File size: 1,564 Bytes
169fa69
dcdbe08
 
 
169fa69
 
add0ccf
 
dcdbe08
d21103d
6583693
169fa69
 
56fb6b2
3b77fca
56fb6b2
 
 
 
 
 
 
3b77fca
56fb6b2
3b77fca
 
 
56fb6b2
3b77fca
56fb6b2
3b77fca
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
02d0d41
56fb6b2
02d0d41
3b77fca
 
 
 
 
56fb6b2
02d0d41
3b77fca
 
 
 
 
 
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
---
language:
- en
- el
tags:
- translation
widget:
- text: "'Katerina' is the best name for a girl."
license: apache-2.0
pipeline_tag: translation

---

## English to Greek NMT from Hellenic Army Academy (SSE) and Technical University of Crete (TUC)

* source languages: en
* target languages: el
* licence: apache-2.0
* dataset: opus, ccmatric
* model: transformer(fairseq)
* pre-processing: tokenization+BPE segmentation
* metrics: bleu, chrf

### Model description

Trained using the Fairseq framework, transformer_iwslt_de_en architecture.\
BPE segmentation (20k codes).\
Mixed-case model. 

### How to use

```
from transformers import FSMTTokenizer, FSMTForConditionalGeneration

mname = " <your_downloaded_model_folderpath_here> "

tokenizer = FSMTTokenizer.from_pretrained(mname)
model = FSMTForConditionalGeneration.from_pretrained(mname)

text = " Katerina, is the best name for a girl."

encoded = tokenizer.encode(text, return_tensors='pt')

outputs = model.generate(encoded, num_beams=5, num_return_sequences=5, early_stopping=True)
for i, output in enumerate(outputs):
    i += 1
    print(f"{i}: {output.tolist()}")
    
    decoded = tokenizer.decode(output, skip_special_tokens=True)
    print(f"{i}: {decoded}")
```


## Training data

Consolidated corpus from Opus and CC-Matrix (~6.6GB in total)


## Eval results


Results on Tatoeba testset (EN-EL): 

| BLEU | chrF  |
| ------ | ------ |
| 76.9 |  0.733 |


Results on XNLI parallel (EN-EL): 

| BLEU | chrF  |
| ------ | ------ |
| 65.4 |  0.624 |

### BibTeX entry and citation info
TODO