app.py

import ctranslate2
import gradio as gr
from huggingface_hub import snapshot_download
from sentencepiece import SentencePieceProcessor

title = "Mesolitica t5-base-standard-bahasa Translation Demo"
description = """
<p>
Translator using <a href='https://huggingface.co/spaces/mesolitica/malaysian-translation/' target='_blank'>Mesolitica Malaysian Translation model</a>. This demo application uses  
CTranslate2 optimized version of it: <a href="https://huggingface.co/santhosh/mesolitica-t5-base-standard-bahasa-cased-ct2">santhosh/mesolitica-t5-base-standard-bahasa-cased-ct2</a>,
</p>
"""


model_name = "santhosh/mesolitica-t5-base-standard-bahasa-cased-ct2"
model_path = snapshot_download(model_name)

tokenizer = SentencePieceProcessor()
tokenizer.load(f"{model_path}/sentencepiece.model")
translator = ctranslate2.Translator(model_path)

map_lang = {"en": "Inggeris", "jv": "Jawa", "bjn": "Banjarese", "ms": "Melayu", "id": "Indonesia"}


def translate(input_text, target_language):
    input_tokens = tokenizer.encode(
        f"f'terjemah ke {map_lang[target_language]}: {input_text}", out_type=str
    )
    results = translator.translate_batch(
        [input_tokens],
        batch_type="tokens",
        max_input_length=6144,
        max_decoding_length=6144,
        max_batch_size=1024,
        beam_size=1,
    )
    translated_sentence = tokenizer.decode(results[0].hypotheses[0])
    return translated_sentence


def translate_interface(input_text, target_language):
    translated_text = translate(input_text, target_language)
    return translated_text


input_text = gr.Textbox(
    label="Input Text",
    value="Imagine a world in which every single person on the planet is given free access to the sum of all human knowledge.",
)
languages = [
    ("English", "en"),
    ("Bahasa Melayu", "ms"),
    ("Indonesian", "id"),
    ("Banjarese", "bjn"),
    ("Jawa", "jv"),
]
target_language = gr.Dropdown(languages, value="en", label="Target Language")
output_text = gr.Textbox(label="Translated Text")

gr.Interface(
    title=title,
    description=description,
    fn=translate_interface,
    inputs=[input_text, target_language],
    outputs=output_text,
).launch()