martimfasantos commited on
Commit
5aa79ce
1 Parent(s): 3dc1d21

End of training

Browse files
README.md CHANGED
@@ -15,18 +15,6 @@ should probably proofread and complete it, then remove this comment. -->
15
  # cpo-xcomet-xl_xxl-inc7b-10p-shuff-5e-7-full-tiny
16
 
17
  This model is a fine-tuned version of [TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T](https://huggingface.co/TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T) on the Unbabel/TowerAligned-v0.1 dataset.
18
- It achieves the following results on the evaluation set:
19
- - Loss: 2.5861
20
- - Nll Loss: 0.9631
21
- - Logps/best: -93.9342
22
- - Rewards/chosen: -9.3934
23
- - Rewards/rejected: -8.9617
24
- - Rewards/accuracies: 0.4760
25
- - Rewards/margins: -0.4317
26
- - Logps/rejected: -89.6171
27
- - Logps/chosen: -93.9342
28
- - Logits/rejected: -1.8016
29
- - Logits/chosen: -1.9358
30
 
31
  ## Model description
32
 
@@ -49,8 +37,11 @@ The following hyperparameters were used during training:
49
  - train_batch_size: 1
50
  - eval_batch_size: 4
51
  - seed: 42
52
- - gradient_accumulation_steps: 16
53
- - total_train_batch_size: 16
 
 
 
54
  - optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-08
55
  - lr_scheduler_type: linear
56
  - lr_scheduler_warmup_ratio: 0.1
@@ -58,41 +49,11 @@ The following hyperparameters were used during training:
58
 
