just1nseo commited on
Commit
15f0f76
1 Parent(s): a18ae73

Model save

Browse files
README.md ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: alignment-handbook/zephyr-7b-sft-full
3
+ library_name: peft
4
+ license: apache-2.0
5
+ tags:
6
+ - trl
7
+ - dpo
8
+ - generated_from_trainer
9
+ model-index:
10
+ - name: zephyr-dpop-qlora-uf-oursuf6k-5e-7
11
+ results: []
12
+ ---
13
+
14
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
15
+ should probably proofread and complete it, then remove this comment. -->
16
+
17
+ # zephyr-dpop-qlora-uf-oursuf6k-5e-7
18
+
19
+ This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) on the None dataset.
20
+ It achieves the following results on the evaluation set:
21
+ - Loss: 0.6903
22
+ - Positive Losses: 0.0373
23
+ - Dpo Losses: 0.6851
24
+ - Rewards/chosen: 0.0599
25
+ - Rewards/rejected: 0.0432
26
+ - Rewards/accuracies: 0.6160
27
+ - Rewards/margins: 0.0167
28
+ - Rewards/margins Max: 0.0930
29
+ - Rewards/margins Min: -0.0511
30
+ - Rewards/margins Std: 0.0481
31
+ - Logps/rejected: -254.2623
32
+ - Logps/chosen: -278.6050
33
+ - Logits/rejected: -2.7602
34
+ - Logits/chosen: -2.7988
35
+
36
+ ## Model description
37
+
38
+ More information needed
39
+
40
+ ## Intended uses & limitations
41
+
42
+ More information needed
43
+
44
+ ## Training and evaluation data
45
+
46
+ More information needed
47
+
48
+ ## Training procedure
49
+
50
+ ### Training hyperparameters
51
+
52
+ The following hyperparameters were used during training:
53
+ - learning_rate: 5e-07
54
+ - train_batch_size: 4
55
+ - eval_batch_size: 8
56
+ - seed: 42
57
+ - distributed_type: multi-GPU
58
+ - num_devices: 2
59
+ - gradient_accumulation_steps: 2
60
+ - total_train_batch_size: 16
61
+ - total_eval_batch_size: 16
62
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
63
+ - lr_scheduler_type: cosine
64
+ - lr_scheduler_warmup_ratio: 0.1
65
+ - num_epochs: 1
66
+
67
+ ### Training results
68
+
69
+ | Training Loss | Epoch | Step | Validation Loss | Positive Losses | Dpo Losses | 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 |
70
+ |:-------------:|:-----:|:----:|:---------------:|:---------------:|:----------:|:--------------:|:----------------:|:------------------:|:---------------:|:-------------------:|:-------------------:|:-------------------:|:--------------:|:------------:|:---------------:|:-------------:|
71
+ | 0.6918 | 0.15 | 100 | 0.6930 | 0.0047 | 0.6922 | 0.0135 | 0.0117 | 0.6000 | 0.0019 | 0.0130 | -0.0081 | 0.0070 | -257.4135 | -283.2392 | -2.7718 | -2.8105 |
72
+ | 0.6867 | 0.29 | 200 | 0.6917 | 0.0124 | 0.6900 | 0.0299 | 0.0235 | 0.6100 | 0.0063 | 0.0373 | -0.0214 | 0.0194 | -256.2250 | -281.6053 | -2.7690 | -2.8075 |
73
+ | 0.6825 | 0.44 | 300 | 0.6909 | 0.0225 | 0.6879 | 0.0441 | 0.0333 | 0.625 | 0.0108 | 0.0619 | -0.0341 | 0.0319 | -255.2539 | -280.1855 | -2.7634 | -2.8021 |
74
+ | 0.6808 | 0.58 | 400 | 0.6905 | 0.0314 | 0.6861 | 0.0543 | 0.0397 | 0.6170 | 0.0147 | 0.0808 | -0.0436 | 0.0415 | -254.6122 | -279.1606 | -2.7608 | -2.7994 |
75
+ | 0.6769 | 0.73 | 500 | 0.6902 | 0.0330 | 0.6855 | 0.0585 | 0.0426 | 0.6120 | 0.0160 | 0.0885 | -0.0487 | 0.0458 | -254.3216 | -278.7400 | -2.7587 | -2.7975 |
76
+ | 0.67 | 0.88 | 600 | 0.6903 | 0.0373 | 0.6851 | 0.0599 | 0.0432 | 0.6160 | 0.0167 | 0.0930 | -0.0511 | 0.0481 | -254.2623 | -278.6050 | -2.7602 | -2.7988 |
77
+
78
+
79
+ ### Framework versions
80
+
81
+ - PEFT 0.7.1
82
+ - Transformers 4.39.0.dev0
83
+ - Pytorch 2.1.2+cu121
84
+ - Datasets 2.14.6
85
+ - Tokenizers 0.15.2
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fca3ad12a2bf4af258944c96da659fcc8df319458672dc9d1a65daef756d32ce
3
  size 671150064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:824ff4b0773838199a21e37e84916de704e6a91d73f2cea127d597e120ab6935
3
  size 671150064
all_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 0.6833745032026056,
4
+ "train_runtime": 9125.7542,
5
+ "train_samples": 10941,
6
+ "train_samples_per_second": 1.199,
7
+ "train_steps_per_second": 0.075
8
+ }
runs/Jul24_06-16-31_notebook-deployment-48-7d9b6c99-khd85/events.out.tfevents.1721801834.notebook-deployment-48-7d9b6c99-khd85.3314908.0 CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8beb56f33e266d0d83289c8e8269f6351b4460bacdc43a6532a8109df61af854
3
- size 72077
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f3d76600712c5de53578285f3c6a322de9e0844b1cba5441b84da4d4abfb38df
3
+ size 80391
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 0.6833745032026056,
4
+ "train_runtime": 9125.7542,
5
+ "train_samples": 10941,
6
+ "train_samples_per_second": 1.199,
7
+ "train_steps_per_second": 0.075
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,1536 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 1.0,
5
+ "eval_steps": 100,
6
+ "global_step": 684,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "dpo_losses": 0.6931471824645996,
13
+ "epoch": 0.0,
14
+ "grad_norm": 1.9721656718078973,
15
+ "learning_rate": 7.246376811594203e-09,
16
+ "logits/chosen": -2.8746490478515625,
17
+ "logits/rejected": -2.840811252593994,
18
+ "logps/chosen": -227.73272705078125,
19
+ "logps/rejected": -174.71890258789062,
20
+ "loss": 0.6931,
21
+ "positive_losses": 0.0,
22
+ "rewards/accuracies": 0.0,
23
+ "rewards/chosen": 0.0,
24
+ "rewards/margins": 0.0,
25
+ "rewards/margins_max": 0.0,
26
+ "rewards/margins_min": 0.0,
27
+ "rewards/margins_std": 0.0,
28
+ "rewards/rejected": 0.0,
29
+ "step": 1
30
+ },
31
+ {
32
+ "dpo_losses": 0.6933290958404541,
33
+ "epoch": 0.01,
34
+ "grad_norm": 25.64726309944567,
35
+ "learning_rate": 7.246376811594203e-08,
36
+ "logits/chosen": -2.7698583602905273,
37
+ "logits/rejected": -2.726212501525879,
38
+ "logps/chosen": -244.640380859375,
39
+ "logps/rejected": -240.38246154785156,
40
+ "loss": 0.6981,
41
+ "positive_losses": 0.04526931792497635,
42
+ "rewards/accuracies": 0.2638888955116272,
43
+ "rewards/chosen": -7.686785829719156e-05,
44
+ "rewards/margins": -0.0003627391706686467,
45
+ "rewards/margins_max": 0.0012993874261155725,
46
+ "rewards/margins_min": -0.0022027057129889727,
47
+ "rewards/margins_std": 0.0015639414777979255,
48
+ "rewards/rejected": 0.00028587126871570945,
49
+ "step": 10
50
+ },
51
+ {
52
+ "dpo_losses": 0.6930510401725769,
53
+ "epoch": 0.03,
54
+ "grad_norm": 20.592262133277757,
55
+ "learning_rate": 1.4492753623188405e-07,
56
+ "logits/chosen": -2.897423028945923,
57
+ "logits/rejected": -2.8587498664855957,
58
+ "logps/chosen": -317.21636962890625,
59
+ "logps/rejected": -239.84121704101562,
60
+ "loss": 0.6999,
61
+ "positive_losses": 0.07797051966190338,
62
+ "rewards/accuracies": 0.5249999761581421,
63
+ "rewards/chosen": 0.0003467574715614319,
64
+ "rewards/margins": 0.00019531897851265967,
65
+ "rewards/margins_max": 0.0034703321289271116,
66
+ "rewards/margins_min": -0.0027894906233996153,
67
+ "rewards/margins_std": 0.002780328970402479,
68
+ "rewards/rejected": 0.000151438478496857,
69
+ "step": 20
70
+ },
71
+ {
72
+ "dpo_losses": 0.6931239366531372,
73
+ "epoch": 0.04,
74
+ "grad_norm": 16.747192147468148,
75
+ "learning_rate": 2.1739130434782607e-07,
76
+ "logits/chosen": -2.811279535293579,
77
+ "logits/rejected": -2.790393114089966,
78
+ "logps/chosen": -264.90753173828125,
79
+ "logps/rejected": -223.55331420898438,
80
+ "loss": 0.6988,
81
+ "positive_losses": 0.052257489413022995,
82
+ "rewards/accuracies": 0.4625000059604645,
83
+ "rewards/chosen": 0.0009221597574651241,
84
+ "rewards/margins": 4.864180664299056e-05,
85
+ "rewards/margins_max": 0.002964555751532316,
86
+ "rewards/margins_min": -0.0025322353467345238,
87
+ "rewards/margins_std": 0.0024663216900080442,
88
+ "rewards/rejected": 0.0008735180017538369,
89
+ "step": 30
90
+ },
91
+ {
92
+ "dpo_losses": 0.6930278539657593,
93
+ "epoch": 0.06,
94
+ "grad_norm": 12.398083667958339,
95
+ "learning_rate": 2.898550724637681e-07,
96
+ "logits/chosen": -2.864264965057373,
97
