Hugging Face's logo

wzhouad
/
zephyr-7b-dpo-full

Text Generation
Transformers
Safetensors
llama
trl
dpo
Generated from Trainer
conversational
Inference Endpoints
text-generation-inference
Model card Files Community
wzhouad commited on
Commit
25d6e11
•
1 Parent(s): c69cd50

Model save

Browse files
Files changed (7) hide show
  1. README.md +29 -18
  2. all_results.json +5 -5
  3. model-00001-of-00003.safetensors +1 -1
  4. model-00002-of-00003.safetensors +1 -1
  5. model-00003-of-00003.safetensors +1 -1
  6. train_results.json +5 -5
  7. trainer_state.json +0 -0
README.md CHANGED
@@ -17,15 +17,18 @@ should probably proofread and complete it, then remove this comment. -->
17
 
18
  This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on the None dataset.
19
  It achieves the following results on the evaluation set:
20
- - Loss: 0.0558
21
- - Rewards/chosen: -2.1552
22
- - Rewards/rejected: -2.8915
23
- - Rewards/accuracies: 0.7188
24
- - Rewards/margins: 0.7363
25
- - Logps/rejected: -546.5007
26
- - Logps/chosen: -472.5589
27
- - Logits/rejected: -2.1374
28
- - Logits/chosen: -2.1641
 
 
 
29
 
30
  ## Model description
31
 
@@ -47,11 +50,11 @@ The following hyperparameters were used during training:
47
  - learning_rate: 5e-07
48
  - train_batch_size: 8
49
  - eval_batch_size: 8
50
- - seed: 5
51
  - distributed_type: multi-GPU
52
  - num_devices: 8
53
- - gradient_accumulation_steps: 4
54
- - total_train_batch_size: 256
55
  - total_eval_batch_size: 64
56
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
57
  - lr_scheduler_type: cosine
@@ -60,12 +63,20 @@ The following hyperparameters were used during training:
60
 
61
  ### Training results
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
- | 0.0773 | 0.23 | 100 | 0.0837 | -1.4384 | -1.8521 | 0.6602 | 0.4137 | -442.5617 | -400.8813 | -2.4089 | -2.4262 |
66
- | 0.0481 | 0.45 | 200 | 0.0602 | -1.9981 | -2.5808 | 0.6836 | 0.5827 | -515.4323 | -456.8460 | -2.2208 | -2.2428 |
67
- | 0.0368 | 0.68 | 300 | 0.0517 | -2.2669 | -3.0039 | 0.6953 | 0.7370 | -557.7418 | -483.7301 | -2.1438 | -2.1703 |
68
- | 0.0426 | 0.91 | 400 | 0.0558 | -2.1552 | -2.8915 | 0.7188 | 0.7363 | -546.5007 | -472.5589 | -2.1374 | -2.1641 |
 
 
 
 
 
 
 
 
69
 
70
 
71
  ### Framework versions
 
17
 
18
  This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on the None dataset.
19
  It achieves the following results on the evaluation set:
20
+ - Loss: 0.0050
21
+ - Rewards/chosen: -1.4001
22
+ - Rewards/rejected: -1.6028
23
+ - Rewards/accuracies: 0.6101
24
+ - Rewards/margins: 0.2027
25
+ - Logps/rejected: -311.4186
26
+ - Logps/chosen: -284.5843
27
+ - Logits/rejected: -2.3597
28
+ - Logits/chosen: -2.3712
29
+ - Debug/policy Weights: 0.0065
30
+ - Debug/losses: 0.0043
31
+ - Debug/raw Losses: 0.6467
32
 
33
  ## Model description
34
 
 
50
  - learning_rate: 5e-07
51
  - train_batch_size: 8
52
  - eval_batch_size: 8
53
+ - seed: 42
54
  - distributed_type: multi-GPU
55
  - num_devices: 8
56
+ - gradient_accumulation_steps: 2
57
+ - total_train_batch_size: 128
58
  - total_eval_batch_size: 64
59
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
60
  - lr_scheduler_type: cosine
 
