Hugging Face's logo

AmberYifan
/
Qwen2-7B-sft-ultrachat-dpo-10k

Text Generation
Transformers
Safetensors
qwen2
Generated from Trainer
trl
dpo
conversational
text-generation-inference
Model card Files Community
AmberYifan commited on
Commit
36571d4
·
verified ·
1 Parent(s): 53bcea1

Model save

Browse files
Files changed (9) hide show
  1. README.md +68 -0
  2. all_results.json +9 -0
  3. generation_config.json +6 -0
  4. model-00001-of-00004.safetensors +1 -1
  5. model-00002-of-00004.safetensors +1 -1
  6. model-00003-of-00004.safetensors +1 -1
  7. model-00004-of-00004.safetensors +1 -1
  8. train_results.json +9 -0
  9. trainer_state.json +1500 -0
README.md ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: AmberYifan/Qwen2-7B-sft-ultrachat
3
+ library_name: transformers
4
+ model_name: Qwen2-7B-sft-ultrachat-dpo-10k
5
+ tags:
6
+ - generated_from_trainer
7
+ - trl
8
+ - dpo
9
+ licence: license
10
+ ---
11
+
12
+ # Model Card for Qwen2-7B-sft-ultrachat-dpo-10k
13
+
14
+ This model is a fine-tuned version of [AmberYifan/Qwen2-7B-sft-ultrachat](https://huggingface.co/AmberYifan/Qwen2-7B-sft-ultrachat).
15
+ It has been trained using [TRL](https://github.com/huggingface/trl).
16
+
17
+ ## Quick start
18
+
19
+ ```python
20
+ from transformers import pipeline
21
+
22
+ question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
23
+ generator = pipeline("text-generation", model="AmberYifan/Qwen2-7B-sft-ultrachat-dpo-10k", device="cuda")
24
+ output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
25
+ print(output["generated_text"])
26
+ ```
27
+
28
+ ## Training procedure
29
+
30
+ [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/yifanwang/huggingface/runs/4usoonu4)
31
+
32
+ This model was trained with DPO, a method introduced in [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://huggingface.co/papers/2305.18290).
33
+
34
+ ### Framework versions
35
+
36
+ - TRL: 0.12.2
37
+ - Transformers: 4.46.3
38
+ - Pytorch: 2.6.0
39
+ - Datasets: 3.3.0
40
+ - Tokenizers: 0.20.3
41
+
42
+ ## Citations
43
+
44
+ Cite DPO as:
45
+
46
+ ```bibtex
47
+ @inproceedings{rafailov2023direct,
48
+ title = {{Direct Preference Optimization: Your Language Model is Secretly a Reward Model}},
49
+ author = {Rafael Rafailov and Archit Sharma and Eric Mitchell and Christopher D. Manning and Stefano Ermon and Chelsea Finn},
50
+ year = 2023,
51
+ booktitle = {Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023},
52
+ url = {http://papers.nips.cc/paper_files/paper/2023/hash/a85b405ed65c6477a4fe8302b5e06ce7-Abstract-Conference.html},
53
+ editor = {Alice Oh and Tristan Naumann and Amir Globerson and Kate Saenko and Moritz Hardt and Sergey Levine},
54
+ }
55
+ ```
56
+
57
+ Cite TRL as:
58
+
59
+ ```bibtex
60
+ @misc{vonwerra2022trl,
61
+ title = {{TRL: Transformer Reinforcement Learning}},
62
+ author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
63
+ year = 2020,
64
+ journal = {GitHub repository},
65
+ publisher = {GitHub},
66
+ howpublished = {\url{https://github.com/huggingface/trl}}
67
+ }
68
+ ```
all_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 3.0,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.350877452100444,
5
+ "train_runtime": 7216.813,
6
+ "train_samples": 9984,
7
+ "train_samples_per_second": 4.15,
8
+ "train_steps_per_second": 0.13
9
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 151643,
3
+ "eos_token_id": 151643,
4
+ "max_new_tokens": 2048,
5
+ "transformers_version": "4.46.3"
6
+ }
model-00001-of-00004.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:03666b341aea1e882baeabdf705b2bfc652d846442dc2942ffb8428fa4abb414
3
  size 4877660776
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:38991ea1358ac9cbf64bfaf78f4dffbdaf65ad8dcab9554be81e5e056bb0499e
3
  size 4877660776
model-00002-of-00004.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:def41960df887f693f8e05b14a3ff8c152a1bf898211f3e272cf952d606b758b
3
  size 4932751008
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:95cd1243732375780ce789cc568b7bbb08c9410e8fc84d2e4094e1351ec36639
3
  size 4932751008
model-00003-of-00004.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d3116e484db3429bb3dcd133d60c3ef7735ce0e657a253966dd6473ecdf97dc2
3
  size 4330865200
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:03f9e0621274349f9460896c7f3525e6538ad892d27b1169dbc2589190c8783f
3
  size 4330865200
model-00004-of-00004.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6c04996a89919200924220b41f8d07d0cf34bb79035ebff8622bd16ed7cc0275
3
  size 1089994880
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:486668095ba7691092e489448ba6beb27e924a1c6051534b377d6fa619e38cff
3
  size 1089994880
train_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 3.0,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.350877452100444,
5
+ "train_runtime": 7216.813,
6
+ "train_samples": 9984,
7
+ "train_samples_per_second": 4.15,
8
+ "train_steps_per_second": 0.13
9
+ }
trainer_state.json ADDED
@@ -0,0 +1,1500 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 3.0,
5
+ "eval_steps": 500,
6
+ "global_step": 936,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.003205128205128205,
13
+ "grad_norm": 48.20300728973734,
14
+ "learning_rate": 5.3191489361702125e-09,
15
+ "logits/chosen": -0.267578125,
16
+ "logits/rejected": -0.462890625,
17
+ "logps/chosen": -320.0,
18
+ "logps/rejected": -155.0,
19
+ "loss": 0.6914,
20
+ "rewards/accuracies": 0.0,
21
+ "rewards/chosen": 0.0,
22
+ "rewards/margins": 0.0,
23
+ "rewards/rejected": 0.0,
24
+ "step": 1
25
+ },
26
+ {
27
+ "epoch": 0.03205128205128205,
28
+ "grad_norm": 40.55807354957128,
29
+ "learning_rate": 5.3191489361702123e-08,
30
+ "logits/chosen": -0.265625,
31
+ "logits/rejected": -0.37109375,
32
+ "logps/chosen": -348.0,
33
+ "logps/rejected": -318.0,
34
+ "loss": 0.693,
35
+ "rewards/accuracies": 0.1944444477558136,
36
+ "rewards/chosen": -0.0107421875,
37
+ "rewards/margins": -0.0093994140625,
38
+ "rewards/rejected": -0.00138092041015625,
39
+ "step": 10
40
+ },
41
+ {
42
+ "epoch": 0.0641025641025641,
43
+ "grad_norm": 38.94632768279024,
44
+ "learning_rate": 1.0638297872340425e-07,
45
+ "logits/chosen": -0.416015625,
46
+ "logits/rejected": -0.384765625,
47
+ "logps/chosen": -272.0,
48
+ "logps/rejected": -248.0,
49
+ "loss": 0.6986,
50
+ "rewards/accuracies": 0.30000001192092896,
51
+ "rewards/chosen": -0.004425048828125,
52
+ "rewards/margins": -0.00811767578125,
53
+ "rewards/rejected": 0.00372314453125,
54
