Week 4：AI / LLM 基础认知

# app/llm/basic_call.py
import httpx
import json
from pathlib import Path

# 从 .env 读取 API Key（推荐用 python-dotenv）
API_KEY = "your-api-key-here"      # 或者读 .env
BASE_URL = "https://api.siliconflow.cn/v1"  # 硅基流动示例

async def call_llm(prompt: str, model: str = "Qwen/Qwen2.5-7B-Instruct") -> str:
    """最简单的 LLM 调用"""
    async with httpx.AsyncClient(timeout=30) as client:
        resp = await client.post(
            f"{BASE_URL}/chat/completions",
            headers={
                "Authorization": f"Bearer {API_KEY}",
                "Content-Type": "application/json",
            },
            json={
                "model": model,
                "messages": [
                    {"role": "system", "content": "你是一个专业的交易助手。"},
                    {"role": "user", "content": prompt},
                ],
                "temperature": 0.7,  # 0-2，越大越发散
            },
        )
        resp.raise_for_status()
        data = resp.json()
        return data["choices"][0]["message"]["content"]

# 测试
async def main():
    result = await call_llm("帮我分析一下今天的交易策略")
    print(result)

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())

# app/llm/function_calling.py
import httpx
import json
from typing import Any

API_KEY = "your-api-key-here"
BASE_URL = "https://api.siliconflow.cn/v1"

# ===== 第一步：定义你的 Python 函数 =====
def get_stock_price(symbol: str) -> dict:
    """获取股票/加密货币的当前价格（Mock）"""
    mock_prices = {"BTC": 99999.99, "ETH": 3500.0, "AAPL": 210.5}
    return {"symbol": symbol, "price": mock_prices.get(symbol.upper(), 0), "currency": "USD"}

def place_order(symbol: str, side: str, quantity: int) -> dict:
    """下单交易（Mock）"""
    return {
        "order_id": "ORD12345",
        "symbol": symbol,
        "side": side,
        "quantity": quantity,
        "status": "pending",
    }

# ===== 第二步：把函数注册成 LLM 能理解的 schema =====
TOOLS = [
    {
        "type": "function",
        "function": {
            "name": "get_stock_price",
            "description": "获取股票或加密货币的当前价格",
            "parameters": {
                "type": "object",
                "properties": {
                    "symbol": {"type": "string", "description": "交易标的代码，如 BTC, ETH, AAPL"},
                },
                "required": ["symbol"],
            },
        },
    },
    {
        "type": "function",
        "function": {
            "name": "place_order",
            "description": "下单买入或卖出",
            "parameters": {
                "type": "object",
                "properties": {
                    "symbol": {"type": "string", "description": "交易标的代码"},
                    "side": {"type": "string", "enum": ["buy", "sell"], "description": "买入或卖出"},
                    "quantity": {"type": "integer", "description": "数量"},
                },
                "required": ["symbol", "side", "quantity"],
            },
        },
    },
]

# ===== 第三步：映射函数名到真实函数 =====
FUNCTION_MAP = {
    "get_stock_price": get_stock_price,
    "place_order": place_order,
}

# ===== 第四步：完整的 Agent 对话流程 =====
async def agent_chat(user_input: str) -> str:
    messages = [
        {"role": "system", "content": "你是一个专业的交易助手。必要时调用工具获取真实数据。"},
        {"role": "user", "content": user_input},
    ]

    async with httpx.AsyncClient(timeout=30) as client:
        # 第一次调用：让 LLM 决定要不要调工具
        resp = await client.post(
            f"{BASE_URL}/chat/completions",
            headers={"Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json"},
            json={
                "model": "Qwen/Qwen2.5-7B-Instruct",
                "messages": messages,
                "tools": TOOLS,
                "tool_choice": "auto",
            },
        )
        resp.raise_for_status()
        data = resp.json()
        message = data["choices"][0]["message"]

        # LLM 要调工具
        if message.get("tool_calls"):
            # 把 LLM 的 tool_call 消息加入对话历史
            messages.append(message)

            # 执行每个工具调用
            for tool_call in message["tool_calls"]:
                func_name = tool_call["function"]["name"]
                func_args = json.loads(tool_call["function"]["arguments"])

                # 执行真实的 Python 函数
                func = FUNCTION_MAP.get(func_name)
                if func:
                    result = func(**func_args)
                else:
                    result = {"error": f"未知函数: {func_name}"}

                # 把工具结果喂回对话历史
                messages.append({
                    "role": "tool",
                    "tool_call_id": tool_call["id"],
                    "content": json.dumps(result, ensure_ascii=False),
                })

            # 第二次调用：LLM 根据工具结果生成最终回答
            resp2 = await client.post(
                f"{BASE_URL}/chat/completions",
                headers={"Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json"},
                json={
                    "model": "Qwen/Qwen2.5-7B-Instruct",
                    "messages": messages,
                },
            )
            resp2.raise_for_status()
            return resp2.json()["choices"][0]["message"]["content"]

        # LLM 直接回答（没调工具）
        return message["content"]

# 测试
async def main():
    # 测试 1：会触发工具调用
    result1 = await agent_chat("BTC 现在多少钱？")
    print("Q1:", result1)

    # 测试 2：可能触发工具调用
    result2 = await agent_chat("帮我买 0.1 个 ETH")
    print("Q2:", result2)

    # 测试 3：不需要工具
    result3 = await agent_chat("你好，你是做什么的？")
    print("Q3:", result3)

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())

# ❌ 反例：模糊的 Prompt
prompt = "分析一下"

# ✅ 正例：清晰、有角色、有约束、有输出格式
prompt = """
你是一个专业的量化交易分析师。

请基于以下数据，给出你的分析：
<data>
BTC 2024-01-01 至 2024-01-07 的日线数据
</data>

你的回答必须包含：
1. 价格走势概述（2-3 句话）
2. 关键技术指标（MA、RSI）的解读
3. 风险提示
4. 建议操作方向（买/卖/观望）

用简洁的专业语言回答，不要超过 300 字。
"""