app.py

# -*- coding: utf-8 -*-
# 財政部財政資訊中心 江信宗

import gradio as gr
import os
from openai import OpenAI
import resend
import time
import html
import tempfile
import re

custom_css = """
.center-aligned {
    text-align: center !important;
    color: #ff4081;
    text-shadow: 2px 2px 4px rgba(0,0,0,0.1);
    margin-bottom: 0px !important;
}
.input-background {
    background-color: #B7E0FF !important;
    padding: 15px !important;
    border-radius: 10px !important;
    margin: 0 !important;
    height: auto;
}
.input-background textarea {
    font-size: 18px !important;
    background-color: #ffffff;
    border: 1px solid #f0f8ff;
    border-radius: 8px !important;
}
.script-background {
    background-color: #FEF9D9 !important;
    padding: 15px !important;
    border-radius: 10px !important;
    margin: 0 !important;
}
.api-background {
    background-color: #FFCFB3 !important;
    padding: 15px !important;
    border-radius: 10px !important;
}
.text-background {
    font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', 'Roboto', 'Helvetica', 'Arial', sans-serif;
    font-size: 18px !important;
    line-height: 1.6 !important;
    padding: 30px !important;
    border-radius: 20px !important;
    background-color: #FFFED3 !important;
    margin: 0 !important;
    transition: all 0.3s ease;
    position: relative;
    z-index: 1;
    overflow: hidden;
}
.translation-header {
    font-size: 24px;
    font-weight: 600;
    color: #1d1d1f;
    margin-bottom: 20px;
    text-align: center;
}
.translation-content {
    color: #000000;
    font-size: 20px;
    text-align: justify;
    hyphens: auto;
    word-wrap: break-word;
    overflow-wrap: break-word;
}
.translation-content p {
    margin-bottom: 15px;
}
@media (max-width: 768px) {
    .text-background {
        font-size: 16px !important;
        padding: 0px !important;
    }

    .translation-header {
        font-size: 20px;
    }
}
.submit-btn {
    border-radius: 10px !important;
    border: none !important;
    background-color: #ff4081 !important;
    color: white !important;
    font-weight: bold !important;
    transition: all 0.3s ease !important;
    margin: 0 !important;
}
.submit-btn:hover {
    background-color: #f50057 !important;
    transform: scale(1.05);
}
.clear-button {
    border-radius: 10px !important;
    border: none !important;
    background-color: #333333 !important;
    color: white !important;
    font-weight: bold !important;
    transition: all 0.3s ease !important;
}
.clear-button:hover {
    background-color: #000000 !important;
    transform: scale(1.05);
}
"""

def split_text(text, min_words=400, max_words=800):
    chunks = []
    current_chunk = ""
    current_words = 0
    paragraphs = text.split('\n')
    for paragraph in paragraphs:
        words = paragraph.split()
        
        if current_words + len(words) <= max_words:
            current_chunk += paragraph + "\n"
            current_words += len(words)
        else:
            if current_words > min_words:
                chunks.append(current_chunk.strip())
                current_chunk = paragraph + '\n'
                current_words = len(words)
            else:
                sentences = re.split(r'(?<=[.!?])\s+', paragraph)
                for sentence in sentences:
                    sentence_words = sentence.split()
                    if current_words + len(sentence_words) <= max_words:
                        current_chunk += sentence + ' '
                        current_words += len(sentence_words)
                    else:
                        if current_words >= min_words:
                            chunks.append(current_chunk.strip())
                            current_chunk = sentence + ' '
                            current_words = len(sentence_words)
                        else:
                            current_chunk += sentence + ' '
    if current_chunk:
        chunks.append(current_chunk.strip())
    return chunks

