Hugging Face's logo

AmberYifan
/
Llama-2-7b-sft-dpo-4k

Text Generation
Transformers
Safetensors
llama
Generated from Trainer
trl
dpo
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
AmberYifan commited on
Commit
3b47868
·
verified ·
1 Parent(s): 5764ada

Model save

Browse files
Files changed (5) hide show
  1. README.md +68 -0
  2. all_results.json +9 -0
  3. generation_config.json +10 -0
  4. train_results.json +9 -0
  5. trainer_state.json +690 -0
README.md ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: AmberYifan/llama2-7b-sft-ultrachat-safeRLHF
3
+ library_name: transformers
4
+ model_name: Llama-2-7b-sft-dpo-4k
5
+ tags:
6
+ - generated_from_trainer
7
+ - trl
8
+ - dpo
9
+ licence: license
10
+ ---
11
+
12
+ # Model Card for Llama-2-7b-sft-dpo-4k
13
+
14
+ This model is a fine-tuned version of [AmberYifan/llama2-7b-sft-ultrachat-safeRLHF](https://huggingface.co/AmberYifan/llama2-7b-sft-ultrachat-safeRLHF).
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/Llama-2-7b-sft-dpo-4k", 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/joe2o7ob)
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: 2.5.1+cu118
39
+ - Datasets: 3.2.0
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": 0.5475334398674242,
5
+ "train_runtime": 2411.1864,
6
+ "train_samples": 4199,
7
+ "train_samples_per_second": 5.224,
8
+ "train_steps_per_second": 0.164
9
+ }
generation_config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 1,
3
+ "do_sample": true,
4
+ "eos_token_id": 2,
5
+ "max_length": 4096,
6
+ "pad_token_id": 0,
7
+ "temperature": 0.6,
8
+ "top_p": 0.9,
9
+ "transformers_version": "4.46.3"
10
+ }
train_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 3.0,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.5475334398674242,
5
+ "train_runtime": 2411.1864,
6
+ "train_samples": 4199,
7
+ "train_samples_per_second": 5.224,
8
+ "train_steps_per_second": 0.164
9
+ }
trainer_state.json ADDED
@@ -0,0 +1,690 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 3.0,
5
+ "eval_steps": 500,
6
+ "global_step": 396,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.007575757575757576,
13
+ "grad_norm": 15.596037687491558,
14
+ "learning_rate": 1.25e-08,
15
+ "logits/chosen": -1.5390625,
16
+ "logits/rejected": -1.578125,
17
+ "logps/chosen": -166.0,
18
+ "logps/rejected": -140.0,
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": 0.07575757575757576,
28
+ "grad_norm": 15.63845059404282,
29
+ "learning_rate": 1.25e-07,
30
+ "logits/chosen": -1.59375,
31
+ "logits/rejected": -1.546875,
32
+ "logps/chosen": -146.0,
33
+ "logps/rejected": -132.0,
34
+ "loss": 0.6918,
35
+ "rewards/accuracies": 0.1805555522441864,
36
+ "rewards/chosen": -0.006622314453125,
37
+ "rewards/margins": -0.0020904541015625,
38
+ "rewards/rejected": -0.0045166015625,
39
+ "step": 10
40
+ },
41
+ {
42