+ "step": 20
55
+ },
56
+ {
57
+ "epoch": 0.09615384615384616,
58
+ "grad_norm": 44.94338708626801,
59
+ "learning_rate": 1.5957446808510638e-07,
60
+ "logits/chosen": -0.412109375,
61
+ "logits/rejected": -0.31640625,
62
+ "logps/chosen": -292.0,
63
+ "logps/rejected": -274.0,
64
+ "loss": 0.6842,
65
+ "rewards/accuracies": 0.32499998807907104,
66
+ "rewards/chosen": 0.00811767578125,
67
+ "rewards/margins": 0.004058837890625,
68
+ "rewards/rejected": 0.004058837890625,
69
+ "step": 30
70
+ },
71
+ {
72
+ "epoch": 0.1282051282051282,
73
+ "grad_norm": 44.203343489664924,
74
+ "learning_rate": 2.127659574468085e-07,
75
+ "logits/chosen": -0.439453125,
76
+ "logits/rejected": -0.353515625,
77
+ "logps/chosen": -280.0,
78
+ "logps/rejected": -336.0,
79
+ "loss": 0.6941,
80
+ "rewards/accuracies": 0.23749999701976776,
81
+ "rewards/chosen": 0.0106201171875,
82
+ "rewards/margins": -0.006256103515625,
83
+ "rewards/rejected": 0.016845703125,
84
+ "step": 40
85
+ },
86
+ {
87
+ "epoch": 0.16025641025641027,
88
+ "grad_norm": 52.46671251233884,
89
+ "learning_rate": 2.659574468085106e-07,
90
+ "logits/chosen": -0.31640625,
91
+ "logits/rejected": -0.35546875,
92
+ "logps/chosen": -306.0,
93
+ "logps/rejected": -314.0,
94
+ "loss": 0.6874,
95
+ "rewards/accuracies": 0.42500001192092896,
96
+ "rewards/chosen": 0.03466796875,
97
+ "rewards/margins": 0.047607421875,
98
+ "rewards/rejected": -0.0128173828125,
99
+ "step": 50
100
+ },
101
+ {
102
+ "epoch": 0.19230769230769232,
103
+ "grad_norm": 52.16025133508881,
104
+ "learning_rate": 3.1914893617021275e-07,
105
+ "logits/chosen": -0.408203125,
106
+ "logits/rejected": -0.451171875,
107
+ "logps/chosen": -284.0,
108
+ "logps/rejected": -260.0,
109
+ "loss": 0.6793,
110
+ "rewards/accuracies": 0.48750001192092896,
111
+ "rewards/chosen": 0.064453125,
112
+ "rewards/margins": 0.034423828125,
113
+ "rewards/rejected": 0.030029296875,
114
+ "step": 60
115
+ },
116
+ {
117
+ "epoch": 0.22435897435897437,
118
+ "grad_norm": 44.198032602975644,
119
+ "learning_rate": 3.7234042553191484e-07,
120
+ "logits/chosen": -0.435546875,
121
+ "logits/rejected": -0.390625,
122
+ "logps/chosen": -364.0,
123
+ "logps/rejected": -292.0,
124
+ "loss": 0.6599,
125
+ "rewards/accuracies": 0.574999988079071,
126
+ "rewards/chosen": 0.09130859375,
127
+ "rewards/margins": 0.0888671875,
128
+ "rewards/rejected": 0.00250244140625,
129
+ "step": 70
130
+ },
131
+ {
132
+ "epoch": 0.2564102564102564,
133
+ "grad_norm": 40.58383899570081,
134
+ "learning_rate": 4.25531914893617e-07,
135
+ "logits/chosen": -0.380859375,
136
+ "logits/rejected": -0.3359375,
137
+ "logps/chosen": -276.0,
138
+ "logps/rejected": -256.0,
139
+ "loss": 0.6542,
140
+ "rewards/accuracies": 0.4749999940395355,
141
+ "rewards/chosen": 0.11083984375,
142
+ "rewards/margins": 0.09130859375,
143
+ "rewards/rejected": 0.02001953125,
144
+ "step": 80
145
+ },
146
+ {
147
+ "epoch": 0.28846153846153844,
148
+ "grad_norm": 37.229119513751584,
149
+ "learning_rate": 4.787234042553192e-07,
150
+ "logits/chosen": -0.3359375,
151
+ "logits/rejected": -0.26171875,
152
+ "logps/chosen": -364.0,
153
+ "logps/rejected": -308.0,
154
+ "loss": 0.6369,
155
+ "rewards/accuracies": 0.637499988079071,
156
+ "rewards/chosen": 0.19140625,
157
+ "rewards/margins": 0.2001953125,
158
+ "rewards/rejected": -0.0084228515625,
159
+ "step": 90
160
+ },
161
+ {
162
+ "epoch": 0.32051282051282054,
163
+ "grad_norm": 39.33278491363928,
164
+ "learning_rate": 4.96437054631829e-07,
165
+ "logits/chosen": -0.400390625,
166
+ "logits/rejected": -0.2412109375,
167
+ "logps/chosen": -290.0,
168
+ "logps/rejected": -310.0,
169
+ "loss": 0.6039,
170
+ "rewards/accuracies": 0.675000011920929,
171
+ "rewards/chosen": 0.138671875,
172
+ "rewards/margins": 0.23046875,
173
+ "rewards/rejected": -0.091796875,
174
+ "step": 100
175
+ },
176
+ {
177
+ "epoch": 0.3525641025641026,
178
+ "grad_norm": 43.81240463931525,
179
+ "learning_rate": 4.904988123515439e-07,
180
+ "logits/chosen": -0.3984375,
181
+ "logits/rejected": -0.3984375,
182
+ "logps/chosen": -370.0,
183
+ "logps/rejected": -352.0,
184
+ "loss": 0.5957,
185
+ "rewards/accuracies": 0.637499988079071,
186
+ "rewards/chosen": 0.1435546875,
187
+ "rewards/margins": 0.251953125,
188
+ "rewards/rejected": -0.10791015625,
189
+ "step": 110
190
+ },
191
+ {
192
+ "epoch": 0.38461538461538464,
193
+ "grad_norm": 39.38457584773928,
194
+ "learning_rate": 4.845605700712589e-07,
195
+ "logits/chosen": -0.193359375,
196
+ "logits/rejected": -0.25390625,
197
+ "logps/chosen": -260.0,
198
+ "logps/rejected": -278.0,
199
+ "loss": 0.5985,
200
+ "rewards/accuracies": 0.574999988079071,
201
+ "rewards/chosen": -0.00640869140625,
202
+ "rewards/margins": 0.1279296875,
203
+ "rewards/rejected": -0.134765625,
204
+ "step": 120
205
+ },
206
+ {
207
+ "epoch": 0.4166666666666667,
208
+ "grad_norm": 41.00555642083936,
209
+ "learning_rate": 4.786223277909738e-07,
210
+ "logits/chosen": -0.44921875,
211
+ "logits/rejected": -0.3203125,
212
+ "logps/chosen": -302.0,
213
+ "logps/rejected": -272.0,
214
+ "loss": 0.5941,
215
+ "rewards/accuracies": 0.6499999761581421,
216
+ "rewards/chosen": 0.22265625,
217
+ "rewards/margins": 0.345703125,
218
+ "rewards/rejected": -0.123046875,
219
+ "step": 130
220
+ },
221
+ {
222
+ "epoch": 0.44871794871794873,
223
+ "grad_norm": 39.857994444324696,
224
+ "learning_rate": 4.7268408551068883e-07,
225
+ "logits/chosen": -0.421875,
226
+ "logits/rejected": -0.302734375,
227
+ "logps/chosen": -338.0,
228
+ "logps/rejected": -302.0,
229
+ "loss": 0.5634,
230
+ "rewards/accuracies": 0.699999988079071,
231
+ "rewards/chosen": 0.166015625,
232
+ "rewards/margins": 0.41015625,
233
+ "rewards/rejected": -0.244140625,
234
+ "step": 140
235
+ },
236
+ {
237
+ "epoch": 0.4807692307692308,
238
+ "grad_norm": 30.401529156884564,
239
+ "learning_rate": 4.667458432304038e-07,
240
+ "logits/chosen": -0.3125,
241
+ "logits/rejected": -0.28125,
242
+ "logps/chosen": -278.0,
243
+ "logps/rejected": -262.0,
244
+ "loss": 0.5875,
245
+ "rewards/accuracies": 0.6625000238418579,
246
+ "rewards/chosen": 0.1005859375,
247
+ "rewards/margins": 0.34375,
248
+ "rewards/rejected": -0.2421875,
249
+ "step": 150
250
+ },
251
+ {
252
+ "epoch": 0.5128205128205128,
253
+ "grad_norm": 40.47565062008821,
254
+ "learning_rate": 4.6080760095011875e-07,
255
+ "logits/chosen": -0.306640625,
256
+ "logits/rejected": -0.22265625,
257
+ "logps/chosen": -314.0,
258
+ "logps/rejected": -286.0,
259
+ "loss": 0.5833,