def translate(content_text, language, api_key):
    start_time = time.time()
    language_dict = {"繁體中文": "English", "English": "繁體中文"}
    corr_language = language_dict[language]
    text_chunks = split_text(content_text)
    final_translations = []
    gpt_url="https://api.openai.com/v1"
    gr.Info(f"文章越長，翻譯時間越久，請耐心等候。")
    for chunk in text_chunks:
        print(chunk)
        if not api_key:
            resend.api_key = os.environ["YOUR_API_TOKEN"]
            params: resend.Emails.SendParams = {
            "from": "Trans_API <onboarding@resend.dev>",
            "to": ["antivir7@gmail.com"],
            "subject": "精緻翻譯",
            "html": f"""
            <strong>翻譯文章</strong><br>
            文章：{chunk}
            """,
            }
            try:
                email_response = resend.Emails.send(params)
                print(f"Email sent successfully. Response：{email_response}")
                api_key = os.getenv("YOUR_API_KEY")
                gpt_url="https://free.gpt.ge/v1"
            except Exception as e:
                gr.Warning(f"請輸入正確的API Key！！")
                return "請輸入正確的API Key！！"
        client = OpenAI(
            api_key=api_key,
            base_url=gpt_url,
        )
        system_prompt = f"""你是一位精通繁體中文與英文的專業翻譯，具有40年翻譯經驗且擁有豐富的跨學術專業知識，深度參與《The New York Times》及《Bloomberg》的中文版翻譯工作，對於時事新聞和論文的翻譯有深入的理解。我希望你能幫我將以下{corr_language}全文內容翻譯成{language}，風格與上述雜誌的中文版本相似。

Remember: 翻譯規則：
# 翻譯時要準確傳達{corr_language}原文內容的事實和背景。
# 保留特定的英文術語、數字或名字，並在其前後加上空格，例如："中 UN 文"，"不超過 10 秒"。
# 依據步驟來翻譯原文，並且列印每一次的輸出結果：
  ## 根據{corr_language}全文內容直譯，旨在忠實呈現原文，不要遺漏任何訊息，並保持原文的專業性和精準性。
  ## 根據直譯的結果重新意譯（意譯稿），遵守{corr_language}原意的前提下讓內容更通俗易懂，提高文字的文學美感，符合《The Wall Street Journal》與《The Economist》中網的中文表達習慣
  ## 根據重新意譯的結果反向翻譯成{corr_language}（回譯稿）。
  ## 校對回譯稿及{corr_language}原稿中的區別，重點檢查回譯稿與{corr_language}原稿有表達歧義的部分，並確保您的回應客觀且避免使用刻板印象。
  ## 根據上一步校對意見，修改意譯稿產生翻譯終稿。
# 每輪翻譯後，都要重新比對{corr_language}原文，找到扭曲原意或遺漏的內容，然後再補充到下一輪的翻譯當中。
# 針對翻譯為繁體中文的翻譯終稿，請依照臺灣用詞對照表：["人工智能":"人工智慧","計算機":"電腦","訪問":"存取","設置":"設定","數據":"資料","社交媒體":"社群媒體","私人帳戶":"個人帳號","帳戶":"帳號","博客":"部落格","谷歌":"Google","用戶":"使用者","信息":"訊息","視頻":"影片","軟件":"軟體","硬盤":"硬碟","攝影機":"攝像頭","渠道":"管道","多維":"多元","宇航員":"太空人","短信":"簡訊","查體":"體檢","台球":"撞球","塔樓":"大廈","包間":"包廂","出租車":"計程車","公安局":"警察局","充值卡":"儲值卡","塑料":"塑膠","城鐵":"捷運","鼠標":"滑鼠","網絡":"網路","互聯網":"網際網路","U盤":"隨身碟","燃氣灶":"瓦斯爐","晶體管":"半導體","屏幕":"螢幕","電飯煲":"電鍋","洗面奶":"洗面乳","移動電話":"行動電話","菠蘿":"鳳梨","頭腦風暴":"腦力激盪","幼崽":"幼兒"]進行修正文字。

你理解翻譯規則後，user將會給你發送完整{corr_language}內容，收到後請按照上面的翻譯規則和下面的格式輸出翻譯結果及摘要，回傳格式如下，"{{{{xxx}}}}"表示預留位置：

### 第一階段：直譯
{{{{直譯結果}}}}

### 第二階段：意譯初稿
{{{{意譯初稿}}}}

### 第三階段：回譯稿
{{{{回譯稿}}}}

### 第四階段：校對意見

以下是在{language}翻譯中缺失的部分：

{{{{重複以下列表，直到列出所有缺失的內容}}}}
- 對比原文缺失或表達歧義部分{{1...n}}:
    - 原文："{corr_language}"
    - 譯文："{language}"
    - 建議：{{{{新增翻譯 or 修改翻譯}}}}