+ "epoch": 0.15151515151515152,
43
+ "grad_norm": 15.541972149316718,
44
+ "learning_rate": 2.5e-07,
45
+ "logits/chosen": -1.5703125,
46
+ "logits/rejected": -1.5703125,
47
+ "logps/chosen": -135.0,
48
+ "logps/rejected": -139.0,
49
+ "loss": 0.6929,
50
+ "rewards/accuracies": 0.1875,
51
+ "rewards/chosen": -0.0137939453125,
52
+ "rewards/margins": -0.015625,
53
+ "rewards/rejected": 0.00188446044921875,
54
+ "step": 20
55
+ },
56
+ {
57
+ "epoch": 0.22727272727272727,
58
+ "grad_norm": 16.167428018856896,
59
+ "learning_rate": 3.75e-07,
60
+ "logits/chosen": -1.578125,
61
+ "logits/rejected": -1.5625,
62
+ "logps/chosen": -127.5,
63
+ "logps/rejected": -119.0,
64
+ "loss": 0.6891,
65
+ "rewards/accuracies": 0.4124999940395355,
66
+ "rewards/chosen": -0.006256103515625,
67
+ "rewards/margins": 0.010986328125,
68
+ "rewards/rejected": -0.0172119140625,
69
+ "step": 30
70
+ },
71
+ {
72
+ "epoch": 0.30303030303030304,
73
+ "grad_norm": 15.146059864609068,
74
+ "learning_rate": 5e-07,
75
+ "logits/chosen": -1.5859375,
76
+ "logits/rejected": -1.5,
77
+ "logps/chosen": -137.0,
78
+ "logps/rejected": -129.0,
79
+ "loss": 0.6851,
80
+ "rewards/accuracies": 0.2874999940395355,
81
+ "rewards/chosen": -0.03564453125,
82
+ "rewards/margins": -0.00186920166015625,
83
+ "rewards/rejected": -0.033935546875,
84
+ "step": 40
85
+ },
86
+ {
87
+ "epoch": 0.3787878787878788,
88
+ "grad_norm": 16.049039283963904,
89
+ "learning_rate": 4.859550561797752e-07,
90
+ "logits/chosen": -1.6328125,
91
+ "logits/rejected": -1.5546875,
92
+ "logps/chosen": -166.0,
93
+ "logps/rejected": -138.0,
94
+ "loss": 0.6814,
95
+ "rewards/accuracies": 0.5375000238418579,
96
+ "rewards/chosen": -0.04931640625,
97
+ "rewards/margins": 0.04736328125,
98
+ "rewards/rejected": -0.0966796875,
99
+ "step": 50
100
+ },
101
+ {
102
+ "epoch": 0.45454545454545453,
103
+ "grad_norm": 16.507555023132472,
104
+ "learning_rate": 4.7191011235955054e-07,
105
+ "logits/chosen": -1.5625,
106
+ "logits/rejected": -1.5,
107
+ "logps/chosen": -146.0,
108
+ "logps/rejected": -131.0,
109
+ "loss": 0.671,
110
+ "rewards/accuracies": 0.5625,
111
+ "rewards/chosen": -0.0888671875,
112
+ "rewards/margins": 0.047119140625,
113
+ "rewards/rejected": -0.1357421875,
114
+ "step": 60
115
+ },
116
+ {
117
+ "epoch": 0.5303030303030303,
118
+ "grad_norm": 15.118251247338051,
119
+ "learning_rate": 4.5786516853932584e-07,
120
+ "logits/chosen": -1.5390625,
121
+ "logits/rejected": -1.5390625,
122
+ "logps/chosen": -146.0,
123
+ "logps/rejected": -129.0,
124
+ "loss": 0.6756,
125
+ "rewards/accuracies": 0.4749999940395355,
126
+ "rewards/chosen": -0.1279296875,
127
+ "rewards/margins": 0.049072265625,
128
+ "rewards/rejected": -0.1767578125,
129
+ "step": 70
130
+ },
131
+ {
132
+ "epoch": 0.6060606060606061,
133
+ "grad_norm": 14.88927933446616,
134
+ "learning_rate": 4.438202247191011e-07,
