src/translate/Translate.py

from nltk.tokenize import sent_tokenize
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch
import src.exception.Exception.Exception as ExceptionCustom
# Use a pipeline as a high-level helper
from transformers import pipeline

METHOD = "TRANSLATE"

# Load models and tokenizers
tokenizerROMENG = AutoTokenizer.from_pretrained("BlackKakapo/opus-mt-ro-en")
modelROMENG = AutoModelForSeq2SeqLM.from_pretrained("BlackKakapo/opus-mt-ro-en")

tokenizerENGROM = AutoTokenizer.from_pretrained("BlackKakapo/opus-mt-en-ro")
modelENGROM = AutoModelForSeq2SeqLM.from_pretrained("BlackKakapo/opus-mt-en-ro")

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

def paraphraseTranslateMethod(requestValue: str, model: str):
    exception = ExceptionCustom.checkForException(requestValue, METHOD)
    if exception:
        return "", exception

    tokenized_sent_list = sent_tokenize(requestValue)
    result_value = []

    for SENTENCE in tokenized_sent_list:
        if model == 'roen':
            input_ids = tokenizerROMENG(SENTENCE, return_tensors='pt').to(device)
            output = modelROMENG.generate(
                input_ids=input_ids.input_ids,
                do_sample=True,
                max_length=512,
                top_k=90,
                top_p=0.97,
                early_stopping=False
            )
            result = tokenizerROMENG.batch_decode(output, skip_special_tokens=True)[0]
        else:
            input_ids = tokenizerENGROM(SENTENCE, return_tensors='pt').to(device)
            output = modelENGROM.generate(
                input_ids=input_ids.input_ids,
                do_sample=True,
                max_length=512,
                top_k=90,
                top_p=0.97,
                early_stopping=False
            )
            result = tokenizerENGROM.batch_decode(output, skip_special_tokens=True)[0]
        result_value.append(result)

    return " ".join(result_value).strip(), model

def gemma(requestValue: str, model: str = 'Gargaz/gemma-2b-romanian-better'):  
    pipe = pipeline("text-generation", model="Gargaz/gemma-2b-romanian-better")
    messages = [
        {"role": "user", "content": f"Translate the following text to Romanian using a formal tone and provide only translation: {requestValue}"},
    ]
    return pipe(messages)