ "rewards/margins_min": 0.04202253371477127,
2347
+ "rewards/margins_std": 0.06075497344136238,
2348
+ "rewards/rejected": -0.017155908048152924,
2349
+ "step": 1290
2350
+ },
2351
+ {
2352
+ "epoch": 0.96,
2353
+ "grad_norm": 0.51953125,
2354
+ "learning_rate": 2.865691370028761e-09,
2355
+ "logits/chosen": -0.11016593128442764,
2356
+ "logits/rejected": 0.7786885499954224,
2357
+ "logps/chosen": -310.4071960449219,
2358
+ "logps/rejected": -242.39505004882812,
2359
+ "loss": 0.6487,
2360
+ "rewards/accuracies": 1.0,
2361
+ "rewards/chosen": 0.0797882154583931,
2362
+ "rewards/margins": 0.09522275626659393,
2363
+ "rewards/margins_max": 0.13675549626350403,
2364
+ "rewards/margins_min": 0.05369000881910324,
2365
+ "rewards/margins_std": 0.0587361678481102,
2366
+ "rewards/rejected": -0.015434542670845985,
2367
+ "step": 1300
2368
+ },
2369
+ {
2370
+ "epoch": 0.96,
2371
+ "grad_norm": 0.482421875,
2372
+ "learning_rate": 1.977768926776896e-09,
2373
+ "logits/chosen": 0.07193834334611893,
2374
+ "logits/rejected": 0.7082736492156982,
2375
+ "logps/chosen": -269.1424255371094,
2376
+ "logps/rejected": -258.5067443847656,
2377
+ "loss": 0.6549,
2378
+ "rewards/accuracies": 0.949999988079071,
2379
+ "rewards/chosen": 0.07826878130435944,
2380
+ "rewards/margins": 0.0938304141163826,
2381
+ "rewards/margins_max": 0.13868080079555511,
2382
+ "rewards/margins_min": 0.04898003116250038,
2383
+ "rewards/margins_std": 0.06342802941799164,
2384
+ "rewards/rejected": -0.01556164026260376,
2385
+ "step": 1310
2386
+ },
2387
+ {
2388
+ "epoch": 0.97,
2389
+ "grad_norm": 0.474609375,
2390
+ "learning_rate": 1.2534955605274233e-09,
2391
+ "logits/chosen": -0.014053803868591785,
2392
+ "logits/rejected": 0.777262806892395,
2393
+ "logps/chosen": -242.13796997070312,
2394
+ "logps/rejected": -218.4778594970703,
2395
+ "loss": 0.6505,
2396
+ "rewards/accuracies": 0.9750000238418579,
2397
+ "rewards/chosen": 0.07983841001987457,
2398
+ "rewards/margins": 0.09085123986005783,
2399
+ "rewards/margins_max": 0.1351737529039383,
2400
+ "rewards/margins_min": 0.046528737992048264,
2401
+ "rewards/margins_std": 0.06268148869276047,
2402
+ "rewards/rejected": -0.011012842878699303,
2403
+ "step": 1320
2404
+ },
2405
+ {
2406
+ "epoch": 0.98,
2407
+ "grad_norm": 0.48828125,
2408
+ "learning_rate": 6.933491585542351e-10,
2409
+ "logits/chosen": 0.03265099227428436,
2410
+ "logits/rejected": 0.7704384326934814,
2411
+ "logps/chosen": -293.1032409667969,
2412
+ "logps/rejected": -263.2946472167969,
2413
+ "loss": 0.6471,
2414
+ "rewards/accuracies": 0.9624999761581421,
2415
+ "rewards/chosen": 0.08791947364807129,
2416
+ "rewards/margins": 0.10431689023971558,
2417
+ "rewards/margins_max": 0.15510039031505585,
2418
+ "rewards/margins_min": 0.05353338643908501,
2419
