Update app.py

app.py

@@ -2,260 +2,480 @@ import gradio as gr
 import requests
 import json
 import os
+import re
+import sqlite3
+import pandas as pd
+from datetime import datetime
 from datasets import load_dataset
 from sentence_transformers import SentenceTransformer, util
 import torch
 from huggingface_hub import hf_hub_download
-import re
+from typing import List, Dict, Tuple, Optional
 
-# --- 配置區 ---
-HF_TOKEN = os.environ.get("HF_TOKEN") 
+# ==================== 配置區 ====================
+HF_TOKEN = os.environ.get("HF_TOKEN", "您的_HuggingFace_Token")
 DATASET_REPO_ID = "Paul720810/Text-to-SQL-Softline"
-SIMILARITY_THRESHOLD = 0.75  # 進一步降低閾值
+SIMILARITY_THRESHOLD = 0.75  # 相似度閾值
 
-# 多個備用模型（保證至少有一個可用）
+# 多個備用LLM模型（保證可用性）
 LLM_MODELS = [
-    "https://api-inference.huggingface.co/models/gpt2",  # 最基礎的模型，保證可用
-    "https://api-inference.huggingface.co/models/distilgpt2",
+    "https://api-inference.huggingface.co/models/gpt2",
+    "https://api-inference.huggingface.co/models/distilgpt2", 
     "https://api-inference.huggingface.co/models/microsoft/DialoGPT-small"
 ]
 
-print("--- [1/5] 開始初始化應用 ---")
+# 數據庫連接配置（可選）
+DB_CONFIG = {
+    "enabled": False,  # 設置為True啟用真實數據庫連接
+    "path": "您的數據庫路徑.db",
+    "test_queries": True  # 是否啟用SQL測試功能
+}
+
+print("=" * 50)
+print("🚀 智能 Text-to-SQL 系統啟動中...")
+print("=" * 50)
 
-# --- 1. 載入知識庫 ---
-questions = []
-sql_answers = []
-schema_data = {}
+# ==================== 工具函數 ====================
+def get_current_time():
+    """獲取當前時間字符串"""
+    return datetime.now().strftime("%Y-%m-%d %H:%M:%S")
 
-try:
-    print(f"--- [2/5] 正在從 '{DATASET_REPO_ID}' 載入知識庫... ---")
-    raw_dataset = load_dataset(DATASET_REPO_ID, token=HF_TOKEN)['train']
+def safe_json_load(data, default=None):
+    """安全的JSON解析"""
+    try:
+        return json.loads(data) if isinstance(data, str) else data
+    except (json.JSONDecodeError, TypeError):
+        return default
+
+def validate_sql(sql_query: str) -> Dict:
+    """驗證SQL語句的安全性"""
+    security_issues = []
     
-    # 解析 messages 格式
-    print("--- > 解析 messages 格式...")
+    # 檢查危險操作
+    dangerous_keywords = ['DROP', 'DELETE', 'INSERT', 'UPDATE', 'ALTER', 'TRUNCATE', 'EXEC', 'EXECUTE']
+    for keyword in dangerous_keywords:
+        if f" {keyword} " in sql_query.upper():
+            security_issues.append(f"發現危險操作: {keyword}")
     
-    for item in raw_dataset:
-        try:
-            if 'messages' in item and len(item['messages']) >= 2:
-                user_content = item['messages'][0]['content']
-                assistant_content = item['messages'][1]['content']
-                
-                # 提取問題
-                question_match = re.search(r'指令:\s*(.*?)(?:\n|$)', user_content)
-                question = question_match.group(1).strip() if question_match else user_content
-                
-                # 提取SQL
-                sql_match = re.search(r'SQL查詢:\s*(.*?)(?:\n|$)', assistant_content, re.DOTALL)
-                if sql_match:
-                    sql_query = sql_match.group(1).strip()
-                    sql_query = re.sub(r'^sql\s*', '', sql_query)
-                    sql_query = re.sub(r'```sql|```', '', sql_query).strip()
-                else:
-                    sql_query = assistant_content
-                
-                questions.append(question)
-                sql_answers.append(sql_query)
-                
-        except Exception as e:
-            continue
+    # 檢查基本語法
+    if "SELECT" not in sql_query.upper():
+        security_issues.append("缺少SELECT語句")
     
