app.py

# -*- coding: utf-8 -*-

import gradio as gr
import operator
import torch
import os
from transformers import BertTokenizer, BertForMaskedLM

# 使用私有模型和分詞器
model_name_or_path = "DeepLearning101/Corrector101zhTW"
# auth_token = os.getenv("Corrector101zhTW")  # 從環境變量中獲取 token

# tokenizer = BertTokenizer.from_pretrained(model_name_or_path, use_auth_token=auth_token)
# model = BertForMaskedLM.from_pretrained(model_name_or_path, use_auth_token=auth_token)
# tokenizer = BertTokenizer.from_pretrained(model_name_or_path)
# model = BertForMaskedLM.from_pretrained(model_name_or_path)

model_name_or_path = "DeepLearning101/Corrector101zhTW"

try:
    tokenizer = BertTokenizer.from_pretrained(model_name_or_path)
    model = BertForMaskedLM.from_pretrained(model_name_or_path)
except Exception as e:
    

def ai_text(text):
    with torch.no_grad():
        outputs = model(**tokenizer([text], padding=True, return_tensors='pt'))

    def to_highlight(corrected_sent, errs):
        output = [{"entity": "糾錯", "word": err[1], "start": err[2], "end": err[3]} for i, err in
                  enumerate(errs)]
        return {"text": corrected_sent, "entities": output}

    def get_errors(corrected_text, origin_text):
        sub_details = []
        for i, ori_char in enumerate(origin_text):
            if ori_char in [' ', '“', '”', '‘', '’', '琊', '\n', '…', '—', '擤']:
                # add unk word
                corrected_text = corrected_text[:i] + ori_char + corrected_text[i:]
                continue
            if i >= len(corrected_text):
                continue
            if ori_char != corrected_text[i]:
                if ori_char.lower() == corrected_text[i]:
                    # pass english upper char
                    corrected_text = corrected_text[:i] + ori_char + corrected_text[i + 1:]
                    continue
                sub_details.append((ori_char, corrected_text[i], i, i + 1))
        sub_details = sorted(sub_details, key=operator.itemgetter(2))
        return corrected_text, sub_details

    _text = tokenizer.decode(torch.argmax(outputs.logits[0], dim=-1), skip_special_tokens=True).replace(' ', '')
    corrected_text = _text[:len(text)]
    corrected_text, details = get_errors(corrected_text, text)
    print(text, ' => ', corrected_text, details)
    return corrected_text + ' ' + str(details)


if __name__ == '__main__':

    examples = [
        ['你究輸入利的手機門號跟生分證就可以了。'],
        ['這裡是客服中新，很高性為您服物，請問金天有什麼須要幫忙'],
        ['因為我們這邊是按天術比例計蒜給您的，其實不會有態大的穎響。也就是您用前面的資非的廢率來做計算'],
        ['我來看以下，他的時價是多少？起實您就可以直皆就不用到門事'],
        ['因為你現在月富是六九九嘛，我幫擬減衣百塊，兒且也不會江速'],
    ]

    inputs=[gr.Textbox(lines=2, label="欲校正的文字")],
    outputs=[gr.Textbox(lines=2, label="修正後的文字")],
    gr.Interface(
        inputs='text',
        outputs='text',
    title="客服ASR文本AI糾錯系統",
    description="""
    <a href="https://www.twman.org" target='_blank'>TonTon Huang Ph.D. @ 2024/04 </a><br>
    輸入ASR文本，糾正同音字/詞錯誤<br>
    Masked Language Model (MLM) as correction BERT  
    """,    examples=examples
    ).launch()