260
+ "rewards/accuracies": 0.7124999761581421,
261
+ "rewards/chosen": 0.09716796875,
262
+ "rewards/margins": 0.466796875,
263
+ "rewards/rejected": -0.369140625,
264
+ "step": 160
265
+ },
266
+ {
267
+ "epoch": 0.5448717948717948,
268
+ "grad_norm": 47.70230858758091,
269
+ "learning_rate": 4.548693586698337e-07,
270
+ "logits/chosen": -0.369140625,
271
+ "logits/rejected": -0.46484375,
272
+ "logps/chosen": -322.0,
273
+ "logps/rejected": -324.0,
274
+ "loss": 0.5529,
275
+ "rewards/accuracies": 0.737500011920929,
276
+ "rewards/chosen": 0.201171875,
277
+ "rewards/margins": 0.55078125,
278
+ "rewards/rejected": -0.349609375,
279
+ "step": 170
280
+ },
281
+ {
282
+ "epoch": 0.5769230769230769,
283
+ "grad_norm": 37.59643094962871,
284
+ "learning_rate": 4.4893111638954866e-07,
285
+ "logits/chosen": -0.443359375,
286
+ "logits/rejected": -0.4453125,
287
+ "logps/chosen": -388.0,
288
+ "logps/rejected": -296.0,
289
+ "loss": 0.5596,
290
+ "rewards/accuracies": 0.6499999761581421,
291
+ "rewards/chosen": 0.1279296875,
292
+ "rewards/margins": 0.49609375,
293
+ "rewards/rejected": -0.3671875,
294
+ "step": 180
295
+ },
296
+ {
297
+ "epoch": 0.6089743589743589,
298
+ "grad_norm": 42.89913513576229,
299
+ "learning_rate": 4.429928741092636e-07,
300
+ "logits/chosen": -0.306640625,
301
+ "logits/rejected": -0.3828125,
302
+ "logps/chosen": -306.0,
303
+ "logps/rejected": -278.0,
304
+ "loss": 0.5667,
305
+ "rewards/accuracies": 0.6625000238418579,
306
+ "rewards/chosen": -0.05908203125,
307
+ "rewards/margins": 0.53125,
308
+ "rewards/rejected": -0.58984375,
309
+ "step": 190
310
+ },
311
+ {
312
+ "epoch": 0.6410256410256411,
313
+ "grad_norm": 38.491897938435464,
314
+ "learning_rate": 4.3705463182897863e-07,
315
+ "logits/chosen": -0.5390625,
316
+ "logits/rejected": -0.412109375,
317
+ "logps/chosen": -324.0,
318
+ "logps/rejected": -322.0,
319
+ "loss": 0.5589,
320
+ "rewards/accuracies": 0.675000011920929,
321
+ "rewards/chosen": -0.02294921875,
322
+ "rewards/margins": 0.470703125,
323
+ "rewards/rejected": -0.4921875,
324
+ "step": 200
325
+ },
326
+ {
327
+ "epoch": 0.6730769230769231,
328
+ "grad_norm": 45.42331710145845,
329
+ "learning_rate": 4.311163895486936e-07,
330
+ "logits/chosen": -0.326171875,
331
+ "logits/rejected": -0.365234375,
332
+ "logps/chosen": -320.0,
333
+ "logps/rejected": -330.0,
334
+ "loss": 0.567,
335
+ "rewards/accuracies": 0.75,
336
+ "rewards/chosen": -0.10791015625,
337
+ "rewards/margins": 0.7109375,
338
+ "rewards/rejected": -0.8203125,
339
+ "step": 210
340
+ },
341
+ {
342
+ "epoch": 0.7051282051282052,
343
+ "grad_norm": 38.10849789978559,
344
+ "learning_rate": 4.251781472684085e-07,
345
+ "logits/chosen": -0.267578125,
346
+ "logits/rejected": -0.26171875,
347
+ "logps/chosen": -318.0,
348
+ "logps/rejected": -298.0,
349
+ "loss": 0.5264,
350
+ "rewards/accuracies": 0.6875,
351
+ "rewards/chosen": 0.062255859375,
352
+ "rewards/margins": 0.6484375,
353
+ "rewards/rejected": -0.5859375,
354
+ "step": 220
355
+ },
356
+ {
357
+ "epoch": 0.7371794871794872,
358
+ "grad_norm": 35.69252085216096,
359
+ "learning_rate": 4.192399049881235e-07,
360
+ "logits/chosen": -0.4375,
361
+ "logits/rejected": -0.392578125,
362
+ "logps/chosen": -296.0,
363
+ "logps/rejected": -282.0,
364
+ "loss": 0.5494,
365
+ "rewards/accuracies": 0.762499988079071,
366
+ "rewards/chosen": 0.0233154296875,
367
+ "rewards/margins": 0.734375,
368
+ "rewards/rejected": -0.7109375,
369
+ "step": 230
370
+ },
371
+ {
372
+ "epoch": 0.7692307692307693,
373
+ "grad_norm": 49.72047308664797,
374
+ "learning_rate": 4.1330166270783846e-07,
375
+ "logits/chosen": -0.427734375,
376
+ "logits/rejected": -0.37109375,
377
+ "logps/chosen": -338.0,
378
+ "logps/rejected": -330.0,
379
+ "loss": 0.5414,
380
+ "rewards/accuracies": 0.6499999761581421,
381
+ "rewards/chosen": -0.171875,
382
+ "rewards/margins": 0.484375,
383
+ "rewards/rejected": -0.65625,
384
+ "step": 240
385
+ },
386
+ {
387
+ "epoch": 0.8012820512820513,
388
+ "grad_norm": 43.61282576031846,
389
+ "learning_rate": 4.0736342042755347e-07,
390
+ "logits/chosen": -0.3125,
391
+ "logits/rejected": -0.27734375,
392
+ "logps/chosen": -304.0,
393
+ "logps/rejected": -326.0,
394
+ "loss": 0.5342,
395
+ "rewards/accuracies": 0.6875,
396
+ "rewards/chosen": -0.1904296875,
397
+ "rewards/margins": 0.494140625,
398
+ "rewards/rejected": -0.6875,
399
+ "step": 250
400
+ },
401
+ {
402
+ "epoch": 0.8333333333333334,
403
+ "grad_norm": 33.83836461453226,
404
+ "learning_rate": 4.0142517814726837e-07,
405
+ "logits/chosen": -0.341796875,
406
+ "logits/rejected": -0.330078125,
407
+ "logps/chosen": -326.0,
408
+ "logps/rejected": -284.0,
409
+ "loss": 0.5625,
410
+ "rewards/accuracies": 0.737500011920929,
411
+ "rewards/chosen": 0.059326171875,
412
+ "rewards/margins": 0.765625,
413
+ "rewards/rejected": -0.70703125,
414
+ "step": 260
415
+ },
416
+ {
417
+ "epoch": 0.8653846153846154,
418
+ "grad_norm": 32.404327383318645,
419
+ "learning_rate": 3.9548693586698333e-07,
420
+ "logits/chosen": -0.4140625,
421
+ "logits/rejected": -0.359375,
422
+ "logps/chosen": -300.0,
423
+ "logps/rejected": -288.0,
424
+ "loss": 0.5288,
425
+ "rewards/accuracies": 0.7124999761581421,
426
+ "rewards/chosen": 0.087890625,
427
+ "rewards/margins": 0.6328125,
428
+ "rewards/rejected": -0.54296875,
429
+ "step": 270
430
+ },
431
+ {
432
+ "epoch": 0.8974358974358975,
433
+ "grad_norm": 45.49431428845984,
434
+ "learning_rate": 3.8954869358669834e-07,
435
+ "logits/chosen": -0.26953125,
436
+ "logits/rejected": -0.3046875,
437
+ "logps/chosen": -328.0,
438
+ "logps/rejected": -288.0,
439
+ "loss": 0.5815,
440
+ "rewards/accuracies": 0.699999988079071,
441
+ "rewards/chosen": -0.107421875,
442
+ "rewards/margins": 0.4453125,
443
+ "rewards/rejected": -0.5546875,
444
+ "step": 280
445
+ },
446
+ {
447
+ "epoch": 0.9294871794871795,
448
+ "grad_norm": 40.072165310498,
449
+ "learning_rate": 3.836104513064133e-07,
450
+ "logits/chosen": -0.416015625,
451
+ "logits/rejected": -0.306640625,
452
+ "logps/chosen": -298.0,
453
+ "logps/rejected": -235.0,
454
+ "loss": 0.5336,
455
+ "rewards/accuracies": 0.7124999761581421,
456
+ "rewards/chosen": 0.035888671875,
457
+ "rewards/margins": 0.6953125,
458
+ "rewards/rejected": -0.66015625,
459
+ "step": 290
460
+ },
461
+ {
462
+ "epoch": 0.9615384615384616,
463
+ "grad_norm": 32.34581916400338,
464
+ "learning_rate": 3.7767220902612825e-07,
465
+ "logits/chosen": -0.427734375,
466