+ "logits/rejected": -2.834386110305786,
98
+ "logps/chosen": -304.97265625,
99
+ "logps/rejected": -266.8782043457031,
100
+ "loss": 0.6961,
101
+ "positive_losses": 0.031120896339416504,
102
+ "rewards/accuracies": 0.4625000059604645,
103
+ "rewards/chosen": 0.0022245876025408506,
104
+ "rewards/margins": 0.0002419196825940162,
105
+ "rewards/margins_max": 0.004220693372189999,
106
+ "rewards/margins_min": -0.0033918179105967283,
107
+ "rewards/margins_std": 0.0034401887096464634,
108
+ "rewards/rejected": 0.001982667949050665,
109
+ "step": 40
110
+ },
111
+ {
112
+ "dpo_losses": 0.6931858658790588,
113
+ "epoch": 0.07,
114
+ "grad_norm": 8.878267358365656,
115
+ "learning_rate": 3.6231884057971015e-07,
116
+ "logits/chosen": -2.870177984237671,
117
+ "logits/rejected": -2.8384270668029785,
118
+ "logps/chosen": -260.7277526855469,
119
+ "logps/rejected": -234.89218139648438,
120
+ "loss": 0.6942,
121
+ "positive_losses": 0.022727489471435547,
122
+ "rewards/accuracies": 0.4124999940395355,
123
+ "rewards/chosen": 0.003678619861602783,
124
+ "rewards/margins": -7.445671508321539e-05,
125
+ "rewards/margins_max": 0.0031701656989753246,
126
+ "rewards/margins_min": -0.0034985982347279787,
127
+ "rewards/margins_std": 0.0029860776849091053,
128
+ "rewards/rejected": 0.003753076074644923,
129
+ "step": 50
130
+ },
131
+ {
132
+ "dpo_losses": 0.6925040483474731,
133
+ "epoch": 0.09,
134
+ "grad_norm": 1.6719366044432857,
135
+ "learning_rate": 4.3478260869565214e-07,
136
+ "logits/chosen": -2.891150951385498,
137
+ "logits/rejected": -2.843611478805542,
138
+ "logps/chosen": -280.95947265625,
139
+ "logps/rejected": -248.34915161132812,
140
+ "loss": 0.6931,
141
+ "positive_losses": 0.0025458037853240967,
142
+ "rewards/accuracies": 0.6625000238418579,
143
+ "rewards/chosen": 0.0058168042451143265,
144
+ "rewards/margins": 0.0012899416033178568,
145
+ "rewards/margins_max": 0.005339757539331913,
146
+ "rewards/margins_min": -0.002916422439739108,
147
+ "rewards/margins_std": 0.003642101539298892,
148
+ "rewards/rejected": 0.004526862408965826,
149
+ "step": 60
150
+ },
151
+ {
152
+ "dpo_losses": 0.6921008229255676,
153
+ "epoch": 0.1,
154
+ "grad_norm": 1.9262001556938266,
155
+ "learning_rate": 4.999967381905813e-07,
156
+ "logits/chosen": -2.7916743755340576,
157
+ "logits/rejected": -2.7574045658111572,
158
+ "logps/chosen": -290.14361572265625,
159
+ "logps/rejected": -212.0419158935547,
160
+ "loss": 0.6926,
161
+ "positive_losses": 0.004826736636459827,
162
+ "rewards/accuracies": 0.675000011920929,
163
+ "rewards/chosen": 0.006598347332328558,
164
+ "rewards/margins": 0.002098249038681388,
165
+ "rewards/margins_max": 0.006628250237554312,
166
+ "rewards/margins_min": -0.0018369384342804551,
167
+ "rewards/margins_std": 0.00378109747543931,
168
+ "rewards/rejected": 0.004500098060816526,
169
+ "step": 70
170
+ },
171
+ {
172
+ "dpo_losses": 0.6920569539070129,
173
+ "epoch": 0.12,
174
+ "grad_norm": 13.284239668101833,
175
+ "learning_rate": 4.996054240392509e-07,
176
+ "logits/chosen": -2.7785937786102295,
177
+ "logits/rejected": -2.759201765060425,
178
+ "logps/chosen": -264.8086242675781,
179
+ "logps/rejected": -243.39212036132812,
180
+ "loss": 0.6931,
181
+ "positive_losses": 0.014845085330307484,
182
+ "rewards/accuracies": 0.6499999761581421,
183
+ "rewards/chosen": 0.007849474437534809,
184
+ "rewards/margins": 0.0021913917735219,
185
+ "rewards/margins_max": 0.008834488689899445,
186
+ "rewards/margins_min": -0.00322969863191247,
187
+ "rewards/margins_std": 0.005265987012535334,
188
+ "rewards/rejected": 0.0056580835953354836,
189
+ "step": 80
190
+ },
191
+ {
192
+ "dpo_losses": 0.6912720203399658,
193
+ "epoch": 0.13,
194
+ "grad_norm": 2.317442586572389,
195
+ "learning_rate": 4.985629178361649e-07,
196
+ "logits/chosen": -2.8908092975616455,
197
+ "logits/rejected": -2.8579392433166504,
198
+ "logps/chosen": -272.96112060546875,
199
+ "logps/rejected": -245.47073364257812,
200
+ "loss": 0.6918,
201
+ "positive_losses": 0.008871316909790039,
202
+ "rewards/accuracies": 0.737500011920929,
203
+ "rewards/chosen": 0.009552950970828533,
204
+ "rewards/margins": 0.0037618838250637054,
205
+ "rewards/margins_max": 0.010290957987308502,
206
+ "rewards/margins_min": -0.002331961877644062,
207
+ "rewards/margins_std": 0.005662451032549143,
208
+ "rewards/rejected": 0.00579106854274869,
209
+ "step": 90
210
+ },
211
+ {
212
+ "dpo_losses": 0.6900182962417603,
213
+ "epoch": 0.15,
214
+ "grad_norm": 12.043613650426611,
215
+ "learning_rate": 4.968719393609756e-07,
216
+ "logits/chosen": -2.8450093269348145,
217
+ "logits/rejected": -2.7789740562438965,
218
+ "logps/chosen": -337.09600830078125,
219
+ "logps/rejected": -243.02517700195312,
220
+ "loss": 0.6918,
221
+ "positive_losses": 0.017842482775449753,
222
+ "rewards/accuracies": 0.7250000238418579,
223
+ "rewards/chosen": 0.012230691500008106,
224
+ "rewards/margins": 0.0062872739508748055,
225
+ "rewards/margins_max": 0.01721397042274475,
226
+ "rewards/margins_min": -0.002463764976710081,
227
+ "rewards/margins_std": 0.008927479386329651,
228
+ "rewards/rejected": 0.0059434170834720135,
229
+ "step": 100
230
+ },
231
+ {
232
+ "epoch": 0.15,
233
+ "eval_dpo_losses": 0.6922104358673096,
234
+ "eval_logits/chosen": -2.8104522228240967,
235
+ "eval_logits/rejected": -2.7718498706817627,
236
+ "eval_logps/chosen": -283.23919677734375,
237
+ "eval_logps/rejected": -257.4134826660156,
238
+ "eval_loss": 0.6930333375930786,
239
+ "eval_positive_losses": 0.004661041311919689,
240
+ "eval_rewards/accuracies": 0.6000000238418579,
241
+ "eval_rewards/chosen": 0.013542202301323414,
242
+ "eval_rewards/margins": 0.001887724851258099,
243
+ "eval_rewards/margins_max": 0.013032433576881886,
244
+ "eval_rewards/margins_min": -0.008088048547506332,
245
+ "eval_rewards/margins_std": 0.006959096062928438,
246
+ "eval_rewards/rejected": 0.011654475703835487,
247
+ "eval_runtime": 439.0202,
248
+ "eval_samples_per_second": 4.556,
249
+ "eval_steps_per_second": 0.285,
250
+ "step": 100
251
+ },
252
+ {
253
+ "dpo_losses": 0.6904088258743286,
254
+ "epoch": 0.16,
255
+ "grad_norm": 5.183696409472315,
256
+ "learning_rate": 4.945369001834514e-07,
257
+ "logits/chosen": -2.8782482147216797,
258
+ "logits/rejected": -2.8235890865325928,
259
+ "logps/chosen": -287.38250732421875,
260
+ "logps/rejected": -222.830078125,
261
+ "loss": 0.6908,
262
+ "positive_losses": 0.0031826018821448088,
263
+ "rewards/accuracies": 0.7749999761581421,
264
+ "rewards/chosen": 0.01270085759460926,
265
+ "rewards/margins": 0.00550065329298377,
266
+ "rewards/margins_max": 0.014992681331932545,
267
+ "rewards/margins_min": -0.0012518208241090178,
268
+ "rewards/margins_std": 0.007309773005545139,
269
+ "rewards/rejected": 0.0072002047672867775,
270
+ "step": 110
271
+ },
272
+ {
273
+ "dpo_losses": 0.6896742582321167,
274
+ "epoch": 0.18,
275
+ "grad_norm": 1.9553746809300425,
276
+ "learning_rate": 4.915638921541951e-07,
277
+ "logits/chosen": -2.8301899433135986,
278
+ "logits/rejected": -2.8140978813171387,
279
+ "logps/chosen": -256.82879638671875,
280
+ "logps/rejected": -236.7211456298828,
281
+ "loss": 0.6898,
282
+ "positive_losses": 0.0022869110107421875,
283
+ "rewards/accuracies": 0.7875000238418579,
284
+ "rewards/chosen": 0.013588182628154755,
285
+ "rewards/margins": 0.00698423758149147,
286
+ "rewards/margins_max": 0.018451116979122162,
287
+ "rewards/margins_min": -0.001020266441628337,
288
+ "rewards/margins_std": 0.008837779983878136,
289
+ "rewards/rejected": 0.006603945046663284,
290
+ "step": 120
291
+ },
292
+ {
293
+ "dpo_losses": 0.6898840665817261,
294
+ "epoch": 0.19,
295
+ "grad_norm": 1.6022519000216686,
296
+ "learning_rate": 4.879606715117018e-07,
297
+ "logits/chosen": -2.881978988647461,
298
+ "logits/rejected": -2.82822847366333,
299
+ "logps/chosen": -274.6767578125,
300
+ "logps/rejected": -234.8880615234375,
301
+ "loss": 0.6897,
302