135
+ "logits/chosen": -1.546875,
136
+ "logits/rejected": -1.5390625,
137
+ "logps/chosen": -135.0,
138
+ "logps/rejected": -129.0,
139
+ "loss": 0.6703,
140
+ "rewards/accuracies": 0.550000011920929,
141
+ "rewards/chosen": -0.1318359375,
142
+ "rewards/margins": 0.04248046875,
143
+ "rewards/rejected": -0.173828125,
144
+ "step": 80
145
+ },
146
+ {
147
+ "epoch": 0.6818181818181818,
148
+ "grad_norm": 14.799638412304567,
149
+ "learning_rate": 4.297752808988764e-07,
150
+ "logits/chosen": -1.53125,
151
+ "logits/rejected": -1.5703125,
152
+ "logps/chosen": -151.0,
153
+ "logps/rejected": -142.0,
154
+ "loss": 0.6579,
155
+ "rewards/accuracies": 0.6000000238418579,
156
+ "rewards/chosen": -0.138671875,
157
+ "rewards/margins": 0.06689453125,
158
+ "rewards/rejected": -0.2060546875,
159
+ "step": 90
160
+ },
161
+ {
162
+ "epoch": 0.7575757575757576,
163
+ "grad_norm": 14.614214826436204,
164
+ "learning_rate": 4.157303370786517e-07,
165
+ "logits/chosen": -1.546875,
166
+ "logits/rejected": -1.5625,
167
+ "logps/chosen": -146.0,
168
+ "logps/rejected": -142.0,
169
+ "loss": 0.6571,
170
+ "rewards/accuracies": 0.612500011920929,
171
+ "rewards/chosen": -0.181640625,
172
+ "rewards/margins": 0.0791015625,
173
+ "rewards/rejected": -0.259765625,
174
+ "step": 100
175
+ },
176
+ {
177
+ "epoch": 0.8333333333333334,
178
+ "grad_norm": 14.469109228087035,
179
+ "learning_rate": 4.0168539325842696e-07,
180
+ "logits/chosen": -1.5390625,
181
+ "logits/rejected": -1.5625,
182
+ "logps/chosen": -143.0,
183
+ "logps/rejected": -143.0,
184
+ "loss": 0.6619,
185
+ "rewards/accuracies": 0.5375000238418579,
186
+ "rewards/chosen": -0.2353515625,
187
+ "rewards/margins": 0.05029296875,
188
+ "rewards/rejected": -0.28515625,
189
+ "step": 110
190
+ },
191
+ {
192
+ "epoch": 0.9090909090909091,
193
+ "grad_norm": 14.828181252975622,
194
+ "learning_rate": 3.876404494382022e-07,
195
+ "logits/chosen": -1.5,
196
+ "logits/rejected": -1.5078125,
197
+ "logps/chosen": -139.0,
198
+ "logps/rejected": -130.0,
199
+ "loss": 0.6482,
200
+ "rewards/accuracies": 0.550000011920929,
201
+ "rewards/chosen": -0.302734375,
202
+ "rewards/margins": 0.0732421875,
203
+ "rewards/rejected": -0.376953125,
204
+ "step": 120
205
+ },
206
+ {
207
+ "epoch": 0.9848484848484849,
208
+ "grad_norm": 14.708424671919467,
209
+ "learning_rate": 3.735955056179775e-07,
210
+ "logits/chosen": -1.484375,
211
+ "logits/rejected": -1.46875,
212
+ "logps/chosen": -130.0,
213
+ "logps/rejected": -119.0,
214
+ "loss": 0.6459,
215
+ "rewards/accuracies": 0.6000000238418579,
216
+ "rewards/chosen": -0.248046875,
217
+ "rewards/margins": 0.1552734375,
218
+ "rewards/rejected": -0.40234375,
219
+ "step": 130
220
+ },
221
+ {
222
+ "epoch": 1.0,
223
+ "eval_logits/chosen": -1.484375,
224
+ "eval_logits/rejected": -1.4765625,
225
+ "eval_logps/chosen": -141.0,
