NM-development/nllb-ce-rus-v0

This is fine tuned NLLB-200 model for Chechen-Russian translation, presented in paper The first open machine translation system for the Chechen language.

The language token for the Chechen language is ce_Cyrl, while for all the other languages included in NLLB-200, the tokens are composed of three letters (i.e. rus_Cyrl for Russian).

Here is an example of how the model can be used in the code:

import torch
from transformers import AutoModelForSeq2SeqLM
from transformers import NllbTokenizer

model_nllb = AutoModelForSeq2SeqLM.from_pretrained('NM-development/nllb-ce-rus-v0').cuda()
tokenizer_nllb = NllbTokenizer.from_pretrained('NM-development/nllb-ce-rus-v0')

def translate(text, model, tokenizer, src_lang='rus_Cyrl', tgt_lang='eng_Latn', a=16, b=1.5, max_input_length=1024, **kwargs):
    model.eval()
    with torch.no_grad():
        tokenizer.src_lang = src_lang
        tokenizer.tgt_lang = tgt_lang
        inputs = tokenizer(text, return_tensors='pt', padding=True, truncation=True, max_length=max_input_length)
        result = model.generate(
            **inputs.to(model.device),
            forced_bos_token_id=tokenizer.convert_tokens_to_ids(tgt_lang),
            max_new_tokens=int(a + b * inputs.input_ids.shape[1]),
            **kwargs
        )
    return tokenizer.batch_decode(result, skip_special_tokens=True)


text = "Стигална кӀел къахьоьгуш, ша мел динчу хӀуманах буьсун болу хӀун пайда оьцу адамо?"
translate(text, model_nllb, tokenizer_nllb, src_lang='ce_Cyrl', tgt_lang='rus_Cyrl')[0]
# 'Что пользы человеку от того, что он трудился под солнцем и что сделал?'