优化 LLM API 开发成本的实用指南

import openai

# 配置客户端指向 n1n.ai 聚合网关
client = openai.OpenAI(
    base_url="https://api.n1n.ai/v1",
    api_key="YOUR_N1N_API_KEY"
)

def route_and_query(user_prompt: str) -> str:
    # 步骤 1：分析复杂度（意图分类）
    classification_prompt = f"""Classify the complexity of the following user request.
Respond with exactly one word: 'SIMPLE' or 'COMPLEX'.
Request: {user_prompt}"""

    router_response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": classification_prompt}],
        temperature=0.0,
        max_tokens=5
    )

    complexity = router_response.choices[0].message.content.strip().upper()

    # 步骤 2：动态路由选择
    if complexity == "SIMPLE":
        # 路由到高性价比模型
        selected_model = "deepseek-v3"
    else:
        # 路由到高能力推理模型
        selected_model = "claude-3-5-sonnet"

    print(f"Routing request to: {selected_model}")

    # 步骤 3：执行最终请求
    final_response = client.chat.completions.create(
        model=selected_model,
        messages=[{"role": "user", "content": user_prompt}],
        temperature=0.7
    )
    return final_response.choices[0].message.content

传统历史记录: [第 1 轮] -> [第 2 轮] -> [第 3 轮] -> [第 4 轮] ... (Token 数量线性增长)
摘要内存模式: [1-3 轮的摘要信息] -> [第 4 轮]                    (Token 数量保持平稳)

// 低效格式 (消耗 35 个 Token)
{
  "user_identification_number": 10293,
  "user_account_status_active": true
}

// 高效格式 (仅消耗 15 个 Token)
{
  "id": 10293,
  "active": true
}

import time

class TokenBudgetManager:
    def __init__(self, daily_budget_usd: float):
        self.daily_budget = daily_budget_usd
        self.current_spend = 0.0
        # 预设模型单价
        self.prices = {
            "deepseek-v3": {"input": 0.14 / 1e6, "output": 0.28 / 1e6},
            "claude-3-5-sonnet": {"input": 3.00 / 1e6, "output": 15.00 / 1e6}
        }

    def track_usage(self, model: str, input_tokens: int, output_tokens: int):
        if model not in self.prices:
            raise ValueError("Model pricing not configured.")

        cost = (input_tokens * self.prices[model]["input"]) + (output_tokens * self.prices[model]["output"])
        self.current_spend += cost
        print(f"[监控] 本次消耗: ${cost:.6f} | 今日累计支出: ${self.current_spend:.4f}")

        if self.current_spend >= self.daily_budget:
            self.trigger_budget_alert()

    def trigger_budget_alert(self):
        print("[警告] 已超出每日 API 预算上限！正在限制后续请求...")
        # 在此处实现降级或限流逻辑

# 使用示例
tracker = TokenBudgetManager(daily_budget_usd=10.0)

# 模拟 API 调用追踪
tracker.track_usage("claude-3-5-sonnet", input_tokens=1500, output_tokens=500)
tracker.track_usage("deepseek-v3", input_tokens=10000, output_tokens=2000)