lole25 commited on
Commit
44ca9b5
1 Parent(s): 0e5b437

Model save

Browse files
README.md ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ library_name: peft
4
+ tags:
5
+ - trl
6
+ - dpo
7
+ - generated_from_trainer
8
+ base_model: microsoft/phi-2
9
+ model-index:
10
+ - name: phi-2-gpo-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
+ # phi-2-gpo-iter-1
18
+
19
+ This model is a fine-tuned version of [microsoft/phi-2](https://huggingface.co/microsoft/phi-2) on the None dataset.
20
+ It achieves the following results on the evaluation set:
21
+ - Loss: 0.0108
22
+ - Rewards/chosen: -0.0004
23
+ - Rewards/rejected: -0.0003
24
+ - Rewards/accuracies: 0.5
25
+ - Rewards/margins: -0.0000
26
+ - Logps/rejected: -278.6819
27
+ - Logps/chosen: -306.4212
28
+ - Logits/rejected: 0.0927
29
+ - Logits/chosen: -0.0062
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: 4
50
+ - eval_batch_size: 4
51
+ - seed: 42
52
+ - distributed_type: multi-GPU
53
+ - gradient_accumulation_steps: 4
54
+ - total_train_batch_size: 16
55
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
56
+ - lr_scheduler_type: cosine
57
+ - lr_scheduler_warmup_ratio: 0.1
58
+ - num_epochs: 2
59
+
60
+ ### Training results
61
+
62
+ | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
63
+ |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
64
+ | 0.0104 | 1.6 | 100 | 0.0107 | -0.0005 | -0.0009 | 0.5145 | 0.0005 | -278.7423 | -306.4289 | 0.0889 | -0.0100 |
65
+
66
+
67
+ ### Framework versions
68
+
69
+ - PEFT 0.7.1
70
+ - Transformers 4.36.2
71
+ - Pytorch 2.1.2+cu118
72
+ - Datasets 2.14.6
73
+ - Tokenizers 0.15.2
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7539c67dd5a385cd6683ad4e587215a04095fdb2a6e88be0bd64d983614173d2
3
  size 41977616
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3e1ab81c9f24de5d5babb839ae8effd891adbc3e4eb5a25dfb767ae52ad45abe
3
  size 41977616
all_results.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.98,
3
+ "eval_logits/chosen": -0.006220105104148388,
4
+ "eval_logits/rejected": 0.09270644932985306,
5
+ "eval_logps/chosen": -306.42120361328125,
6
+ "eval_logps/rejected": -278.68194580078125,
7
+ "eval_loss": 0.010818002745509148,
8
+ "eval_rewards/accuracies": 0.5,
9
+ "eval_rewards/chosen": -0.0003898772120010108,
10
+ "eval_rewards/margins": -4.452037319424562e-05,
11
+ "eval_rewards/rejected": -0.00034535673330537975,
12
+ "eval_runtime": 710.2034,
13
+ "eval_samples": 2000,
14
+ "eval_samples_per_second": 2.816,
15
+ "eval_steps_per_second": 0.704,
16
+ "train_loss": 0.010306676939850854,
17
+ "train_runtime": 2255.3236,
18
+ "train_samples": 61135,
19
+ "train_samples_per_second": 0.887,
20
+ "train_steps_per_second": 0.055
21
+ }
eval_results.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.98,
3
+ "eval_logits/chosen": -0.006220105104148388,
4
+ "eval_logits/rejected": 0.09270644932985306,
5
+ "eval_logps/chosen": -306.42120361328125,
6
+ "eval_logps/rejected": -278.68194580078125,
7
+ "eval_loss": 0.010818002745509148,
8
+ "eval_rewards/accuracies": 0.5,
9
+ "eval_rewards/chosen": -0.0003898772120010108,
10
+ "eval_rewards/margins": -4.452037319424562e-05,
11
+ "eval_rewards/rejected": -0.00034535673330537975,
12
+ "eval_runtime": 710.2034,
13
+ "eval_samples": 2000,
14
+ "eval_samples_per_second": 2.816,
15
+ "eval_steps_per_second": 0.704
16
+ }
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.98,
3
+ "train_loss": 0.010306676939850854,
4
+ "train_runtime": 2255.3236,
5
+ "train_samples": 61135,
6
+ "train_samples_per_second": 0.887,
7
+ "train_steps_per_second": 0.055
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,228 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 1.984,
