Toadoum/ngambay-fr-v1-ct2

CTranslate2 weights for French → Ngambay.

Source: Toadoum/ngambay-fr-v1

Quick usage

import ctranslate2
from transformers import AutoTokenizer
tok = AutoTokenizer.from_pretrained("Toadoum/ngambay-fr-v1")
gen = ctranslate2.Generator("Toadoum/ngambay-fr-v1-ct2", device="cuda", compute_type="float16")  # or cpu/int8
texts = ["Bonjour, comment allez-vous ?"]
enc = tok._build_translation_inputs(texts, src_lang="fr_Latn", tgt_lang="sba_Latn", return_tensors=None)
batches = [tok.convert_ids_to_tokens(ids) for ids in enc["input_ids"]]
bos_id = enc.get("forced_bos_token_id", tok.bos_token_id)
tgt_prefix = [tok.convert_ids_to_tokens([bos_id])[0]]
res = gen.generate_batch(batches, target_prefix=[tgt_prefix]*len(batches), beam_size=1, sampling_topk=0, max_length=256)
print(tok.decode(tok.convert_tokens_to_ids(res[0].sequences[0]), skip_special_tokens=True))