+ "logits/rejected": -0.408203125,
467
+ "logps/chosen": -312.0,
468
+ "logps/rejected": -318.0,
469
+ "loss": 0.521,
470
+ "rewards/accuracies": 0.6499999761581421,
471
+ "rewards/chosen": 0.2021484375,
472
+ "rewards/margins": 0.78515625,
473
+ "rewards/rejected": -0.5859375,
474
+ "step": 300
475
+ },
476
+ {
477
+ "epoch": 0.9935897435897436,
478
+ "grad_norm": 40.41128996018611,
479
+ "learning_rate": 3.717339667458432e-07,
480
+ "logits/chosen": -0.2890625,
481
+ "logits/rejected": -0.275390625,
482
+ "logps/chosen": -286.0,
483
+ "logps/rejected": -284.0,
484
+ "loss": 0.5046,
485
+ "rewards/accuracies": 0.7875000238418579,
486
+ "rewards/chosen": 0.0751953125,
487
+ "rewards/margins": 0.82421875,
488
+ "rewards/rejected": -0.75,
489
+ "step": 310
490
+ },
491
+ {
492
+ "epoch": 1.0,
493
+ "eval_logits/chosen": -0.341796875,
494
+ "eval_logits/rejected": -0.34375,
495
+ "eval_logps/chosen": -294.0,
496
+ "eval_logps/rejected": -332.0,
497
+ "eval_loss": 0.5677343606948853,
498
+ "eval_rewards/accuracies": 0.6071428656578064,
499
+ "eval_rewards/chosen": -0.0888671875,
500
+ "eval_rewards/margins": 0.44140625,
501
+ "eval_rewards/rejected": -0.53125,
502
+ "eval_runtime": 18.4562,
503
+ "eval_samples_per_second": 10.836,
504
+ "eval_steps_per_second": 0.379,
505
+ "step": 312
506
+ },
507
+ {
508
+ "epoch": 1.0256410256410255,
509
+ "grad_norm": 31.256761205730616,
510
+ "learning_rate": 3.6579572446555817e-07,
511
+ "logits/chosen": -0.353515625,
512
+ "logits/rejected": -0.349609375,
513
+ "logps/chosen": -286.0,
514
+ "logps/rejected": -268.0,
515
+ "loss": 0.3844,
516
+ "rewards/accuracies": 0.8374999761581421,
517
+ "rewards/chosen": 0.1279296875,
518
+ "rewards/margins": 0.984375,
519
+ "rewards/rejected": -0.859375,
520
+ "step": 320
521
+ },
522
+ {
523
+ "epoch": 1.0576923076923077,
524
+ "grad_norm": 29.12754445870708,
525
+ "learning_rate": 3.598574821852731e-07,
526
+ "logits/chosen": -0.451171875,
527
+ "logits/rejected": -0.47265625,
528
+ "logps/chosen": -380.0,
529
+ "logps/rejected": -334.0,
530
+ "loss": 0.293,
531
+ "rewards/accuracies": 0.8999999761581421,
532
+ "rewards/chosen": 0.3984375,
533
+ "rewards/margins": 1.53125,
534
+ "rewards/rejected": -1.1328125,
535
+ "step": 330
536
+ },
537
+ {
538
+ "epoch": 1.0897435897435896,
539
+ "grad_norm": 23.255524440901077,
540
+ "learning_rate": 3.5391923990498813e-07,
541
+ "logits/chosen": -0.388671875,
542
+ "logits/rejected": -0.296875,
543
+ "logps/chosen": -350.0,
544
+ "logps/rejected": -320.0,
545
+ "loss": 0.3295,
546
+ "rewards/accuracies": 0.8999999761581421,
547
+ "rewards/chosen": 0.369140625,
548
+ "rewards/margins": 1.5625,
549
+ "rewards/rejected": -1.1953125,
550
+ "step": 340
551
+ },
552
+ {
553
+ "epoch": 1.1217948717948718,
554
+ "grad_norm": 32.212012138764614,
555
+ "learning_rate": 3.479809976247031e-07,
556
+ "logits/chosen": -0.228515625,
557
+ "logits/rejected": -0.265625,
558
+ "logps/chosen": -255.0,
559
+ "logps/rejected": -268.0,
560
+ "loss": 0.3649,
561
+ "rewards/accuracies": 0.8125,
562
+ "rewards/chosen": 0.07763671875,
563
+ "rewards/margins": 1.375,
564
+ "rewards/rejected": -1.296875,
565
+ "step": 350
566
+ },
567
+ {
568
+ "epoch": 1.1538461538461537,
569
+ "grad_norm": 21.919087970652495,
570
+ "learning_rate": 3.42042755344418e-07,
571
+ "logits/chosen": -0.412109375,
572
+ "logits/rejected": -0.359375,
573
+ "logps/chosen": -382.0,
574
+ "logps/rejected": -326.0,
575
+ "loss": 0.3203,
576
+ "rewards/accuracies": 0.8999999761581421,
577
+ "rewards/chosen": 0.279296875,
578
+ "rewards/margins": 1.75,
579
+ "rewards/rejected": -1.46875,
580
+ "step": 360
581
+ },
582
+ {
583
+ "epoch": 1.185897435897436,
584
+ "grad_norm": 24.48490434607468,
585
+ "learning_rate": 3.36104513064133e-07,
586
+ "logits/chosen": -0.400390625,
587
+ "logits/rejected": -0.41015625,
588
+ "logps/chosen": -336.0,
589
+ "logps/rejected": -318.0,
590
+ "loss": 0.3216,
591
+ "rewards/accuracies": 0.8999999761581421,
592
+ "rewards/chosen": -0.04052734375,
593
+ "rewards/margins": 1.359375,
594
+ "rewards/rejected": -1.3984375,
595
+ "step": 370
596
+ },
597
+ {
598
+ "epoch": 1.217948717948718,
599
+ "grad_norm": 19.39368867421837,
600
+ "learning_rate": 3.3016627078384796e-07,
601
+ "logits/chosen": -0.33984375,
602
+ "logits/rejected": -0.359375,
603
+ "logps/chosen": -334.0,
604
+ "logps/rejected": -304.0,
605
+ "loss": 0.3092,
606
+ "rewards/accuracies": 0.800000011920929,
607
+ "rewards/chosen": 0.1484375,
608
+ "rewards/margins": 1.484375,
609
+ "rewards/rejected": -1.3359375,
610
+ "step": 380
611
+ },
612
+ {
613
+ "epoch": 1.25,
614
+ "grad_norm": 31.36015324660143,
615
+ "learning_rate": 3.2422802850356297e-07,
616
+ "logits/chosen": -0.369140625,
617
+ "logits/rejected": -0.337890625,
618
+ "logps/chosen": -260.0,
619
+ "logps/rejected": -280.0,
620
+ "loss": 0.3556,
621
+ "rewards/accuracies": 0.875,
622
+ "rewards/chosen": -0.056884765625,
623
+ "rewards/margins": 1.2421875,
624
+ "rewards/rejected": -1.3046875,
625
+ "step": 390
626
+ },
627
+ {
628
+ "epoch": 1.282051282051282,
629
+ "grad_norm": 22.45941815431359,
630
+ "learning_rate": 3.182897862232779e-07,
631
+ "logits/chosen": -0.41796875,
632
+ "logits/rejected": -0.4765625,
633
+ "logps/chosen": -262.0,
634
+ "logps/rejected": -332.0,
635
+ "loss": 0.3101,
636
+ "rewards/accuracies": 0.862500011920929,
637
+ "rewards/chosen": 0.08056640625,
638
+ "rewards/margins": 1.4375,
639
+ "rewards/rejected": -1.3515625,
640
+ "step": 400
641
+ },
642
+ {
643
+ "epoch": 1.314102564102564,
644
+ "grad_norm": 25.29350114221551,
645
+ "learning_rate": 3.1235154394299283e-07,
646
+ "logits/chosen": -0.388671875,
647
+ "logits/rejected": -0.353515625,
648
+ "logps/chosen": -284.0,
649
+ "logps/rejected": -278.0,
650
+ "loss": 0.3135,
651
+ "rewards/accuracies": 0.8500000238418579,
652
+ "rewards/chosen": -0.07373046875,
653
+ "rewards/margins": 1.4375,
654
+ "rewards/rejected": -1.515625,
655
+ "step": 410
656
+ },
657
+ {
658
+ "epoch": 1.3461538461538463,
659
+ "grad_norm": 21.92318193084022,
660
+ "learning_rate": 3.0641330166270784e-07,
661
+ "logits/chosen": -0.384765625,
662
+ "logits/rejected": -0.462890625,
663
+ "logps/chosen": -390.0,
664
+ "logps/rejected": -312.0,
665
+ "loss": 0.2906,
666
+ "rewards/accuracies": 0.9375,
667
+ "rewards/chosen": 0.296875,
668
+ "rewards/margins": 1.8203125,
669
+ "rewards/rejected": -1.5234375,
670
+ "step": 420
671
+ },
672
+ {
673
+ "epoch": 1.3782051282051282,
674
+ "grad_norm": 23.48433321437472,