226
+ "eval_logps/rejected": -129.0,
227
+ "eval_loss": 0.6321874856948853,
228
+ "eval_rewards/accuracies": 0.6428571343421936,
229
+ "eval_rewards/chosen": -0.265625,
230
+ "eval_rewards/margins": 0.1767578125,
231
+ "eval_rewards/rejected": -0.443359375,
232
+ "eval_runtime": 12.2418,
233
+ "eval_samples_per_second": 16.337,
234
+ "eval_steps_per_second": 0.572,
235
+ "step": 132
236
+ },
237
+ {
238
+ "epoch": 1.0606060606060606,
239
+ "grad_norm": 12.113429780384491,
240
+ "learning_rate": 3.5955056179775277e-07,
241
+ "logits/chosen": -1.515625,
242
+ "logits/rejected": -1.453125,
243
+ "logps/chosen": -147.0,
244
+ "logps/rejected": -141.0,
245
+ "loss": 0.5898,
246
+ "rewards/accuracies": 0.737500011920929,
247
+ "rewards/chosen": -0.2314453125,
248
+ "rewards/margins": 0.236328125,
249
+ "rewards/rejected": -0.466796875,
250
+ "step": 140
251
+ },
252
+ {
253
+ "epoch": 1.1363636363636362,
254
+ "grad_norm": 11.88094972407373,
255
+ "learning_rate": 3.4550561797752807e-07,
256
+ "logits/chosen": -1.5703125,
257
+ "logits/rejected": -1.515625,
258
+ "logps/chosen": -149.0,
259
+ "logps/rejected": -150.0,
260
+ "loss": 0.5547,
261
+ "rewards/accuracies": 0.7749999761581421,
262
+ "rewards/chosen": -0.1630859375,
263
+ "rewards/margins": 0.333984375,
264
+ "rewards/rejected": -0.49609375,
265
+ "step": 150
266
+ },
267
+ {
268
+ "epoch": 1.2121212121212122,
269
+ "grad_norm": 12.356915867599595,
270
+ "learning_rate": 3.314606741573033e-07,
271
+ "logits/chosen": -1.5703125,
272
+ "logits/rejected": -1.5390625,
273
+ "logps/chosen": -142.0,
274
+ "logps/rejected": -130.0,
275
+ "loss": 0.5473,
276
+ "rewards/accuracies": 0.762499988079071,
277
+ "rewards/chosen": -0.2412109375,
278
+ "rewards/margins": 0.369140625,
279
+ "rewards/rejected": -0.609375,
280
+ "step": 160
281
+ },
282
+ {
283
+ "epoch": 1.2878787878787878,
284
+ "grad_norm": 12.507248800803001,
285
+ "learning_rate": 3.1741573033707863e-07,
286
+ "logits/chosen": -1.5078125,
287
+ "logits/rejected": -1.484375,
288
+ "logps/chosen": -143.0,
289
+ "logps/rejected": -131.0,
290
+ "loss": 0.5301,
291
+ "rewards/accuracies": 0.824999988079071,
292
+ "rewards/chosen": -0.33984375,
293
+ "rewards/margins": 0.435546875,
294
+ "rewards/rejected": -0.7734375,
295
+ "step": 170
296
+ },
297
+ {
298
+ "epoch": 1.3636363636363638,
299
+ "grad_norm": 12.914463649706578,
300
+ "learning_rate": 3.0337078651685393e-07,
301
+ "logits/chosen": -1.5234375,
302
+ "logits/rejected": -1.4921875,
303
+ "logps/chosen": -152.0,
304
+ "logps/rejected": -144.0,
305
+ "loss": 0.5235,
306
+ "rewards/accuracies": 0.8125,
307
+ "rewards/chosen": -0.2470703125,
308
+ "rewards/margins": 0.44140625,
309
+ "rewards/rejected": -0.6875,
310
+ "step": 180
311
+ },
312
+ {
313
+ "epoch": 1.4393939393939394,
314
+ "grad_norm": 14.04555120726488,
315
