File size: 2,774 Bytes
cad0f31
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
84
85
86
87
88
89
90
---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- un_multi
metrics:
- bleu
model-index:
- name: opus-mt-en-ar-finetuned-en-to-ar
  results:
  - task:
      name: Sequence-to-sequence Language Modeling
      type: text2text-generation
    dataset:
      name: un_multi
      type: un_multi
      args: ar-en
    metrics:
    - name: Bleu
      type: bleu
      value: 64.6767
---

<!-- 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. -->

# opus-mt-en-ar-finetuned-en-to-ar

This model is a fine-tuned version of [Helsinki-NLP/opus-mt-en-ar](https://huggingface.co/Helsinki-NLP/opus-mt-en-ar) on the un_multi dataset.
It achieves the following results on the evaluation set:
- Loss: 0.8133
- Bleu: 64.6767
- Gen Len: 17.595

## 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: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 16
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step | Validation Loss | Bleu    | Gen Len |
|:-------------:|:-----:|:----:|:---------------:|:-------:|:-------:|
| No log        | 1.0   | 50   | 0.7710          | 64.3416 | 17.4    |
| No log        | 2.0   | 100  | 0.7569          | 63.9546 | 17.465  |
| No log        | 3.0   | 150  | 0.7570          | 64.7484 | 17.385  |
| No log        | 4.0   | 200  | 0.7579          | 65.4073 | 17.305  |
| No log        | 5.0   | 250  | 0.7624          | 64.8939 | 17.325  |
| No log        | 6.0   | 300  | 0.7696          | 65.1257 | 17.45   |
| No log        | 7.0   | 350  | 0.7747          | 65.527  | 17.395  |
| No log        | 8.0   | 400  | 0.7791          | 65.1357 | 17.52   |
| No log        | 9.0   | 450  | 0.7900          | 65.3812 | 17.415  |
| 0.3982        | 10.0  | 500  | 0.7925          | 65.7346 | 17.39   |
| 0.3982        | 11.0  | 550  | 0.7951          | 65.1267 | 17.62   |
| 0.3982        | 12.0  | 600  | 0.8040          | 64.6874 | 17.495  |
| 0.3982        | 13.0  | 650  | 0.8069          | 64.7788 | 17.52   |
| 0.3982        | 14.0  | 700  | 0.8105          | 64.6701 | 17.585  |
| 0.3982        | 15.0  | 750  | 0.8120          | 64.7111 | 17.58   |
| 0.3982        | 16.0  | 800  | 0.8133          | 64.6767 | 17.595  |


### Framework versions

- Transformers 4.19.2
- Pytorch 1.11.0+cu113
- Datasets 2.2.2
- Tokenizers 0.12.1