+ "positive_losses": 0.007077217102050781,
303
+ "rewards/accuracies": 0.6625000238418579,
304
+ "rewards/chosen": 0.014239470474421978,
305
+ "rewards/margins": 0.006568121258169413,
306
+ "rewards/margins_max": 0.017170870676636696,
307
+ "rewards/margins_min": -0.0028213211335241795,
308
+ "rewards/margins_std": 0.00886800978332758,
309
+ "rewards/rejected": 0.00767135014757514,
310
+ "step": 130
311
+ },
312
+ {
313
+ "dpo_losses": 0.6880620718002319,
314
+ "epoch": 0.2,
315
+ "grad_norm": 3.638881806293095,
316
+ "learning_rate": 4.837366386472174e-07,
317
+ "logits/chosen": -2.8916850090026855,
318
+ "logits/rejected": -2.8278439044952393,
319
+ "logps/chosen": -296.95111083984375,
320
+ "logps/rejected": -253.7423553466797,
321
+ "loss": 0.6901,
322
+ "positive_losses": 0.002955913543701172,
323
+ "rewards/accuracies": 0.8125,
324
+ "rewards/chosen": 0.01859217882156372,
325
+ "rewards/margins": 0.010244466364383698,
326
+ "rewards/margins_max": 0.024400196969509125,
327
+ "rewards/margins_min": -0.0004118576762266457,
328
+ "rewards/margins_std": 0.011427570134401321,
329
+ "rewards/rejected": 0.008347713388502598,
330
+ "step": 140
331
+ },
332
+ {
333
+ "dpo_losses": 0.6885315179824829,
334
+ "epoch": 0.22,
335
+ "grad_norm": 6.527916454528557,
336
+ "learning_rate": 4.789028135801918e-07,
337
+ "logits/chosen": -2.852397918701172,
338
+ "logits/rejected": -2.8406243324279785,
339
+ "logps/chosen": -297.78564453125,
340
+ "logps/rejected": -271.2729187011719,
341
+ "loss": 0.6914,
342
+ "positive_losses": 0.054250240325927734,
343
+ "rewards/accuracies": 0.6499999761581421,
344
+ "rewards/chosen": 0.01786804385483265,
345
+ "rewards/margins": 0.009341379627585411,
346
+ "rewards/margins_max": 0.03060457669198513,
347
+ "rewards/margins_min": -0.008363897912204266,
348
+ "rewards/margins_std": 0.01767081953585148,
349
+ "rewards/rejected": 0.008526663295924664,
350
+ "step": 150
351
+ },
352
+ {
353
+ "dpo_losses": 0.6880606412887573,
354
+ "epoch": 0.23,
355
+ "grad_norm": 8.44151922500754,
356
+ "learning_rate": 4.7347180720830627e-07,
357
+ "logits/chosen": -2.8739285469055176,
358
+ "logits/rejected": -2.7969882488250732,
359
+ "logps/chosen": -300.0534973144531,
360
+ "logps/rejected": -260.5088195800781,
361
+ "loss": 0.6899,
362
+ "positive_losses": 0.006076812744140625,
363
+ "rewards/accuracies": 0.800000011920929,
364
+ "rewards/chosen": 0.01983806863427162,
365
+ "rewards/margins": 0.01025958452373743,
366
+ "rewards/margins_max": 0.028022507205605507,
367
+ "rewards/margins_min": -0.0024425422307103872,
368
+ "rewards/margins_std": 0.013813845813274384,
369
+ "rewards/rejected": 0.009578483179211617,
370
+ "step": 160
371
+ },
372
+ {
373
+ "dpo_losses": 0.684917151927948,
374
+ "epoch": 0.25,
375
+ "grad_norm": 1.8673882157090276,
376
+ "learning_rate": 4.6745778840708107e-07,
377
+ "logits/chosen": -2.8020739555358887,
378
+ "logits/rejected": -2.7900993824005127,
379
+ "logps/chosen": -295.7058410644531,
380
+ "logps/rejected": -225.8823699951172,
381
+ "loss": 0.6858,
382
+ "positive_losses": 0.0014563084114342928,
383
+ "rewards/accuracies": 0.800000011920929,
384
+ "rewards/chosen": 0.026786331087350845,
385
+ "rewards/margins": 0.01665376126766205,
386
+ "rewards/margins_max": 0.03810211643576622,
387
+ "rewards/margins_min": -0.0024380278773605824,
388
+ "rewards/margins_std": 0.017996035516262054,
389
+ "rewards/rejected": 0.010132570751011372,
390
+ "step": 170
391
+ },
392
+ {
393
+ "dpo_losses": 0.6871532201766968,
394
+ "epoch": 0.26,
395
+ "grad_norm": 2.603750597458806,
396
+ "learning_rate": 4.6087644706489703e-07,
397
+ "logits/chosen": -2.833789587020874,
398
+ "logits/rejected": -2.7939248085021973,
399
+ "logps/chosen": -294.24456787109375,
400
+ "logps/rejected": -254.28988647460938,
401
+ "loss": 0.6889,
402
+ "positive_losses": 0.024321984499692917,
403
+ "rewards/accuracies": 0.75,
404
+ "rewards/chosen": 0.022618165239691734,
405
+ "rewards/margins": 0.01210461463779211,
406
+ "rewards/margins_max": 0.030625376850366592,
407
+ "rewards/margins_min": -0.004277624189853668,
408
+ "rewards/margins_std": 0.015746023505926132,
409
+ "rewards/rejected": 0.010513553395867348,
410
+ "step": 180
411
+ },
412
+ {
413
+ "dpo_losses": 0.686254620552063,
414
+ "epoch": 0.28,
415
+ "grad_norm": 1.7335945641686348,
416
+ "learning_rate": 4.537449531498687e-07,
417
+ "logits/chosen": -2.7748355865478516,
418
+ "logits/rejected": -2.725888729095459,
419
+ "logps/chosen": -280.95428466796875,
420
+ "logps/rejected": -228.15097045898438,
421
+ "loss": 0.6878,
422
+ "positive_losses": 0.012805747799575329,
423
+ "rewards/accuracies": 0.7875000238418579,
424
+ "rewards/chosen": 0.028492039069533348,
425
+ "rewards/margins": 0.013932852074503899,
426
+ "rewards/margins_max": 0.03530807048082352,
427
+ "rewards/margins_min": -0.002935736905783415,
428
+ "rewards/margins_std": 0.017064960673451424,
429
+ "rewards/rejected": 0.0145591851323843,
430
+ "step": 190
431
+ },
432
+ {
433
+ "dpo_losses": 0.6861851811408997,
434
+ "epoch": 0.29,
435
+ "grad_norm": 1.5625317379589145,
436
+ "learning_rate": 4.4608191191535736e-07,
437
+ "logits/chosen": -2.8794567584991455,
438
+ "logits/rejected": -2.835148334503174,
439
+ "logps/chosen": -274.84735107421875,
440
+ "logps/rejected": -242.3157958984375,
441
+ "loss": 0.6867,
442
+ "positive_losses": 0.004458713345229626,
443
+ "rewards/accuracies": 0.6875,
444
+ "rewards/chosen": 0.02746543660759926,
445
+ "rewards/margins": 0.014083576388657093,
446
+ "rewards/margins_max": 0.03864169120788574,
447
+ "rewards/margins_min": -0.005826378706842661,
448
+ "rewards/margins_std": 0.019664574414491653,
449
+ "rewards/rejected": 0.01338185928761959,
450
+ "step": 200
451
+ },
452
+ {
453
+ "epoch": 0.29,
454
+ "eval_dpo_losses": 0.6900327801704407,
455
+ "eval_logits/chosen": -2.8075411319732666,
456
+ "eval_logits/rejected": -2.7689805030822754,
457
+ "eval_logps/chosen": -281.6053161621094,
458
+ "eval_logps/rejected": -256.2250061035156,
459
+ "eval_loss": 0.6917006373405457,
460
+ "eval_positive_losses": 0.012367370538413525,
461
+ "eval_rewards/accuracies": 0.6100000143051147,
462
+ "eval_rewards/chosen": 0.02988099865615368,
463
+ "eval_rewards/margins": 0.006341494619846344,
464
+ "eval_rewards/margins_max": 0.03729286044836044,
465
+ "eval_rewards/margins_min": -0.02144700102508068,
466
+ "eval_rewards/margins_std": 0.01944614388048649,
467
+ "eval_rewards/rejected": 0.023539502173662186,
468
+ "eval_runtime": 457.2097,
469
+ "eval_samples_per_second": 4.374,
470
+ "eval_steps_per_second": 0.273,
471
+ "step": 200
472
+ },
473
+ {
474
+ "dpo_losses": 0.6849207878112793,
475
+ "epoch": 0.31,
476
+ "grad_norm": 6.776479704467401,
477
+ "learning_rate": 4.379073153609896e-07,
478
+ "logits/chosen": -2.85437273979187,
479
+ "logits/rejected": -2.8159098625183105,
480
+ "logps/chosen": -298.1737365722656,
481
+ "logps/rejected": -258.9627685546875,
482
+ "loss": 0.6892,
483
+ "positive_losses": 0.024309396743774414,
484
+ "rewards/accuracies": 0.737500011920929,
485
+ "rewards/chosen": 0.031262680888175964,
486
+ "rewards/margins": 0.016705606132745743,
487
+ "rewards/margins_max": 0.0462987944483757,
488
+ "rewards/margins_min": -0.009217771701514721,
489
+ "rewards/margins_std": 0.024988915771245956,
490
+ "rewards/rejected": 0.014557072892785072,
491
+ "step": 210
492
+ },
493
+ {
494
+ "dpo_losses": 0.6835730075836182,
495
+ "epoch": 0.32,
496
+ "grad_norm": 1.7559811016317117,
497
+ "learning_rate": 4.292424900758128e-07,
498
+ "logits/chosen": -2.8128745555877686,
499
+ "logits/rejected": -2.7835748195648193,
500
+ "logps/chosen": -278.7519836425781,
501
+ "logps/rejected": -221.44998168945312,
502
+ "loss": 0.687,
503
+ "positive_losses": 0.01152725238353014,
504
+ "rewards/accuracies": 0.7875000238418579,
505
+ "rewards/chosen": 0.032679036259651184,
506
+ "rewards/margins": 0.019430814310908318,
507
+ "rewards/margins_max": 0.04703899472951889,