-    print(f"--- > 成功解析 {len(questions)} 條問答範例 ---")
+    if "FROM" not in sql_query.upper():
+        security_issues.append("缺少FROM子句")
     
-    # 載入Schema
-    try:
-        schema_file_path = hf_hub_download(
-            repo_id=DATASET_REPO_ID, 
-            filename="sqlite_schema_FULL.json", 
-            repo_type='dataset', 
-            token=HF_TOKEN
-        )
-        with open(schema_file_path, 'r', encoding='utf-8') as f:
-            schema_data = json.load(f)
-    except Exception as e:
-        print(f"警告: 無法載入Schema文件: {e}")
-
-except Exception as e:
-    print(f"錯誤: 載入數據集失敗: {e}")
-    questions = ["示例問題"]
-    sql_answers = ["SELECT '系統就緒' AS status;"]
-
-# --- 2. 初始化檢索模型 ---
-print("--- [3/5] 正在載入句向量模型... ---")
-embedder = SentenceTransformer('all-MiniLM-L6-v2', device='cpu')
-
-# 計算問題向量
-if questions:
-    print(f"--- [4/5] 正在為 {len(questions)} 個問題計算向量... ---")
-    question_embeddings = embedder.encode(questions, convert_to_tensor=True, show_progress_bar=False)  # 關閉進度條
-    print("--- > 向量計算完���！ ---")
-else:
-    print("--- [4/5] 警告：沒有可用的問題 ---")
-    question_embeddings = torch.Tensor([])
+    return {
+        "valid": len(security_issues) == 0,
+        "issues": security_issues,
+        "is_safe": len([i for i in security_issues if '危險' in i]) == 0
+    }
 
-# --- 3. 構建DDL ---
-def build_schema_context(schema_dict):
-    if not schema_dict:
-        return "/* 無Schema信息 */"
+def execute_test_query(sql_query: str) -> Tuple[bool, str]:
+    """執行測試查詢（可選功能）"""
+    if not DB_CONFIG["enabled"]:
+        return False, "數據庫連接未啟用"
     
-    context = "/* 數據庫表結構 */\n"
-    for table_name, columns in schema_dict.items():
-        if isinstance(columns, list):
-            context += f"\n-- 表: {table_name}\n"
-            for col in columns:
-                col_name = col.get('name', 'unknown')
-                col_type = col.get('type', 'TEXT')
-                col_desc = col.get('description', '')
-                context += f"--   {col_name} ({col_type}) - {col_desc}\n"
-    return context
-
-SCHEMA_CONTEXT = build_schema_context(schema_data)
+    try:
+        validation = validate_sql(sql_query)
+        if not validation["is_safe"]:
+            return False, f"SQL安全檢查失敗: {', '.join(validation['issues'])}"
+        
+        # 連接數據庫並執行
+        conn = sqlite3.connect(DB_CONFIG["path"])
+        df = pd.read_sql_query(sql_query, conn)
+        conn.close()
+        
+        if len(df) == 0:
+            return True, "✅ SQL執行成功，但返回0條數據\n💡 可能原因: 條件太嚴格或數據不存在"
+        else:
+            sample_info = f"✅ SQL執行成功，返回 {len(df)} 條數據\n"
+            sample_info += f"📊 前3條數據:\n{df.head(3).to_string()}"
+            return True, sample_info
+            
+    except Exception as e:
+        return False, f"❌ SQL執行錯誤: {str(e)}"
 
