DeepSeek V4 实战教程：百万Token上下文时代的开发新范式 #

from openai import OpenAI

client = OpenAI(
    api_key="your-api-key",
    base_url="https://api.deepseek.com"
)

# Flash 版 - 极低成本
response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[
        {"role": "system", "content": "你是一个专业的编程助手。"},
        {"role": "user", "content": "用 Python 实现一个 LRU 缓存"}
    ],
    max_tokens=4096
)

print(response.choices[0].message.content)

# 读取大型代码库，让模型理解整个项目
import os

def load_codebase(directory, extensions=('.py', '.js', '.ts', '.go')):
    """将整个代码库拼成一个超长上下文"""
    code_content = ""
    for root, dirs, files in os.walk(directory):
        for f in files:
            if f.endswith(extensions):
                filepath = os.path.join(root, f)
                with open(filepath, 'r', encoding='utf-8') as fh:
                    code_content += f"\n--- {filepath} ---\n{fh.read()}\n"
    return code_content

# 加载代码库
codebase = load_codebase("./my-project")

# 让模型基于整个代码库回答问题
response = client.chat.completions.create(
    model="deepseek-v4-pro",  # Pro版，更强的理解能力
    messages=[
        {
            "role": "system",
            "content": "你是一个高级软件架构师。以下是完整代码库：\n" + codebase
        },
        {
            "role": "user",
            "content": "分析这个项目的架构，指出潜在的性能瓶颈和改进建议。"
        }
    ]
)

import json

def agent_loop(task, max_iterations=20):
    """低成本 Agent 循环 - Flash版每次调用几分钱"""
    messages = [
        {"role": "system", "content": "你是一个自主执行任务的AI代理。"},
        {"role": "user", "content": task}
    ]

    for i in range(max_iterations):
        response = client.chat.completions.create(
            model="deepseek-v4-flash",  # 用Flash版跑循环，成本极低
            messages=messages,
            tools=[{
                "type": "function",
                "function": {
                    "name": "execute_code",
                    "description": "执行Python代码",
                    "parameters": {
                        "type": "object",
                        "properties": {
                            "code": {"type": "string", "description": "要执行的代码"}
                        },
                        "required": ["code"]
                    }
                }
            }],
            temperature=0.1
        )

        msg = response.choices[0].message
        messages.append(msg)

        # 检查是否完成
        if msg.tool_calls:
            for tool_call in msg.tool_calls:
                # 处理工具调用...
                result = execute_code(tool_call.function.arguments)
                messages.append({
                    "role": "tool",
                    "tool_call_id": tool_call.id,
                    "content": result
                })
        else:
            # 模型直接给出最终答案
            return msg.content

    return "达到最大迭代次数"