508
+ "rewards/margins_min": -0.004503333941102028,
509
+ "rewards/margins_std": 0.02318265102803707,
510
+ "rewards/rejected": 0.013248221948742867,
511
+ "step": 220
512
+ },
513
+ {
514
+ "dpo_losses": 0.6845378279685974,
515
+ "epoch": 0.34,
516
+ "grad_norm": 5.2855521291360335,
517
+ "learning_rate": 4.201100415996597e-07,
518
+ "logits/chosen": -2.7867586612701416,
519
+ "logits/rejected": -2.7540698051452637,
520
+ "logps/chosen": -262.8429260253906,
521
+ "logps/rejected": -259.90594482421875,
522
+ "loss": 0.6873,
523
+ "positive_losses": 0.010059595108032227,
524
+ "rewards/accuracies": 0.737500011920929,
525
+ "rewards/chosen": 0.033603329211473465,
526
+ "rewards/margins": 0.017468122765421867,
527
+ "rewards/margins_max": 0.05101371556520462,
528
+ "rewards/margins_min": -0.005049402825534344,
529
+ "rewards/margins_std": 0.025209274142980576,
530
+ "rewards/rejected": 0.016135206446051598,
531
+ "step": 230
532
+ },
533
+ {
534
+ "dpo_losses": 0.6822143793106079,
535
+ "epoch": 0.35,
536
+ "grad_norm": 7.1123081067417635,
537
+ "learning_rate": 4.1053379544787557e-07,
538
+ "logits/chosen": -2.8566319942474365,
539
+ "logits/rejected": -2.8126063346862793,
540
+ "logps/chosen": -285.74066162109375,
541
+ "logps/rejected": -246.8583221435547,
542
+ "loss": 0.6875,
543
+ "positive_losses": 0.0023063658736646175,
544
+ "rewards/accuracies": 0.8125,
545
+ "rewards/chosen": 0.03864200785756111,
546
+ "rewards/margins": 0.02219047024846077,
547
+ "rewards/margins_max": 0.04993893578648567,
548
+ "rewards/margins_min": -0.002027359325438738,
549
+ "rewards/margins_std": 0.023801075294613838,
550
+ "rewards/rejected": 0.01645154133439064,
551
+ "step": 240
552
+ },
553
+ {
554
+ "dpo_losses": 0.6844903230667114,
555
+ "epoch": 0.37,
556
+ "grad_norm": 2.225597078597608,
557
+ "learning_rate": 4.0053873495326964e-07,
558
+ "logits/chosen": -2.8150901794433594,
559
+ "logits/rejected": -2.798675537109375,
560
+ "logps/chosen": -243.178955078125,
561
+ "logps/rejected": -226.9827117919922,
562
+ "loss": 0.6875,
563
+ "positive_losses": 0.0484701506793499,
564
+ "rewards/accuracies": 0.7250000238418579,
565
+ "rewards/chosen": 0.033011239022016525,
566
+ "rewards/margins": 0.017598429694771767,
567
+ "rewards/margins_max": 0.047145940363407135,
568
+ "rewards/margins_min": -0.0065514100715518,
569
+ "rewards/margins_std": 0.02364528551697731,
570
+ "rewards/rejected": 0.015412810258567333,
571
+ "step": 250
572
+ },
573
+ {
574
+ "dpo_losses": 0.6823347806930542,
575
+ "epoch": 0.38,
576
+ "grad_norm": 1.8954330972040565,
577
+ "learning_rate": 3.9015093608745143e-07,
578
+ "logits/chosen": -2.87410306930542,
579
+ "logits/rejected": -2.808983087539673,
580
+ "logps/chosen": -283.3040771484375,
581
+ "logps/rejected": -240.4110565185547,
582
+ "loss": 0.6824,
583
+ "positive_losses": 0.011256242170929909,
584
+ "rewards/accuracies": 0.824999988079071,
585
+ "rewards/chosen": 0.03804592788219452,
586
+ "rewards/margins": 0.021955588832497597,
587
+ "rewards/margins_max": 0.05776814743876457,
588
+ "rewards/margins_min": -0.004212898667901754,
589
+ "rewards/margins_std": 0.028291797265410423,
590
+ "rewards/rejected": 0.016090339049696922,
591
+ "step": 260
592
+ },
593
+ {
594
+ "dpo_losses": 0.681152880191803,
595
+ "epoch": 0.39,
596
+ "grad_norm": 2.3462040578320438,
597
+ "learning_rate": 3.79397499431599e-07,
598
+ "logits/chosen": -2.838275909423828,
599
+ "logits/rejected": -2.7828450202941895,
600
+ "logps/chosen": -291.1107177734375,
601
+ "logps/rejected": -261.1417236328125,
602
+ "loss": 0.6844,
603
+ "positive_losses": 0.031092902645468712,
604
+ "rewards/accuracies": 0.762499988079071,
605
+ "rewards/chosen": 0.04311501979827881,
606
+ "rewards/margins": 0.024433860555291176,
607
+ "rewards/margins_max": 0.05960015207529068,
608
+ "rewards/margins_min": -0.005230102688074112,
609
+ "rewards/margins_std": 0.028690297156572342,
610
+ "rewards/rejected": 0.018681161105632782,
611
+ "step": 270
612
+ },
613
+ {
614
+ "dpo_losses": 0.6834903955459595,
615
+ "epoch": 0.41,
616
+ "grad_norm": 1.678928576392419,
617
+ "learning_rate": 3.6830647947413694e-07,
618
+ "logits/chosen": -2.906564235687256,
619
+ "logits/rejected": -2.8589329719543457,
620
+ "logps/chosen": -273.8109436035156,
621
+ "logps/rejected": -246.07925415039062,
622
+ "loss": 0.6848,
623
+ "positive_losses": 0.03764519840478897,
624
+ "rewards/accuracies": 0.762499988079071,
625
+ "rewards/chosen": 0.03491830825805664,
626
+ "rewards/margins": 0.019722772762179375,
627
+ "rewards/margins_max": 0.05492109805345535,
628
+ "rewards/margins_min": -0.009231898002326488,
629
+ "rewards/margins_std": 0.029070580378174782,
630
+ "rewards/rejected": 0.015195533633232117,
631
+ "step": 280
632
+ },
633
+ {
634
+ "dpo_losses": 0.6819528937339783,
635
+ "epoch": 0.42,
636
+ "grad_norm": 1.5933364630845321,
637
+ "learning_rate": 3.5690681141977837e-07,
638
+ "logits/chosen": -2.8415896892547607,
639
+ "logits/rejected": -2.795182704925537,
640
+ "logps/chosen": -252.76455688476562,
641
+ "logps/rejected": -203.42364501953125,
642
+ "loss": 0.6836,
643
+ "positive_losses": 0.024924278259277344,
644
+ "rewards/accuracies": 0.762499988079071,
645
+ "rewards/chosen": 0.039224088191986084,
646
+ "rewards/margins": 0.022878160700201988,
647
+ "rewards/margins_max": 0.06510117650032043,
648
+ "rewards/margins_min": -0.010645577684044838,
649
+ "rewards/margins_std": 0.034357957541942596,
650
+ "rewards/rejected": 0.016345927491784096,
651
+ "step": 290
652
+ },
653
+ {
654
+ "dpo_losses": 0.6809358596801758,
655
+ "epoch": 0.44,
656
+ "grad_norm": 10.347091457783785,
657
+ "learning_rate": 3.4522823570088067e-07,
658
+ "logits/chosen": -2.840087413787842,
659
+ "logits/rejected": -2.8272616863250732,
660
+ "logps/chosen": -259.23687744140625,
661
+ "logps/rejected": -251.6184844970703,
662
+ "loss": 0.6825,
663
+ "positive_losses": 0.014734840020537376,
664
+ "rewards/accuracies": 0.675000011920929,
665
+ "rewards/chosen": 0.03966939076781273,
666
+ "rewards/margins": 0.02503836713731289,
667
+ "rewards/margins_max": 0.07522507011890411,
668
+ "rewards/margins_min": -0.010506678372621536,
669
+ "rewards/margins_std": 0.03809111937880516,
670
+ "rewards/rejected": 0.01463102363049984,
671
+ "step": 300
672
+ },
673
+ {
674
+ "epoch": 0.44,
675
+ "eval_dpo_losses": 0.6878849864006042,
676
+ "eval_logits/chosen": -2.8020598888397217,
677
+ "eval_logits/rejected": -2.763356924057007,
678
+ "eval_logps/chosen": -280.1855163574219,
679
+ "eval_logps/rejected": -255.25390625,
680
+ "eval_loss": 0.69089674949646,
681
+ "eval_positive_losses": 0.022481659427285194,
682
+ "eval_rewards/accuracies": 0.625,
683
+ "eval_rewards/chosen": 0.04407881200313568,
684
+ "eval_rewards/margins": 0.010828724130988121,
685
+ "eval_rewards/margins_max": 0.061860937625169754,
686
+ "eval_rewards/margins_min": -0.03410045802593231,
687
+ "eval_rewards/margins_std": 0.03193770721554756,
688
+ "eval_rewards/rejected": 0.03325008600950241,
689
+ "eval_runtime": 439.4476,
690
+ "eval_samples_per_second": 4.551,
691
+ "eval_steps_per_second": 0.284,
692
+ "step": 300
693
+ },
694
+ {
695
+ "dpo_losses": 0.6795910596847534,
696
+ "epoch": 0.45,
697
+ "grad_norm": 1.8194274060306528,
698
+ "learning_rate": 3.3330122038805277e-07,
699
+ "logits/chosen": -2.861931562423706,
700
+ "logits/rejected": -2.8135368824005127,
701
+ "logps/chosen": -271.8392639160156,
702
+ "logps/rejected": -225.01708984375,
703
+ "loss": 0.6837,
704
+ "positive_losses": 0.02801961824297905,
705
+ "rewards/accuracies": 0.8500000238418579,
706
+ "rewards/chosen": 0.04437270388007164,
707
+ "rewards/margins": 0.027612268924713135,
708
+ "rewards/margins_max": 0.07083487510681152,
709
+ "rewards/margins_min": -0.0025140452198684216,
710
+ "rewards/margins_std": 0.033803679049015045,
711
+ "rewards/rejected": 0.016760434955358505,
712
+ "step": 310
713
+ },
714
+ {
715
