File size: 1,860 Bytes
35af852
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
language:
- en
- el
tags:
- translation

widget:
- text: "Δεν υπάρχει τίποτε πιο άνισο από την ίση μεταχείριση των ανίσων."
license: apache-2.0
metrics:
- bleu
---


## Greek to English NMT (lower-case output)
## By the Hellenic Army Academy (SSE) and the Technical University of Crete (TUC)

* source languages: el
* target languages: en
* licence: apache-2.0
* dataset: Opus, CCmatrix
* model: transformer(fairseq)
* pre-processing: tokenization + BPE segmentation
* metrics: bleu, chrf
* output: lowercase only, for mixed-cased model use this: https://huggingface.co/lighteternal/SSE-TUC-mt-el-en-cased


### Model description

Trained using the Fairseq framework, transformer_iwslt_de_en architecture.\
BPE segmentation (10k codes).\
Lower-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 = "Δεν υπάρχει τίποτε πιο άνισο από την ίση μεταχείριση των ανίσων."

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 (EL-EN): 

| BLEU | chrF  |
| ------ | ------ |
| 79.3 |  0.795 |


Results on XNLI parallel (EL-EN): 

| BLEU | chrF  |
| ------ | ------ |
| 66.2 |  0.623 |

### BibTeX entry and citation info
TODO