5
+ "eval_steps": 100,
6
+ "global_step": 124,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.02,
13
+ "learning_rate": 3.846153846153847e-07,
14
+ "logits/chosen": 0.4176342189311981,
15
+ "logits/rejected": 0.3709704279899597,
16
+ "logps/chosen": -211.92245483398438,
17
+ "logps/rejected": -192.94674682617188,
18
+ "loss": 0.0102,
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": 0.16,
27
+ "learning_rate": 3.846153846153847e-06,
28
+ "logits/chosen": 0.11221448332071304,
29
+ "logits/rejected": 0.2316199392080307,
30
+ "logps/chosen": -163.23516845703125,
31
+ "logps/rejected": -129.18174743652344,
32
+ "loss": 0.0101,
33
+ "rewards/accuracies": 0.4097222089767456,
34
+ "rewards/chosen": 0.002231667749583721,
35
+ "rewards/margins": 0.0012476635165512562,
36
+ "rewards/rejected": 0.000984004233032465,
37
+ "step": 10
38
+ },
39
+ {
40
+ "epoch": 0.32,
41
+ "learning_rate": 4.951096619903317e-06,
42
+ "logits/chosen": 0.11361704021692276,
43
+ "logits/rejected": 0.05081893131136894,
44
+ "logps/chosen": -177.82867431640625,
45
+ "logps/rejected": -143.34634399414062,
46
+ "loss": 0.0103,
47
+ "rewards/accuracies": 0.40625,
48
+ "rewards/chosen": -0.0008938458049669862,
49
+ "rewards/margins": 0.00015756110951770097,
50
+ "rewards/rejected": -0.0010514067253097892,
51
+ "step": 20
52
+ },
53
+ {
54
+ "epoch": 0.48,
55
+ "learning_rate": 4.716164218065246e-06,
56
+ "logits/chosen": 0.1756063550710678,
57
+ "logits/rejected": 0.23870661854743958,
58
+ "logps/chosen": -182.95594787597656,
59
+ "logps/rejected": -149.51669311523438,
60
+ "loss": 0.0104,
61
+ "rewards/accuracies": 0.38749998807907104,
62
+ "rewards/chosen": 0.0036168531514704227,
63
+ "rewards/margins": 8.136676478898153e-05,
64
+ "rewards/rejected": 0.0035354860592633486,
65
+ "step": 30
66
+ },
67
+ {
68
+ "epoch": 0.64,
69
+ "learning_rate": 4.3048902348863116e-06,
70
+ "logits/chosen": 0.23890891671180725,
71
+ "logits/rejected": 0.15510627627372742,
72
+ "logps/chosen": -198.42312622070312,
73
+ "logps/rejected": -165.66856384277344,
74
+ "loss": 0.0103,
75
+ "rewards/accuracies": 0.41874998807907104,
76
+ "rewards/chosen": -0.000657144351862371,
77
+ "rewards/margins": 0.0006801211275160313,
78
+ "rewards/rejected": -0.0013372651301324368,
79
+ "step": 40
80
+ },
81
+ {
82
+ "epoch": 0.8,
83
+ "learning_rate": 3.7500000000000005e-06,
84
+ "logits/chosen": 0.0918453186750412,
85
+ "logits/rejected": 0.22310352325439453,
86
+ "logps/chosen": -167.31259155273438,
87
+ "logps/rejected": -152.1429443359375,
88
+ "loss": 0.0103,
89
+ "rewards/accuracies": 0.4000000059604645,
90
+ "rewards/chosen": -0.0004377079603727907,
91
+ "rewards/margins": 0.00018744784756563604,
92
+ "rewards/rejected": -0.0006251559825614095,
93
+ "step": 50
94
+ },
95
+ {
96
+ "epoch": 0.96,
97
+ "learning_rate": 3.0956464785579125e-06,
98
+ "logits/chosen": 0.17539598047733307,
99
+ "logits/rejected": 0.15246877074241638,
100
+ "logps/chosen": -172.65084838867188,
101
+ "logps/rejected": -149.18576049804688,
102
+ "loss": 0.0102,
103
+ "rewards/accuracies": 0.48124998807907104,
104
+ "rewards/chosen": 0.00030969156068749726,
105
+ "rewards/margins": 0.0009550409158691764,
106
+ "rewards/rejected": -0.0006453495589084923,
107
+ "step": 60
108
+ },
109
+ {
110
+ "epoch": 1.12,
111
+ "learning_rate": 2.39389699200963e-06,
112
+ "logits/chosen": 0.2547686994075775,
113
+ "logits/rejected": 0.22828814387321472,
114
+ "logps/chosen": -180.34805297851562,
115
+ "logps/rejected": -161.528076171875,
116
+ "loss": 0.0105,
117
+ "rewards/accuracies": 0.375,
118
+ "rewards/chosen": -0.0031430914532393217,
119
