|
import gradio as gr |
|
|
|
from translation import Translator, LANGUAGES, MODEL_URL |
|
|
|
LANGUAGES_LIST = list(LANGUAGES.keys()) |
|
|
|
def translate_wrapper(text, src, trg, by_sentence=True, preprocess=True, random=False, num_beams=4): |
|
src_lang = LANGUAGES.get(src) |
|
tgt_lang = LANGUAGES.get(trg) |
|
result = translator.translate( |
|
text=text, |
|
src_lang=src_lang, |
|
tgt_lang=tgt_lang, |
|
do_sample=random, |
|
num_beams=int(num_beams), |
|
by_sentence=by_sentence, |
|
preprocess=preprocess, |
|
) |
|
return result |
|
|
|
article = f""" |
|
This is the demo for a NLLB-200-600M model fine-tuned for a few (mostly new) languages. |
|
The model itself is available at https://huggingface.co/{MODEL_URL} |
|
If you want to host in on your own backend, consider running this dockerized app: https://github.com/slone-nlp/nllb-docker-demo. |
|
""" |
|
|
|
interface = gr.Interface( |
|
translate_wrapper, |
|
[ |
|
gr.Textbox(label="Text to Translate", lines=2, placeholder='Enter text to translate'), |
|
gr.Dropdown(LANGUAGES_LIST, type="value", label='Source Language', value=LANGUAGES_LIST[0], description='Select the source language'), |
|
gr.Dropdown(LANGUAGES_LIST, type="value", label='Target Language', value=LANGUAGES_LIST[1], description='Select the target language'), |
|
gr.Checkbox(label="Translate by Sentence", value=True, description='If checked, the text will be translated sentence by sentence'), |
|
gr.Checkbox(label="Apply Text Preprocessing", value=True, description='If checked, the text will be preprocessed before translation'), |
|
gr.Checkbox(label="Randomize", value=False, description='If checked, the translation will use random sampling'), |
|
gr.Slider(minimum=1, maximum=5, step=1, label="Number of Beams", value=4, description='Select the number of beams for the translation'), |
|
], |
|
"text", |
|
title='Erzya-Russian Translation', |
|
article=article, |
|
) |
|
|
|
if __name__ == '__main__': |
|
translator = Translator() |
|
interface.launch() |