59
  ### Training results
60
 
61
- | Training Loss | Epoch | Step | Validation Loss | Nll Loss | Logps/best | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
62
- |:-------------:|:------:|:----:|:---------------:|:--------:|:----------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
63
- | 2.9692 | 0.1063 | 100 | 2.8057 | 1.0649 | -103.2722 | -10.3272 | -9.7340 | 0.4600 | -0.5932 | -97.3398 | -103.2722 | -1.8268 | -1.9640 |
64
- | 3.0031 | 0.2127 | 200 | 2.7925 | 1.0588 | -102.7081 | -10.2708 | -9.6860 | 0.4580 | -0.5848 | -96.8598 | -102.7081 | -1.8260 | -1.9630 |
65
- | 2.7823 | 0.3190 | 300 | 2.7675 | 1.0456 | -101.5052 | -10.1505 | -9.5824 | 0.4600 | -0.5681 | -95.8241 | -101.5052 | -1.8242 | -1.9611 |
66
- | 2.8692 | 0.4254 | 400 | 2.7382 | 1.0320 | -100.2503 | -10.0250 | -9.4771 | 0.4580 | -0.5479 | -94.7710 | -100.2503 | -1.8210 | -1.9575 |
67
- | 3.1882 | 0.5317 | 500 | 2.7126 | 1.0203 | -99.1754 | -9.9175 | -9.3884 | 0.4580 | -0.5291 | -93.8843 | -99.1754 | -1.8185 | -1.9548 |
68
- | 2.8104 | 0.6381 | 600 | 2.6920 | 1.0101 | -98.2426 | -9.8243 | -9.3098 | 0.4640 | -0.5144 | -93.0982 | -98.2426 | -1.8161 | -1.9520 |
69
- | 3.127 | 0.7444 | 700 | 2.6735 | 1.0018 | -97.4806 | -9.7481 | -9.2466 | 0.4680 | -0.5015 | -92.4660 | -97.4806 | -1.8148 | -1.9505 |
70
- | 2.5488 | 0.8508 | 800 | 2.6599 | 0.9951 | -96.8657 | -9.6866 | -9.1952 | 0.4640 | -0.4914 | -91.9516 | -96.8657 | -1.8113 | -1.9467 |
71
- | 2.9106 | 0.9571 | 900 | 2.6461 | 0.9892 | -96.3278 | -9.6328 | -9.1532 | 0.4700 | -0.4795 | -91.5325 | -96.3278 | -1.8107 | -1.9459 |
72
- | 2.7349 | 1.0635 | 1000 | 2.6355 | 0.9845 | -95.8978 | -9.5898 | -9.1181 | 0.4660 | -0.4717 | -91.1811 | -95.8978 | -1.8091 | -1.9442 |
73
- | 2.607 | 1.1698 | 1100 | 2.6258 | 0.9802 | -95.4924 | -9.5492 | -9.0851 | 0.4660 | -0.4642 | -90.8505 | -95.4924 | -1.8079 | -1.9429 |
74
- | 2.5949 | 1.2762 | 1200 | 2.6187 | 0.9772 | -95.2189 | -9.5219 | -9.0638 | 0.4660 | -0.4581 | -90.6378 | -95.2189 | -1.8067 | -1.9415 |
75
- | 3.0028 | 1.3825 | 1300 | 2.6133 | 0.9745 | -94.9713 | -9.4971 | -9.0415 | 0.4660 | -0.4556 | -90.4151 | -94.9713 | -1.8062 | -1.9409 |
76
- | 2.5891 | 1.4889 | 1400 | 2.6075 | 0.9720 | -94.7468 | -9.4747 | -9.0242 | 0.4700 | -0.4505 | -90.2418 | -94.7468 | -1.8062 | -1.9409 |
77
- | 2.5647 | 1.5952 | 1500 | 2.6035 | 0.9701 | -94.5637 | -9.4564 | -9.0103 | 0.4640 | -0.4460 | -90.1033 | -94.5637 | -1.8044 | -1.9389 |
78
- | 2.566 | 1.7016 | 1600 | 2.5974 | 0.9682 | -94.3869 | -9.4387 | -8.9984 | 0.4660 | -0.4403 | -89.9837 | -94.3869 | -1.8037 | -1.9382 |
79
- | 2.4615 | 1.8079 | 1700 | 2.5960 | 0.9672 | -94.3052 | -9.4305 | -8.9899 | 0.4760 | -0.4406 | -89.8987 | -94.3052 | -1.8029 | -1.9373 |
80
- | 2.5336 | 1.9143 | 1800 | 2.5936 | 0.9662 | -94.2071 | -9.4207 | -8.9834 | 0.4700 | -0.4373 | -89.8344 | -94.2071 | -1.8026 | -1.9369 |
81
- | 2.7186 | 2.0206 | 1900 | 2.5908 | 0.9653 | -94.1252 | -9.4125 | -8.9777 | 0.4820 | -0.4349 | -89.7766 | -94.1252 | -1.8021 | -1.9364 |
82
- | 2.6496 | 2.1270 | 2000 | 2.5912 | 0.9646 | -94.0712 | -9.4071 | -8.9704 | 0.4700 | -0.4367 | -89.7039 | -94.0712 | -1.8020 | -1.9363 |
83
- | 2.4786 | 2.2333 | 2100 | 2.5882 | 0.9642 | -94.0305 | -9.4031 | -8.9690 | 0.4820 | -0.4340 | -89.6904 | -94.0305 | -1.8022 | -1.9365 |
84
- | 2.5261 | 2.3396 | 2200 | 2.5871 | 0.9636 | -93.9762 | -9.3976 | -8.9653 | 0.4720 | -0.4323 | -89.6528 | -93.9762 | -1.8015 | -1.9357 |
85
- | 2.4197 | 2.4460 | 2300 | 2.5855 | 0.9634 | -93.9605 | -9.3961 | -8.9653 | 0.4720 | -0.4308 | -89.6529 | -93.9605 | -1.8015 | -1.9357 |
86
- | 2.9723 | 2.5523 | 2400 | 2.5863 | 0.9633 | -93.9504 | -9.3950 | -8.9631 | 0.4760 | -0.4319 | -89.6311 | -93.9504 | -1.8016 | -1.9358 |
87
- | 2.4721 | 2.6587 | 2500 | 2.5864 | 0.9634 | -93.9537 | -9.3954 | -8.9651 | 0.4740 | -0.4302 | -89.6514 | -93.9537 | -1.8014 | -1.9356 |
88
- | 2.8984 | 2.7650 | 2600 | 2.5856 | 0.9630 | -93.9154 | -9.3915 | -8.9610 | 0.4700 | -0.4305 | -89.6100 | -93.9154 | -1.8014 | -1.9356 |
89
- | 3.0422 | 2.8714 | 2700 | 2.5848 | 0.9630 | -93.9148 | -9.3915 | -8.9617 | 0.4800 | -0.4298 | -89.6169 | -93.9148 | -1.8015 | -1.9357 |
90
- | 2.6226 | 2.9777 | 2800 | 2.5861 | 0.9631 | -93.9342 | -9.3934 | -8.9617 | 0.4760 | -0.4317 | -89.6171 | -93.9342 | -1.8016 | -1.9358 |
91
 
