Hugging Face's logo

LeeSB
/
zephyr-7b-dpo-qlora

PEFT
TensorBoard
Safetensors
trl
dpo
Generated from Trainer
Model card Files Metrics Community
LeeSB commited on
Commit
8d391f1
•
1 Parent(s): ea3d501

Model save

Browse files
Files changed (6) hide show
  1. README.md +19 -47
  2. adapter_model.safetensors +1 -1
  3. all_results.json +5 -5
  4. runs/Apr07_01-31-00_allennlp-cirrascale-68.reviz.ai2.in/events.out.tfevents.1712478752.allennlp-cirrascale-68.reviz.ai2.in.89885.0 +2 -2
  5. train_results.json +5 -5
  6. trainer_state.json +0 -0
README.md CHANGED
@@ -18,15 +18,15 @@ should probably proofread and complete it, then remove this comment. -->
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: 0.9740
22
- - Rewards/chosen: -5.1257
23
- - Rewards/rejected: -6.0686
24
- - Rewards/accuracies: 0.5950
25
- - Rewards/margins: 0.9430
26
- - Logps/rejected: -870.1065
27
- - Logps/chosen: -797.7003
28
- - Logits/rejected: -1.8366
29
- - Logits/chosen: -1.9539
30
 
31
  ## Model description
32
 
@@ -50,8 +50,9 @@ The following hyperparameters were used during training:
50
  - eval_batch_size: 8
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
@@ -61,43 +62,14 @@ The following hyperparameters were used during training:
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
- | 0.6698 | 0.03 | 100 | 0.6901 | -0.0365 | -0.0430 | 0.5625 | 0.0065 | -267.5472 | -288.7829 | -2.5622 | -2.6958 |
65
- | 0.5864 | 0.05 | 200 | 0.6712 | -0.1514 | -0.2064 | 0.5915 | 0.0551 | -283.8881 | -300.2703 | -2.5837 | -2.7210 |
66
- | 0.4604 | 0.08 | 300 | 0.6635 | -0.3454 | -0.4581 | 0.5755 | 0.1127 | -309.0577 | -319.6734 | -2.6465 | -2.7867 |
67
- | 0.3274 | 0.11 | 400 | 0.6736 | -1.4021 | -1.7630 | 0.5865 | 0.3609 | -439.5495 | -425.3439 | -2.3318 | -2.4653 |
68
- | 0.2403 | 0.13 | 500 | 0.6997 | -2.2874 | -2.7446 | 0.5985 | 0.4573 | -537.7081 | -513.8699 | -2.0456 | -2.1684 |
69
- | 0.2586 | 0.16 | 600 | 0.7062 | -2.3888 | -2.8539 | 0.5840 | 0.4651 | -548.6315 | -524.0146 | -1.9324 | -2.0492 |
70
- | 0.2338 | 0.19 | 700 | 0.7076 | -2.4279 | -2.8802 | 0.5865 | 0.4523 | -551.2661 | -527.9254 | -1.9258 | -2.0429 |
71
- | 0.2163 | 0.21 | 800 | 0.7139 | -3.2505 | -3.7850 | 0.6030 | 0.5345 | -641.7469 | -610.1845 | -1.8626 | -1.9782 |
72
- | 0.2297 | 0.24 | 900 | 0.7277 | -3.6449 | -4.1924 | 0.6015 | 0.5475 | -682.4832 | -649.6269 | -1.9512 | -2.0713 |
73
- | 0.1739 | 0.27 | 1000 | 0.7613 | -3.7042 | -4.3355 | 0.6010 | 0.6313 | -696.7919 | -655.5536 | -1.9477 | -2.0703 |
74
- | 0.1868 | 0.29 | 1100 | 0.9361 | -5.1247 | -5.8779 | 0.5725 | 0.7532 | -851.0316 | -797.6036 | -1.7606 | -1.8779 |
75
- | 0.191 | 0.32 | 1200 | 0.8650 | -4.3479 | -5.0050 | 0.5785 | 0.6570 | -763.7404 | -719.9283 | -1.8935 | -2.0135 |
76
- | 0.1594 | 0.35 | 1300 | 0.9262 | -5.0384 | -5.8517 | 0.5940 | 0.8132 | -848.4104 | -788.9777 | -1.8286 | -1.9478 |
77