+ "learning_rate": 2.893258426966292e-07,
316
+ "logits/chosen": -1.515625,
317
+ "logits/rejected": -1.5,
318
+ "logps/chosen": -144.0,
319
+ "logps/rejected": -137.0,
320
+ "loss": 0.5319,
321
+ "rewards/accuracies": 0.7749999761581421,
322
+ "rewards/chosen": -0.515625,
323
+ "rewards/margins": 0.423828125,
324
+ "rewards/rejected": -0.9375,
325
+ "step": 190
326
+ },
327
+ {
328
+ "epoch": 1.5151515151515151,
329
+ "grad_norm": 12.20920671777009,
330
+ "learning_rate": 2.752808988764045e-07,
331
+ "logits/chosen": -1.5234375,
332
+ "logits/rejected": -1.5390625,
333
+ "logps/chosen": -152.0,
334
+ "logps/rejected": -155.0,
335
+ "loss": 0.5072,
336
+ "rewards/accuracies": 0.8500000238418579,
337
+ "rewards/chosen": -0.40625,
338
+ "rewards/margins": 0.5234375,
339
+ "rewards/rejected": -0.9296875,
340
+ "step": 200
341
+ },
342
+ {
343
+ "epoch": 1.5909090909090908,
344
+ "grad_norm": 12.279499640810553,
345
+ "learning_rate": 2.612359550561798e-07,
346
+ "logits/chosen": -1.4765625,
347
+ "logits/rejected": -1.453125,
348
+ "logps/chosen": -138.0,
349
+ "logps/rejected": -137.0,
350
+ "loss": 0.5292,
351
+ "rewards/accuracies": 0.762499988079071,
352
+ "rewards/chosen": -0.447265625,
353
+ "rewards/margins": 0.478515625,
354
+ "rewards/rejected": -0.92578125,
355
+ "step": 210
356
+ },
357
+ {
358
+ "epoch": 1.6666666666666665,
359
+ "grad_norm": 14.135915136852619,
360
+ "learning_rate": 2.4719101123595505e-07,
361
+ "logits/chosen": -1.5,
362
+ "logits/rejected": -1.4609375,
363
+ "logps/chosen": -142.0,
364
+ "logps/rejected": -133.0,
365
+ "loss": 0.5232,
366
+ "rewards/accuracies": 0.75,
367
+ "rewards/chosen": -0.5625,
368
+ "rewards/margins": 0.37109375,
369
+ "rewards/rejected": -0.93359375,
370
+ "step": 220
371
+ },
372
+ {
373
+ "epoch": 1.7424242424242424,
374
+ "grad_norm": 13.704058332324305,
375
+ "learning_rate": 2.331460674157303e-07,
376
+ "logits/chosen": -1.5546875,
377
+ "logits/rejected": -1.546875,
378
+ "logps/chosen": -147.0,
379
+ "logps/rejected": -145.0,
380
+ "loss": 0.5218,
381
+ "rewards/accuracies": 0.8125,
382
+ "rewards/chosen": -0.4375,
383
+ "rewards/margins": 0.5390625,
384
+ "rewards/rejected": -0.9765625,
385
+ "step": 230
386
+ },
387
+ {
388
+ "epoch": 1.8181818181818183,
389
+ "grad_norm": 12.255502416027877,
390
+ "learning_rate": 2.191011235955056e-07,
391
+ "logits/chosen": -1.5859375,
392
+ "logits/rejected": -1.5546875,
393
+ "logps/chosen": -163.0,
394
+ "logps/rejected": -152.0,
395
+ "loss": 0.4985,
396
+ "rewards/accuracies": 0.875,
397
+ "rewards/chosen": -0.349609375,
398
+ "rewards/margins": 0.625,
399
+ "rewards/rejected": -0.97265625,
400
+ "step": 240
401
+ },
402
+ {
403
+ "epoch": 1.893939393939394,
404
+ "grad_norm": 13.196400764481162,
405
+ "learning_rate": 2.0505617977528089e-07,
406
+ "logits/chosen": -1.5546875,
407
+ "logits/rejected": -1.453125,
408