+ "dpo_losses": 0.6798790693283081,
716
+ "epoch": 0.47,
717
+ "grad_norm": 1.8110670147086625,
718
+ "learning_rate": 3.2115688170243734e-07,
719
+ "logits/chosen": -2.8523640632629395,
720
+ "logits/rejected": -2.82257080078125,
721
+ "logps/chosen": -270.28253173828125,
722
+ "logps/rejected": -248.6698455810547,
723
+ "loss": 0.6802,
724
+ "positive_losses": 0.009492230601608753,
725
+ "rewards/accuracies": 0.762499988079071,
726
+ "rewards/chosen": 0.04459889978170395,
727
+ "rewards/margins": 0.027110621333122253,
728
+ "rewards/margins_max": 0.06967566907405853,
729
+ "rewards/margins_min": -0.005822637118399143,
730
+ "rewards/margins_std": 0.03437542915344238,
731
+ "rewards/rejected": 0.017488282173871994,
732
+ "step": 320
733
+ },
734
+ {
735
+ "dpo_losses": 0.6773503422737122,
736
+ "epoch": 0.48,
737
+ "grad_norm": 9.150401672552029,
738
+ "learning_rate": 3.088269028370435e-07,
739
+ "logits/chosen": -2.875394821166992,
740
+ "logits/rejected": -2.8434793949127197,
741
+ "logps/chosen": -295.0433349609375,
742
+ "logps/rejected": -246.5893096923828,
743
+ "loss": 0.6835,
744
+ "positive_losses": 0.03743886947631836,
745
+ "rewards/accuracies": 0.7875000238418579,
746
+ "rewards/chosen": 0.04838422313332558,
747
+ "rewards/margins": 0.03234843909740448,
748
+ "rewards/margins_max": 0.08645691722631454,
749
+ "rewards/margins_min": -0.006109377834945917,
750
+ "rewards/margins_std": 0.04178870469331741,
751
+ "rewards/rejected": 0.016035789623856544,
752
+ "step": 330
753
+ },
754
+ {
755
+ "dpo_losses": 0.6774032115936279,
756
+ "epoch": 0.5,
757
+ "grad_norm": 4.795115194677247,
758
+ "learning_rate": 2.9634345129891294e-07,
759
+ "logits/chosen": -2.8238093852996826,
760
+ "logits/rejected": -2.785935163497925,
761
+ "logps/chosen": -287.4772033691406,
762
+ "logps/rejected": -254.5933074951172,
763
+ "loss": 0.6804,
764
+ "positive_losses": 0.03258161619305611,
765
+ "rewards/accuracies": 0.762499988079071,
766
+ "rewards/chosen": 0.04798700287938118,
767
+ "rewards/margins": 0.032188571989536285,
768
+ "rewards/margins_max": 0.0826764851808548,
769
+ "rewards/margins_min": -0.006551785860210657,
770
+ "rewards/margins_std": 0.04137192294001579,
771
+ "rewards/rejected": 0.015798425301909447,
772
+ "step": 340
773
+ },
774
+ {
775
+ "dpo_losses": 0.6810469627380371,
776
+ "epoch": 0.51,
777
+ "grad_norm": 11.724114006745417,
778
+ "learning_rate": 2.8373909498776744e-07,
779
+ "logits/chosen": -2.8722732067108154,
780
+ "logits/rejected": -2.854133129119873,
781
+ "logps/chosen": -279.33917236328125,
782
+ "logps/rejected": -276.32232666015625,
783
+ "loss": 0.6845,
784
+ "positive_losses": 0.04124107211828232,
785
+ "rewards/accuracies": 0.762499988079071,
786
+ "rewards/chosen": 0.04524795338511467,
787
+ "rewards/margins": 0.024712715297937393,
788
+ "rewards/margins_max": 0.06821637600660324,
789
+ "rewards/margins_min": -0.0043562790378928185,
790
+ "rewards/margins_std": 0.03368216007947922,
791
+ "rewards/rejected": 0.020535241812467575,
792
+ "step": 350
793
+ },
794
+ {
795
+ "dpo_losses": 0.6761634945869446,
796
+ "epoch": 0.53,
797
+ "grad_norm": 5.401380341327095,
798
+ "learning_rate": 2.710467172300768e-07,
799
+ "logits/chosen": -2.7806248664855957,
800
+ "logits/rejected": -2.74116587638855,
801
+ "logps/chosen": -318.1079406738281,
802
+ "logps/rejected": -248.9733428955078,
803
+ "loss": 0.6774,
804
+ "positive_losses": 0.012074279598891735,
805
+ "rewards/accuracies": 0.7875000238418579,
806
+ "rewards/chosen": 0.05469100549817085,
807
+ "rewards/margins": 0.03459729254245758,
808
+ "rewards/margins_max": 0.0798027291893959,
809
+ "rewards/margins_min": -0.0013110924046486616,
810
+ "rewards/margins_std": 0.03574884682893753,
811
+ "rewards/rejected": 0.020093712955713272,
812
+ "step": 360
813
+ },
814
+ {
815
+ "dpo_losses": 0.6761563420295715,
816
+ "epoch": 0.54,
817
+ "grad_norm": 8.370611151073765,
818
+ "learning_rate": 2.582994309902146e-07,
819
+ "logits/chosen": -2.8219902515411377,
820
+ "logits/rejected": -2.7716636657714844,
821
+ "logps/chosen": -304.6541748046875,
822
+ "logps/rejected": -254.8093719482422,
823
+ "loss": 0.6842,
824
+ "positive_losses": 0.04906577989459038,
825
+ "rewards/accuracies": 0.737500011920929,
826
+ "rewards/chosen": 0.054839521646499634,
827
+ "rewards/margins": 0.03491091728210449,
828
+ "rewards/margins_max": 0.09758211672306061,
829
+ "rewards/margins_min": -0.014001704752445221,
830
+ "rewards/margins_std": 0.05021950602531433,
831
+ "rewards/rejected": 0.019928600639104843,
832
+ "step": 370
833
+ },
834
+ {
835
+ "dpo_losses": 0.6737654805183411,
836
+ "epoch": 0.56,
837
+ "grad_norm": 2.3064692447314554,
838
+ "learning_rate": 2.455304924825151e-07,
839
+ "logits/chosen": -2.837787389755249,
840
+ "logits/rejected": -2.831657886505127,
841
+ "logps/chosen": -275.2360534667969,
842
+ "logps/rejected": -255.569580078125,
843
+ "loss": 0.6738,
844
+ "positive_losses": 0.01725444756448269,
845
+ "rewards/accuracies": 0.8125,
846
+ "rewards/chosen": 0.05640103667974472,
847
+ "rewards/margins": 0.0399165078997612,
848
+ "rewards/margins_max": 0.09945692121982574,
849
+ "rewards/margins_min": -0.001225404441356659,
850
+ "rewards/margins_std": 0.045562874525785446,
851
+ "rewards/rejected": 0.016484536230564117,
852
+ "step": 380
853
+ },
854
+ {
855
+ "dpo_losses": 0.6766735911369324,
856
+ "epoch": 0.57,
857
+ "grad_norm": 1.805419113024878,
858
+ "learning_rate": 2.3277321440960732e-07,
859
+ "logits/chosen": -2.838366985321045,
860
+ "logits/rejected": -2.793659210205078,
861
+ "logps/chosen": -266.8939208984375,
862
+ "logps/rejected": -261.4732666015625,
863
+ "loss": 0.6807,
864
+ "positive_losses": 0.028429364785552025,
865
+ "rewards/accuracies": 0.7250000238418579,
866
+ "rewards/chosen": 0.0527314729988575,
867
+ "rewards/margins": 0.034005872905254364,
868
+ "rewards/margins_max": 0.10273897647857666,
869
+ "rewards/margins_min": -0.011970164254307747,
870
+ "rewards/margins_std": 0.05079951137304306,
871
+ "rewards/rejected": 0.018725596368312836,
872
+ "step": 390
873
+ },
874
+ {
875
+ "dpo_losses": 0.6763077974319458,
876
+ "epoch": 0.58,
877
+ "grad_norm": 6.733018778387388,
878
+ "learning_rate": 2.2006087905337698e-07,
879
+ "logits/chosen": -2.868297815322876,
880
+ "logits/rejected": -2.8423171043395996,
881
+ "logps/chosen": -238.1276092529297,
882
+ "logps/rejected": -231.619384765625,
883
+ "loss": 0.6808,
884
+ "positive_losses": 0.045419443398714066,
885
+ "rewards/accuracies": 0.7749999761581421,
886
+ "rewards/chosen": 0.05248277261853218,
887
+ "rewards/margins": 0.03450237587094307,
888
+ "rewards/margins_max": 0.09071175009012222,
889
+ "rewards/margins_min": -0.006746978498995304,
890
+ "rewards/margins_std": 0.04475879669189453,
891
+ "rewards/rejected": 0.01798040233552456,
892
+ "step": 400
893
+ },
894
+ {
895
+ "epoch": 0.58,
896
+ "eval_dpo_losses": 0.6860782504081726,
897
+ "eval_logits/chosen": -2.7994227409362793,
898
+ "eval_logits/rejected": -2.760826826095581,
899
+ "eval_logps/chosen": -279.1606140136719,
900
+ "eval_logps/rejected": -254.61215209960938,
901
+ "eval_loss": 0.6905239224433899,
902
+ "eval_positive_losses": 0.0314272977411747,
903
+ "eval_rewards/accuracies": 0.6169999837875366,
904
+ "eval_rewards/chosen": 0.05432803928852081,
905
+ "eval_rewards/margins": 0.014660396613180637,
906
+ "eval_rewards/margins_max": 0.08083328604698181,
907
+ "eval_rewards/margins_min": -0.043623533099889755,
908
+ "eval_rewards/margins_std": 0.04150667414069176,
909
+ "eval_rewards/rejected": 0.03966763988137245,
910
+ "eval_runtime": 428.0845,
911
+ "eval_samples_per_second": 4.672,
912
+ "eval_steps_per_second": 0.292,
913
+ "step": 400
914
+ },
915
+ {
916
+ "dpo_losses": 0.6729803085327148,
917
+ "epoch": 0.6,
918
+ "grad_norm": 6.255311463771333,
919
+ "learning_rate": 2.0742665144529372e-07,
920
+ "logits/chosen": -2.852710723876953,
921
+ "logits/rejected": -2.812171459197998,
