|
--- |
|
language: |
|
- ca |
|
- fr |
|
|
|
tags: |
|
- translation |
|
|
|
library_name: opennmt |
|
license: mit |
|
metrics: |
|
- bleu |
|
|
|
inference: false |
|
--- |
|
|
|
### Introduction |
|
|
|
French - Catalan translation model based on OpenNMT. These are the same models that we have in production at https://www.softcatala.org/traductor/. |
|
|
|
|
|
### Usage |
|
|
|
|
|
```bash |
|
pip3 install ctranslate2 pyonmttok |
|
``` |
|
|
|
Simple translation using Python: |
|
|
|
|
|
|
|
```python |
|
import ctranslate2 |
|
import pyonmttok |
|
from huggingface_hub import snapshot_download |
|
model_dir = snapshot_download(repo_id="softcatala/translate-fra-cat", revision="main") |
|
|
|
tokenizer=pyonmttok.Tokenizer(mode="none", sp_model_path = model_dir + "/sp_m.model") |
|
tokenized=tokenizer.tokenize("Salut monde") |
|
|
|
import ctranslate2 |
|
translator = ctranslate2.Translator(model_dir) |
|
translated = translator.translate_batch([tokenized[0]]) |
|
print(tokenizer.detokenize(translated[0][0]['tokens'])) |
|
Hola món! |
|
``` |
|
|
|
## Benchmarks |
|
|
|
| testset | BLEU | |
|
|---------------------------------------|-------| |
|
| test dataset (from train/dev/test) | 43.0 | |
|
| Flores200 dataset | 33.8 | |
|
|
|
## Additional information |
|
* https://github.com/Softcatala/nmt-models |
|
* https://github.com/Softcatala/parallel-catalan-corpus |
|
|
|
|