- | 0.1899 | 0.37 | 1400 | 0.7746 | -3.7073 | -4.4417 | 0.6080 | 0.7344 | -707.4120 | -655.8643 | -1.9178 | -2.0372 |
78
- | 0.1972 | 0.4 | 1500 | 0.9740 | -5.8928 | -6.9457 | 0.6130 | 1.0529 | -957.8102 | -874.4113 | -1.7423 | -1.8571 |
79
- | 0.1712 | 0.43 | 1600 | 0.8206 | -3.9552 | -4.6548 | 0.5935 | 0.6996 | -728.7280 | -680.6552 | -1.8792 | -1.9979 |
80
- | 0.2211 | 0.45 | 1700 | 0.8185 | -4.3265 | -5.1186 | 0.6120 | 0.7921 | -775.1051 | -717.7828 | -1.8551 | -1.9732 |
81
- | 0.1773 | 0.48 | 1800 | 0.9661 | -5.2583 | -6.2006 | 0.5950 | 0.9423 | -883.3090 | -810.9665 | -1.8040 | -1.9210 |
82
- | 0.2611 | 0.51 | 1900 | 0.8358 | -3.9291 | -4.7135 | 0.6035 | 0.7844 | -734.5935 | -678.0472 | -1.8683 | -1.9849 |
83
- | 0.1584 | 0.53 | 2000 | 0.9012 | -4.5759 | -5.4125 | 0.6010 | 0.8366 | -804.4951 | -742.7199 | -1.8457 | -1.9624 |
84
- | 0.173 | 0.56 | 2100 | 0.9585 | -5.3270 | -6.2039 | 0.5905 | 0.8769 | -883.6325 | -817.8319 | -1.8311 | -1.9477 |
85
- | 0.1348 | 0.59 | 2200 | 1.0276 | -5.6514 | -6.6786 | 0.6010 | 1.0272 | -931.1091 | -850.2776 | -1.8167 | -1.9336 |
86
- | 0.1724 | 0.61 | 2300 | 0.9419 | -5.1420 | -6.0555 | 0.6030 | 0.9135 | -868.7937 | -799.3293 | -1.8558 | -1.9729 |
87
- | 0.1652 | 0.64 | 2400 | 0.9893 | -5.2806 | -6.2527 | 0.5960 | 0.9721 | -888.5103 | -813.1928 | -1.8445 | -1.9630 |
88
- | 0.1393 | 0.67 | 2500 | 1.0265 | -5.6807 | -6.7469 | 0.5985 | 1.0662 | -937.9327 | -853.2055 | -1.8254 | -1.9432 |
89
- | 0.2064 | 0.69 | 2600 | 0.9616 | -4.9490 | -5.8748 | 0.5960 | 0.9259 | -850.7297 | -780.0315 | -1.8539 | -1.9722 |
90
- | 0.1235 | 0.72 | 2700 | 1.0443 | -5.6025 | -6.6246 | 0.5925 | 1.0222 | -925.7085 | -845.3807 | -1.8262 | -1.9443 |
91
- | 0.1229 | 0.75 | 2800 | 1.0563 | -5.7201 | -6.7637 | 0.5915 | 1.0437 | -939.6196 | -857.1437 | -1.8211 | -1.9388 |
92
- | 0.1734 | 0.77 | 2900 | 1.0244 | -5.4916 | -6.5069 | 0.5940 | 1.0154 | -913.9391 | -834.2927 | -1.8266 | -1.9443 |
93
- | 0.2791 | 0.8 | 3000 | 0.9939 | -5.2451 | -6.2104 | 0.5940 | 0.9654 | -884.2896 | -809.6420 | -1.8336 | -1.9511 |
94
- | 0.2041 | 0.83 | 3100 | 0.9621 | -4.9990 | -5.9242 | 0.5955 | 0.9253 | -855.6674 | -785.0303 | -1.8434 | -1.9610 |
95
- | 0.1699 | 0.85 | 3200 | 0.9698 | -5.0798 | -6.0155 | 0.5950 | 0.9357 | -864.7924 | -793.1119 | -1.8420 | -1.9592 |
96
- | 0.1894 | 0.88 | 3300 | 0.9731 | -5.1171 | -6.0599 | 0.5975 | 0.9428 | -869.2389 | -796.8451 | -1.8393 | -1.9565 |
97
- | 0.1929 | 0.91 | 3400 | 0.9734 | -5.1189 | -6.0616 | 0.5945 | 0.9427 | -869.4042 | -797.0283 | -1.8394 | -1.9565 |
98
- | 0.1222 | 0.94 | 3500 | 0.9742 | -5.1242 | -6.0674 | 0.5955 | 0.9432 | -869.9834 | -797.5565 | -1.8377 | -1.9550 |
99
- | 0.1486 | 0.96 | 3600 | 0.9741 | -5.1260 | -6.0694 | 0.5955 | 0.9434 | -870.1893 | -797.7342 | -1.8378 | -1.9550 |
100
- | 0.1384 | 0.99 | 3700 | 0.9740 | -5.1257 | -6.0686 | 0.5950 | 0.9430 | -870.1065 | -797.7003 | -1.8366 | -1.9539 |
101
 