922
+ "logps/chosen": -307.93133544921875,
923
+ "logps/rejected": -281.7948303222656,
924
+ "loss": 0.6784,
925
+ "positive_losses": 0.0057888031005859375,
926
+ "rewards/accuracies": 0.7250000238418579,
927
+ "rewards/chosen": 0.06238215044140816,
928
+ "rewards/margins": 0.04161245375871658,
929
+ "rewards/margins_max": 0.1156817227602005,
930
+ "rewards/margins_min": -0.008250080049037933,
931
+ "rewards/margins_std": 0.05661824345588684,
932
+ "rewards/rejected": 0.020769696682691574,
933
+ "step": 410
934
+ },
935
+ {
936
+ "dpo_losses": 0.6805132627487183,
937
+ "epoch": 0.61,
938
+ "grad_norm": 2.3659410877777494,
939
+ "learning_rate": 1.9490349284263033e-07,
940
+ "logits/chosen": -2.788618564605713,
941
+ "logits/rejected": -2.7586796283721924,
942
+ "logps/chosen": -255.90536499023438,
943
+ "logps/rejected": -242.9296417236328,
944
+ "loss": 0.6804,
945
+ "positive_losses": 0.0345400795340538,
946
+ "rewards/accuracies": 0.7250000238418579,
947
+ "rewards/chosen": 0.051929228007793427,
948
+ "rewards/margins": 0.026662463322281837,
949
+ "rewards/margins_max": 0.08727220445871353,
950
+ "rewards/margins_min": -0.02731608971953392,
951
+ "rewards/margins_std": 0.053510747849941254,
952
+ "rewards/rejected": 0.02526676654815674,
953
+ "step": 420
954
+ },
955
+ {
956
+ "dpo_losses": 0.6737715005874634,
957
+ "epoch": 0.63,
958
+ "grad_norm": 2.271597415804245,
959
+ "learning_rate": 1.8252407473630605e-07,
960
+ "logits/chosen": -2.8645436763763428,
961
+ "logits/rejected": -2.8088572025299072,
962
+ "logps/chosen": -295.20916748046875,
963
+ "logps/rejected": -230.15567016601562,
964
+ "loss": 0.6795,
965
+ "positive_losses": 0.03842503949999809,
966
+ "rewards/accuracies": 0.8125,
967
+ "rewards/chosen": 0.058985937386751175,
968
+ "rewards/margins": 0.03972195088863373,
969
+ "rewards/margins_max": 0.0958792120218277,
970
+ "rewards/margins_min": -0.005095441360026598,
971
+ "rewards/margins_std": 0.044632021337747574,
972
+ "rewards/rejected": 0.01926398277282715,
973
+ "step": 430
974
+ },
975
+ {
976
+ "dpo_losses": 0.6767727136611938,
977
+ "epoch": 0.64,
978
+ "grad_norm": 10.115009795187461,
979
+ "learning_rate": 1.7032069361469764e-07,
980
+ "logits/chosen": -2.7652392387390137,
981
+ "logits/rejected": -2.7494540214538574,
982
+ "logps/chosen": -226.9882354736328,
983
+ "logps/rejected": -269.02301025390625,
984
+ "loss": 0.6792,
985
+ "positive_losses": 0.009693240746855736,
986
+ "rewards/accuracies": 0.7250000238418579,
987
+ "rewards/chosen": 0.05702495574951172,
988
+ "rewards/margins": 0.03372834250330925,
989
+ "rewards/margins_max": 0.09970168769359589,
990
+ "rewards/margins_min": -0.013438252732157707,
991
+ "rewards/margins_std": 0.051222048699855804,
992
+ "rewards/rejected": 0.02329660952091217,
993
+ "step": 440
994
+ },
995
+ {
996
+ "dpo_losses": 0.6781430244445801,
997
+ "epoch": 0.66,
998
+ "grad_norm": 1.5440271538041583,
999
+ "learning_rate": 1.58325186705788e-07,
1000
+ "logits/chosen": -2.8817787170410156,
1001
+ "logits/rejected": -2.8590903282165527,
1002
+ "logps/chosen": -267.78729248046875,
1003
+ "logps/rejected": -251.52542114257812,
1004
+ "loss": 0.6868,
1005
+ "positive_losses": 0.08534412086009979,
1006
+ "rewards/accuracies": 0.7749999761581421,
1007
+ "rewards/chosen": 0.05013534426689148,
1008
+ "rewards/margins": 0.031011879444122314,
1009
+ "rewards/margins_max": 0.09007111191749573,
1010
+ "rewards/margins_min": -0.02042573131620884,
1011
+ "rewards/margins_std": 0.04950300231575966,
1012
+ "rewards/rejected": 0.019123462960124016,
1013
+ "step": 450
1014
+ },
1015
+ {
1016
+ "dpo_losses": 0.675542950630188,
1017
+ "epoch": 0.67,
1018
+ "grad_norm": 1.6924810012359601,
1019
+ "learning_rate": 1.4656884891747395e-07,
1020
+ "logits/chosen": -2.812002182006836,
1021
+ "logits/rejected": -2.765144109725952,
1022
+ "logps/chosen": -282.1943664550781,
1023
+ "logps/rejected": -234.0194854736328,
1024
+ "loss": 0.6808,
1025
+ "positive_losses": 0.03008117713034153,
1026
+ "rewards/accuracies": 0.800000011920929,
1027
+ "rewards/chosen": 0.055878765881061554,
1028
+ "rewards/margins": 0.03611396998167038,
1029
+ "rewards/margins_max": 0.09502451121807098,
1030
+ "rewards/margins_min": -0.008020835928618908,
1031
+ "rewards/margins_std": 0.04623667150735855,
1032
+ "rewards/rejected": 0.019764794036746025,
1033
+ "step": 460
1034
+ },
1035
+ {
1036
+ "dpo_losses": 0.6700197458267212,
1037
+ "epoch": 0.69,
1038
+ "grad_norm": 1.8078995560874291,
1039
+ "learning_rate": 1.3508235119272466e-07,
1040
+ "logits/chosen": -2.8690173625946045,
1041
+ "logits/rejected": -2.823881149291992,
1042
+ "logps/chosen": -322.7649230957031,
1043
+ "logps/rejected": -289.1722106933594,
1044
+ "loss": 0.6782,
1045
+ "positive_losses": 0.07821483910083771,
1046
+ "rewards/accuracies": 0.7749999761581421,
1047
+ "rewards/chosen": 0.06522538512945175,
1048
+ "rewards/margins": 0.04820011183619499,
1049
+ "rewards/margins_max": 0.1358465701341629,
1050
+ "rewards/margins_min": -0.011710671707987785,
1051
+ "rewards/margins_std": 0.06825564056634903,
1052
+ "rewards/rejected": 0.017025265842676163,
1053
+ "step": 470
1054
+ },
1055
+ {
1056
+ "dpo_losses": 0.6712939143180847,
1057
+ "epoch": 0.7,
1058
+ "grad_norm": 1.451487777642733,
1059
+ "learning_rate": 1.2389566049259336e-07,
1060
+ "logits/chosen": -2.811398983001709,
1061
+ "logits/rejected": -2.7891058921813965,
1062
+ "logps/chosen": -238.219970703125,
1063
+ "logps/rejected": -230.36062622070312,
1064
+ "loss": 0.6767,
1065
+ "positive_losses": 0.054373931139707565,
1066
+ "rewards/accuracies": 0.7875000238418579,
1067
+ "rewards/chosen": 0.06273017823696136,
1068
+ "rewards/margins": 0.0451803021132946,
1069
+ "rewards/margins_max": 0.10969402641057968,
1070
+ "rewards/margins_min": -0.0036571747623384,
1071
+ "rewards/margins_std": 0.051760654896497726,
1072
+ "rewards/rejected": 0.017549879848957062,
1073
+ "step": 480
1074
+ },
1075
+ {
1076
+ "dpo_losses": 0.6667095422744751,
1077
+ "epoch": 0.72,
1078
+ "grad_norm": 2.410585831948769,
1079
+ "learning_rate": 1.1303796161583762e-07,
1080
+ "logits/chosen": -2.9242846965789795,
1081
+ "logits/rejected": -2.840134859085083,
1082
+ "logps/chosen": -316.83929443359375,
1083
+ "logps/rejected": -256.23675537109375,
1084
+ "loss": 0.6785,
1085
+ "positive_losses": 0.06970572471618652,
1086
+ "rewards/accuracies": 0.800000011920929,
1087
+ "rewards/chosen": 0.0768619254231453,
1088
+ "rewards/margins": 0.05491393804550171,
1089
+ "rewards/margins_max": 0.13214218616485596,
1090
+ "rewards/margins_min": -0.004780332557857037,
1091
+ "rewards/margins_std": 0.06225153058767319,
1092
+ "rewards/rejected": 0.02194797992706299,
1093
+ "step": 490
1094
+ },
1095
+ {
1096
+ "dpo_losses": 0.6667904853820801,
1097
+ "epoch": 0.73,
1098
+ "grad_norm": 1.9094755680052469,
1099
+ "learning_rate": 1.0253758105911167e-07,
1100
+ "logits/chosen": -2.886962890625,
1101
+ "logits/rejected": -2.840977430343628,
1102
+ "logps/chosen": -331.50872802734375,
1103
+ "logps/rejected": -285.4704895019531,
1104
+ "loss": 0.6769,
1105
+ "positive_losses": 0.03339347988367081,
1106
+ "rewards/accuracies": 0.875,
1107
+ "rewards/chosen": 0.07617944478988647,
1108
+ "rewards/margins": 0.0544399730861187,
1109
+ "rewards/margins_max": 0.13042126595973969,
1110
+ "rewards/margins_min": -0.00018090903176926076,
1111
+ "rewards/margins_std": 0.05866100266575813,
1112
+ "rewards/rejected": 0.021739467978477478,
1113
+ "step": 500
1114
+ },
1115
+ {
1116
+ "epoch": 0.73,
1117
+ "eval_dpo_losses": 0.6854846477508545,
1118
+ "eval_logits/chosen": -2.797536611557007,
1119
+ "eval_logits/rejected": -2.7586684226989746,
1120
+ "eval_logps/chosen": -278.739990234375,
1121
+ "eval_logps/rejected": -254.3216094970703,
1122
+ "eval_loss": 0.6902235150337219,
1123
+ "eval_positive_losses": 0.032978080213069916,
1124
+ "eval_rewards/accuracies": 0.6119999885559082,
1125
