TravelContentCreator/domain/aigc/shared/llm_client.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

"""
LLM 客户端
封装 LLM 调用，提供统一接口
"""

import logging
from typing import Dict, Any, Optional, Callable, AsyncIterator

logger = logging.getLogger(__name__)


class LLMClient:
    """
    LLM 客户端
    
    封装底层 AI Agent，提供简化的调用接口
    """
    
    def __init__(self, ai_agent=None):
        """
        初始化 LLM 客户端
        
        Args:
            ai_agent: 底层 AI Agent 实例
        """
        self._ai_agent = ai_agent
    
    def set_agent(self, ai_agent):
        """设置 AI Agent"""
        self._ai_agent = ai_agent
    
    async def generate(
        self,
        prompt: str = None,
        system_prompt: Optional[str] = None,
        user_prompt: Optional[str] = None,
        temperature: Optional[float] = None,
        max_tokens: Optional[int] = None,
        top_p: Optional[float] = None,
        presence_penalty: Optional[float] = None,
        **kwargs
    ) -> tuple:
        """
        生成文本
        
        Args:
            prompt: 用户提示词 (兼容旧接口)
            system_prompt: 系统提示词
            user_prompt: 用户提示词 (新接口)
            temperature: 温度参数
            max_tokens: 最大 token 数
            top_p: top_p 参数
            presence_penalty: 存在惩罚
            **kwargs: 其他参数
            
        Returns:
            tuple: (result, input_tokens, output_tokens, time_cost)
        """
        # 兼容两种调用方式
        actual_user_prompt = user_prompt or prompt
        if not actual_user_prompt:
            raise ValueError("必须提供 prompt 或 user_prompt")
        
        if not self._ai_agent:
            raise RuntimeError("AI Agent 未初始化")
        
        # 构建参数
        params = {}
        if temperature is not None:
            params['temperature'] = temperature
        if max_tokens is not None:
            params['max_tokens'] = max_tokens
        if top_p is not None:
            params['top_p'] = top_p
        if presence_penalty is not None:
            params['presence_penalty'] = presence_penalty
        params.update(kwargs)
        
        try:
            result, input_tokens, output_tokens, time_cost = await self._ai_agent.generate_text(
                system_prompt=system_prompt or "",
                user_prompt=actual_user_prompt,
                use_stream=True,
                stage="aigc_engine",
                **params
            )
            
            logger.debug(f"LLM 调用完成: input={input_tokens}, output={output_tokens}, time={time_cost:.2f}s")
            return result, input_tokens, output_tokens, time_cost
            
        except Exception as e:
            logger.error(f"LLM 调用失败: {e}")
            raise
    
    async def generate_stream(
        self,
        prompt: str,
        system_prompt: Optional[str] = None,
        callback: Optional[Callable[[str], None]] = None,
        **kwargs
    ) -> str:
        """
        流式生成文本
        
        Args:
            prompt: 用户提示词
            system_prompt: 系统提示词
            callback: 流式回调函数
            **kwargs: 其他参数
            
        Returns:
            完整的生成文本
        """
        if not self._ai_agent:
            raise RuntimeError("AI Agent 未初始化")
        
        try:
            result, _, _, _ = await self._ai_agent.generate_text(
                system_prompt=system_prompt or "",
                user_prompt=prompt,
                use_stream=True,
                stage="aigc_engine_stream",
                **kwargs
            )
            return result
            
        except Exception as e:
            logger.error(f"LLM 流式调用失败: {e}")
            raise
    
    async def generate_with_retry(
        self,
        prompt: str,
        system_prompt: Optional[str] = None,
        max_retries: int = 3,
        **kwargs
    ) -> str:
        """
        带重试的生成
        
        Args:
            prompt: 用户提示词
            system_prompt: 系统提示词
            max_retries: 最大重试次数
            **kwargs: 其他参数
            
        Returns:
            生成的文本
        """
        last_error = None
        
        for attempt in range(max_retries):
            try:
                return await self.generate(prompt, system_prompt, **kwargs)
            except Exception as e:
                last_error = e
                logger.warning(f"LLM 调用失败 (尝试 {attempt + 1}/{max_retries}): {e}")
                
                if attempt < max_retries - 1:
                    import asyncio
                    await asyncio.sleep(1 * (attempt + 1))  # 递增等待
        
        raise last_error