+ "logps/chosen": -162.0,
409
+ "logps/rejected": -127.0,
410
+ "loss": 0.4993,
411
+ "rewards/accuracies": 0.824999988079071,
412
+ "rewards/chosen": -0.53125,
413
+ "rewards/margins": 0.65234375,
414
+ "rewards/rejected": -1.1796875,
415
+ "step": 250
416
+ },
417
+ {
418
+ "epoch": 1.9696969696969697,
419
+ "grad_norm": 15.136911810706158,
420
+ "learning_rate": 1.9101123595505617e-07,
421
+ "logits/chosen": -1.5390625,
422
+ "logits/rejected": -1.5078125,
423
+ "logps/chosen": -148.0,
424
+ "logps/rejected": -145.0,
425
+ "loss": 0.5393,
426
+ "rewards/accuracies": 0.75,
427
+ "rewards/chosen": -0.6328125,
428
+ "rewards/margins": 0.46484375,
429
+ "rewards/rejected": -1.09375,
430
+ "step": 260
431
+ },
432
+ {
433
+ "epoch": 2.0,
434
+ "eval_logits/chosen": -1.484375,
435
+ "eval_logits/rejected": -1.46875,
436
+ "eval_logps/chosen": -144.0,
437
+ "eval_logps/rejected": -133.0,
438
+ "eval_loss": 0.6057812571525574,
439
+ "eval_rewards/accuracies": 0.6607142686843872,
440
+ "eval_rewards/chosen": -0.5625,
441
+ "eval_rewards/margins": 0.3515625,
442
+ "eval_rewards/rejected": -0.9140625,
443
+ "eval_runtime": 14.4828,
444
+ "eval_samples_per_second": 13.81,
445
+ "eval_steps_per_second": 0.483,
446
+ "step": 264
447
+ },
448
+ {
449
+ "epoch": 2.0454545454545454,
450
+ "grad_norm": 11.094892248853895,
451
+ "learning_rate": 1.7696629213483144e-07,
452
+ "logits/chosen": -1.5,
453
+ "logits/rejected": -1.4453125,
454
+ "logps/chosen": -148.0,
455
+ "logps/rejected": -142.0,
456
+ "loss": 0.4837,
457
+ "rewards/accuracies": 0.8500000238418579,
458
+ "rewards/chosen": -0.58984375,
459
+ "rewards/margins": 0.62890625,
460
+ "rewards/rejected": -1.21875,
461
+ "step": 270
462
+ },
463
+ {
464
+ "epoch": 2.121212121212121,
465
+ "grad_norm": 10.64557888461514,
466
+ "learning_rate": 1.6292134831460675e-07,
467
+ "logits/chosen": -1.5546875,
468
+ "logits/rejected": -1.4765625,
469
+ "logps/chosen": -150.0,
470
+ "logps/rejected": -134.0,
471
+ "loss": 0.4358,
472
+ "rewards/accuracies": 0.8374999761581421,
473
+ "rewards/chosen": -0.5078125,
474
+ "rewards/margins": 0.734375,
475
+ "rewards/rejected": -1.2421875,
476
+ "step": 280
477
+ },
478
+ {
479
+ "epoch": 2.196969696969697,
480
+ "grad_norm": 11.580320333983588,
481
+ "learning_rate": 1.4887640449438203e-07,
482
+ "logits/chosen": -1.5625,
483
+ "logits/rejected": -1.5390625,
484
+ "logps/chosen": -165.0,
485
+ "logps/rejected": -160.0,
486
+ "loss": 0.4185,
487
+ "rewards/accuracies": 0.8125,
488
+ "rewards/chosen": -0.546875,
489
+ "rewards/margins": 0.765625,
490
+ "rewards/rejected": -1.3125,
491
+ "step": 290
492
+ },
493
+ {
494
+ "epoch": 2.2727272727272725,
495
+ "grad_norm": 15.056725052177573,
496
+ "learning_rate": 1.3483146067415728e-07,
497
+ "logits/chosen": -1.546875,
498
+ "logits/rejected": -1.5078125,
499
+ "logps/chosen": -158.0,
