Hugging Face's logo

YYYYYYibo
/
vanilla_dpo_iter_1

PEFT
Safetensors
mistral
alignment-handbook
Generated from Trainer
trl
dpo
Model card Files Community
YYYYYYibo commited on
Commit
d312438
·
verified ·
1 Parent(s): 114a2e0

Model save

Browse files
Files changed (4) hide show
  1. README.md +75 -0
  2. all_results.json +8 -0
  3. train_results.json +8 -0
  4. trainer_state.json +270 -0
README.md ADDED
@@ -0,0 +1,75 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: peft
4
+ tags:
5
+ - trl
6
+ - dpo
7
+ - generated_from_trainer
8
+ base_model: alignment-handbook/zephyr-7b-sft-full
9
+ model-index:
10
+ - name: vanilla_dpo_iter_1
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
+ # vanilla_dpo_iter_1
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: 0.6274
22
+ - Rewards/chosen: -0.0991
23
+ - Rewards/rejected: -0.2700
24
+ - Rewards/accuracies: 0.6800
25
+ - Rewards/margins: 0.1709
26
+ - Logps/rejected: -284.5185
27
+ - Logps/chosen: -293.9490
28
+ - Logits/rejected: -2.5507
29
+ - Logits/chosen: -2.6341
30
+
31
+ ## Model description
32
+
33
+ More information needed
34
+
35
+ ## Intended uses & limitations
36
+
37
+ More information needed
38
+
39
+ ## Training and evaluation data
40
+
41
+ More information needed
42
+
43
+ ## Training procedure
44
+
45
+ ### Training hyperparameters
46
+
47
+ The following hyperparameters were used during training:
48
+ - learning_rate: 5e-06
49
+ - train_batch_size: 2
50
+ - eval_batch_size: 2
51
+ - seed: 42
52
+ - distributed_type: multi-GPU
53
+ - num_devices: 4
54
+ - gradient_accumulation_steps: 16
55
+ - total_train_batch_size: 128
56
+ - total_eval_batch_size: 8
57
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
58
+ - lr_scheduler_type: cosine
59
+ - lr_scheduler_warmup_ratio: 0.1
60
+ - num_epochs: 1
61
+
62
+ ### Training results
63
+
64
+ | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
65
+ |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
66
+ | 0.6186 | 0.63 | 100 | 0.6274 | -0.0991 | -0.2700 | 0.6800 | 0.1709 | -284.5185 | -293.9490 | -2.5507 | -2.6341 |
67
+
68
+
69
+ ### Framework versions
70
+
71
+ - PEFT 0.7.1
72
+ - Transformers 4.36.2
73
+ - Pytorch 2.1.2+cu121
74
+ - Datasets 2.14.6
75
+ - Tokenizers 0.15.2
all_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 0.6436299775381508,
4
+ "train_runtime": 7284.0023,
5
+ "train_samples": 20378,
6
+ "train_samples_per_second": 2.798,
7
+ "train_steps_per_second": 0.022
8
+ }
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 0.6436299775381508,
4
+ "train_runtime": 7284.0023,
5
+ "train_samples": 20378,
6
+ "train_samples_per_second": 2.798,
7
+ "train_steps_per_second": 0.022
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,270 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.9984301412872841,
5
+ "eval_steps": 100,
6
+ "global_step": 159,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.01,
13
+ "learning_rate": 3.125e-07,
14
+ "logits/chosen": -2.872462511062622,
15
+ "logits/rejected": -2.8200550079345703,
16
+ "logps/chosen": -279.5750732421875,
17
+ "logps/rejected": -283.5914306640625,
18
+ "loss": 0.6931,
19
+ "rewards/accuracies": 0.0,
20
+ "rewards/chosen": 0.0,
21
+ "rewards/margins": 0.0,
22
+ "rewards/rejected": 0.0,
23
+ "step": 1
24
+ },
25
+ {
26
+ "epoch": 0.06,
27
+ "learning_rate": 3.125e-06,
28
+ "logits/chosen": -2.7532310485839844,
29
+ "logits/rejected": -2.7174949645996094,
30
+ "logps/chosen": -283.4659423828125,
31
+ "logps/rejected": -245.05682373046875,
32
+ "loss": 0.6929,
33
+ "rewards/accuracies": 0.4652777910232544,
34
+ "rewards/chosen": -0.00040846053161658347,
35
+ "rewards/margins": 0.0003327548620291054,
36
+ "rewards/rejected": -0.0007412154227495193,
37
+ "step": 10
38
+ },
39
+ {
40
+ "epoch": 0.13,
41
+ "learning_rate": 4.9903533134293035e-06,
42
+ "logits/chosen": -2.7988767623901367,