-# --- 4. LLM調用函數（多模型備用）---
-def call_llm_api(prompt, model_urls=LLM_MODELS):
-    headers = {"Authorization": f"Bearer {HF_TOKEN}"}
-    payload = {
-        "inputs": prompt,
-        "parameters": {
-            "max_new_tokens": 150,
-            "temperature": 0.1,
-            "do_sample": False
-        }
-    }
-    
-    # 嘗試所有備用模型
-    for model_url in model_urls:
+# ==================== 數據加載模塊 ====================
+class DataLoader:
+    def __init__(self, hf_token: str):
+        self.hf_token = hf_token
+        self.questions = []
+        self.sql_answers = []
+        self.schema_data = {}
+        
+    def load_dataset(self) -> bool:
+        """加載問答數據集"""
         try:
-            response = requests.post(model_url, headers=headers, json=payload, timeout=15)
+            print(f"[{get_current_time()}] 正在加載數據集 '{DATASET_REPO_ID}'...")
+            raw_dataset = load_dataset(DATASET_REPO_ID, token=self.hf_token)['train']
             
-            if response.status_code == 200:
-                result = response.json()
-                if isinstance(result, list) and len(result) > 0:
-                    return result[0]['generated_text'].strip()
+            print("正在解析 messages 格式...")
+            for item in raw_dataset:
+                try:
+                    if 'messages' in item and len(item['messages']) >= 2:
+                        user_content = item['messages'][0]['content']
+                        assistant_content = item['messages'][1]['content']
+                        
+                        # 提取問題
+                        question_match = re.search(r'指令:\s*(.*?)(?:\n|$)', user_content)
+                        question = question_match.group(1).strip() if question_match else user_content
+                        
+                        # 提取SQL
+                        sql_match = re.search(r'SQL查詢:\s*(.*?)(?:\n|$)', assistant_content, re.DOTALL)
+                        if sql_match:
+                            sql_query = sql_match.group(1).strip()
+                            sql_query = re.sub(r'^sql\s*', '', sql_query)
+                            sql_query = re.sub(r'```sql|```', '', sql_query).strip()
+                        else:
+                            sql_query = assistant_content
+                        
+                        self.questions.append(question)
+                        self.sql_answers.append(sql_query)
+                        
+                except Exception as e:
+                    continue
             
+            print(f"成功解析 {len(self.questions)} 條問答範例")
+            return True
+            
+        except Exception as e:
+            print(f"數據集加載失敗: {e}")
+            self.questions = ["系統初始化問題"]
+            self.sql_answers = ["SELECT '數據庫連接就緒' AS status;"]
+            return False
+    
+    def load_schema(self) -> bool:
+        """加載數據庫Schema"""
+        try:
+            schema_file_path = hf_hub_download(
+                repo_id=DATASET_REPO_ID, 
+                filename="sqlite_schema_FULL.json", 
+                repo_type='dataset', 
+                token=self.hf_token
+            )
+            with open(schema_file_path, 'r', encoding='utf-8') as f:
+                self.schema_data = safe_json_load(f.read(), {})
+            print("Schema加載成功")
+            return True
         except Exception as e:
-            print(f"模型 {model_url} 失敗: {e}")
-            continue
+            print(f"Schema加載失敗: {e}")
+            self.schema_data = {}
+            return False
     
-    return None  # 所有模型都失敗
+    def build_schema_context(self) -> str:
+        """構建Schema上下文"""
+        if not self.schema_data:
+            return "/* 無Schema信息 */"
+        
+        context = "/* 數據庫表結構 */\n"
+        for table_name, columns in self.schema_data.items():
+            if isinstance(columns, list):
+                context += f"\n-- 表: {table_name}\n"
+                for col in columns:
+                    col_name = col.get('name', 'unknown')
+                    col_type = col.get('type', 'TEXT')
+                    col_desc = col.get('description', '')
+                    context += f"--   {col_name} ({col_type}) - {col_desc}\n"
+        return context
 