63
 
64
  ### Training results
65
 
66
+ | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen | Debug/policy Weights | Debug/losses | Debug/raw Losses |
67
+ |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|:--------------------:|:------------:|:----------------:|
68
+ | 0.0493 | 0.08 | 100 | 0.0465 | -0.1430 | -0.1669 | 0.5905 | 0.0239 | -167.8270 | -158.8738 | -2.7118 | -2.7198 | 0.0662 | 0.0453 | 0.6827 |
69
+ | 0.008 | 0.16 | 200 | 0.0085 | -0.8731 | -0.9349 | 0.5849 | 0.0618 | -244.6274 | -231.8869 | -2.5966 | -2.6043 | 0.0119 | 0.0079 | 0.6736 |
70
+ | 0.007 | 0.24 | 300 | 0.0084 | -0.9513 | -1.0922 | 0.6110 | 0.1409 | -260.3554 | -239.7077 | -2.5912 | -2.6003 | 0.0117 | 0.0076 | 0.6574 |
71
+ | 0.0065 | 0.32 | 400 | 0.0079 | -1.0231 | -1.1828 | 0.6315 | 0.1597 | -269.4169 | -246.8811 | -2.6498 | -2.6594 | 0.0111 | 0.0074 | 0.6524 |
72
+ | 0.0054 | 0.4 | 500 | 0.0054 | -1.2147 | -1.3633 | 0.6054 | 0.1486 | -287.4665 | -266.0439 | -2.5605 | -2.5698 | 0.0072 | 0.0047 | 0.6539 |
73
+ | 0.0045 | 0.48 | 600 | 0.0047 | -1.3549 | -1.5390 | 0.6147 | 0.1841 | -305.0333 | -280.0664 | -2.4873 | -2.4979 | 0.0061 | 0.0039 | 0.6487 |
74
+ | 0.0058 | 0.56 | 700 | 0.0061 | -1.2583 | -1.4259 | 0.6045 | 0.1676 | -293.7197 | -270.4000 | -2.4961 | -2.5069 | 0.0079 | 0.0050 | 0.6482 |
75
+ | 0.0039 | 0.64 | 800 | 0.0037 | -1.5176 | -1.6919 | 0.5896 | 0.1743 | -320.3220 | -296.3322 | -2.4092 | -2.4203 | 0.0048 | 0.0032 | 0.6584 |
76
+ | 0.005 | 0.72 | 900 | 0.0049 | -1.3883 | -1.5704 | 0.6063 | 0.1820 | -308.1698 | -283.4043 | -2.3689 | -2.3800 | 0.0064 | 0.0042 | 0.6486 |
77
+ | 0.0044 | 0.8 | 1000 | 0.0049 | -1.4264 | -1.6227 | 0.5989 | 0.1963 | -313.4052 | -287.2113 | -2.3493 | -2.3607 | 0.0063 | 0.0041 | 0.6497 |
78
+ | 0.0059 | 0.88 | 1100 | 0.0051 | -1.3862 | -1.5910 | 0.6110 | 0.2047 | -310.2328 | -283.1982 | -2.3569 | -2.3684 | 0.0067 | 0.0044 | 0.6455 |
79
+ | 0.0043 | 0.96 | 1200 | 0.0050 | -1.4001 | -1.6028 | 0.6101 | 0.2027 | -311.4186 | -284.5843 | -2.3597 | -2.3712 | 0.0065 | 0.0043 | 0.6467 |
80
 
81
 
82
  ### Framework versions
all_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
  "epoch": 1.0,
3
- "train_loss": 0.0789936019568081,
4
- "train_runtime": 7437.2558,
5
- "train_samples": 113028,
6
- "train_samples_per_second": 15.198,
7
- "train_steps_per_second": 0.059
8
  }
 
1
  {
2
  "epoch": 1.0,
3
+ "train_loss": 0.01215437763655309,
4
+ "train_runtime": 11339.6145,
5
+ "train_samples": 160800,
6
+ "train_samples_per_second": 14.18,
7
+ "train_steps_per_second": 0.111
8
  }
model-00001-of-00003.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:05b6e046bbaff69b33f93e8f211e5a4305ba2e7727c1ef4668e7b6192cd8289e
3
  size 4943162336
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f07ea723b5a4f4136e492ae69d3665813ff7d23e6749a21988a9a3eeaf0c0f3a
3
  size 4943162336
model-00002-of-00003.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3df89a8522aeb38b85eeb27d82b91dcd6d8e4ecdc0d4a72fe3fcac20445570be
3
  size 4999819336
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:620339b4fe05468a571491ab152f8ed5a30a27d4ae4bdadc41e57f2a212d0e5b
3
  size 4999819336
model-00003-of-00003.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f65ba9fbcf2434548e9470bfa3725e9520f866a8b314aa3cc31ab9cb00065766
3
  size 4540516344
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:148235695f6e08d4c98d258464fe682c7a309502a667a5f257f0c2c4e5cc08e5
3
  size 4540516344
train_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
  "epoch": 1.0,
3
- "train_loss": 0.0789936019568081,
4
- "train_runtime": 7437.2558,
5
- "train_samples": 113028,
6
- "train_samples_per_second": 15.198,
7
- "train_steps_per_second": 0.059
8
  }
 
1
  {
2
  "epoch": 1.0,
3
+ "train_loss": 0.01215437763655309,
4
+ "train_runtime": 11339.6145,
5
+ "train_samples": 160800,
6
+ "train_samples_per_second": 14.18,
7
+ "train_steps_per_second": 0.111
8
  }
trainer_state.json CHANGED
The diff for this file is too large to render. See raw diff