Hugging Face's logo

DUAL-GPO-2
/
phi-2-gpo-renew2-i0

PEFT
TensorBoard
Safetensors
phi
alignment-handbook
Generated from Trainer
trl
dpo
custom_code
Model card Files Metrics Community
BraylonDash commited on
Commit
2f18277
•
1 Parent(s): d4507d9

Model save

Browse files
Files changed (8) hide show
  1. README.md +110 -0
  2. adapter_model.safetensors +1 -1
  3. all_results.json +21 -0
  4. eval_results.json +16 -0
  5. runs/Apr09_13-55-32_gpu4-119-5/events.out.tfevents.1712634984.gpu4-119-5.259196.0 +2 -2
  6. runs/Apr09_13-55-32_gpu4-119-5/events.out.tfevents.1712691265.gpu4-119-5.259196.1 +3 -0
  7. train_results.json +8 -0
  8. trainer_state.json +0 -0
README.md ADDED
@@ -0,0 +1,110 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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-renew2-i0
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-renew2-i0
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: 0.0346
22
+ - Rewards/chosen: -0.0264
23
+ - Rewards/rejected: -0.0854
24
+ - Rewards/accuracies: 0.6290
25
+ - Rewards/margins: 0.0591
26
+ - Logps/rejected: -252.3589
27
+ - Logps/chosen: -280.1829
28
+ - Logits/rejected: 1.0402
29
+ - Logits/chosen: 0.9379
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: 1
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
+ | 0.0659 | 0.03 | 100 | 0.0536 | -0.0002 | -0.0008 | 0.4745 | 0.0005 | -243.8923 | -277.5683 | 1.0635 | 0.9711 |
65
+ | 0.0597 | 0.05 | 200 | 0.0518 | 0.0035 | -0.0015 | 0.5880 | 0.0050 | -243.9651 | -277.1979 | 1.0617 | 0.9688 |
66
+ | 0.0564 | 0.08 | 300 | 0.0475 | 0.0104 | -0.0081 | 0.6175 | 0.0185 | -244.6272 | -276.5096 | 1.0440 | 0.9499 |
67
+ | 0.0402 | 0.1 | 400 | 0.0438 | 0.0017 | -0.0309 | 0.6325 | 0.0326 | -246.9109 | -277.3771 | 0.9932 | 0.8995 |
68
+ | 0.0421 | 0.13 | 500 | 0.0411 | -0.0415 | -0.0810 | 0.6195 | 0.0395 | -251.9139 | -281.6956 | 0.9295 | 0.8362 |
69
+ | 0.0439 | 0.16 | 600 | 0.0395 | -0.0701 | -0.1168 | 0.6175 | 0.0468 | -255.5005 | -284.5547 | 0.9520 | 0.8607 |
70
+ | 0.0363 | 0.18 | 700 | 0.0390 | -0.0362 | -0.0808 | 0.6310 | 0.0446 | -251.8926 | -281.1619 | 0.9895 | 0.8949 |
71
+ | 0.0402 | 0.21 | 800 | 0.0382 | -0.0514 | -0.1006 | 0.6220 | 0.0491 | -253.8720 | -282.6901 | 0.9937 | 0.9001 |
72
+ | 0.0381 | 0.24 | 900 | 0.0376 | -0.0554 | -0.1099 | 0.6315 | 0.0545 | -254.8047 | -283.0851 | 1.0465 | 0.9534 |
73
+ | 0.0421 | 0.26 | 1000 | 0.0374 | -0.0408 | -0.0930 | 0.6270 | 0.0522 | -253.1114 | -281.6268 | 1.0399 | 0.9448 |
74
+ | 0.0393 | 0.29 | 1100 | 0.0370 | -0.0576 | -0.1053 | 0.6285 | 0.0478 | -254.3491 | -283.3031 | 1.0557 | 0.9609 |
75
+ | 0.0533 | 0.31 | 1200 | 0.0369 | -0.0606 | -0.1154 | 0.6210 | 0.0548 | -255.3544 | -283.6022 | 1.0368 | 0.9417 |
76
+ | 0.0392 | 0.34 | 1300 | 0.0367 | -0.0207 | -0.0714 | 0.6120 | 0.0508 | -250.9576 | -279.6129 | 1.0634 | 0.9660 |
77