-# --- 5. 核心邏輯 ---
-def get_sql_query(user_question: str):
-    if not user_question:
-        return "請輸入您的問題。", "日誌：用戶未輸入問題。"
+# ==================== LLM模塊 ====================
+class LLMClient:
+    def __init__(self, hf_token: str):
+        self.hf_token = hf_token
     
-    log_messages = []
+    def call_llm_api(self, prompt: str, model_urls: List[str] = LLM_MODELS) -> Optional[str]:
+        """調用LLM API（多模型備用）"""
+        headers = {"Authorization": f"Bearer {self.hf_token}"}
+        payload = {
+            "inputs": prompt,
+            "parameters": {
+                "max_new_tokens": 200,
+                "temperature": 0.1,
+                "do_sample": False
+            }
+        }
+        
+        for model_url in model_urls:
+            try:
+                response = requests.post(model_url, headers=headers, json=payload, timeout=20)
+                
+                if response.status_code == 200:
+                    result = response.json()
+                    if isinstance(result, list) and len(result) > 0:
+                        generated_text = result[0]['generated_text'].strip()
+                        # 清理輸出
+                        generated_text = re.sub(r'^```sql|```$', '', generated_text).strip()
+                        return generated_text
+                
+            except Exception as e:
+                print(f"模型 {model_url} 調用失敗: {e}")
+                continue
+        
+        return None
+
+# ==================== 檢索模塊 ====================
+class RetrievalSystem:
+    def __init__(self):
+        self.embedder = SentenceTransformer('all-MiniLM-L6-v2', device='cpu')
+        self.question_embeddings = None
     
-    # 1. 首先嘗試檢索
-    if len(questions) > 0 and len(question_embeddings) > 0:
+    def compute_embeddings(self, questions: List[str]) -> None:
+        """計算問題向量"""
+        if questions:
+            print(f"正在為 {len(questions)} 個問題計算向量...")
+            self.question_embeddings = self.embedder.encode(questions, convert_to_tensor=True, show_progress_bar=False)
+            print("向量計算完成")
+        else:
+            self.question_embeddings = torch.Tensor([])
+    
+    def retrieve_similar(self, user_question: str, top_k: int = 3) -> List[Dict]:
+        """檢索相似問題"""
+        if self.question_embeddings is None or len(self.question_embeddings) == 0:
+            return []
+        
         try:
-            question_embedding = embedder.encode(user_question, convert_to_tensor=True)
-            hits = util.semantic_search(question_embedding, question_embeddings, top_k=3)
+            question_embedding = self.embedder.encode(user_question, convert_to_tensor=True)
+            hits = util.semantic_search(question_embedding, self.question_embeddings, top_k=top_k)
+            return hits[0] if hits and hits[0] else []
+        except Exception as e:
+            print(f"檢索失敗: {e}")
+            return []
+
+# ==================== 主系統 ====================
+class TextToSQLSystem:
+    def __init__(self, hf_token: str):
+        self.hf_token = hf_token
+        self.data_loader = DataLoader(hf_token)
+        self.llm_client = LLMClient(hf_token)
+        self.retrieval_system = RetrievalSystem()
+        
+        # 初始化組件
+        self.initialize_system()
+    
+    def initialize_system(self):
+        """初始化系統組件"""
+        print("正在初始化系統組件...")
+        
+        # 加載數據
+        self.data_loader.load_dataset()
+        self.data_loader.load_schema()
+        
+        # 初始化檢索系統
+        self.retrieval_system.compute_embeddings(self.data_loader.questions)
+        
+        self.schema_context = self.data_loader.build_schema_context()
+        print("系統初始化完成")
+    
+    def generate_sql(self, user_question: str) -> Tuple[str, str]:
+        """生成SQL查詢（主函數）"""
+        log_messages = [f"🕒 開始處理: {get_current_time()}"]
+        
+        if not user_question or user_question.strip() == "":
+            return "請輸入您的問題。", "錯誤: 問題為空"
+        
+        # 1. 嘗試檢索相似問題
+        if len(self.data_loader.questions) > 0:
+            hits = self.retrieval_system.retrieve_similar(user_question)
             