500
+ "logps/rejected": -149.0,
501
+ "loss": 0.458,
502
+ "rewards/accuracies": 0.8374999761581421,
503
+ "rewards/chosen": -0.435546875,
504
+ "rewards/margins": 0.75,
505
+ "rewards/rejected": -1.1875,
506
+ "step": 300
507
+ },
508
+ {
509
+ "epoch": 2.3484848484848486,
510
+ "grad_norm": 12.006511122339738,
511
+ "learning_rate": 1.2078651685393259e-07,
512
+ "logits/chosen": -1.515625,
513
+ "logits/rejected": -1.453125,
514
+ "logps/chosen": -154.0,
515
+ "logps/rejected": -144.0,
516
+ "loss": 0.4472,
517
+ "rewards/accuracies": 0.8374999761581421,
518
+ "rewards/chosen": -0.5546875,
519
+ "rewards/margins": 0.65625,
520
+ "rewards/rejected": -1.2109375,
521
+ "step": 310
522
+ },
523
+ {
524
+ "epoch": 2.4242424242424243,
525
+ "grad_norm": 13.746362722926833,
526
+ "learning_rate": 1.0674157303370785e-07,
527
+ "logits/chosen": -1.5390625,
528
+ "logits/rejected": -1.5234375,
529
+ "logps/chosen": -140.0,
530
+ "logps/rejected": -145.0,
531
+ "loss": 0.4728,
532
+ "rewards/accuracies": 0.8500000238418579,
533
+ "rewards/chosen": -0.58203125,
534
+ "rewards/margins": 0.6015625,
535
+ "rewards/rejected": -1.1796875,
536
+ "step": 320
537
+ },
538
+ {
539
+ "epoch": 2.5,
540
+ "grad_norm": 10.459882513773048,
541
+ "learning_rate": 9.269662921348314e-08,
542
+ "logits/chosen": -1.515625,
543
+ "logits/rejected": -1.46875,
544
+ "logps/chosen": -153.0,
545
+ "logps/rejected": -149.0,
546
+ "loss": 0.4405,
547
+ "rewards/accuracies": 0.862500011920929,
548
+ "rewards/chosen": -0.5703125,
549
+ "rewards/margins": 0.79296875,
550
+ "rewards/rejected": -1.3671875,
551
+ "step": 330
552
+ },
553
+ {
554
+ "epoch": 2.5757575757575757,
555
+ "grad_norm": 13.750139706707387,
556
+ "learning_rate": 7.865168539325842e-08,
557
+ "logits/chosen": -1.46875,
558
+ "logits/rejected": -1.5,
559
+ "logps/chosen": -147.0,
560
+ "logps/rejected": -143.0,
561
+ "loss": 0.4422,
562
+ "rewards/accuracies": 0.824999988079071,
563
+ "rewards/chosen": -0.515625,
564
+ "rewards/margins": 0.69140625,
565
+ "rewards/rejected": -1.2109375,
566
+ "step": 340
567
+ },
568
+ {
569
+ "epoch": 2.6515151515151514,
570
+ "grad_norm": 10.53559766204164,
571
+ "learning_rate": 6.460674157303371e-08,
572
+ "logits/chosen": -1.5390625,
573
+ "logits/rejected": -1.4921875,
574
+ "logps/chosen": -155.0,
575
+ "logps/rejected": -142.0,
576
+ "loss": 0.4407,
577
+ "rewards/accuracies": 0.8374999761581421,
578
+ "rewards/chosen": -0.54296875,
579
+ "rewards/margins": 0.7109375,
580
+ "rewards/rejected": -1.25,
581
+ "step": 350
582
+ },
583
+ {
584
+ "epoch": 2.7272727272727275,
585
+ "grad_norm": 10.932011907954834,
586
+ "learning_rate": 5.056179775280899e-08,
587
+ "logits/chosen": -1.5078125,
588
+ "logits/rejected": -1.453125,
589
+ "logps/chosen": -142.0,
590
+ "logps/rejected": -139.0,
591
+ "loss": 0.4395,
592