+ "rewards/margins": -0.0006838366389274597,
120
+ "rewards/rejected": -0.002459254814311862,
121
+ "step": 70
122
+ },
123
+ {
124
+ "epoch": 1.28,
125
+ "learning_rate": 1.700590188571887e-06,
126
+ "logits/chosen": 0.24805791676044464,
127
+ "logits/rejected": 0.14473767578601837,
128
+ "logps/chosen": -168.27664184570312,
129
+ "logps/rejected": -145.95330810546875,
130
+ "loss": 0.0107,
131
+ "rewards/accuracies": 0.38749998807907104,
132
+ "rewards/chosen": -0.001294322544708848,
133
+ "rewards/margins": -0.0014985360903665423,
134
+ "rewards/rejected": 0.00020421366207301617,
135
+ "step": 80
136
+ },
137
+ {
138
+ "epoch": 1.44,
139
+ "learning_rate": 1.0708929268538034e-06,
140
+ "logits/chosen": 0.21019116044044495,
141
+ "logits/rejected": 0.0877661406993866,
142
+ "logps/chosen": -181.98239135742188,
143
+ "logps/rejected": -144.48709106445312,
144
+ "loss": 0.0103,
145
+ "rewards/accuracies": 0.375,
146
+ "rewards/chosen": -0.0010133858304470778,
147
+ "rewards/margins": 0.00020684293122030795,
148
+ "rewards/rejected": -0.0012202286161482334,
149
+ "step": 90
150
+ },
151
+ {
152
+ "epoch": 1.6,
153
+ "learning_rate": 5.549106142039018e-07,
154
+ "logits/chosen": 0.17486083507537842,
155
+ "logits/rejected": 0.10187114775180817,
156
+ "logps/chosen": -171.99929809570312,
157
+ "logps/rejected": -156.14585876464844,
158
+ "loss": 0.0104,
159
+ "rewards/accuracies": 0.3812499940395355,
160
+ "rewards/chosen": -0.00023761784541420639,
161
+ "rewards/margins": -0.00010095858306158334,
162
+ "rewards/rejected": -0.00013665916048921645,
163
+ "step": 100
164
+ },
165
+ {
166
+ "epoch": 1.6,
167
+ "eval_logits/chosen": -0.010001667775213718,
168
+ "eval_logits/rejected": 0.08886812627315521,
169
+ "eval_logps/chosen": -306.4288635253906,
170
+ "eval_logps/rejected": -278.7423095703125,
171
+ "eval_loss": 0.01069465558975935,
172
+ "eval_rewards/accuracies": 0.5145000219345093,
173
+ "eval_rewards/chosen": -0.00046658667270094156,
174
+ "eval_rewards/margins": 0.00048252198030240834,
175
+ "eval_rewards/rejected": -0.0009491087403148413,
176
+ "eval_runtime": 710.6081,
177
+ "eval_samples_per_second": 2.814,
178
+ "eval_steps_per_second": 0.704,
179
+ "step": 100
180
+ },
181
+ {
182
+ "epoch": 1.76,
183
+ "learning_rate": 1.937002879188285e-07,
184
+ "logits/chosen": 0.17259174585342407,
185
+ "logits/rejected": 0.22864675521850586,
186
+ "logps/chosen": -194.3363494873047,
187
+ "logps/rejected": -159.630126953125,
188
+ "loss": 0.0099,
189
+ "rewards/accuracies": 0.4375,
190
+ "rewards/chosen": -0.00021430714696180075,
191
+ "rewards/margins": 0.0019922677893191576,
192
+ "rewards/rejected": -0.002206574659794569,
193
+ "step": 110
194
+ },
195
+ {
196
+ "epoch": 1.92,
197
+ "learning_rate": 1.6003680950742728e-08,
198
+ "logits/chosen": 0.14167846739292145,
199
+ "logits/rejected": 0.09132746607065201,
200
+ "logps/chosen": -186.17977905273438,
201
+ "logps/rejected": -150.71856689453125,
202
+ "loss": 0.0102,
203
+ "rewards/accuracies": 0.46875,
204
+ "rewards/chosen": -0.0018472172087058425,
205
+ "rewards/margins": 0.0006628064438700676,
206
+ "rewards/rejected": -0.002510023768991232,
207
+ "step": 120
208
+ },
209
+ {
210
+ "epoch": 1.98,
211
+ "step": 124,
212
+ "total_flos": 0.0,
213
+ "train_loss": 0.010306676939850854,
214
+ "train_runtime": 2255.3236,
215
+ "train_samples_per_second": 0.887,
216
+ "train_steps_per_second": 0.055
217
+ }
218
+ ],
219
+ "logging_steps": 10,
220
+ "max_steps": 124,
221
+ "num_input_tokens_seen": 0,
222
+ "num_train_epochs": 2,
223
+ "save_steps": 100,
224
+ "total_flos": 0.0,
225
+ "train_batch_size": 4,
226
+ "trial_name": null,
227
+ "trial_params": null
228
+ }