-            if hits and hits[0]:
-                best_hit = hits[0][0]
+            if hits:
+                best_hit = hits[0]
                 similarity_score = best_hit['score']
-                similar_question = questions[best_hit['corpus_id']]
+                similar_question = self.data_loader.questions[best_hit['corpus_id']]
                 
-                log_messages.append(f"檢索到相似問題: '{similar_question}' (相似度: {similarity_score:.3f})")
+                log_messages.append(f"🔍 檢索到相似問題: '{similar_question}'")
+                log_messages.append(f"📊 相似度: {similarity_score:.3f}")
                 
                 if similarity_score > SIMILARITY_THRESHOLD:
-                    sql_result = sql_answers[best_hit['corpus_id']]
-                    log_messages.append(f"相似度 > {SIMILARITY_THRESHOLD}，直接返回")
+                    sql_result = self.data_loader.sql_answers[best_hit['corpus_id']]
+                    log_messages.append(f"✅ 相似度 > {SIMILARITY_THRESHOLD}，直接返回預先SQL")
+                    
+                    # 驗證SQL安全性
+                    validation = validate_sql(sql_result)
+                    if not validation["is_safe"]:
+                        log_messages.append(f"⚠️ 安全警告: {', '.join(validation['issues'])}")
+                    
                     return sql_result, "\n".join(log_messages)
                 else:
-                    log_messages.append(f"相似度低於閾值 {SIMILARITY_THRESHOLD}")
+                    log_messages.append(f"ℹ️ 相似度低於閾值 {SIMILARITY_THRESHOLD}")
+        
+        # 2. LLM生成模式
+        log_messages.append("🤖 進入LLM生成模式...")
+        
+        prompt = self.build_llm_prompt(user_question)
+        generated_sql = self.llm_client.call_llm_api(prompt)
+        
+        if generated_sql:
+            # 清理和驗證生成的SQL
+            generated_sql = re.sub(r'^```sql|```$', '', generated_sql).strip()
+            validation = validate_sql(generated_sql)
+            
+            if validation["valid"]:
+                log_messages.append("✅ LLM生成成功")
+                if validation["issues"]:
+                    log_messages.append(f"ℹ️ 驗證提示: {', '.join(validation['issues'])}")
             else:
-                log_messages.append("檢索失敗：找不到相似問題")
+                log_messages.append("⚠️ LLM生成可能存在问题")
                 
-        except Exception as e:
-            log_messages.append(f"檢索過程出錯: {e}")
-    
-    # 2. 檢索失敗或相似度低，嘗試LLM
-    log_messages.append("嘗試LLM生成...")
+            return generated_sql, "\n".join(log_messages)
+        else:
+            # 3. 備用方案
+            log_messages.append("❌ 所有LLM模型都失敗，啟用備用方案")
+            backup_sql = self.generate_backup_sql(user_question)
+            return backup_sql, "\n".join(log_messages)
     
-    # 構建簡單提示詞
-    prompt = f"""請為這個問題生成SQL查詢：
+    def build_llm_prompt(self, user_question: str) -> str:
+        """構建LLM提示詞"""
+        return f"""你是一個SQL專家。請根據以下數據庫結構生成SQL查詢。
+
+{self.schema_context}
 
-數據庫結構：
-{SCHEMA_CONTEXT}
+請為以下問題生成準確的SQL查詢：
+{user_question}
 
-問題：{user_question}
+要求：
+1. 只輸出SQL語句
+2. 不要任何解釋
+3. 使用正確的語法
 
-SQL："""
+SQL查詢："""
     