+ | 0.0432 | 0.37 | 1400 | 0.0367 | -0.0146 | -0.0629 | 0.6260 | 0.0483 | -250.1082 | -279.0112 | 1.0463 | 0.9482 |
78
+ | 0.0304 | 0.39 | 1500 | 0.0359 | -0.0523 | -0.1062 | 0.6360 | 0.0539 | -254.4339 | -282.7773 | 1.0471 | 0.9496 |
79
+ | 0.0436 | 0.42 | 1600 | 0.0359 | -0.0322 | -0.0845 | 0.6340 | 0.0522 | -252.2616 | -280.7699 | 1.0586 | 0.9585 |
80
+ | 0.0405 | 0.44 | 1700 | 0.0355 | -0.0531 | -0.1105 | 0.6335 | 0.0575 | -254.8697 | -282.8529 | 1.0312 | 0.9322 |
81
+ | 0.0352 | 0.47 | 1800 | 0.0354 | -0.0369 | -0.0956 | 0.6220 | 0.0586 | -253.3721 | -281.2394 | 1.0533 | 0.9539 |
82
+ | 0.0392 | 0.5 | 1900 | 0.0355 | -0.0281 | -0.0860 | 0.6210 | 0.0579 | -252.4193 | -280.3594 | 1.0498 | 0.9508 |
83
+ | 0.0368 | 0.52 | 2000 | 0.0354 | -0.0231 | -0.0770 | 0.6300 | 0.0539 | -251.5159 | -279.8615 | 1.0563 | 0.9577 |
84
+ | 0.0326 | 0.55 | 2100 | 0.0352 | -0.0360 | -0.0915 | 0.6300 | 0.0555 | -252.9630 | -281.1432 | 1.0751 | 0.9760 |
85
+ | 0.0368 | 0.58 | 2200 | 0.0352 | -0.0391 | -0.0965 | 0.6345 | 0.0574 | -253.4691 | -281.4595 | 1.0642 | 0.9640 |
86
+ | 0.0315 | 0.6 | 2300 | 0.0351 | -0.0252 | -0.0801 | 0.6330 | 0.0549 | -251.8242 | -280.0628 | 1.0685 | 0.9676 |
87
+ | 0.0341 | 0.63 | 2400 | 0.0352 | -0.0240 | -0.0803 | 0.6320 | 0.0563 | -251.8426 | -279.9447 | 1.0420 | 0.9405 |
88
+ | 0.0488 | 0.65 | 2500 | 0.0350 | -0.0321 | -0.0918 | 0.6340 | 0.0597 | -252.9968 | -280.7594 | 1.0394 | 0.9378 |
89
+ | 0.0279 | 0.68 | 2600 | 0.0349 | -0.0383 | -0.0996 | 0.6315 | 0.0613 | -253.7721 | -281.3765 | 1.0361 | 0.9350 |
90
+ | 0.0427 | 0.71 | 2700 | 0.0348 | -0.0312 | -0.0911 | 0.6310 | 0.0600 | -252.9290 | -280.6644 | 1.0336 | 0.9319 |
91
+ | 0.0331 | 0.73 | 2800 | 0.0349 | -0.0291 | -0.0872 | 0.6290 | 0.0581 | -252.5369 | -280.4611 | 1.0354 | 0.9335 |
92
+ | 0.0415 | 0.76 | 2900 | 0.0349 | -0.0298 | -0.0883 | 0.6315 | 0.0585 | -252.6469 | -280.5276 | 1.0248 | 0.9228 |
93
+ | 0.0404 | 0.79 | 3000 | 0.0349 | -0.0268 | -0.0859 | 0.6295 | 0.0590 | -252.4009 | -280.2291 | 1.0305 | 0.9277 |
94
+ | 0.0362 | 0.81 | 3100 | 0.0348 | -0.0264 | -0.0849 | 0.6305 | 0.0585 | -252.3079 | -280.1861 | 1.0296 | 0.9270 |
95
+ | 0.0412 | 0.84 | 3200 | 0.0348 | -0.0274 | -0.0861 | 0.6260 | 0.0587 | -252.4237 | -280.2876 | 1.0338 | 0.9313 |
96
+ | 0.0485 | 0.86 | 3300 | 0.0347 | -0.0242 | -0.0824 | 0.6270 | 0.0582 | -252.0546 | -279.9648 | 1.0359 | 0.9336 |
97
+ | 0.0376 | 0.89 | 3400 | 0.0346 | -0.0264 | -0.0854 | 0.6310 | 0.0590 | -252.3589 | -280.1902 | 1.0377 | 0.9354 |
98
+ | 0.0352 | 0.92 | 3500 | 0.0346 | -0.0266 | -0.0856 | 0.6260 | 0.0590 | -252.3726 | -280.2037 | 1.0418 | 0.9392 |
99
+ | 0.0379 | 0.94 | 3600 | 0.0347 | -0.0263 | -0.0852 | 0.6315 | 0.0589 | -252.3377 | -280.1781 | 1.0414 | 0.9390 |
100
+ | 0.0361 | 0.97 | 3700 | 0.0346 | -0.0266 | -0.0856 | 0.6310 | 0.0590 | -252.3741 | -280.2047 | 1.0399 | 0.9377 |
101
