File size: 2,601 Bytes
894c861
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
---
language:
- ko
- en
base_model: facebook/mbart-large-50-many-to-many-mmt
tags:
- generated_from_trainer
metrics:
- bleu
model-index:
- name: ko-en_mbartLarge_exp10p
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# ko-en_mbartLarge_exp10p

This model is a fine-tuned version of [facebook/mbart-large-50-many-to-many-mmt](https://huggingface.co/facebook/mbart-large-50-many-to-many-mmt) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 1.1770
- Bleu: 27.7431
- Gen Len: 18.6157

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- total_eval_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine_with_restarts
- lr_scheduler_warmup_steps: 1000
- num_epochs: 40

### Training results

| Training Loss | Epoch | Step  | Validation Loss | Bleu    | Gen Len |
|:-------------:|:-----:|:-----:|:---------------:|:-------:|:-------:|
| 1.5087        | 0.46  | 2000  | 1.4383          | 21.689  | 18.6869 |
| 1.3739        | 0.93  | 4000  | 1.3328          | 23.8363 | 18.7463 |
| 1.2585        | 1.39  | 6000  | 1.2720          | 24.7319 | 18.4624 |
| 1.2355        | 1.86  | 8000  | 1.2356          | 26.1612 | 18.484  |
| 1.0973        | 2.32  | 10000 | 1.2074          | 26.6567 | 18.554  |
| 1.1157        | 2.78  | 12000 | 1.2069          | 26.4733 | 18.8044 |
| 0.9631        | 3.25  | 14000 | 1.1901          | 27.1062 | 18.6803 |
| 1.0223        | 3.71  | 16000 | 1.2280          | 26.3038 | 18.7993 |
| 0.8621        | 4.18  | 18000 | 1.2185          | 26.8035 | 18.6679 |
| 0.866         | 4.64  | 20000 | 1.1770          | 27.7431 | 18.6157 |
| 0.7063        | 5.11  | 22000 | 1.2176          | 27.7268 | 18.6026 |
| 0.7504        | 5.57  | 24000 | 1.2268          | 27.053  | 18.5299 |
| 0.6986        | 6.03  | 26000 | 1.2739          | 27.5119 | 18.7806 |
| 0.6193        | 6.5   | 28000 | 1.2745          | 27.3877 | 18.5109 |


### Framework versions

- Transformers 4.34.0
- Pytorch 2.1.0+cu121
- Datasets 2.14.5
- Tokenizers 0.14.1