-    generated_sql = call_llm_api(prompt)
+    def generate_backup_sql(self, user_question: str) -> str:
+        """生成備用SQL"""
+        user_question_lower = user_question.lower()
+        
+        if any(kw in user_question_lower for kw in ['銷售', '業績', '金額', '收入']):
+            return "SELECT product_name, SUM(sales_amount) as total_sales FROM sales GROUP BY product_name ORDER BY total_sales DESC LIMIT 10;"
+        elif any(kw in user_question_lower for kw in ['客戶', '買家', '用戶']):
+            return "SELECT customer_name, COUNT(*) as order_count, SUM(order_amount) as total_spent FROM orders GROUP BY customer_name ORDER BY total_spent DESC;"
+        elif any(kw in user_question_lower for kw in ['時間', '日期', '最近', '月份']):
+            return "SELECT strftime('%Y-%m', order_date) as month, COUNT(*) as orders, SUM(order_amount) as revenue FROM orders WHERE order_date >= date('now', '-6 months') GROUP BY month ORDER BY month DESC;"
+        elif any(kw in user_question_lower for kw in ['產品', '商品', '項目']):
+            return "SELECT product_name, category, stock_quantity, price FROM products WHERE stock_quantity > 0 ORDER BY price DESC;"
+        else:
+            return "SELECT '請重試或提供更詳細的問題' AS status;"
+
+# ==================== 初始化系統 ====================
+print("正在初始化Text-to-SQL系統...")
+text_to_sql_system = TextToSQLSystem(HF_TOKEN)
+
+# ==================== Gradio界面 ====================
+def process_query(user_question: str, test_query: bool = False) -> Tuple[str, str, str]:
+    """處理用戶查詢"""
+    sql_result, log_message = text_to_sql_system.generate_sql(user_question)
+    
+    # SQL調試信息
+    debug_info = ""
+    validation = validate_sql(sql_result)
     
-    if generated_sql:
-        # 清理輸出
-        generated_sql = re.sub(r'^```sql|```$', '', generated_sql).strip()
-        log_messages.append("LLM生成成功！")
-        return generated_sql, "\n".join(log_messages)
+    if not validation["valid"]:
+        debug_info = "❌ SQL驗證失敗:\n" + "\n".join(validation["issues"])
     else:
-        # 3. LLM也失敗，提供智能備用答案
-        log_messages.append("所有LLM模型都失敗，提供備用答案")
+        debug_info = "✅ SQL語法驗證通過"
         
-        # 基於問題內容提供有意義的備用SQL
-        if any(keyword in user_question.lower() for keyword in ['銷售', '業績', '金額']):
-            backup_sql = "SELECT product_name, SUM(sales_amount) as total_sales FROM sales GROUP BY product_name ORDER BY total_sales DESC LIMIT 10;"
-        elif any(keyword in user_question.lower() for keyword in ['客戶', '買家', '用戶']):
-            backup_sql = "SELECT customer_name, COUNT(*) as order_count FROM orders GROUP BY customer_name ORDER BY order_count DESC;"
-        elif any(keyword in user_question.lower() for keyword in ['時間', '日期', '最近']):
-            backup_sql = "SELECT DATE(order_date) as day, COUNT(*) as orders FROM orders WHERE order_date >= DATE('now', '-7 days') GROUP BY day ORDER BY day DESC;"
-        else:
-            backup_sql = "SELECT '請重試或聯繫管理員' AS status;"
+        if validation["issues"]:
+            debug_info += "\nℹ️ 提示: " + ", ".join(validation["issues"])
         
-        return backup_sql, "\n".join(log_messages)
-
-# --- 6. 創建界面 ---
-print("--- [5/5] 正在創建 Web 界面... ---")
+        # 如果啟用測試功能
+        if test_query and DB_CONFIG["test_queries"]:
+            success, test_result = execute_test_query(sql_result)
+            debug_info += f"\n\n🔧 測試結果:\n{test_result}"
+    
+    return sql_result, debug_info, log_message
 
