daily-news-app/app/services/llm_service.py

"""
LLM 服務模組
支援 Google Gemini、OpenAI、Ollama
"""
import time
from typing import Optional
import httpx

from app.core.config import settings


def get_llm_client():
    """取得 LLM 客戶端"""
    provider = settings.llm_provider
    
    if provider == "gemini":
        import google.generativeai as genai
        genai.configure(api_key=settings.gemini_api_key)
        return genai
    elif provider == "openai":
        from openai import OpenAI
        return OpenAI(api_key=settings.openai_api_key)
    elif provider == "ollama":
        return None  # 使用 httpx 直接呼叫
    
    raise ValueError(f"不支援的 LLM 提供者: {provider}")


def generate_summary(group, articles: list) -> str:
    """
    產生 AI 摘要
    
    Args:
        group: 群組物件（包含 ai_background, ai_prompt）
        articles: 新聞文章列表
    
    Returns:
        綜合摘要文字
    """
    if not articles:
        return "無相關新聞可供摘要。"
    
    # 組合文章內容
    articles_text = ""
    for i, article in enumerate(articles, 1):
        articles_text += f"""
---
新聞 {i}：{article.title}
來源：{article.source.name if article.source else '未知'}
內容：{article.content[:1000] if article.content else article.summary or '無內容'}
---
"""
    
    # 建立 Prompt
    system_prompt = f"""你是一位專業的產業分析師，負責彙整每日新聞並產出精闘的綜合分析報告。

背景資訊：
{group.ai_background or '無特定背景資訊'}

摘要方向：
{group.ai_prompt or '請綜合分析以下新聞的重點、趨勢與潛在影響。'}
"""
    
    user_prompt = f"""請根據以下 {len(articles)} 則新聞，產出一份繁體中文的綜合分析報告：

{articles_text}

請注意：
1. 使用繁體中文
2. 整合相關主題，避免逐條列舉
3. 突出重要趨勢與影響
4. 控制在 500 字以內
"""
    
    provider = settings.llm_provider
    
    try:
        if provider == "gemini":
            import google.generativeai as genai
            genai.configure(api_key=settings.gemini_api_key)
            model = genai.GenerativeModel(settings.gemini_model or "gemini-1.5-pro")
            response = model.generate_content(
                f"{system_prompt}\n\n{user_prompt}",
                generation_config={
                    "temperature": 0.7,
                    "max_output_tokens": 2048,
                    "top_p": 0.95,
                    "top_k": 40
                }
            )
            return response.text
        
        elif provider == "openai":
            from openai import OpenAI
            client = OpenAI(api_key=settings.openai_api_key)
            response = client.chat.completions.create(
                model=settings.openai_model or "gpt-4o",
                messages=[
                    {"role": "system", "content": system_prompt},
                    {"role": "user", "content": user_prompt}
                ],
                max_tokens=2048,
                temperature=0.7
            )
            return response.choices[0].message.content
        
        elif provider == "ollama":
            response = httpx.post(
                f"{settings.ollama_endpoint}/api/generate",
                json={
                    "model": settings.ollama_model or "llama3",
                    "prompt": f"{system_prompt}\n\n{user_prompt}",
                    "stream": False,
                    "options": {
                        "temperature": 0.7,
                        "num_predict": 2048,
                        "top_p": 0.9,
                        "top_k": 40
                    }
                },
                timeout=120
            )
            return response.json().get("response", "")
        
    except Exception as e:
        return f"摘要產生失敗：{str(e)}"


def test_llm_connection(provider: str, model: str) -> dict:
    """
    測試 LLM 連線
    
    Returns:
        {"success": bool, "response_time_ms": int, "message": str}
    """
    start_time = time.time()
    
    try:
        if provider == "gemini":
            import google.generativeai as genai
            genai.configure(api_key=settings.gemini_api_key)
            gen_model = genai.GenerativeModel(model)
            response = gen_model.generate_content(
                "Hello",
                generation_config={"max_output_tokens": 10}
            )
            elapsed = int((time.time() - start_time) * 1000)
            return {"success": True, "response_time_ms": elapsed}
        
        elif provider == "openai":
            from openai import OpenAI
            client = OpenAI(api_key=settings.openai_api_key)
            response = client.chat.completions.create(
                model=model,
                messages=[{"role": "user", "content": "Hello"}],
                max_tokens=10
            )
            elapsed = int((time.time() - start_time) * 1000)
            return {"success": True, "response_time_ms": elapsed}
        
        elif provider == "ollama":
            response = httpx.post(
                f"{settings.ollama_endpoint}/api/generate",
                json={"model": model, "prompt": "Hello", "stream": False},
                timeout=30
            )
            elapsed = int((time.time() - start_time) * 1000)
            if response.status_code == 200:
                return {"success": True, "response_time_ms": elapsed}
            return {"success": False, "message": f"HTTP {response.status_code}"}
        
        return {"success": False, "message": f"不支援的提供者: {provider}"}
        
    except Exception as e:
        elapsed = int((time.time() - start_time) * 1000)
        return {"success": False, "response_time_ms": elapsed, "message": str(e)}