92
 
93
  ### Framework versions
94
 
95
- - Transformers 4.41.2
96
- - Pytorch 2.1.2
97
  - Datasets 2.20.0
98
  - Tokenizers 0.19.1
 
15
  # cpo-xcomet-xl_xxl-inc7b-10p-shuff-5e-7-full-tiny
16
 
17
  This model is a fine-tuned version of [TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T](https://huggingface.co/TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T) on the Unbabel/TowerAligned-v0.1 dataset.
 
 
 
 
 
 
 
 
 
 
 
 
18
 
19
  ## Model description
20
 
 
37
  - train_batch_size: 1
38
  - eval_batch_size: 4
39
  - seed: 42
40
+ - distributed_type: multi-GPU
41
+ - num_devices: 2
42
+ - gradient_accumulation_steps: 32
43
+ - total_train_batch_size: 64
44
+ - total_eval_batch_size: 8
45
  - optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-08
46
  - lr_scheduler_type: linear
47
  - lr_scheduler_warmup_ratio: 0.1
 
49
 
50
  ### Training results
51
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
52
 
53
 
54
  ### Framework versions
55
 
56
+ - Transformers 4.43.3
57
+ - Pytorch 2.3.1+cu121
58
  - Datasets 2.20.0
59
  - Tokenizers 0.19.1
config.json CHANGED
@@ -23,7 +23,7 @@
23
  "rope_theta": 10000.0,
24
  "tie_word_embeddings": false,
25
  "torch_dtype": "bfloat16",
26
- "transformers_version": "4.41.2",
27
  "use_cache": true,
28
  "vocab_size": 32000
29
  }
 
23
  "rope_theta": 10000.0,
24
  "tie_word_embeddings": false,
25
  "torch_dtype": "bfloat16",
26
+ "transformers_version": "4.43.3",
27
  "use_cache": true,
28
  "vocab_size": 32000
29
  }
generation_config.json CHANGED
@@ -3,5 +3,5 @@
3
  "eos_token_id": 2,
4
  "max_length": 2048,
5
  "pad_token_id": 0,
6
- "transformers_version": "4.41.2"
7
  }
 
3
  "eos_token_id": 2,
4
  "max_length": 2048,
5
  "pad_token_id": 0,
6
+ "transformers_version": "4.43.3"
7
  }
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3fb4cdee2aca37b087aa8a80d84c22c9fd6779a5ee8b36627d3c56bc2c19a4e9
3
  size 2200119864
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6560a55925799a780991b83fd8faafd504848f8222e09053f1db22ce254f482c
3
  size 2200119864
tokenizer_config.json CHANGED
@@ -1,6 +1,7 @@
1
  {
2
  "add_bos_token": true,
3
  "add_eos_token": false,
 
4
  "added_tokens_decoder": {
5
  "0": {
6
  "content": "<unk>",
 
1
  {
2
  "add_bos_token": true,
3
  "add_eos_token": false,
4
+ "add_prefix_space": null,
5
  "added_tokens_decoder": {
6
  "0": {
7
  "content": "<unk>",
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0679d255030fbbcaf5dbce3c1e7b54b9f5cf09b8c52dbfc1f17740f4625e9995
3
- size 5240
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89285b365dd86f355e8eded496655e0b322b563886ca14547adc8bf7a770493a
3
+ size 6456