+ "eval_rewards/chosen": 0.058533985167741776,
1126
+ "eval_rewards/margins": 0.015960821881890297,
1127
+ "eval_rewards/margins_max": 0.08850042521953583,
1128
+ "eval_rewards/margins_min": -0.048725374042987823,
1129
+ "eval_rewards/margins_std": 0.045832596719264984,
1130
+ "eval_rewards/rejected": 0.04257315769791603,
1131
+ "eval_runtime": 448.6936,
1132
+ "eval_samples_per_second": 4.457,
1133
+ "eval_steps_per_second": 0.279,
1134
+ "step": 500
1135
+ },
1136
+ {
1137
+ "dpo_losses": 0.6742907166481018,
1138
+ "epoch": 0.75,
1139
+ "grad_norm": 2.2304663983815316,
1140
+ "learning_rate": 9.242191311637049e-08,
1141
+ "logits/chosen": -2.8333396911621094,
1142
+ "logits/rejected": -2.8009631633758545,
1143
+ "logps/chosen": -277.5825500488281,
1144
+ "logps/rejected": -235.39108276367188,
1145
+ "loss": 0.6774,
1146
+ "positive_losses": 0.007343196775764227,
1147
+ "rewards/accuracies": 0.7749999761581421,
1148
+ "rewards/chosen": 0.05940967798233032,
1149
+ "rewards/margins": 0.038841020315885544,
1150
+ "rewards/margins_max": 0.1059766411781311,
1151
+ "rewards/margins_min": -0.00883492175489664,
1152
+ "rewards/margins_std": 0.052775926887989044,
1153
+ "rewards/rejected": 0.02056865207850933,
1154
+ "step": 510
1155
+ },
1156
+ {
1157
+ "dpo_losses": 0.6684087514877319,
1158
+ "epoch": 0.76,
1159
+ "grad_norm": 1.9114464768987025,
1160
+ "learning_rate": 8.271734841028552e-08,
1161
+ "logits/chosen": -2.8523058891296387,
1162
+ "logits/rejected": -2.7778074741363525,
1163
+ "logps/chosen": -335.83966064453125,
1164
+ "logps/rejected": -264.36016845703125,
1165
+ "loss": 0.6743,
1166
+ "positive_losses": 0.0079803466796875,
1167
+ "rewards/accuracies": 0.887499988079071,
1168
+ "rewards/chosen": 0.06848612427711487,
1169
+ "rewards/margins": 0.05102965235710144,
1170
+ "rewards/margins_max": 0.12273881584405899,
1171
+ "rewards/margins_min": -0.0025417390279471874,
1172
+ "rewards/margins_std": 0.05730118602514267,
1173
+ "rewards/rejected": 0.017456475645303726,
1174
+ "step": 520
1175
+ },
1176
+ {
1177
+ "dpo_losses": 0.6773437261581421,
1178
+ "epoch": 0.77,
1179
+ "grad_norm": 1.8982933837442546,
1180
+ "learning_rate": 7.344920504212243e-08,
1181
+ "logits/chosen": -2.839207172393799,
1182
+ "logits/rejected": -2.7776246070861816,
1183
+ "logps/chosen": -260.9156799316406,
1184
+ "logps/rejected": -216.26376342773438,
1185
+ "loss": 0.6801,
1186
+ "positive_losses": 0.04813060909509659,
1187
+ "rewards/accuracies": 0.7749999761581421,
1188
+ "rewards/chosen": 0.054224640130996704,
1189
+ "rewards/margins": 0.032528605312108994,
1190
+ "rewards/margins_max": 0.09149853885173798,
1191
+ "rewards/margins_min": -0.01699809357523918,
1192
+ "rewards/margins_std": 0.04981226474046707,
1193
+ "rewards/rejected": 0.02169603481888771,
1194
+ "step": 530
1195
+ },
1196
+ {
1197
+ "dpo_losses": 0.6705015301704407,
1198
+ "epoch": 0.79,
1199
+ "grad_norm": 1.723774449740064,
1200
+ "learning_rate": 6.46416625397067e-08,
1201
+ "logits/chosen": -2.834272861480713,
1202
+ "logits/rejected": -2.7763218879699707,
1203
+ "logps/chosen": -320.51885986328125,
1204
+ "logps/rejected": -295.53875732421875,
1205
+ "loss": 0.6751,
1206
+ "positive_losses": 0.018300628289580345,
1207
+ "rewards/accuracies": 0.8374999761581421,
1208
+ "rewards/chosen": 0.06741219758987427,
1209
+ "rewards/margins": 0.04666105657815933,
1210
+ "rewards/margins_max": 0.10739126056432724,
1211
+ "rewards/margins_min": -0.008476030081510544,
1212
+ "rewards/margins_std": 0.05081728845834732,
1213
+ "rewards/rejected": 0.020751142874360085,
1214
+ "step": 540
1215
+ },
1216
+ {
1217
+ "dpo_losses": 0.6752179861068726,
1218
+ "epoch": 0.8,
1219
+ "grad_norm": 1.6108967521538597,
1220
+ "learning_rate": 5.6317698775795344e-08,
1221
+ "logits/chosen": -2.8729920387268066,
1222
+ "logits/rejected": -2.833665370941162,
1223
+ "logps/chosen": -278.38275146484375,
1224
+ "logps/rejected": -270.0639953613281,
1225
+ "loss": 0.6723,
1226
+ "positive_losses": 0.005997705273330212,
1227
+ "rewards/accuracies": 0.75,
1228
+ "rewards/chosen": 0.057433564215898514,
1229
+ "rewards/margins": 0.037055402994155884,
1230
+ "rewards/margins_max": 0.09951488673686981,
1231
+ "rewards/margins_min": -0.01219971850514412,
1232
+ "rewards/margins_std": 0.051766276359558105,
1233
+ "rewards/rejected": 0.02037816308438778,
1234
+ "step": 550
1235
+ },
1236
+ {
1237
+ "dpo_losses": 0.6663364171981812,
1238
+ "epoch": 0.82,
1239
+ "grad_norm": 1.9255604336867507,
1240
+ "learning_rate": 4.849903002143113e-08,
1241
+ "logits/chosen": -2.9117777347564697,
1242
+ "logits/rejected": -2.8610081672668457,
1243
+ "logps/chosen": -328.8406066894531,
1244
+ "logps/rejected": -272.1076965332031,
1245
+ "loss": 0.6731,
1246
+ "positive_losses": 0.038059137761592865,
1247
+ "rewards/accuracies": 0.7749999761581421,
1248
+ "rewards/chosen": 0.07473967969417572,
1249
+ "rewards/margins": 0.055439990013837814,
1250
+ "rewards/margins_max": 0.12707814574241638,
1251
+ "rewards/margins_min": -0.002053080592304468,
1252
+ "rewards/margins_std": 0.05929755046963692,
1253
+ "rewards/rejected": 0.019299697130918503,
1254
+ "step": 560
1255
+ },
1256
+ {
1257
+ "dpo_losses": 0.6710665822029114,
1258
+ "epoch": 0.83,
1259
+ "grad_norm": 1.733507781542446,
1260
+ "learning_rate": 4.1206054290670535e-08,
1261
+ "logits/chosen": -2.8613619804382324,
1262
+ "logits/rejected": -2.82637357711792,
1263
+ "logps/chosen": -266.23773193359375,
1264
+ "logps/rejected": -258.0359802246094,
1265
+ "loss": 0.6756,
1266
+ "positive_losses": 0.013837432488799095,
1267
+ "rewards/accuracies": 0.7875000238418579,
1268
+ "rewards/chosen": 0.06855504214763641,
1269
+ "rewards/margins": 0.04551316797733307,
1270
+ "rewards/margins_max": 0.1095261201262474,
1271
+ "rewards/margins_min": -0.004014958627521992,
1272
+ "rewards/margins_std": 0.0510232150554657,
1273
+ "rewards/rejected": 0.02304188162088394,
1274
+ "step": 570
1275
+ },
1276
+ {
1277
+ "dpo_losses": 0.676629900932312,
1278
+ "epoch": 0.85,
1279
+ "grad_norm": 8.907703588890858,
1280
+ "learning_rate": 3.44577981244944e-08,
1281
+ "logits/chosen": -2.8228065967559814,
1282
+ "logits/rejected": -2.809900999069214,
1283
+ "logps/chosen": -257.9283752441406,
1284
+ "logps/rejected": -256.25799560546875,
1285
+ "loss": 0.6858,
1286
+ "positive_losses": 0.03807668760418892,
1287
+ "rewards/accuracies": 0.625,
1288
+ "rewards/chosen": 0.05910138040781021,
1289
+ "rewards/margins": 0.034723442047834396,
1290
+ "rewards/margins_max": 0.12428979575634003,
1291
+ "rewards/margins_min": -0.03289930149912834,
1292
+ "rewards/margins_std": 0.07077684998512268,
1293
+ "rewards/rejected": 0.024377940222620964,
1294
+ "step": 580
1295
+ },
1296
+ {
1297
+ "dpo_losses": 0.6756445169448853,
1298
+ "epoch": 0.86,
1299
+ "grad_norm": 8.494851741114475,
1300
+ "learning_rate": 2.8271866952734814e-08,
1301
+ "logits/chosen": -2.8694353103637695,
1302
+ "logits/rejected": -2.844571113586426,
1303
+ "logps/chosen": -302.86883544921875,
1304
+ "logps/rejected": -284.4591369628906,
1305
+ "loss": 0.6829,
1306
+ "positive_losses": 0.07926483452320099,
1307
+ "rewards/accuracies": 0.699999988079071,
1308
+ "rewards/chosen": 0.05832204222679138,
1309
+ "rewards/margins": 0.03624555841088295,
1310
+ "rewards/margins_max": 0.10978803783655167,
1311
+ "rewards/margins_min": -0.017319655045866966,
1312
+ "rewards/margins_std": 0.05776820331811905,
1313
+ "rewards/rejected": 0.022076481953263283,
1314
+ "step": 590
1315
+ },
1316
+ {
1317
+ "dpo_losses": 0.6685220003128052,
1318
+ "epoch": 0.88,
1319
+ "grad_norm": 2.0828943807385567,
1320
+ "learning_rate": 2.2664399163518782e-08,
1321
+ "logits/chosen": -2.7986605167388916,
1322
+ "logits/rejected": -2.781226396560669,
1323
+ "logps/chosen": -289.7279968261719,
1324
+ "logps/rejected": -246.7861328125,
1325
+ "loss": 0.67,
1326
+ "positive_losses": 0.0,
1327
+ "rewards/accuracies": 0.762499988079071,
1328