43
+ "logits/rejected": -2.745978832244873,
44
+ "logps/chosen": -275.3612365722656,
45
+ "logps/rejected": -259.2863464355469,
46
+ "loss": 0.6891,
47
+ "rewards/accuracies": 0.6875,
48
+ "rewards/chosen": 0.0042456588707864285,
49
+ "rewards/margins": 0.009360184893012047,
50
+ "rewards/rejected": -0.005114525556564331,
51
+ "step": 20
52
+ },
53
+ {
54
+ "epoch": 0.19,
55
+ "learning_rate": 4.882681251368549e-06,
56
+ "logits/chosen": -2.750326633453369,
57
+ "logits/rejected": -2.6968815326690674,
58
+ "logps/chosen": -287.40570068359375,
59
+ "logps/rejected": -264.3677978515625,
60
+ "loss": 0.6796,
61
+ "rewards/accuracies": 0.653124988079071,
62
+ "rewards/chosen": 0.019095653668045998,
63
+ "rewards/margins": 0.02568042278289795,
64
+ "rewards/rejected": -0.006584773771464825,
65
+ "step": 30
66
+ },
67
+ {
68
+ "epoch": 0.25,
69
+ "learning_rate": 4.660472094042121e-06,
70
+ "logits/chosen": -2.753675937652588,
71
+ "logits/rejected": -2.7094359397888184,
72
+ "logps/chosen": -269.2148742675781,
73
+ "logps/rejected": -266.2108459472656,
74
+ "loss": 0.6689,
75
+ "rewards/accuracies": 0.625,
76
+ "rewards/chosen": 0.008650553412735462,
77
+ "rewards/margins": 0.03987262770533562,
78
+ "rewards/rejected": -0.031222078949213028,
79
+ "step": 40
80
+ },
81
+ {
82
+ "epoch": 0.31,
83
+ "learning_rate": 4.33440758555951e-06,
84
+ "logits/chosen": -2.700374126434326,
85
+ "logits/rejected": -2.640312910079956,
86
+ "logps/chosen": -279.00579833984375,
87
+ "logps/rejected": -280.64959716796875,
88
+ "loss": 0.657,
89
+ "rewards/accuracies": 0.684374988079071,
90
+ "rewards/chosen": -0.0027772835455834866,
91
+ "rewards/margins": 0.06927188485860825,
92
+ "rewards/rejected": -0.07204917818307877,
93
+ "step": 50
94
+ },
95
+ {
96
+ "epoch": 0.38,
97
+ "learning_rate": 3.92016186682789e-06,
98
+ "logits/chosen": -2.695119857788086,
99
+ "logits/rejected": -2.624582529067993,
100
+ "logps/chosen": -274.4857482910156,
101
+ "logps/rejected": -262.78082275390625,
102
+ "loss": 0.6506,
103
+ "rewards/accuracies": 0.7093750238418579,
104
+ "rewards/chosen": 0.0005743972724303603,
105
+ "rewards/margins": 0.09640298783779144,
106
+ "rewards/rejected": -0.095828577876091,
107
+ "step": 60
108
+ },
109
+ {
110
+ "epoch": 0.44,
111
+ "learning_rate": 3.437648009023905e-06,
112
+ "logits/chosen": -2.645156145095825,
113
+ "logits/rejected": -2.5854098796844482,
114
+ "logps/chosen": -281.52117919921875,
115
+ "logps/rejected": -287.34521484375,
116
+ "loss": 0.6423,
117
+ "rewards/accuracies": 0.706250011920929,
118
+ "rewards/chosen": -0.0025214666966348886,
119
+ "rewards/margins": 0.1480122059583664,
120
+ "rewards/rejected": -0.15053364634513855,
121
+ "step": 70
122
+ },
123
+ {
124
+ "epoch": 0.5,
125
+ "learning_rate": 2.9100607788275547e-06,
126
+ "logits/chosen": -2.600132703781128,
127
+ "logits/rejected": -2.5598268508911133,
128
+ "logps/chosen": -278.39520263671875,
129
+ "logps/rejected": -274.490478515625,
130
+ "loss": 0.6365,
131
+ "rewards/accuracies": 0.684374988079071,
132
+ "rewards/chosen": -0.08650828897953033,
133
+ "rewards/margins": 0.12828321754932404,
134
+ "rewards/rejected": -0.21479150652885437,
135
+ "step": 80
136
+ },
137
+ {
138
+ "epoch": 0.57,
139
+ "learning_rate": 2.3627616503391813e-06,
140
+ "logits/chosen": -2.6346802711486816,
141
+ "logits/rejected": -2.561859130859375,
142
+ "logps/chosen": -267.5691223144531,
143
+ "logps/rejected": -296.9002380371094,
144
+ "loss": 0.625,
145
+ "rewards/accuracies": 0.684374988079071,
146
+ "rewards/chosen": -0.10211894661188126,
147
+ "rewards/margins": 0.14463460445404053,
148
+ "rewards/rejected": -0.2467535436153412,
149
+ "step": 90
150
+ },
151
+ {
152
+ "epoch": 0.63,
153
+ "learning_rate": 1.8220596619089576e-06,
154
+ "logits/chosen": -2.628725528717041,
155
+ "logits/rejected": -2.5589451789855957,
