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
669880e
•
1 Parent(s): 0f070c2

Model save

Browse files
Files changed (12) hide show
  1. README.md +34 -31
  2. all_results.json +6 -5
  3. config.json +2 -1
  4. generation_config.json +1 -1
  5. model-00001-of-00003.safetensors +1 -1
  6. model-00002-of-00003.safetensors +1 -1
  7. model-00003-of-00003.safetensors +1 -1
  8. tokenizer.json +1 -0
  9. tokenizer_config.json +2 -1
  10. train_results.json +6 -5
  11. trainer_state.json +0 -0
  12. training_args.bin +2 -2
README.md CHANGED
@@ -2,31 +2,34 @@
2
  license: mit
3
  base_model: HuggingFaceH4/mistral-7b-sft-beta
4
  tags:
5
- - alignment-handbook
 
6
  - generated_from_trainer
7
- datasets:
8
- - HuggingFaceH4/hh-rlhf-h4
9
  model-index:
10
- - name: dpo_hh
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
- # dpo_hh
 
18
 
19
- This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on the HuggingFaceH4/hh-rlhf-h4 dataset.
20
  It achieves the following results on the evaluation set:
21
- - Loss: 0.5440
22
- - Rewards/chosen: -2.2940
23
- - Rewards/rejected: -3.0054
24
- - Rewards/accuracies: 0.7090
25
- - Rewards/margins: 0.7114
26
- - Logps/rejected: -451.6765
27
- - Logps/chosen: -373.9785
28
- - Logits/rejected: 0.3244
29
- - Logits/chosen: 0.0742
 
 
 
30
 
31
  ## Model description
32
 
@@ -61,25 +64,25 @@ The following hyperparameters were used during training:
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
- | 0.6789 | 0.08 | 100 | 0.6770 | -0.1062 | -0.1422 | 0.5914 | 0.0360 | -165.3552 | -155.1927 | -2.7255 | -2.7337 |
67
- | 0.6062 | 0.16 | 200 | 0.6079 | -1.0212 | -1.3873 | 0.6670 | 0.3660 | -289.8622 | -246.6971 | -2.3696 | -2.3856 |
68
- | 0.5965 | 0.24 | 300 | 0.5907 | -1.3779 | -1.8008 | 0.6623 | 0.4229 | -331.2100 | -282.3621 | -2.2450 | -2.2656 |
69
- | 0.5729 | 0.32 | 400 | 0.5711 | -1.6763 | -2.2404 | 0.6828 | 0.5640 | -375.1720 | -312.2064 | -1.2920 | -1.3760 |
70
- | 0.5645 | 0.4 | 500 | 0.5639 | -2.0721 | -2.6869 | 0.6987 | 0.6147 | -419.8194 | -351.7883 | -0.6091 | -0.7860 |
71
- | 0.5513 | 0.48 | 600 | 0.5582 | -2.9237 | -3.5389 | 0.7108 | 0.6152 | -505.0223 | -436.9386 | 0.1224 | -0.1054 |
72
- | 0.5571 | 0.56 | 700 | 0.5559 | -2.7971 | -3.5456 | 0.7043 | 0.7485 | -505.6961 | -424.2823 | 0.2980 | 0.0356 |
73
- | 0.5609 | 0.64 | 800 | 0.5469 | -2.4314 | -3.0831 | 0.7108 | 0.6517 | -459.4439 | -387.7092 | 0.1922 | -0.0312 |
74
- | 0.5514 | 0.72 | 900 | 0.5474 | -2.4774 | -3.2082 | 0.6996 | 0.7308 | -471.9533 | -392.3096 | 0.5382 | 0.2860 |
75
- | 0.527 | 0.8 | 1000 | 0.5454 | -2.5040 | -3.2071 | 0.7080 | 0.7031 | -471.8454 | -394.9711 | 0.6372 | 0.3871 |
76
- | 0.5487 | 0.88 | 1100 | 0.5444 | -2.2851 | -2.9963 | 0.7090 | 0.7112 | -450.7599 | -373.0831 | 0.4336 | 0.1858 |
77
- | 0.5483 | 0.96 | 1200 | 0.5440 | -2.2940 | -3.0054 | 0.7090 | 0.7114 | -451.6765 | -373.9785 | 0.3244 | 0.0742 |
78
 
79
 
80
  ### Framework versions
81
 
82
- - Transformers 4.35.2
83
  - Pytorch 2.1.2+cu121
84
  - Datasets 2.14.6
85
- - Tokenizers 0.14.1
 
2
  license: mit
3
  base_model: HuggingFaceH4/mistral-7b-sft-beta
4
  tags:
5
+ - trl
6
+ - dpo
7
  - generated_from_trainer
 
 
8
  model-index:
9
+ - name: zephyr-7b-dpo-full
10
  results: []
11
  ---
12
 
13
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
14
  should probably proofread and complete it, then remove this comment. -->
15
 
16
+ [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/sanqiang/wdpo/runs/p3jwj9bo)
17
+ # zephyr-7b-dpo-full
18
 