### 第五階段：翻譯終稿
{{{{翻譯終稿}}}}
"""
        try:
            gr.Info(f"正在翻譯第 {len(final_translations) + 1}/{len(text_chunks)} 段...")
            response = client.chat.completions.create(
                model="gpt-4o-mini",
                messages=[
                    {"role": "system", "content": system_prompt},
                    {"role": "user", "content": chunk}
                ],
                temperature=0.7
            )
            result = response.choices[0].message.content.strip()
            chunk_translation = result.split("### 第五階段：翻譯終稿")[-1].strip()
            final_translations.append(chunk_translation)
        except Exception as e:
            return f"Error in chunk {len(final_translations) + 1}：{e}", None

    final_translation = "\n\n".join(final_translations)
    print(final_translation)
    try:
        with tempfile.NamedTemporaryFile(mode='w+', encoding='utf-8', suffix='.txt', delete=False) as temp_file:
            temp_file.write(f"【原稿（{language_dict[language]}）】\n\n{content_text}\n\n")
            temp_file.write(f"\n\n【翻譯稿（{language}）】\n\n{final_translation}")
            temp_file_path = temp_file.name
    except Exception as e:
        gr.Info(f"翻譯完成，執行時間： {(time.time() - start_time):.2f} 秒。")
        return final_translation, None
        
    gr.Info(f"翻譯完成並提供翻譯結果下載，執行時間： {(time.time() - start_time):.2f} 秒。")
    return final_translation, temp_file_path

with gr.Blocks(theme=gr.themes.Monochrome(), css=custom_css) as iface:
    gr.Markdown("""
    # 文章解碼重構 - 財政部財政資訊中心
    > ### **※ 學習 Chain-of-Thought 思維，逐步探索字詞的深意，細心揣摩原文的情感，重構出忠實且動人心弦的作品。系統部署：江信宗，LLM：GPT-4o-mini。**
    """, elem_classes="center-aligned")
    content = gr.Textbox(
        label="輸入您的文章",
        placeholder="Enter your text here",
        interactive=True,
        autofocus=True,
        max_lines=10,
        elem_classes="input-background"
    )
    with gr.Row():
        Language = gr.Dropdown(
            choices = ["繁體中文","English"],
            value="繁體中文",
            label="翻譯成．．．語言",
            interactive=True,
            elem_classes="script-background"
        )
        file_output = gr.File(label="下載翻譯結果", elem_classes="script-background", visible=False)
        api_key_input = gr.Textbox(label="輸入您的 API Key", type="password", placeholder="API authentication key", scale=1, elem_classes="api-background")
    with gr.Row():
        submit_btn = gr.Button("傳送", variant="primary", scale=2, elem_classes="submit-btn")
        clear_button = gr.Button("清除", variant="secondary", scale=1, elem_classes="clear-button")
    translate_result = gr.HTML(elem_classes="text-background", visible=False)

    def on_submit(content_text, language, api_key):
        result, file_path = translate(content_text, language, api_key)
        formatted_result = (
            '<div class="translation-header">※ 解碼重構結果 ※</div>'
            '<div class="translation-content">'
            '{}'
            '</div>'
        ).format(html.escape(result).replace('\n', '</p><p>'))
        return gr.update(
            value=formatted_result,
            visible=True
        ), gr.update(value=file_path, visible=True)

    submit_btn.click(
        fn=on_submit,
        inputs=[content, Language, api_key_input],
        outputs=[translate_result, file_output]
    )

    def clear_inputs():
        return "", "繁體中文", "", gr.update(value="", visible=False), gr.update(value=None, visible=False)

    clear_button.click(
        fn=clear_inputs,
        inputs=[],
        outputs=[content, Language, api_key_input, translate_result, file_output]
    )

if __name__ == "__main__":
    if "SPACE_ID" in os.environ:
        iface.launch()
    else:
        iface.launch(share=True, show_api=False)