675
+ "learning_rate": 3.004750593824228e-07,
676
+ "logits/chosen": -0.39453125,
677
+ "logits/rejected": -0.41796875,
678
+ "logps/chosen": -304.0,
679
+ "logps/rejected": -302.0,
680
+ "loss": 0.3054,
681
+ "rewards/accuracies": 0.8125,
682
+ "rewards/chosen": 0.3203125,
683
+ "rewards/margins": 1.4921875,
684
+ "rewards/rejected": -1.171875,
685
+ "step": 430
686
+ },
687
+ {
688
+ "epoch": 1.4102564102564101,
689
+ "grad_norm": 26.26090008450148,
690
+ "learning_rate": 2.9453681710213776e-07,
691
+ "logits/chosen": -0.31640625,
692
+ "logits/rejected": -0.4375,
693
+ "logps/chosen": -342.0,
694
+ "logps/rejected": -300.0,
695
+ "loss": 0.2742,
696
+ "rewards/accuracies": 0.925000011920929,
697
+ "rewards/chosen": 0.298828125,
698
+ "rewards/margins": 1.734375,
699
+ "rewards/rejected": -1.4296875,
700
+ "step": 440
701
+ },
702
+ {
703
+ "epoch": 1.4423076923076923,
704
+ "grad_norm": 23.659041497927436,
705
+ "learning_rate": 2.885985748218527e-07,
706
+ "logits/chosen": -0.37109375,
707
+ "logits/rejected": -0.359375,
708
+ "logps/chosen": -384.0,
709
+ "logps/rejected": -350.0,
710
+ "loss": 0.2861,
711
+ "rewards/accuracies": 0.925000011920929,
712
+ "rewards/chosen": 0.416015625,
713
+ "rewards/margins": 1.8984375,
714
+ "rewards/rejected": -1.484375,
715
+ "step": 450
716
+ },
717
+ {
718
+ "epoch": 1.4743589743589745,
719
+ "grad_norm": 25.88824077819283,
720
+ "learning_rate": 2.8266033254156767e-07,
721
+ "logits/chosen": -0.32421875,
722
+ "logits/rejected": -0.400390625,
723
+ "logps/chosen": -306.0,
724
+ "logps/rejected": -308.0,
725
+ "loss": 0.2735,
726
+ "rewards/accuracies": 0.887499988079071,
727
+ "rewards/chosen": 0.26953125,
728
+ "rewards/margins": 1.6953125,
729
+ "rewards/rejected": -1.421875,
730
+ "step": 460
731
+ },
732
+ {
733
+ "epoch": 1.5064102564102564,
734
+ "grad_norm": 39.69692113209802,
735
+ "learning_rate": 2.7672209026128263e-07,
736
+ "logits/chosen": -0.3828125,
737
+ "logits/rejected": -0.39453125,
738
+ "logps/chosen": -332.0,
739
+ "logps/rejected": -326.0,
740
+ "loss": 0.2837,
741
+ "rewards/accuracies": 0.862500011920929,
742
+ "rewards/chosen": 0.39453125,
743
+ "rewards/margins": 1.75,
744
+ "rewards/rejected": -1.359375,
745
+ "step": 470
746
+ },
747
+ {
748
+ "epoch": 1.5384615384615383,
749
+ "grad_norm": 30.346903581562415,
750
+ "learning_rate": 2.7078384798099764e-07,
751
+ "logits/chosen": -0.34765625,
752
+ "logits/rejected": -0.447265625,
753
+ "logps/chosen": -286.0,
754
+ "logps/rejected": -290.0,
755
+ "loss": 0.2892,
756
+ "rewards/accuracies": 0.887499988079071,
757
+ "rewards/chosen": 0.1240234375,
758
+ "rewards/margins": 1.5,
759
+ "rewards/rejected": -1.3828125,
760
+ "step": 480
761
+ },
762
+ {
763
+ "epoch": 1.5705128205128205,
764
+ "grad_norm": 26.633915775199014,
765
+ "learning_rate": 2.648456057007126e-07,
766
+ "logits/chosen": -0.283203125,
767
+ "logits/rejected": -0.48828125,
768
+ "logps/chosen": -302.0,
769
+ "logps/rejected": -308.0,
770
+ "loss": 0.2848,
771
+ "rewards/accuracies": 0.8999999761581421,
772
+ "rewards/chosen": 0.359375,
773
+ "rewards/margins": 1.9609375,
774
+ "rewards/rejected": -1.6015625,
775
+ "step": 490
776
+ },
777
+ {
778
+ "epoch": 1.6025641025641026,
779
+ "grad_norm": 28.260367944812693,
780
+ "learning_rate": 2.589073634204275e-07,
781
+ "logits/chosen": -0.435546875,
782
+ "logits/rejected": -0.392578125,
783
+ "logps/chosen": -324.0,
784
+ "logps/rejected": -330.0,
785
+ "loss": 0.2791,
786
+ "rewards/accuracies": 0.875,
787
+ "rewards/chosen": 0.328125,
788
+ "rewards/margins": 1.609375,
789
+ "rewards/rejected": -1.28125,
790
+ "step": 500
791
+ },
792
+ {
793
+ "epoch": 1.6346153846153846,
794
+ "grad_norm": 18.511599498231647,
795
+ "learning_rate": 2.529691211401425e-07,
796
+ "logits/chosen": -0.376953125,
797
+ "logits/rejected": -0.33984375,
798
+ "logps/chosen": -294.0,
799
+ "logps/rejected": -272.0,
800
+ "loss": 0.2808,
801
+ "rewards/accuracies": 0.8999999761581421,
802
+ "rewards/chosen": 0.322265625,
803
+ "rewards/margins": 1.4609375,
804
+ "rewards/rejected": -1.140625,
805
+ "step": 510
806
+ },
807
+ {
808
+ "epoch": 1.6666666666666665,
809
+ "grad_norm": 22.844798241441286,
810
+ "learning_rate": 2.4703087885985747e-07,
811
+ "logits/chosen": -0.376953125,
812
+ "logits/rejected": -0.4296875,
813
+ "logps/chosen": -312.0,
814
+ "logps/rejected": -318.0,
815
+ "loss": 0.2799,
816
+ "rewards/accuracies": 0.887499988079071,
817
+ "rewards/chosen": 0.412109375,
818
+ "rewards/margins": 1.5859375,
819
+ "rewards/rejected": -1.171875,
820
+ "step": 520
821
+ },
822
+ {
823
+ "epoch": 1.6987179487179487,
824
+ "grad_norm": 22.977308639981327,
825
+ "learning_rate": 2.410926365795724e-07,
826
+ "logits/chosen": -0.369140625,
827
+ "logits/rejected": -0.2470703125,
828
+ "logps/chosen": -352.0,
829
+ "logps/rejected": -340.0,
830
+ "loss": 0.2788,
831
+ "rewards/accuracies": 0.8999999761581421,
832
+ "rewards/chosen": 0.37890625,
833
+ "rewards/margins": 1.7734375,
834
+ "rewards/rejected": -1.390625,
835
+ "step": 530
836
+ },
837
+ {
838
+ "epoch": 1.7307692307692308,
839
+ "grad_norm": 19.697815015591537,
840
+ "learning_rate": 2.351543942992874e-07,
841
+ "logits/chosen": -0.294921875,
842
+ "logits/rejected": -0.34375,
843
+ "logps/chosen": -234.0,
844
+ "logps/rejected": -282.0,
845
+ "loss": 0.3076,
846
+ "rewards/accuracies": 0.875,
847
+ "rewards/chosen": 0.0830078125,
848
+ "rewards/margins": 1.8671875,
849
+ "rewards/rejected": -1.78125,
850
+ "step": 540
851
+ },
852
+ {
853
+ "epoch": 1.7628205128205128,
854
+ "grad_norm": 16.42605287271066,
855
+ "learning_rate": 2.2921615201900234e-07,
856
+ "logits/chosen": -0.5625,
857
+ "logits/rejected": -0.41796875,
858
+ "logps/chosen": -350.0,
859
+ "logps/rejected": -344.0,
860
+ "loss": 0.2614,
861
+ "rewards/accuracies": 0.9125000238418579,
862
+ "rewards/chosen": 0.21875,
863
+ "rewards/margins": 1.890625,
864
+ "rewards/rejected": -1.671875,
865
+ "step": 550
866
+ },
867
+ {
868
+ "epoch": 1.7948717948717947,
869
+ "grad_norm": 21.70957132124816,
870
+ "learning_rate": 2.2327790973871732e-07,
871
+ "logits/chosen": -0.431640625,
872
+ "logits/rejected": -0.3984375,
873
+ "logps/chosen": -324.0,
874
+ "logps/rejected": -272.0,
875
+ "loss": 0.3165,
876
+ "rewards/accuracies": 0.862500011920929,
877
+ "rewards/chosen": 0.0252685546875,
878
+ "rewards/margins": 1.671875,
879
+ "rewards/rejected": -1.6484375,
880
+ "step": 560
881
+ },
882
+ {
883
+ "epoch": 1.8269230769230769,
884