102
 
103
  ### Framework versions
 
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: 1.0143
22
+ - Rewards/chosen: -3.1357
23
+ - Rewards/rejected: -3.9504
24
+ - Rewards/accuracies: 0.5615
25
+ - Rewards/margins: 0.8147
26
+ - Logps/rejected: -657.2274
27
+ - Logps/chosen: -598.0983
28
+ - Logits/rejected: -1.6249
29
+ - Logits/chosen: -1.7172
30
 
31
  ## Model description
32
 
 
50
  - eval_batch_size: 8
51
  - seed: 42
52
  - distributed_type: multi-GPU
53
+ - num_devices: 4
54
  - total_train_batch_size: 16
55
+ - total_eval_batch_size: 32
56
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
57
  - lr_scheduler_type: cosine
58
  - lr_scheduler_warmup_ratio: 0.1
 
62
 
63
  | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
64
  |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
65
+ | 0.4083 | 0.11 | 100 | 0.6636 | -0.0687 | -0.1539 | 0.5595 | 0.0852 | -277.5772 | -291.3970 | -2.6015 | -2.7100 |
66
+ | 0.0834 | 0.23 | 200 | 0.7288 | -1.4875 | -1.8701 | 0.5536 | 0.3825 | -449.1904 | -433.2774 | -2.0544 | -2.1544 |
67
+ | 0.047 | 0.34 | 300 | 0.9801 | -2.8622 | -3.5385 | 0.5774 | 0.6763 | -616.0311 | -570.7468 | -1.6400 | -1.7298 |
68
+ | 0.0411 | 0.46 | 400 | 0.9389 | -2.7119 | -3.4267 | 0.5536 | 0.7148 | -604.8529 | -555.7119 | -1.6178 | -1.7086 |
69
+ | 0.0541 | 0.57 | 500 | 1.0554 | -3.1586 | -3.9685 | 0.5575 | 0.8099 | -659.0338 | -600.3828 | -1.6283 | -1.7211 |
70
+ | 0.0315 | 0.68 | 600 | 1.0172 | -3.1217 | -3.9425 | 0.5615 | 0.8208 | -656.4329 | -596.6931 | -1.6210 | -1.7132 |
71
+ | 0.0209 | 0.8 | 700 | 1.0112 | -3.1270 | -3.9417 | 0.5615 | 0.8147 | -656.3586 | -597.2280 | -1.6241 | -1.7165 |
72
+ | 0.0141 | 0.91 | 800 | 1.0143 | -3.1357 | -3.9504 | 0.5615 | 0.8147 | -657.2274 | -598.0983 | -1.6249 | -1.7172 |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
73
 
74
 
75
  ### Framework versions
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2781975710c2430ef05417b5a91f9480e91ad8eadf3ee3a8ae8a6af7efd4f6ae
3
  size 671150064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1b43e3c317a00fed6514543de0896f105c4307887866118d4b7601214ab6d6ea
3
  size 671150064
all_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
  "epoch": 1.0,
3
- "train_loss": 0.23287739991377154,
4
- "train_runtime": 68266.2327,
5
- "train_samples": 59881,
6
- "train_samples_per_second": 0.877,
7
- "train_steps_per_second": 0.055
8
  }
 
1
  {
2
  "epoch": 1.0,
3
+ "train_loss": 0.12645735846515935,
4
+ "train_runtime": 4707.0543,
5
+ "train_samples": 14031,
6
+ "train_samples_per_second": 2.981,
7
+ "train_steps_per_second": 0.186
8
  }
runs/Apr07_01-31-00_allennlp-cirrascale-68.reviz.ai2.in/events.out.tfevents.1712478752.allennlp-cirrascale-68.reviz.ai2.in.89885.0 CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c800b4b5f6586ccfc743098e4e03498f053021268ffeb3e980e3556fe48d488c
3
- size 66770
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a7a188ae190b58e3ad7bb684c52c552f30b45d0211de1d3966df2c935bba66b9
3
+ size 71940
train_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
  "epoch": 1.0,
3
- "train_loss": 0.23287739991377154,
4
- "train_runtime": 68266.2327,
5
- "train_samples": 59881,
6
- "train_samples_per_second": 0.877,
7
- "train_steps_per_second": 0.055
8
  }
 
1
  {
2
  "epoch": 1.0,
3
+ "train_loss": 0.12645735846515935,
4
+ "train_runtime": 4707.0543,
5
+ "train_samples": 14031,
6
+ "train_samples_per_second": 2.981,
7
+ "train_steps_per_second": 0.186
8
  }
trainer_state.json CHANGED
The diff for this file is too large to render. See raw diff