-with gr.Blocks(title="智能Text-to-SQL系統") as demo:
-    gr.Markdown("# 🚀 智能 Text-to-SQL 系統")
-    gr.Markdown("輸入自然語言問題，自動生成SQL查詢")
+# 創建界面
+with gr.Blocks(
+    title="智能Text-to-SQL系統",
+    theme=gr.themes.Soft(),
+    css="""
+    .gradio-container { max-width: 1000px; margin: 0 auto; }
+    .success { color: green; }
+    .warning { color: orange; }
+    .error { color: red; }
+    """
+) as demo:
     
-    with gr.Row():
-        question_input = gr.Textbox(
-            label="您的問題", 
-            placeholder="例如：查詢2024年的銷售數據", 
-            lines=2
-        )
+    gr.Markdown("# 🚀 智能 Text-to-SQL 系統")
+    gr.Markdown("輸入自然語言問題，自動生成並驗證SQL查詢")
     
     with gr.Row():
-        submit_btn = gr.Button("生成SQL", variant="primary")
+        with gr.Column(scale=3):
+            question_input = gr.Textbox(
+                label="📝 您的問題",
+                placeholder="例如：查詢2024年銷售額最高的產品",
+                lines=2,
+                max_lines=4
+            )
+            
+            with gr.Row():
+                submit_btn = gr.Button("🚀 生成SQL", variant="primary")
+                test_btn = gr.Button("🔧 測試SQL", variant="secondary")
+                clear_btn = gr.Button("🗑️ 清除", variant="secondary")
     
     with gr.Row():
         sql_output = gr.Code(
-            label="生成的SQL", 
+            label="📊 生成的SQL",
             language="sql",
-            lines=6
+            lines=6,
+            interactive=True
+        )
+    
+    with gr.Row():
+        debug_output = gr.Textbox(
+            label="🔍 SQL調試信息",
+            lines=4,
+            interactive=False
         )
     
     with gr.Row():
         log_output = gr.Textbox(
-            label="執行日誌", 
+            label="📋 執行日誌",
             lines=4,
             interactive=False
         )
     
-    # 綁定事件
+    # 示例問題
+    gr.Examples(
+        examples=[
+            "2024年銷售額最高的5個產品",
+            "最近30天每個客戶的訂單數量",
+            "庫存不足的商品列表",
+            "比較2023年和2024年的月度銷售額",
+            "付款不及時的客戶統計"
+        ],
+        inputs=question_input,
+        label="💡 示例問題"
+    )
+    
+    # 事件處理
     submit_btn.click(
-        fn=get_sql_query,
+        fn=lambda q: process_query(q, False),
         inputs=question_input,
-        outputs=[sql_output, log_output]
+        outputs=[sql_output, debug_output, log_output]
     )
     
-    # 示例
-    gr.Examples(
-        examples=[
-            "2024年銷售額最高的產品",
-            "最近30天的訂單統計",
-            "每個客戶的訂單數量",
-            "庫存不足的商品列表"
-        ],
-        inputs=question_input
+    test_btn.click(
+        fn=lambda q: process_query(q, True),
+        inputs=question_input,
+        outputs=[sql_output, debug_output, log_output]
+    )
+    
+    clear_btn.click(
+        fn=lambda: ["", "", ""],
+        inputs=[],
+        outputs=[sql_output, debug_output, log_output]
     )
 
-print("--- 應用啟動完成 ---")
-print("--- 訪問地址: http://localhost:7860 ---")
+# ==================== 啟動應用 ====================
 if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", server_port=7860, share=False)
+    print("=" * 50)
+    print("🌐 啟動Gradio Web界面...")
+    print("📍 本地訪問: http://localhost:7860")
+    print("🔄 如果需要公網訪問，設置 share=True")
+    print("=" * 50)
+    
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+        share=False,
+        show_error=True
+    )