+ "rewards/margins_std": 0.07181872427463531,
2420
+ "rewards/rejected": -0.016397414728999138,
2421
+ "step": 1330
2422
+ },
2423
+ {
2424
+ "epoch": 0.99,
2425
+ "grad_norm": 0.451171875,
2426
+ "learning_rate": 2.9769931450737694e-10,
2427
+ "logits/chosen": 0.03406764939427376,
2428
+ "logits/rejected": 0.7593650221824646,
2429
+ "logps/chosen": -301.3545227050781,
2430
+ "logps/rejected": -253.1154022216797,
2431
+ "loss": 0.6492,
2432
+ "rewards/accuracies": 0.9624999761581421,
2433
+ "rewards/chosen": 0.06837470084428787,
2434
+ "rewards/margins": 0.08594924956560135,
2435
+ "rewards/margins_max": 0.1240011602640152,
2436
+ "rewards/margins_min": 0.0478973314166069,
2437
+ "rewards/margins_std": 0.05381352826952934,
2438
+ "rewards/rejected": -0.017574548721313477,
2439
+ "step": 1340
2440
+ },
2441
+ {
2442
+ "epoch": 0.99,
2443
+ "grad_norm": 0.412109375,
2444
+ "learning_rate": 6.680708454906425e-11,
2445
+ "logits/chosen": -0.05887039750814438,
2446
+ "logits/rejected": 0.7168238759040833,
2447
+ "logps/chosen": -274.7958679199219,
2448
+ "logps/rejected": -234.88424682617188,
2449
+ "loss": 0.6506,
2450
+ "rewards/accuracies": 0.9750000238418579,
2451
+ "rewards/chosen": 0.06663821637630463,
2452
+ "rewards/margins": 0.08757266402244568,
2453
+ "rewards/margins_max": 0.1257934719324112,
2454
+ "rewards/margins_min": 0.049351852387189865,
2455
+ "rewards/margins_std": 0.05405237525701523,
2456
+ "rewards/rejected": -0.020934434607625008,
2457
+ "step": 1350
2458
+ },
2459
+ {
2460
+ "epoch": 1.0,
2461
+ "eval_logits/chosen": 0.744699239730835,
2462
+ "eval_logits/rejected": 0.8899587988853455,
2463
+ "eval_logps/chosen": -330.823974609375,
2464
+ "eval_logps/rejected": -316.1527404785156,
2465
+ "eval_loss": 0.6946464776992798,
2466
+ "eval_rewards/accuracies": 0.5195000171661377,
2467
+ "eval_rewards/chosen": 0.03155522421002388,
2468
+ "eval_rewards/margins": -0.0017884434200823307,
2469
+ "eval_rewards/margins_max": 0.09517286717891693,
2470
+ "eval_rewards/margins_min": -0.10408221185207367,
2471
+ "eval_rewards/margins_std": 0.06461790204048157,
2472
+ "eval_rewards/rejected": 0.03334366902709007,
2473
+ "eval_runtime": 837.0887,
2474
+ "eval_samples_per_second": 4.778,
2475
+ "eval_steps_per_second": 0.299,
2476
+ "step": 1359
2477
+ },
2478
+ {
2479
+ "epoch": 1.0,
2480
+ "step": 1359,
2481
+ "total_flos": 0.0,
2482
+ "train_loss": 0.6604200788417925,
2483
+ "train_runtime": 10358.1715,
2484
+ "train_samples_per_second": 2.099,
2485
+ "train_steps_per_second": 0.131
2486
+ }
2487
+ ],
2488
+ "logging_steps": 10,
2489
+ "max_steps": 1359,
2490
+ "num_input_tokens_seen": 0,
2491
+ "num_train_epochs": 1,
2492
+ "save_steps": 100,
2493
+ "total_flos": 0.0,
2494
+ "train_batch_size": 2,
2495
+ "trial_name": null,
2496
+ "trial_params": null
2497
+ }