Direction	madlad400-3b-mt-ce-v0	nllb-ce-rus-v0
ce2ru	BLEU: 18.48 chrF2++: 41.38	BLEU: 9.59 chrF2++: 33.71
ru2ce	BLEU: 7.53 chrF2++: 32.93	BLEU: 4.12 chrF2++: 26.48
ce2en	BLEU: 12.37 chrF2++: 33.26	BLEU: 1.36 chrF2++: 9.29
en2ce	BLEU: 5.30 chrF2++: 27.71	BLEU: 2.99 chrF2++: 22.91

Jupyter Notebook setup

You may run the model in Jupyter Notebook with this code:

python

from transformers import T5ForConditionalGeneration, T5Tokenizer
import torch


def translate(text, model, tokenizer, tgt_lang):
    model.eval()
    inputs = tokenizer(f"<2{tgt_lang}> {text}", return_tensors="pt").to(device)
    outputs = model.generate(
      **inputs,
      max_length=256,
      num_beams=5,
      early_stopping=True
    )
    return tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]





device = torch.device("cuda" if torch.cuda.is_available() else "cpu")


model_id = 'NM-development/madlad400-3b-mt-ce-v0'

model = T5ForConditionalGeneration.from_pretrained(
    model_id,
    dtype=torch.bfloat16,
    device_map=None,
).to(device)

tokenizer = T5Tokenizer.from_pretrained(model_id)

text = "После захода солнца батраки сложили сено у сарая и ушли из хутора."
translate(text, model, tokenizer, 'ce')
# Малх бухаделлачул тӀаьхьа, цӀийнан керта ялта а диллина, кӀотар чуьра дӀабахара бацараш.