+ | 0.0298 | 0.99 | 3800 | 0.0347 | -0.0263 | -0.0850 | 0.6275 | 0.0587 | -252.3201 | -280.1767 | 1.0412 | 0.9387 |
102
+
103
+
104
+ ### Framework versions
105
+
106
+ - PEFT 0.7.1
107
+ - Transformers 4.36.2
108
+ - Pytorch 2.1.2
109
+ - Datasets 2.14.6
110
+ - Tokenizers 0.15.2
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f50a2e1c754c03fe99ee73d9a3ee48c169ef3c298ed8446fcf96173a9d014a42
3
  size 167807296
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e08781d9a0271a1068d57908b2717ee4698996351a0446a8c4605d7a05a1b447
3
  size 167807296
all_results.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "eval_logits/chosen": 0.937932550907135,
4
+ "eval_logits/rejected": 1.0401638746261597,
5
+ "eval_logps/chosen": -280.182861328125,
6
+ "eval_logps/rejected": -252.35885620117188,
7
+ "eval_loss": 0.034593481570482254,
8
+ "eval_rewards/accuracies": 0.6290000081062317,
9
+ "eval_rewards/chosen": -0.02636060118675232,
10
+ "eval_rewards/margins": 0.059068720787763596,
11
+ "eval_rewards/rejected": -0.08542931824922562,
12
+ "eval_runtime": 538.4763,
13
+ "eval_samples": 2000,
14
+ "eval_samples_per_second": 3.714,
15
+ "eval_steps_per_second": 0.929,
16
+ "train_loss": 0.038748307645445686,
17
+ "train_runtime": 55741.6245,
18
+ "train_samples": 61135,
19
+ "train_samples_per_second": 1.097,
20
+ "train_steps_per_second": 0.069
21
+ }
eval_results.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "eval_logits/chosen": 0.937932550907135,
4
+ "eval_logits/rejected": 1.0401638746261597,
5
+ "eval_logps/chosen": -280.182861328125,
6
+ "eval_logps/rejected": -252.35885620117188,
7
+ "eval_loss": 0.034593481570482254,
8
+ "eval_rewards/accuracies": 0.6290000081062317,
9
+ "eval_rewards/chosen": -0.02636060118675232,
10
+ "eval_rewards/margins": 0.059068720787763596,
11
+ "eval_rewards/rejected": -0.08542931824922562,
12
+ "eval_runtime": 538.4763,
13
+ "eval_samples": 2000,
14
+ "eval_samples_per_second": 3.714,
15
+ "eval_steps_per_second": 0.929
16
+ }
runs/Apr09_13-55-32_gpu4-119-5/events.out.tfevents.1712634984.gpu4-119-5.259196.0 CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:54c87eaba2f542ac7ef6039f84bc380108f1a4608c61f5260c24f3da38ccdfb1
3
- size 274327
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:564ccc29e55ba44272ff882d00ed168a137bc1f2ca0144fc3e6cd6d7ec5ce261
3
+ size 275949
runs/Apr09_13-55-32_gpu4-119-5/events.out.tfevents.1712691265.gpu4-119-5.259196.1 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ab7cec4c1b0ad70787645e1e681c5666be93b4a593eca6814b234622721bfa24
3
+ size 828
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 0.038748307645445686,
4
+ "train_runtime": 55741.6245,
5
+ "train_samples": 61135,
6
+ "train_samples_per_second": 1.097,
7
+ "train_steps_per_second": 0.069
8
+ }
trainer_state.json ADDED
The diff for this file is too large to render. See raw diff