app.py

import os
import sys

# 1. 強制下載 Spacy 模型
os.system(f"{sys.executable} -m spacy download en_core_web_sm")

# 2. 解決相容性問題的補丁
try:
    import huggingface_hub
    if not hasattr(huggingface_hub, 'HfFolder'):
        class MockHfFolder:
            @staticmethod
            def get_token(): return os.getenv("HF_TOKEN")
            @staticmethod
            def save_token(token): pass
        huggingface_hub.HfFolder = MockHfFolder
except:
    pass

import gradio as gr
from fastcoref import FCoref
from deep_translator import GoogleTranslator

# 關鍵修正：將 model_name 改為 model_name_or_path
print("🚀 [System] 正在進行阻塞式權重同步（約 300MB）...")
try:
    # 使用相容性最高的參數名稱
    model = FCoref(model_name_or_path='biu-nlp/f-coref', device='cpu')
    print("✅ [System] 模型載入完成，介面即將開啟。")
except TypeError:
    # 萬一連上面的都不行，就用最原始的 positional argument
    model = FCoref('biu-nlp/f-coref', device='cpu')
    print("✅ [System] 使用原始參數模式載入完成。")

def coref_chat(user_input):
    if not user_input.strip():
        return "請輸入內容", "等待輸入..."
    
    try:
        # 1. 偵測語系並統一轉換為英文供模型運算
        has_chinese = any('\u4e00' <= char <= '\u9fff' for char in user_input)
        
        if has_chinese:
            working_text = GoogleTranslator(source='zh-CN', target='en').translate(user_input)
            mode_notice = "【模式：中文 ➔ 英文解析】"
        else:
            working_text = user_input
            mode_notice = "【模式：純英文解析】"

        # 2. 執行指代消解
        preds = model.predict(texts=[working_text])
        clusters = preds[0].get_clusters()
        
        # 3. 整理輸出格式 (嚴格按照你的要求排版)
        result = f"✨ {mode_notice}\n"
        result += f"📝 英文邏輯空間: {working_text}\n"
        
        # --- 英文翻譯中文 ---
        try:
            translation_back = GoogleTranslator(source='en', target='zh-TW').translate(working_text)
            result += f"📖 中文句子: {translation_back}\n"
        except:
            result += f"📖 中文句子: (翻譯暫時無法讀取)\n"
            
        result += "---------------------------------\n"
        
        if not clusters:
            result += "🔍 分析結果：指代關係明確，或模型判定關聯度未達門檻。"
        else:
            result += "🎯【偵測到之實體鏈 (Entity Chains)】:\n"
            for i, cluster in enumerate(clusters):
                cluster_str_en = ' ↔ '.join(cluster)
                # 實體鏈個別翻譯
                try:
                    translated_items = [GoogleTranslator(source='en', target='zh-TW').translate(item) for item in cluster]
                    cluster_str_zh = ' ↔ '.join(translated_items)
                except:
                    cluster_str_zh = "(鏈結翻譯失敗)"
                
                result += f" 🔗 鏈結 {i+1} (繁中): {cluster_str_zh}\n"
                result += f"    └─ (原文): {cluster_str_en}\n"

        return user_input, result

    except Exception as e:
        return user_input, f"運行錯誤: {str(e)}"

# 介面設定
demo = gr.Interface(
    fn=coref_chat,
    inputs=gr.Textbox(label="輸入文本", lines=3, placeholder="輸入中文或英文段落..."),
    outputs=[gr.Textbox(label="原始輸入"), gr.Textbox(label="AI 語意分析報告")],
    title="AI 跨語言指代消解系統 (Stable Version)"
)

if __name__ == "__main__":
    demo.launch()