File size: 1,655 Bytes
a2f3ce5 |
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:
- el
- en
tags:
- translation
widget:
- text: "Κάνω διδακτορικό στην υπολογιστική γλωσσολογία."
license: apache-2.0
metrics:
- bleu
---
## Greek to English NMT from Hellenic Army Academy (SSE) and 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
### 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 = "Κάνω διδακτορικό στην υπολογιστική γλωσσολογία."
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
|