19
+ 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.
20
  It achieves the following results on the evaluation set:
21
+ - Loss: 0.0256
22
+ - Rewards/chosen: -2.0365
23
+ - Rewards/rejected: -2.5297
24
+ - Rewards/accuracies: 0.6950
25
+ - Rewards/margins: 0.4933
26
+ - Logps/rejected: -403.6735
27
+ - Logps/chosen: -347.8913
28
+ - Logits/rejected: -2.1603
29
+ - Logits/chosen: -2.1828
30
+ - Debug/policy Weights: 0.0416
31
+ - Debug/losses: 0.0243
32
+ - Debug/raw Losses: 0.5731
33
 
34
  ## Model description
35
 
 
64
 
65
  ### Training results
66
 
67
+ | 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 |
68
+ |:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|:--------------------:|:------------:|:----------------:|
69
+ | 0.1731 | 0.0796 | 100 | 0.1627 | -0.1434 | -0.1775 | 0.5961 | 0.0341 | -168.4450 | -158.5804 | -2.7045 | -2.7126 | 0.2376 | 0.1613 | 0.6787 |
70
+ | 0.0637 | 0.1592 | 200 | 0.0668 | -0.9118 | -1.1252 | 0.6455 | 0.2134 | -263.2193 | -235.4248 | -2.4769 | -2.4894 | 0.1048 | 0.0652 | 0.6301 |
71
+ | 0.0398 | 0.2388 | 300 | 0.0421 | -1.5345 | -1.8565 | 0.6446 | 0.3220 | -336.3452 | -297.6896 | -2.4777 | -2.4926 | 0.0656 | 0.0401 | 0.6158 |
72
+ | 0.0268 | 0.3183 | 400 | 0.0274 | -1.9929 | -2.3663 | 0.6437 | 0.3735 | -387.3311 | -343.5292 | -2.2480 | -2.2673 | 0.0425 | 0.0260 | 0.6099 |
73
+ | 0.0286 | 0.3979 | 500 | 0.0340 | -1.8450 | -2.2365 | 0.6539 | 0.3916 | -374.3529 | -328.7424 | -2.3185 | -2.3383 | 0.0541 | 0.0326 | 0.6004 |
74
+ | 0.0304 | 0.4775 | 600 | 0.0296 | -1.9424 | -2.3790 | 0.6735 | 0.4366 | -388.5944 | -338.4821 | -2.1888 | -2.2094 | 0.0468 | 0.0278 | 0.5888 |
75
+ | 0.0289 | 0.5571 | 700 | 0.0279 | -1.9248 | -2.3277 | 0.6828 | 0.4030 | -383.4731 | -336.7225 | -2.2155 | -2.2362 | 0.0447 | 0.0266 | 0.5876 |
76
+ | 0.0235 | 0.6367 | 800 | 0.0245 | -2.0777 | -2.5498 | 0.6884 | 0.4720 | -405.6762 | -352.0160 | -2.1066 | -2.1293 | 0.0392 | 0.0231 | 0.5835 |
77
+ | 0.0333 | 0.7163 | 900 | 0.0342 | -1.7749 | -2.2999 | 0.6856 | 0.5250 | -380.6898 | -321.7296 | -2.1171 | -2.1415 | 0.0554 | 0.0321 | 0.5741 |
78
+ | 0.0233 | 0.7959 | 1000 | 0.0238 | -2.2080 | -2.6970 | 0.6950 | 0.4891 | -420.4027 | -365.0407 | -2.1112 | -2.1340 | 0.0381 | 0.0223 | 0.5775 |
79
+ | 0.0253 | 0.8754 | 1100 | 0.0261 | -2.0131 | -2.5002 | 0.6912 | 0.4871 | -400.7220 | -345.5524 | -2.1743 | -2.1963 | 0.0424 | 0.0247 | 0.5737 |
80
+ | 0.0244 | 0.9550 | 1200 | 0.0256 | -2.0365 | -2.5297 | 0.6950 | 0.4933 | -403.6735 | -347.8913 | -2.1603 | -2.1828 | 0.0416 | 0.0243 | 0.5731 |
81
 
82
 
83
  ### Framework versions
84
 
85
+ - Transformers 4.41.0.dev0
86
  - Pytorch 2.1.2+cu121
87
  - Datasets 2.14.6
88
+ - Tokenizers 0.19.1
all_results.json CHANGED
@@ -1,8 +1,9 @@
1
  {
2
- "epoch": 1.0,
3
- "train_loss": 0.5712926928784438,
4
- "train_runtime": 11573.5451,
 
5
  "train_samples": 160800,
6
- "train_samples_per_second": 13.894,
7
- "train_steps_per_second": 0.109
8
  }
 