156
+ "logps/chosen": -290.6697082519531,
157
+ "logps/rejected": -277.4861755371094,
158
+ "loss": 0.6186,
159
+ "rewards/accuracies": 0.737500011920929,
160
+ "rewards/chosen": -0.08957623690366745,
161
+ "rewards/margins": 0.20158691704273224,
162
+ "rewards/rejected": -0.2911631464958191,
163
+ "step": 100
164
+ },
165
+ {
166
+ "epoch": 0.63,
167
+ "eval_logits/chosen": -2.6340699195861816,
168
+ "eval_logits/rejected": -2.550675868988037,
169
+ "eval_logps/chosen": -293.948974609375,
170
+ "eval_logps/rejected": -284.51849365234375,
171
+ "eval_loss": 0.627379298210144,
172
+ "eval_rewards/accuracies": 0.6800000071525574,
173
+ "eval_rewards/chosen": -0.09909208118915558,
174
+ "eval_rewards/margins": 0.17093417048454285,
175
+ "eval_rewards/rejected": -0.27002623677253723,
176
+ "eval_runtime": 384.4479,
177
+ "eval_samples_per_second": 5.202,
178
+ "eval_steps_per_second": 0.65,
179
+ "step": 100
180
+ },
181
+ {
182
+ "epoch": 0.69,
183
+ "learning_rate": 1.3139467229135999e-06,
184
+ "logits/chosen": -2.6164746284484863,
185
+ "logits/rejected": -2.556546688079834,
186
+ "logps/chosen": -296.1142272949219,
187
+ "logps/rejected": -295.5721740722656,
188
+ "loss": 0.6274,
189
+ "rewards/accuracies": 0.668749988079071,
190
+ "rewards/chosen": -0.0782129243016243,
191
+ "rewards/margins": 0.19299447536468506,
192
+ "rewards/rejected": -0.27120739221572876,
193
+ "step": 110
194
+ },
195
+ {
196
+ "epoch": 0.75,
197
+ "learning_rate": 8.628481651367876e-07,
198
+ "logits/chosen": -2.619702100753784,
199
+ "logits/rejected": -2.5185346603393555,
200
+ "logps/chosen": -305.4084167480469,
201
+ "logps/rejected": -269.8453063964844,
202
+ "loss": 0.6209,
203
+ "rewards/accuracies": 0.699999988079071,
204
+ "rewards/chosen": -0.11033356189727783,
205
+ "rewards/margins": 0.20669062435626984,
206
+ "rewards/rejected": -0.3170241713523865,
207
+ "step": 120
208
+ },
209
+ {
210
+ "epoch": 0.82,
211
+ "learning_rate": 4.904486005914027e-07,
212
+ "logits/chosen": -2.639065742492676,
213
+ "logits/rejected": -2.5192837715148926,
214
+ "logps/chosen": -312.64453125,
215
+ "logps/rejected": -290.57025146484375,
216
+ "loss": 0.6221,
217
+ "rewards/accuracies": 0.6875,
218
+ "rewards/chosen": -0.15254992246627808,
219
+ "rewards/margins": 0.21097925305366516,
220
+ "rewards/rejected": -0.36352917551994324,
221
+ "step": 130
222
+ },
223
+ {
224
+ "epoch": 0.88,
225
+ "learning_rate": 2.1464952759020857e-07,
226
+ "logits/chosen": -2.5287649631500244,
227
+ "logits/rejected": -2.4844765663146973,
228
+ "logps/chosen": -285.4928283691406,
229
+ "logps/rejected": -300.5223693847656,
230
+ "loss": 0.6143,
231
+ "rewards/accuracies": 0.6781250238418579,
232
+ "rewards/chosen": -0.14462201297283173,
233
+ "rewards/margins": 0.2161392718553543,
234
+ "rewards/rejected": -0.3607613146305084,
235
+ "step": 140
236
+ },
237
+ {
238
+ "epoch": 0.94,
239
+ "learning_rate": 4.870879364444109e-08,
240
+ "logits/chosen": -2.6168649196624756,
241
+ "logits/rejected": -2.5214154720306396,
242
+ "logps/chosen": -299.5126037597656,
243
+ "logps/rejected": -298.0356750488281,
244
+ "loss": 0.6145,
245
+ "rewards/accuracies": 0.703125,
246
+ "rewards/chosen": -0.1232418566942215,
247
+ "rewards/margins": 0.22777679562568665,
248
+ "rewards/rejected": -0.3510186970233917,
249
+ "step": 150
250
+ },
251
+ {
252
+ "epoch": 1.0,
253
+ "step": 159,
254
+ "total_flos": 0.0,
255
+ "train_loss": 0.6436299775381508,
256
+ "train_runtime": 7284.0023,
257
+ "train_samples_per_second": 2.798,
258
+ "train_steps_per_second": 0.022
259
+ }
260
+ ],
261
+ "logging_steps": 10,
262
+ "max_steps": 159,
263
+ "num_input_tokens_seen": 0,
264
+ "num_train_epochs": 1,
265
+ "save_steps": 100,
266
+ "total_flos": 0.0,
267
+ "train_batch_size": 2,
268
+ "trial_name": null,
269
+ "trial_params": null
270
+ }