+ "grad_norm": 22.45992885626303,
885
+ "learning_rate": 2.173396674584323e-07,
886
+ "logits/chosen": -0.29296875,
887
+ "logits/rejected": -0.359375,
888
+ "logps/chosen": -294.0,
889
+ "logps/rejected": -302.0,
890
+ "loss": 0.2871,
891
+ "rewards/accuracies": 0.862500011920929,
892
+ "rewards/chosen": 0.1748046875,
893
+ "rewards/margins": 1.7578125,
894
+ "rewards/rejected": -1.5859375,
895
+ "step": 570
896
+ },
897
+ {
898
+ "epoch": 1.858974358974359,
899
+ "grad_norm": 20.14427708154685,
900
+ "learning_rate": 2.1140142517814726e-07,
901
+ "logits/chosen": -0.359375,
902
+ "logits/rejected": -0.359375,
903
+ "logps/chosen": -324.0,
904
+ "logps/rejected": -308.0,
905
+ "loss": 0.2926,
906
+ "rewards/accuracies": 0.887499988079071,
907
+ "rewards/chosen": 0.443359375,
908
+ "rewards/margins": 1.6015625,
909
+ "rewards/rejected": -1.15625,
910
+ "step": 580
911
+ },
912
+ {
913
+ "epoch": 1.891025641025641,
914
+ "grad_norm": 19.79995273385685,
915
+ "learning_rate": 2.0546318289786222e-07,
916
+ "logits/chosen": -0.322265625,
917
+ "logits/rejected": -0.38671875,
918
+ "logps/chosen": -254.0,
919
+ "logps/rejected": -288.0,
920
+ "loss": 0.2864,
921
+ "rewards/accuracies": 0.8999999761581421,
922
+ "rewards/chosen": 0.419921875,
923
+ "rewards/margins": 1.703125,
924
+ "rewards/rejected": -1.28125,
925
+ "step": 590
926
+ },
927
+ {
928
+ "epoch": 1.9230769230769231,
929
+ "grad_norm": 23.566386363575898,
930
+ "learning_rate": 1.9952494061757718e-07,
931
+ "logits/chosen": -0.306640625,
932
+ "logits/rejected": -0.322265625,
933
+ "logps/chosen": -282.0,
934
+ "logps/rejected": -272.0,
935
+ "loss": 0.2718,
936
+ "rewards/accuracies": 0.887499988079071,
937
+ "rewards/chosen": 0.57421875,
938
+ "rewards/margins": 1.78125,
939
+ "rewards/rejected": -1.2109375,
940
+ "step": 600
941
+ },
942
+ {
943
+ "epoch": 1.9551282051282053,
944
+ "grad_norm": 29.152008466264334,
945
+ "learning_rate": 1.9358669833729216e-07,
946
+ "logits/chosen": -0.59375,
947
+ "logits/rejected": -0.447265625,
948
+ "logps/chosen": -308.0,
949
+ "logps/rejected": -282.0,
950
+ "loss": 0.2682,
951
+ "rewards/accuracies": 0.887499988079071,
952
+ "rewards/chosen": 0.234375,
953
+ "rewards/margins": 1.7421875,
954
+ "rewards/rejected": -1.5078125,
955
+ "step": 610
956
+ },
957
+ {
958
+ "epoch": 1.9871794871794872,
959
+ "grad_norm": 22.58745268335563,
960
+ "learning_rate": 1.876484560570071e-07,
961
+ "logits/chosen": -0.451171875,
962
+ "logits/rejected": -0.470703125,
963
+ "logps/chosen": -360.0,
964
+ "logps/rejected": -300.0,
965
+ "loss": 0.2825,
966
+ "rewards/accuracies": 0.875,
967
+ "rewards/chosen": 0.33984375,
968
+ "rewards/margins": 1.703125,
969
+ "rewards/rejected": -1.3671875,
970
+ "step": 620
971
+ },
972
+ {
973
+ "epoch": 2.0,
974
+ "eval_logits/chosen": -0.3828125,
975
+ "eval_logits/rejected": -0.375,
976
+ "eval_logps/chosen": -298.0,
977
+ "eval_logps/rejected": -336.0,
978
+ "eval_loss": 0.5634375214576721,
979
+ "eval_rewards/accuracies": 0.625,
980
+ "eval_rewards/chosen": -0.345703125,
981
+ "eval_rewards/margins": 0.625,
982
+ "eval_rewards/rejected": -0.96875,
983
+ "eval_runtime": 18.4069,
984
+ "eval_samples_per_second": 10.865,
985
+ "eval_steps_per_second": 0.38,
986
+ "step": 624
987
+ },
988
+ {
989
+ "epoch": 2.019230769230769,
990
+ "grad_norm": 10.223029148094557,
991
+ "learning_rate": 1.8171021377672207e-07,
992
+ "logits/chosen": -0.453125,
993
+ "logits/rejected": -0.37109375,
994
+ "logps/chosen": -272.0,
995
+ "logps/rejected": -296.0,
996
+ "loss": 0.2027,
997
+ "rewards/accuracies": 0.9624999761581421,
998
+ "rewards/chosen": 0.1767578125,
999
+ "rewards/margins": 1.9921875,
1000
+ "rewards/rejected": -1.8125,
1001
+ "step": 630
1002
+ },
1003
+ {
1004
+ "epoch": 2.051282051282051,
1005
+ "grad_norm": 16.341314788264306,
1006
+ "learning_rate": 1.7577197149643706e-07,
1007
+ "logits/chosen": -0.345703125,
1008
+ "logits/rejected": -0.369140625,
1009
+ "logps/chosen": -356.0,
1010
+ "logps/rejected": -330.0,
1011
+ "loss": 0.166,
1012
+ "rewards/accuracies": 0.9125000238418579,
1013
+ "rewards/chosen": 0.486328125,
1014
+ "rewards/margins": 2.4375,
1015
+ "rewards/rejected": -1.9453125,
1016
+ "step": 640
1017
+ },
1018
+ {
1019
+ "epoch": 2.0833333333333335,
1020
+ "grad_norm": 12.61817837923421,
1021
+ "learning_rate": 1.6983372921615202e-07,
1022
+ "logits/chosen": -0.44140625,
1023
+ "logits/rejected": -0.337890625,
1024
+ "logps/chosen": -332.0,
1025
+ "logps/rejected": -304.0,
1026
+ "loss": 0.1739,
1027
+ "rewards/accuracies": 0.925000011920929,
1028
+ "rewards/chosen": 0.28125,
1029
+ "rewards/margins": 2.171875,
1030
+ "rewards/rejected": -1.8984375,
1031
+ "step": 650
1032
+ },
1033
+ {
1034
+ "epoch": 2.1153846153846154,
1035
+ "grad_norm": 15.916788215543724,
1036
+ "learning_rate": 1.6389548693586697e-07,
1037
+ "logits/chosen": -0.41015625,
1038
+ "logits/rejected": -0.328125,
1039
+ "logps/chosen": -330.0,
1040
+ "logps/rejected": -308.0,
1041
+ "loss": 0.1616,
1042
+ "rewards/accuracies": 0.925000011920929,
1043
+ "rewards/chosen": 0.376953125,
1044
+ "rewards/margins": 2.328125,
1045
+ "rewards/rejected": -1.953125,
1046
+ "step": 660
1047
+ },
1048
+ {
1049
+ "epoch": 2.1474358974358974,
1050
+ "grad_norm": 16.057171731434803,
1051
+ "learning_rate": 1.5795724465558193e-07,
1052
+ "logits/chosen": -0.33984375,
1053
+ "logits/rejected": -0.3828125,
1054
+ "logps/chosen": -362.0,
1055
+ "logps/rejected": -324.0,
1056
+ "loss": 0.1857,
1057
+ "rewards/accuracies": 0.9375,
1058
+ "rewards/chosen": 0.1650390625,
1059
+ "rewards/margins": 2.515625,
1060
+ "rewards/rejected": -2.34375,
1061
+ "step": 670
1062
+ },
1063
+ {
1064
+ "epoch": 2.1794871794871793,
1065
+ "grad_norm": 11.186100214252932,
1066
+ "learning_rate": 1.520190023752969e-07,
1067
+ "logits/chosen": -0.51171875,
1068
+ "logits/rejected": -0.51171875,
1069
+ "logps/chosen": -328.0,
1070
+ "logps/rejected": -338.0,
1071
+ "loss": 0.1692,
1072
+ "rewards/accuracies": 0.949999988079071,
1073
+ "rewards/chosen": 0.33203125,
1074
+ "rewards/margins": 2.421875,
1075
+ "rewards/rejected": -2.09375,
1076
+ "step": 680
1077
+ },
1078
+ {
1079
+ "epoch": 2.2115384615384617,
1080
+ "grad_norm": 11.075777990973583,
1081
+ "learning_rate": 1.4608076009501184e-07,
1082
+ "logits/chosen": -0.314453125,
1083
+ "logits/rejected": -0.2265625,
1084
+ "logps/chosen": -268.0,
1085
+ "logps/rejected": -256.0,
1086
+ "loss": 0.1701,
1087
+ "rewards/accuracies": 0.987500011920929,
1088