1
  {
2
+ "epoch": 0.9996020692399522,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.04845805117374013,
5
+ "train_runtime": 10274.1816,
6
  "train_samples": 160800,
7
+ "train_samples_per_second": 15.651,
8
+ "train_steps_per_second": 0.122
9
  }
config.json CHANGED
@@ -3,6 +3,7 @@
3
  "architectures": [
4
  "MistralForCausalLM"
5
  ],
 
6
  "bos_token_id": 1,
7
  "eos_token_id": 2,
8
  "hidden_act": "silu",
@@ -19,7 +20,7 @@
19
  "sliding_window": 4096,
20
  "tie_word_embeddings": false,
21
  "torch_dtype": "bfloat16",
22
- "transformers_version": "4.35.2",
23
  "use_cache": false,
24
  "vocab_size": 32000
25
  }
 
3
  "architectures": [
4
  "MistralForCausalLM"
5
  ],
6
+ "attention_dropout": 0.0,
7
  "bos_token_id": 1,
8
  "eos_token_id": 2,
9
  "hidden_act": "silu",
 
20
  "sliding_window": 4096,
21
  "tie_word_embeddings": false,
22
  "torch_dtype": "bfloat16",
23
+ "transformers_version": "4.41.0.dev0",
24
  "use_cache": false,
25
  "vocab_size": 32000
26
  }
generation_config.json CHANGED
@@ -2,5 +2,5 @@
2
  "_from_model_config": true,
3
  "bos_token_id": 1,
4
  "eos_token_id": 2,
5
- "transformers_version": "4.35.2"
6
  }
 
2
  "_from_model_config": true,
3
  "bos_token_id": 1,
4
  "eos_token_id": 2,
5
+ "transformers_version": "4.41.0.dev0"
6
  }
model-00001-of-00003.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:18e1cec63bd40f863dc594533ae9ac02d7bcdd4f57a17c1ef5d63193122a0814
3
  size 4943162336
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:784f4cda7f164811e7c3bff436b7a581ee244805e7990bba245355b8652049aa
3
  size 4943162336
model-00002-of-00003.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:565d4244afeda54e7f62be9e162a16c6892085c081422f02c7a001ecce587eb6
3
  size 4999819336
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cad95a625583fe4eff948eda5f56c518a6068f3bcb95042122cef3b9788487c3
3
  size 4999819336
model-00003-of-00003.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0debf1533b3a9f2ffea91ddec7f947ba3d1c43476aedcef3273235a227bb4ce5
3
  size 4540516344
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:51b810878e8730342266d454af4f87ae1547c48ba92cbc706ff45c4962cf89f3
3
  size 4540516344
tokenizer.json CHANGED
@@ -134,6 +134,7 @@
134
  "end_of_word_suffix": null,
135
  "fuse_unk": true,
136
  "byte_fallback": true,
 
137
  "vocab": {
138
  "<unk>": 0,
139
  "<s>": 1,
 
134
  "end_of_word_suffix": null,
135
  "fuse_unk": true,
136
  "byte_fallback": true,
137
+ "ignore_merges": false,
138
  "vocab": {
139
  "<unk>": 0,
140
  "<s>": 1,
tokenizer_config.json CHANGED
@@ -1,4 +1,6 @@
1
  {
 
 
2
  "added_tokens_decoder": {
3
  "0": {
4
  "content": "<unk>",
@@ -34,7 +36,6 @@
34
  "chat_template": "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'system' %}\n{{ '<|system|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n' + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}",
35
  "clean_up_tokenization_spaces": false,
36
  "eos_token": "</s>",
37
- "legacy": true,
38
  "model_max_length": 2048,
39
  "pad_token": "</s>",
40
  "sp_model_kwargs": {},
 
1
  {
2
+ "add_bos_token": true,
3
+ "add_eos_token": false,
4
  "added_tokens_decoder": {
5
  "0": {
6
  "content": "<unk>",
 
36
  "chat_template": "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'system' %}\n{{ '<|system|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n' + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}",
37
  "clean_up_tokenization_spaces": false,
38
  "eos_token": "</s>",
 
39
  "model_max_length": 2048,
40
  "pad_token": "</s>",
41
  "sp_model_kwargs": {},
train_results.json CHANGED
@@ -1,8 +1,9 @@
1
  {
2
- "epoch": 1.0,
3
- "train_loss": 0.5712926928784438,
4
- "train_runtime": 11573.5451,
 
5
  "train_samples": 160800,
6
- "train_samples_per_second": 13.894,
7
- "train_steps_per_second": 0.109
8
  }
 
1
  {
2
+ "epoch": 0.9996020692399522,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.04845805117374013,
5
+ "train_runtime": 10274.1816,
6
  "train_samples": 160800,
7
+ "train_samples_per_second": 15.651,
8
+ "train_steps_per_second": 0.122
9
  }
trainer_state.json CHANGED
The diff for this file is too large to render. See raw diff
 
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:03874d6397994b298ed83a00ae141cef01d671de759bafade848f5d6789711af
3
- size 5944
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a78f94b93b086e56c1a4c9d34b2979b13b6f65cc770836bb1f02ce1951408ca9
3
+ size 6456