+ "rewards/accuracies": 0.887499988079071,
593
+ "rewards/chosen": -0.57421875,
594
+ "rewards/margins": 0.87890625,
595
+ "rewards/rejected": -1.453125,
596
+ "step": 360
597
+ },
598
+ {
599
+ "epoch": 2.8030303030303028,
600
+ "grad_norm": 10.478220370160118,
601
+ "learning_rate": 3.6516853932584266e-08,
602
+ "logits/chosen": -1.53125,
603
+ "logits/rejected": -1.5,
604
+ "logps/chosen": -144.0,
605
+ "logps/rejected": -142.0,
606
+ "loss": 0.447,
607
+ "rewards/accuracies": 0.8374999761581421,
608
+ "rewards/chosen": -0.53515625,
609
+ "rewards/margins": 0.71484375,
610
+ "rewards/rejected": -1.25,
611
+ "step": 370
612
+ },
613
+ {
614
+ "epoch": 2.878787878787879,
615
+ "grad_norm": 11.077008754645805,
616
+ "learning_rate": 2.2471910112359548e-08,
617
+ "logits/chosen": -1.546875,
618
+ "logits/rejected": -1.484375,
619
+ "logps/chosen": -150.0,
620
+ "logps/rejected": -140.0,
621
+ "loss": 0.4625,
622
+ "rewards/accuracies": 0.7749999761581421,
623
+ "rewards/chosen": -0.5234375,
624
+ "rewards/margins": 0.76953125,
625
+ "rewards/rejected": -1.2890625,
626
+ "step": 380
627
+ },
628
+ {
629
+ "epoch": 2.9545454545454546,
630
+ "grad_norm": 10.906822523238372,
631
+ "learning_rate": 8.42696629213483e-09,
632
+ "logits/chosen": -1.5390625,
633
+ "logits/rejected": -1.5390625,
634
+ "logps/chosen": -156.0,
635
+ "logps/rejected": -146.0,
636
+ "loss": 0.4113,
637
+ "rewards/accuracies": 0.875,
638
+ "rewards/chosen": -0.439453125,
639
+ "rewards/margins": 0.82421875,
640
+ "rewards/rejected": -1.265625,
641
+ "step": 390
642
+ },
643
+ {
644
+ "epoch": 3.0,
645
+ "eval_logits/chosen": -1.4765625,
646
+ "eval_logits/rejected": -1.46875,
647
+ "eval_logps/chosen": -144.0,
648
+ "eval_logps/rejected": -133.0,
649
+ "eval_loss": 0.6071093678474426,
650
+ "eval_rewards/accuracies": 0.6607142686843872,
651
+ "eval_rewards/chosen": -0.490234375,
652
+ "eval_rewards/margins": 0.396484375,
653
+ "eval_rewards/rejected": -0.88671875,
654
+ "eval_runtime": 14.7701,
655
+ "eval_samples_per_second": 13.541,
656
+ "eval_steps_per_second": 0.474,
657
+ "step": 396
658
+ },
659
+ {
660
+ "epoch": 3.0,
661
+ "step": 396,
662
+ "total_flos": 0.0,
663
+ "train_loss": 0.5475334398674242,
664
+ "train_runtime": 2411.1864,
665
+ "train_samples_per_second": 5.224,
666
+ "train_steps_per_second": 0.164
667
+ }
668
+ ],
669
+ "logging_steps": 10,
670
+ "max_steps": 396,
671
+ "num_input_tokens_seen": 0,
672
+ "num_train_epochs": 3,
673
+ "save_steps": 500,
674
+ "stateful_callbacks": {
675
+ "TrainerControl": {
676
+ "args": {
677
+ "should_epoch_stop": false,
678
+ "should_evaluate": false,
679
+ "should_log": false,
680
+ "should_save": true,
681
+ "should_training_stop": true
682
+ },
683
+ "attributes": {}
684
+ }
685
+ },
686
+ "total_flos": 0.0,
687
+ "train_batch_size": 8,
688
+ "trial_name": null,
689
+ "trial_params": null
690
+ }