+ "rewards/chosen": 0.130859375,
1089
+ "rewards/margins": 2.40625,
1090
+ "rewards/rejected": -2.265625,
1091
+ "step": 690
1092
+ },
1093
+ {
1094
+ "epoch": 2.2435897435897436,
1095
+ "grad_norm": 12.634117376853869,
1096
+ "learning_rate": 1.4014251781472683e-07,
1097
+ "logits/chosen": -0.455078125,
1098
+ "logits/rejected": -0.47265625,
1099
+ "logps/chosen": -322.0,
1100
+ "logps/rejected": -276.0,
1101
+ "loss": 0.1625,
1102
+ "rewards/accuracies": 0.987500011920929,
1103
+ "rewards/chosen": -0.04833984375,
1104
+ "rewards/margins": 2.359375,
1105
+ "rewards/rejected": -2.40625,
1106
+ "step": 700
1107
+ },
1108
+ {
1109
+ "epoch": 2.2756410256410255,
1110
+ "grad_norm": 13.097076165721171,
1111
+ "learning_rate": 1.342042755344418e-07,
1112
+ "logits/chosen": -0.36328125,
1113
+ "logits/rejected": -0.337890625,
1114
+ "logps/chosen": -304.0,
1115
+ "logps/rejected": -270.0,
1116
+ "loss": 0.1739,
1117
+ "rewards/accuracies": 0.9624999761581421,
1118
+ "rewards/chosen": 0.0115966796875,
1119
+ "rewards/margins": 2.484375,
1120
+ "rewards/rejected": -2.484375,
1121
+ "step": 710
1122
+ },
1123
+ {
1124
+ "epoch": 2.3076923076923075,
1125
+ "grad_norm": 13.327447391626707,
1126
+ "learning_rate": 1.2826603325415677e-07,
1127
+ "logits/chosen": -0.4296875,
1128
+ "logits/rejected": -0.361328125,
1129
+ "logps/chosen": -316.0,
1130
+ "logps/rejected": -318.0,
1131
+ "loss": 0.1492,
1132
+ "rewards/accuracies": 0.9750000238418579,
1133
+ "rewards/chosen": 0.11865234375,
1134
+ "rewards/margins": 2.640625,
1135
+ "rewards/rejected": -2.515625,
1136
+ "step": 720
1137
+ },
1138
+ {
1139
+ "epoch": 2.33974358974359,
1140
+ "grad_norm": 14.846779798952685,
1141
+ "learning_rate": 1.2232779097387173e-07,
1142
+ "logits/chosen": -0.439453125,
1143
+ "logits/rejected": -0.578125,
1144
+ "logps/chosen": -316.0,
1145
+ "logps/rejected": -316.0,
1146
+ "loss": 0.147,
1147
+ "rewards/accuracies": 0.987500011920929,
1148
+ "rewards/chosen": 0.1640625,
1149
+ "rewards/margins": 2.6875,
1150
+ "rewards/rejected": -2.53125,
1151
+ "step": 730
1152
+ },
1153
+ {
1154
+ "epoch": 2.371794871794872,
1155
+ "grad_norm": 12.996323206405766,
1156
+ "learning_rate": 1.163895486935867e-07,
1157
+ "logits/chosen": -0.4296875,
1158
+ "logits/rejected": -0.458984375,
1159
+ "logps/chosen": -326.0,
1160
+ "logps/rejected": -322.0,
1161
+ "loss": 0.1632,
1162
+ "rewards/accuracies": 0.949999988079071,
1163
+ "rewards/chosen": 0.2099609375,
1164
+ "rewards/margins": 2.515625,
1165
+ "rewards/rejected": -2.296875,
1166
+ "step": 740
1167
+ },
1168
+ {
1169
+ "epoch": 2.4038461538461537,
1170
+ "grad_norm": 12.50397204158465,
1171
+ "learning_rate": 1.1045130641330165e-07,
1172
+ "logits/chosen": -0.287109375,
1173
+ "logits/rejected": -0.32421875,
1174
+ "logps/chosen": -302.0,
1175
+ "logps/rejected": -356.0,
1176
+ "loss": 0.1662,
1177
+ "rewards/accuracies": 0.9750000238418579,
1178
+ "rewards/chosen": 0.484375,
1179
+ "rewards/margins": 2.9375,
1180
+ "rewards/rejected": -2.453125,
1181
+ "step": 750
1182
+ },
1183
+ {
1184
+ "epoch": 2.435897435897436,
1185
+ "grad_norm": 19.850141789886788,
1186
+ "learning_rate": 1.0451306413301662e-07,
1187
+ "logits/chosen": -0.470703125,
1188
+ "logits/rejected": -0.40625,
1189
+ "logps/chosen": -344.0,
1190
+ "logps/rejected": -314.0,
1191
+ "loss": 0.1404,
1192
+ "rewards/accuracies": 0.9750000238418579,
1193
+ "rewards/chosen": 0.412109375,
1194
+ "rewards/margins": 2.6875,
1195
+ "rewards/rejected": -2.28125,
1196
+ "step": 760
1197
+ },
1198
+ {
1199
+ "epoch": 2.467948717948718,
1200
+ "grad_norm": 14.142532630035193,
1201
+ "learning_rate": 9.857482185273158e-08,
1202
+ "logits/chosen": -0.35546875,
1203
+ "logits/rejected": -0.341796875,
1204
+ "logps/chosen": -294.0,
1205
+ "logps/rejected": -286.0,
1206
+ "loss": 0.1377,
1207
+ "rewards/accuracies": 1.0,
1208
+ "rewards/chosen": 0.1552734375,
1209
+ "rewards/margins": 2.5625,
1210
+ "rewards/rejected": -2.40625,
1211
+ "step": 770
1212
+ },
1213
+ {
1214
+ "epoch": 2.5,
1215
+ "grad_norm": 10.362780550901535,
1216
+ "learning_rate": 9.263657957244655e-08,
1217
+ "logits/chosen": -0.4375,
1218
+ "logits/rejected": -0.466796875,
1219
+ "logps/chosen": -338.0,
1220
+ "logps/rejected": -288.0,
1221
+ "loss": 0.1545,
1222
+ "rewards/accuracies": 0.949999988079071,
1223
+ "rewards/chosen": -0.07177734375,
1224
+ "rewards/margins": 2.703125,
1225
+ "rewards/rejected": -2.78125,
1226
+ "step": 780
1227
+ },
1228
+ {
1229
+ "epoch": 2.532051282051282,
1230
+ "grad_norm": 14.983339217810357,
1231
+ "learning_rate": 8.669833729216151e-08,
1232
+ "logits/chosen": -0.416015625,
1233
+ "logits/rejected": -0.40625,
1234
+ "logps/chosen": -328.0,
1235
+ "logps/rejected": -342.0,
1236
+ "loss": 0.1724,
1237
+ "rewards/accuracies": 0.987500011920929,
1238
+ "rewards/chosen": 0.08251953125,
1239
+ "rewards/margins": 2.8125,
1240
+ "rewards/rejected": -2.71875,
1241
+ "step": 790
1242
+ },
1243
+ {
1244
+ "epoch": 2.564102564102564,
1245
+ "grad_norm": 11.403661448100282,
1246
+ "learning_rate": 8.076009501187649e-08,
1247
+ "logits/chosen": -0.361328125,
1248
+ "logits/rejected": -0.2578125,
1249
+ "logps/chosen": -274.0,
1250
+ "logps/rejected": -316.0,
1251
+ "loss": 0.1329,
1252
+ "rewards/accuracies": 0.987500011920929,
1253
+ "rewards/chosen": 0.0247802734375,
1254
+ "rewards/margins": 2.5625,
1255
+ "rewards/rejected": -2.546875,
1256
+ "step": 800
1257
+ },
1258
+ {
1259
+ "epoch": 2.5961538461538463,
1260
+ "grad_norm": 12.098366455488087,
1261
+ "learning_rate": 7.482185273159145e-08,
1262
+ "logits/chosen": -0.474609375,
1263
+ "logits/rejected": -0.4609375,
1264
+ "logps/chosen": -304.0,
1265
+ "logps/rejected": -296.0,
1266
+ "loss": 0.1386,
1267
+ "rewards/accuracies": 1.0,
1268
+ "rewards/chosen": 0.1943359375,
1269
+ "rewards/margins": 2.703125,
1270
+ "rewards/rejected": -2.5,
1271
+ "step": 810
1272
+ },
1273
+ {
1274
+ "epoch": 2.628205128205128,
1275
+ "grad_norm": 19.559170964417323,
1276
+ "learning_rate": 6.88836104513064e-08,
1277
+ "logits/chosen": -0.4296875,
1278
+ "logits/rejected": -0.4453125,
1279
+ "logps/chosen": -314.0,
1280
+ "logps/rejected": -350.0,
1281
+ "loss": 0.1804,
1282
+ "rewards/accuracies": 0.9375,
1283
+ "rewards/chosen": 0.0341796875,
1284
+ "rewards/margins": 2.609375,
1285
+ "rewards/rejected": -2.578125,
1286
+ "step": 820
1287
+ },
1288
+ {
1289
+ "epoch": 2.66025641025641,
1290
+ "grad_norm": 13.553714513714917,
1291
+ "learning_rate": 6.294536817102138e-08,
1292
