主流模型对比与选型

综合对比

模型	参数量	上下文	中文	推理	代码	工具调用	开源	价格
GPT-4o	~200B	128K	⭐⭐⭐⭐	⭐⭐⭐⭐⭐	⭐⭐⭐⭐⭐	✅	❌	高
Claude 3.5 Sonnet	-	200K	⭐⭐⭐⭐	⭐⭐⭐⭐⭐	⭐⭐⭐⭐⭐	✅	❌	高
千问 Max	-	32K	⭐⭐⭐⭐⭐	⭐⭐⭐⭐	⭐⭐⭐⭐	✅	❌	中
DeepSeek-V3	671B	128K	⭐⭐⭐⭐⭐	⭐⭐⭐⭐⭐	⭐⭐⭐⭐⭐	✅	部分	低
ChatGLM-4	-	128K	⭐⭐⭐⭐⭐	⭐⭐⭐⭐	⭐⭐⭐⭐	✅	❌	低
Llama 3.1 70B	70B	128K	⭐⭐⭐	⭐⭐⭐⭐	⭐⭐⭐⭐	✅	✅	自托管

金融场景选型建议

# 场景化选型决策树
def select_model(scenario: str, budget: str, privacy: str) -> str:
    """
    scenario: 'qa'(问答) | 'analysis'(分析) | 'code'(代码) | 'agent'(智能体)
    budget: 'low' | 'medium' | 'high'
    privacy: 'public' | 'private'(私有化部署)
    """
    if privacy == "private":
        if scenario == "code":
            return "DeepSeek-Coder-V2 本地部署"
        return "ChatGLM3-6B 量化版 或 Llama3-8B"
    
    if scenario == "agent" and budget != "low":
        return "qwen-max 或 GPT-4o（工具调用最稳定）"
    
    if scenario == "analysis":
        return "qwen-plus 或 DeepSeek-V3（长文本分析）"
    
    if budget == "low":
        return "glm-4-flash（免费）或 qwen-turbo"
    
    return "qwen-plus（综合性价比最高）"

# 金融场景示例
print(select_model("agent", "medium", "public"))    # qwen-max
print(select_model("qa", "low", "public"))          # glm-4-flash
print(select_model("analysis", "high", "private"))  # ChatGLM3-6B

API 接口统一封装

from abc import ABC, abstractmethod
from openai import OpenAI
from zhipuai import ZhipuAI
from typing import List, Dict

class LLMClient(ABC):
    @abstractmethod
    def chat(self, messages: List[Dict], **kwargs) -> str:
        pass

class QwenClient(LLMClient):
    def __init__(self, api_key: str, model: str = "qwen-turbo"):
        self.client = OpenAI(
            api_key=api_key,
            base_url="https://dashscope.aliyuncs.com/compatible-mode/v1"
        )
        self.model = model
    
    def chat(self, messages, **kwargs) -> str:
        resp = self.client.chat.completions.create(
            model=self.model, messages=messages, **kwargs
        )
        return resp.choices[0].message.content

class ChatGLMClient(LLMClient):
    def __init__(self, api_key: str, model: str = "glm-4-flash"):
        self.client = ZhipuAI(api_key=api_key)
        self.model = model
    
    def chat(self, messages, **kwargs) -> str:
        resp = self.client.chat.completions.create(
            model=self.model, messages=messages, **kwargs
        )
        return resp.choices[0].message.content

# 工厂函数
def create_llm(provider: str, **kwargs) -> LLMClient:
    clients = {
        "qwen": QwenClient,
        "chatglm": ChatGLMClient,
    }
    return clients[provider](**kwargs)

# 使用
llm = create_llm("qwen", api_key="sk-xxx", model="qwen-plus")
result = llm.chat([{"role": "user", "content": "你好"}])