Catalan - Occitan (Languedocian variant) translation model for OpenNMT. These are the same models that we have in production at
https://www.softcatala.org/traductor/.
The models are quantified for low latency.
1import ctranslate2
2import pyonmttok
3from huggingface_hub import snapshot_download
4model_dir = snapshot_download(repo_id="softcatala/translate-cat-fra", revision="main")
5
6tokenizer=pyonmttok.Tokenizer(mode="none", sp_model_path = model_dir + "/sp_m.model")
7tokenized=tokenizer.tokenize("Hola amics")
8
9translator = ctranslate2.Translator(model_dir)
10translated = translator.translate_batch([tokenized[0]])
11print(tokenizer.detokenize(translated[0][0]['tokens']))