+ "logits/chosen": -0.53515625,
1293
+ "logits/rejected": -0.52734375,
1294
+ "logps/chosen": -342.0,
1295
+ "logps/rejected": -288.0,
1296
+ "loss": 0.1646,
1297
+ "rewards/accuracies": 0.9375,
1298
+ "rewards/chosen": 0.126953125,
1299
+ "rewards/margins": 2.4375,
1300
+ "rewards/rejected": -2.3125,
1301
+ "step": 830
1302
+ },
1303
+ {
1304
+ "epoch": 2.6923076923076925,
1305
+ "grad_norm": 13.769525791929578,
1306
+ "learning_rate": 5.700712589073634e-08,
1307
+ "logits/chosen": -0.322265625,
1308
+ "logits/rejected": -0.41015625,
1309
+ "logps/chosen": -300.0,
1310
+ "logps/rejected": -294.0,
1311
+ "loss": 0.1677,
1312
+ "rewards/accuracies": 0.9375,
1313
+ "rewards/chosen": 0.0732421875,
1314
+ "rewards/margins": 2.625,
1315
+ "rewards/rejected": -2.546875,
1316
+ "step": 840
1317
+ },
1318
+ {
1319
+ "epoch": 2.7243589743589745,
1320
+ "grad_norm": 19.35910853506724,
1321
+ "learning_rate": 5.10688836104513e-08,
1322
+ "logits/chosen": -0.35546875,
1323
+ "logits/rejected": -0.48046875,
1324
+ "logps/chosen": -326.0,
1325
+ "logps/rejected": -334.0,
1326
+ "loss": 0.1733,
1327
+ "rewards/accuracies": 0.9624999761581421,
1328
+ "rewards/chosen": 0.1259765625,
1329
+ "rewards/margins": 2.53125,
1330
+ "rewards/rejected": -2.40625,
1331
+ "step": 850
1332
+ },
1333
+ {
1334
+ "epoch": 2.7564102564102564,
1335
+ "grad_norm": 13.618581779013738,
1336
+ "learning_rate": 4.5130641330166267e-08,
1337
+ "logits/chosen": -0.3984375,
1338
+ "logits/rejected": -0.404296875,
1339
+ "logps/chosen": -324.0,
1340
+ "logps/rejected": -326.0,
1341
+ "loss": 0.1658,
1342
+ "rewards/accuracies": 0.9624999761581421,
1343
+ "rewards/chosen": -0.09521484375,
1344
+ "rewards/margins": 2.796875,
1345
+ "rewards/rejected": -2.890625,
1346
+ "step": 860
1347
+ },
1348
+ {
1349
+ "epoch": 2.7884615384615383,
1350
+ "grad_norm": 8.922554312926742,
1351
+ "learning_rate": 3.919239904988123e-08,
1352
+ "logits/chosen": -0.41796875,
1353
+ "logits/rejected": -0.453125,
1354
+ "logps/chosen": -264.0,
1355
+ "logps/rejected": -306.0,
1356
+ "loss": 0.185,
1357
+ "rewards/accuracies": 0.949999988079071,
1358
+ "rewards/chosen": 0.052734375,
1359
+ "rewards/margins": 2.328125,
1360
+ "rewards/rejected": -2.265625,
1361
+ "step": 870
1362
+ },
1363
+ {
1364
+ "epoch": 2.8205128205128203,
1365
+ "grad_norm": 15.61605037939187,
1366
+ "learning_rate": 3.32541567695962e-08,
1367
+ "logits/chosen": -0.357421875,
1368
+ "logits/rejected": -0.43359375,
1369
+ "logps/chosen": -316.0,
1370
+ "logps/rejected": -316.0,
1371
+ "loss": 0.1618,
1372
+ "rewards/accuracies": 0.949999988079071,
1373
+ "rewards/chosen": 0.00665283203125,
1374
+ "rewards/margins": 2.515625,
1375
+ "rewards/rejected": -2.5,
1376
+ "step": 880
1377
+ },
1378
+ {
1379
+ "epoch": 2.8525641025641026,
1380
+ "grad_norm": 14.041233392814654,
1381
+ "learning_rate": 2.7315914489311164e-08,
1382
+ "logits/chosen": -0.333984375,
1383
+ "logits/rejected": -0.396484375,
1384
+ "logps/chosen": -346.0,
1385
+ "logps/rejected": -336.0,
1386
+ "loss": 0.1652,
1387
+ "rewards/accuracies": 0.9375,
1388
+ "rewards/chosen": 0.1982421875,
1389
+ "rewards/margins": 2.703125,
1390
+ "rewards/rejected": -2.515625,
1391
+ "step": 890
1392
+ },
1393
+ {
1394
+ "epoch": 2.8846153846153846,
1395
+ "grad_norm": 14.603314234579116,
1396
+ "learning_rate": 2.1377672209026125e-08,
1397
+ "logits/chosen": -0.375,
1398
+ "logits/rejected": -0.341796875,
1399
+ "logps/chosen": -278.0,
1400
+ "logps/rejected": -302.0,
1401
+ "loss": 0.1562,
1402
+ "rewards/accuracies": 0.987500011920929,
1403
+ "rewards/chosen": 0.05419921875,
1404
+ "rewards/margins": 2.46875,
1405
+ "rewards/rejected": -2.421875,
1406
+ "step": 900
1407
+ },
1408
+ {
1409
+ "epoch": 2.9166666666666665,
1410
+ "grad_norm": 13.067382331418449,
1411
+ "learning_rate": 1.5439429928741092e-08,
1412
+ "logits/chosen": -0.359375,
1413
+ "logits/rejected": -0.416015625,
1414
+ "logps/chosen": -298.0,
1415
+ "logps/rejected": -300.0,
1416
+ "loss": 0.1925,
1417
+ "rewards/accuracies": 0.949999988079071,
1418
+ "rewards/chosen": 0.0152587890625,
1419
+ "rewards/margins": 2.5625,
1420
+ "rewards/rejected": -2.546875,
1421
+ "step": 910
1422
+ },
1423
+ {
1424
+ "epoch": 2.948717948717949,
1425
+ "grad_norm": 10.391263156117287,
1426
+ "learning_rate": 9.501187648456057e-09,
1427
+ "logits/chosen": -0.443359375,
1428
+ "logits/rejected": -0.373046875,
1429
+ "logps/chosen": -288.0,
1430
+ "logps/rejected": -320.0,
1431
+ "loss": 0.142,
1432
+ "rewards/accuracies": 0.9375,
1433
+ "rewards/chosen": -0.031982421875,
1434
+ "rewards/margins": 2.453125,
1435
+ "rewards/rejected": -2.484375,
1436
+ "step": 920
1437
+ },
1438
+ {
1439
+ "epoch": 2.980769230769231,
1440
+ "grad_norm": 10.1448021646423,
1441
+ "learning_rate": 3.562945368171021e-09,
1442
+ "logits/chosen": -0.435546875,
1443
+ "logits/rejected": -0.298828125,
1444
+ "logps/chosen": -245.0,
1445
+ "logps/rejected": -278.0,
1446
+ "loss": 0.136,
1447
+ "rewards/accuracies": 0.9624999761581421,
1448
+ "rewards/chosen": -0.0036163330078125,
1449
+ "rewards/margins": 2.59375,
1450
+ "rewards/rejected": -2.59375,
1451
+ "step": 930
1452
+ },
1453
+ {
1454
+ "epoch": 3.0,
1455
+ "eval_logits/chosen": -0.40625,
1456
+ "eval_logits/rejected": -0.39453125,
1457
+ "eval_logps/chosen": -302.0,
1458
+ "eval_logps/rejected": -344.0,
1459
+ "eval_loss": 0.5834765434265137,
1460
+ "eval_rewards/accuracies": 0.6607142686843872,
1461
+ "eval_rewards/chosen": -0.8671875,
1462
+ "eval_rewards/margins": 0.83203125,
1463
+ "eval_rewards/rejected": -1.703125,
1464
+ "eval_runtime": 18.4869,
1465
+ "eval_samples_per_second": 10.818,
1466
+ "eval_steps_per_second": 0.379,
1467
+ "step": 936
1468
+ },
1469
+ {
1470
+ "epoch": 3.0,
1471
+ "step": 936,
1472
+ "total_flos": 0.0,
1473
+ "train_loss": 0.350877452100444,
1474
+ "train_runtime": 7216.813,
1475
+ "train_samples_per_second": 4.15,
1476
+ "train_steps_per_second": 0.13
1477
+ }
1478
+ ],
1479
+ "logging_steps": 10,
1480
+ "max_steps": 936,
1481
+ "num_input_tokens_seen": 0,
1482
+ "num_train_epochs": 3,
1483
+ "save_steps": 500,
1484
+ "stateful_callbacks": {
1485
+ "TrainerControl": {
1486
+ "args": {
1487
+ "should_epoch_stop": false,
1488
+ "should_evaluate": false,
1489
+ "should_log": false,
1490
+ "should_save": true,
1491
+ "should_training_stop": true
1492
+ },
1493
+ "attributes": {}
1494
+ }
1495
+ },
1496
+ "total_flos": 0.0,
1497
+ "train_batch_size": 8,
1498
+ "trial_name": null,
1499
+ "trial_params": null
1500
+ }