app.py

import os
import torch
import gradio as gr
import time
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline

codes_as_string = '''Assamese	asm_Beng
Awadhi	awa_Deva
Bengali	ben_Beng
Bhojpuri	bho_Deva
Standard Tibetan	bod_Tibt
Dzongkha	dzo_Tibt
English	eng_Latn
Gujarati	guj_Gujr
Hindi	hin_Deva
Chhattisgarhi	hne_Deva
Kannada	kan_Knda
Kashmiri (Arabic script)	kas_Arab
Kashmiri (Devanagari script)	kas_Deva
Mizo	lus_Latn
Magahi	mag_Deva
Maithili	mai_Deva
Malayalam	mal_Mlym
Marathi	mar_Deva
Meitei (Bengali script)	mni_Beng
Burmese	mya_Mymr
Nepali	npi_Deva
Odia	ory_Orya
Punjabi	pan_Guru
Sanskrit	san_Deva
Santali	sat_Olck
Sindhi	snd_Arab
Tamil	tam_Taml
Telugu	tel_Telu
Urdu	urd_Arab
Vietnamese	vie_Latn'''



def load_models():
    # build model and tokenizer
    model_name_dict = {
                  'nllb-1.3B': "ychenNLP/nllb-200-distilled-1.3B-easyproject",
                  }

    model_dict = {}

    for call_name, real_name in model_name_dict.items():
        print('\tLoading model: %s' % call_name)
        model = AutoModelForSeq2SeqLM.from_pretrained(real_name)
        tokenizer = AutoTokenizer.from_pretrained("facebook/nllb-200-distilled-600M")
        model_dict[call_name+'_model'] = model
        model_dict[call_name+'_tokenizer'] = tokenizer

    return model_dict


def translation(source, target, text):
    if len(model_dict) == 2:
        model_name = 'nllb-1.3B'

    start_time = time.time()
    source = flores_codes[source]
    target = flores_codes[target]

    model = model_dict[model_name + '_model']
    tokenizer = model_dict[model_name + '_tokenizer']

    translator = pipeline('translation', model=model, tokenizer=tokenizer, src_lang=source, tgt_lang=target)
    output = translator(text, max_length=400)

    end_time = time.time()

    full_output = output
    output = output[0]['translation_text']
    # result = {'inference_time': end_time - start_time,
    #           'source': source,
    #           'target': target,
    #           'result': output,
    #           'full_output': full_output}
    return output


if __name__ == '__main__':
    print('\tinit models')
    codes_as_string = codes_as_string.split('\n')

    flores_codes = {}
    for code in codes_as_string:
        lang, lang_code = code.split('\t')
        flores_codes[lang] = lang_code

    global model_dict

    model_dict = load_models()
    
    # define gradio demo
    lang_codes = list(flores_codes.keys())
    
    inputs = [gr.inputs.Dropdown(lang_codes, default='English', label='Source'),
              gr.inputs.Dropdown(lang_codes, default='Hindi', label='Target'),
              gr.inputs.Textbox(lines=5, label="Input text"),
              ]

    outputs = gr.inputs.Textbox(label="Output text")

    title = "Machine Translation Demo"

    demo_status = "Machine Translation System."
    description = f"{demo_status}"

    gr.Interface(translation,
                 inputs,
                 outputs,
                 title=title,
                 description=description,
                 examples=examples,
                 examples_per_page=50,
                 theme="JohnSmith9982/small_and_pretty"
                 ).launch()