app.py

import gradio as gr
from simpletransformers.seq2seq import Seq2SeqModel, Seq2SeqArgs

# Define the models' paths
BM_MODEL_PATH = "Enutrof/marian-mt-en-pcm"
BBGM_EN_PCM_MODEL_PATH = "NITHUB-AI/marian-mt-bbc-en-pcm"
BBGM_PCM_EN_MODEL_PATH = "NITHUB-AI/marian-mt-bbc-pcm-en"



def load_translator(model_name='Enutrof/marian-mt-en-pcm'):
    '''
    This method loads the sequence to sequence model for translation.
    :return: model
    '''
    pmodel_args = Seq2SeqArgs()
    pmodel_args.max_length = 1024
    pmodel_args.length_penalty = 1
    pmodel_args.num_beams = 50
    pmodel_args.num_return_sequences = 3

    pmodel = Seq2SeqModel(
        encoder_decoder_type="marian",
        encoder_decoder_name=model_name,
        args=pmodel_args,
        use_cuda=False
    )
    return pmodel

#Load models

bm_model = load_translator(BM_MODEL_PATH)
bbgm_en_pcm_model = load_translator(BBGM_EN_PCM_MODEL_PATH)
bbgm_pcm_en_model = load_translator(BBGM_PCM_EN_MODEL_PATH)



# Dictionary to easily select model
models = {
    "BM Model": bm_model,
    "BBGM Model (EN to PCM)": bbgm_en_pcm_model,
    "BBGM Model (PCM to EN)": bbgm_pcm_en_model
}

def translate(model_name, source_sentence):
    if isinstance(source_sentence, str):
        source_sentence = [source_sentence]
    model = models[model_name]
    predictions = model.predict(source_sentence)
    return [i.replace('▁', ' ') for i in predictions[0]]

# Gradio interface
interface = gr.Interface(
    fn=translate,
    inputs=[
        gr.Dropdown(choices=["BM Model", "BBGM Model (EN to PCM)", "BBGM Model (PCM to EN)"], label="Model Selection"),
        gr.Textbox(placeholder="Enter source sentence here...", label="Source Sentence"),
    ],
    outputs=[
        gr.Textbox(label="Predicted 1"),
        gr.Textbox(label="Prediction 2"),
        gr.Textbox(label="Prediction 3"),
    ],
    title='“EHN?”: A Bi-directional English to 🇳🇬 Pidgin Machine Translator'
            '\n'
            'A product of the NITHUB AI Team', # ![NITHUB Logo](https://imgur.com/rNfN7cf)
    description='Type your English/🇳🇬 Pidgin text in the left text box to get 🇳🇬 Pidgin/English translations on the right. '
                '\n'
                '- BM Model: Bible-based Marian Model\n'
                '- BBGM Model: Bible-BBC-GPT3.5Turbo-based Marian Model',
    examples=[
        ['BBGM Model (EN to PCM)', 'Who are you?'],
        ['BBGM Model (EN to PCM)', 'I know every song by that artiste.'],
        ['BBGM Model (EN to PCM)', 'I am lost, please help me find my way to the market.'],
        ['BBGM Model (EN to PCM)', 'Is a personal philosophy of moral relativism, the only way to survive in this ethically complex world, or is it just an excuse to justify doing bad things?'],
        ['BBGM Model (PCM to EN)', 'Wetin Ifihan dey talk about sef?'],
        ['BBGM Model (PCM to EN)', 'Dey don place reward for anyone wey go bring information about di matter.'],
        ['BBGM Model (PCM to EN)', 'Who dey breath?'],
        ['BBGM Model (PCM to EN)', 'Di marriage happun six months after di couple introduction wen dem make dia relationship public in early November, 2021.'],
        ['BM Model', 'Is a personal philosophy of moral relativism, the only way to survive in this ethically complex world, or is it just an excuse to justify doing bad things?'],
        ['BM Model', 'I know every song by that artiste.'],
        ['BM Model', 'They should not be permitted here.'],
        ['BM Model', 'I am lost, please help me find my way to the market.']
    ]
)

interface.launch(enable_queue=True)