app.py

import os, time, re
import gradio as gr
import supabase
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

MODELS = ['VuAI/khi-van', 'VuAI/tuan', 'VuAI/vi2vi_vn98']

# Initialize the Supabase client
supabase_url = os.getenv("supabase_url")
supabase_token = os.getenv("supabase_token")
client = supabase.create_client(supabase_url, supabase_token)

def inference(input_sentence, name_model):
    # Load model
    access_token = os.getenv("access_token")
    tokenizer = AutoTokenizer.from_pretrained(name_model, use_auth_token=access_token)
    model = AutoModelForSeq2SeqLM.from_pretrained(name_model, use_auth_token=access_token)

    # Load input and encode
    input_ids = tokenizer.encode(input_sentence, return_tensors="pt")

    # Generate translation
    translation = model.generate(input_ids=input_ids, max_length = 512)[0]

    # Decode the generated translation
    translated_text = tokenizer.decode(translation, skip_special_tokens=True)

    return translated_text.replace('▁', ' ')

def translator(input_sentence, name_model):
    start_time = time.time()

    input_sentence = input_sentence.replace('【', '[').replace('】', ']').replace('“', '"').replace('”', '"').replace('." ', '."\n').replace('?" ', '?"\n').replace('!" ', '!"\n').replace(': "', ':\n"').replace(', "', ':\n"')
    input_sentence = input_sentence.split('\n')
    
    result = ''
    percent_complete = 0
    for i in input_sentence:
        if i != '' and i != ' ':
            if i.find('"') == 0 and i.find('"',1) == (len(i)-1):
                str_out = '"' + inference(i.replace('"',''), name_model) + '"'
            elif i.find('[') == 0:
                str_out = '[' + inference(i.replace('[','').replace(']',''), name_model) + ']'
            else:
                str_out = inference(i, name_model)
            # st.sidebar.write(str_out)
            result += str_out + '\n'
            
        percent_complete += 1/len(input_sentence)

    title_str = re.findall('^Chương\s\d+:\s.*\n', result)
    if title_str:
        result = result.replace(title_str[0], title_str[0].title())
    
    end_time = time.time()
    process_time = end_time - start_time
    # status_bar.success('Đã dịch thành công! \n\tThời gian dịch: ' + str(process_time) + ' giây', icon="✅")
    
    return result

def words(input_sentence):
    len_inp = len(input_sentence.replace('\n', ' ').replace('  ', ' ').replace('  ', ' ').strip().split(' '))
    return 'Words: ' + str(len_inp) + ' / 9999'
    
def login(username, password):
    # login
    data = client.auth.sign_in_with_password({"email": username, "password": password})
    return data


    
curr_words = 9999
with gr.Blocks(title="VN98 Translator", theme=gr.themes.Default(primary_hue="red", secondary_hue="red")) as demo:
    with gr.Row():
        with gr.Column(scale=20):
            gr.Text('VN98 Translation', show_label=False).style(container=False)
        with gr.Column(scale=1, min_width=50):
            darkBtn = gr.Button("Dark", variant='secondary', elem_id="darkTheme")
    
    with gr.Row():
        with gr.Column():
            inp = gr.Textbox(lines=5, label="Input Text")
            inp.change(fn=words, inputs=inp, outputs=gr.Text(show_label=False, value='Words: 0 / ' + str(curr_words)))
            sel = gr.Dropdown(label="Select Model?", choices=MODELS, value=MODELS[0])
            btn = gr.Button("Run", variant='primary')
        with gr.Column():
            out = gr.Textbox(lines=5, label="Result", interactive=True).style(show_copy_button=True)

    changeTheme = """function changeTheme() {
        body = document.querySelector('body').classList.toggle('dark');
        btn_theme = document.querySelector('#darkTheme');
        btn_theme.innerText = (btn_theme.innerText == 'Light') ? 'Dark' : 'Light';
    }
    """

    btn.click(translator, inputs=[inp, sel], outputs=[out])
    darkBtn.click(fn=None, _js=changeTheme)


if __name__ == "__main__":
    demo.launch(auth=login, auth_message="VN98 Translator.", max_threads=80)