+ "rewards/chosen": 0.07399223744869232,
1329
+ "rewards/margins": 0.051160674542188644,
1330
+ "rewards/margins_max": 0.12970498204231262,
1331
+ "rewards/margins_min": -0.008301710709929466,
1332
+ "rewards/margins_std": 0.06261205673217773,
1333
+ "rewards/rejected": 0.02283155545592308,
1334
+ "step": 600
1335
+ },
1336
+ {
1337
+ "epoch": 0.88,
1338
+ "eval_dpo_losses": 0.6851370930671692,
1339
+ "eval_logits/chosen": -2.798839569091797,
1340
+ "eval_logits/rejected": -2.7602357864379883,
1341
+ "eval_logps/chosen": -278.6050109863281,
1342
+ "eval_logps/rejected": -254.26231384277344,
1343
+ "eval_loss": 0.6902583241462708,
1344
+ "eval_positive_losses": 0.03728185221552849,
1345
+ "eval_rewards/accuracies": 0.6159999966621399,
1346
+ "eval_rewards/chosen": 0.05988401547074318,
1347
+ "eval_rewards/margins": 0.016717953607439995,
1348
+ "eval_rewards/margins_max": 0.09297847747802734,
1349
+ "eval_rewards/margins_min": -0.051056575030088425,
1350
+ "eval_rewards/margins_std": 0.0480518564581871,
1351
+ "eval_rewards/rejected": 0.043166060000658035,
1352
+ "eval_runtime": 428.2792,
1353
+ "eval_samples_per_second": 4.67,
1354
+ "eval_steps_per_second": 0.292,
1355
+ "step": 600
1356
+ },
1357
+ {
1358
+ "dpo_losses": 0.6745679974555969,
1359
+ "epoch": 0.89,
1360
+ "grad_norm": 1.3564104662072647,
1361
+ "learning_rate": 1.7650024000056414e-08,
1362
+ "logits/chosen": -2.7833163738250732,
1363
+ "logits/rejected": -2.7807528972625732,
1364
+ "logps/chosen": -241.8382568359375,
1365
+ "logps/rejected": -243.6163330078125,
1366
+ "loss": 0.6742,
1367
+ "positive_losses": 0.0,
1368
+ "rewards/accuracies": 0.75,
1369
+ "rewards/chosen": 0.06093451380729675,
1370
+ "rewards/margins": 0.038374222815036774,
1371
+ "rewards/margins_max": 0.10369594395160675,
1372
+ "rewards/margins_min": -0.009305914863944054,
1373
+ "rewards/margins_std": 0.05068794637918472,
1374
+ "rewards/rejected": 0.022560294717550278,
1375
+ "step": 610
1376
+ },
1377
+ {
1378
+ "dpo_losses": 0.6677746772766113,
1379
+ "epoch": 0.91,
1380
+ "grad_norm": 7.979260503019614,
1381
+ "learning_rate": 1.3241823394615437e-08,
1382
+ "logits/chosen": -2.908095598220825,
1383
+ "logits/rejected": -2.8685083389282227,
1384
+ "logps/chosen": -304.2915954589844,
1385
+ "logps/rejected": -286.26873779296875,
1386
+ "loss": 0.675,
1387
+ "positive_losses": 0.05257830768823624,
1388
+ "rewards/accuracies": 0.8500000238418579,
1389
+ "rewards/chosen": 0.07303957641124725,
1390
+ "rewards/margins": 0.05244825407862663,
1391
+ "rewards/margins_max": 0.11377520859241486,
1392
+ "rewards/margins_min": -0.004015705548226833,
1393
+ "rewards/margins_std": 0.0548991933465004,
1394
+ "rewards/rejected": 0.020591318607330322,
1395
+ "step": 620
1396
+ },
1397
+ {
1398
+ "dpo_losses": 0.6700986623764038,
1399
+ "epoch": 0.92,
1400
+ "grad_norm": 8.477082267330537,
1401
+ "learning_rate": 9.451297839253913e-09,
1402
+ "logits/chosen": -2.7833220958709717,
1403
+ "logits/rejected": -2.7590456008911133,
1404
+ "logps/chosen": -276.22540283203125,
1405
+ "logps/rejected": -278.2480163574219,
1406
+ "loss": 0.6748,
1407
+ "positive_losses": 0.00507011404260993,
1408
+ "rewards/accuracies": 0.762499988079071,
1409
+ "rewards/chosen": 0.06989948451519012,
1410
+ "rewards/margins": 0.047721296548843384,
1411
+ "rewards/margins_max": 0.11920352280139923,
1412
+ "rewards/margins_min": -0.008280943147838116,
1413
+ "rewards/margins_std": 0.05656985193490982,
1414
+ "rewards/rejected": 0.02217818982899189,
1415
+ "step": 630
1416
+ },
1417
+ {
1418
+ "dpo_losses": 0.6701762080192566,
1419
+ "epoch": 0.94,
1420
+ "grad_norm": 1.4912730505697442,
1421
+ "learning_rate": 6.288336382349463e-09,
1422
+ "logits/chosen": -2.7986907958984375,
1423
+ "logits/rejected": -2.7656140327453613,
1424
+ "logps/chosen": -245.1352996826172,
1425
+ "logps/rejected": -207.5049285888672,
1426
+ "loss": 0.6724,
1427
+ "positive_losses": 0.004144692327827215,
1428
+ "rewards/accuracies": 0.800000011920929,
1429
+ "rewards/chosen": 0.06695154309272766,
1430
+ "rewards/margins": 0.04747752100229263,
1431
+ "rewards/margins_max": 0.12407982349395752,
1432
+ "rewards/margins_min": -0.003674466861411929,
1433
+ "rewards/margins_std": 0.05860821157693863,
1434
+ "rewards/rejected": 0.01947401836514473,
1435
+ "step": 640
1436
+ },
1437
+ {
1438
+ "dpo_losses": 0.6735118627548218,
1439
+ "epoch": 0.95,
1440
+ "grad_norm": 10.343240316228654,
1441
+ "learning_rate": 3.7611908292010665e-09,
1442
+ "logits/chosen": -2.859143018722534,
1443
+ "logits/rejected": -2.8227005004882812,
1444
+ "logps/chosen": -307.39508056640625,
1445
+ "logps/rejected": -258.0931091308594,
1446
+ "loss": 0.6774,
1447
+ "positive_losses": 0.10746382176876068,
1448
+ "rewards/accuracies": 0.7124999761581421,
1449
+ "rewards/chosen": 0.05885057523846626,
1450
+ "rewards/margins": 0.040867820382118225,
1451
+ "rewards/margins_max": 0.11666282266378403,
1452
+ "rewards/margins_min": -0.015247263014316559,
1453
+ "rewards/margins_std": 0.05968106538057327,
1454
+ "rewards/rejected": 0.017982754856348038,
1455
+ "step": 650
1456
+ },
1457
+ {
1458
+ "dpo_losses": 0.6789201498031616,
1459
+ "epoch": 0.96,
1460
+ "grad_norm": 12.216425135481341,
1461
+ "learning_rate": 1.8764542140112527e-09,
1462
+ "logits/chosen": -2.800410509109497,
1463
+ "logits/rejected": -2.8051278591156006,
1464
+ "logps/chosen": -219.88369750976562,
1465
+ "logps/rejected": -224.5011749267578,
1466
+ "loss": 0.6838,
1467
+ "positive_losses": 0.09240980446338654,
1468
+ "rewards/accuracies": 0.699999988079071,
1469
+ "rewards/chosen": 0.0470733642578125,
1470
+ "rewards/margins": 0.02947625145316124,
1471
+ "rewards/margins_max": 0.0936986654996872,
1472
+ "rewards/margins_min": -0.014550586231052876,
1473
+ "rewards/margins_std": 0.048393599689006805,
1474
+ "rewards/rejected": 0.01759711280465126,
1475
+ "step": 660
1476
+ },
1477
+ {
1478
+ "dpo_losses": 0.6718948483467102,
1479
+ "epoch": 0.98,
1480
+ "grad_norm": 7.347929588964824,
1481
+ "learning_rate": 6.390435994127752e-10,
1482
+ "logits/chosen": -2.8291993141174316,
1483
+ "logits/rejected": -2.7946581840515137,
1484
+ "logps/chosen": -279.2850646972656,
1485
+ "logps/rejected": -295.1185607910156,
1486
+ "loss": 0.6767,
1487
+ "positive_losses": 0.03623781353235245,
1488
+ "rewards/accuracies": 0.762499988079071,
1489
+ "rewards/chosen": 0.06117644160985947,
1490
+ "rewards/margins": 0.044048137962818146,
1491
+ "rewards/margins_max": 0.11197330802679062,
1492
+ "rewards/margins_min": -0.004027059301733971,
1493
+ "rewards/margins_std": 0.05286269634962082,
1494
+ "rewards/rejected": 0.01712830550968647,
1495
+ "step": 670
1496
+ },
1497
+ {
1498
+ "dpo_losses": 0.6698868870735168,
1499
+ "epoch": 0.99,
1500
+ "grad_norm": 9.194005207225429,
1501
+ "learning_rate": 5.2187248413465555e-11,
1502
+ "logits/chosen": -2.8184335231781006,
1503
+ "logits/rejected": -2.7909679412841797,
1504
+ "logps/chosen": -296.29473876953125,
1505
+ "logps/rejected": -258.4402160644531,
1506
+ "loss": 0.6757,
1507
+ "positive_losses": 0.03602256625890732,
1508
+ "rewards/accuracies": 0.75,
1509
+ "rewards/chosen": 0.06865200400352478,
1510
+ "rewards/margins": 0.04830945283174515,
1511
+ "rewards/margins_max": 0.1308433711528778,
1512
+ "rewards/margins_min": -0.012822682037949562,
1513
+ "rewards/margins_std": 0.06532330065965652,
1514
+ "rewards/rejected": 0.020342541858553886,
1515
+ "step": 680
1516
+ },
1517
+ {
1518
+ "epoch": 1.0,
1519
+ "step": 684,
1520
+ "total_flos": 0.0,
1521
+ "train_loss": 0.6833745032026056,
1522
+ "train_runtime": 9125.7542,
1523
+ "train_samples_per_second": 1.199,
1524
+ "train_steps_per_second": 0.075
1525
+ }
1526
+ ],
1527
+ "logging_steps": 10,
1528
+ "max_steps": 684,
1529
+ "num_input_tokens_seen": 0,
1530
+ "num_train_epochs": 1,
1531
+ "save_steps": 100,
1532
+ "total_flos": 0.0,
1533
+ "train_batch_size": 4,
1534
+ "trial_name": null,
1535
+ "trial_params": null
1536
+ }