{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "bb4b298d-5e1a-4e98-ab64-934804ad626a",
   "metadata": {},
   "source": [
    "# DeepSeek API 多轮对话 - OpenAI SDK\n",
    "\n",
    "> 指导文章：[DeepSeek API 多轮对话 - OpenAI SDK](https://github.com/Hoper-J/AI-Guide-and-Demos-zh_CN/blob/master/Guide/DeepSeek%20API%20多轮对话%20-%20OpenAI%20SDK.md)\n",
    "> \n",
    "> 在代码运行过程中可能遇到的问题：\n",
    "> 1. 为什么等待时间那么长？\n",
    ">\n",
    ">    答：因为默认参数 `stream=False`，不启用流式输出。对于推理模型来说，非流式输出下的用户体验极其糟糕，故当前代码文件为最后一次非流式输出的代码演示，之后将默认采用流式输出。\n",
    "> 2. 504 报错：对应服务当前的负载高，可以稍后尝试或者切换平台。\n",
    "\n",
    "从下方选择平台开始，替换 `your-api-key` 后点击 `►` 或使用 `Shift + 回车` 运行代码块。\n",
    "\n",
    "**注意**：\n",
    "\n",
    "- 当前代码为非流式输出。\n",
    "\n",
    "在线链接：[Kaggle](https://www.kaggle.com/code/aidemos/deepseek-api-guide-4) | [Colab](https://colab.research.google.com/drive/12HBOxSD-SsVc0qkEhdYBWTRKjQ80gMPw?usp=sharing)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b14b6944-18b4-4b04-be0a-c84f577b4986",
   "metadata": {
    "jp-MarkdownHeadingCollapsed": true
   },
   "source": [
    "# DeepSeek"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d9e23572-b60b-4fa6-ba69-0182088afc44",
   "metadata": {},
   "source": [
    "## 非思考模式"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "cd3bb2c1-57d1-4be7-a9e8-84bc2b9004fa",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:38.225046Z",
     "iopub.status.busy": "2026-07-06T15:27:38.224892Z",
     "iopub.status.idle": "2026-07-06T15:27:38.831788Z",
     "shell.execute_reply": "2026-07-06T15:27:38.830954Z"
    }
   },
   "outputs": [],
   "source": [
    "from openai import OpenAI\n",
    "import os\n",
    "\n",
    "# 临时环境变量配置\n",
    "os.environ[\"DEEPSEEK_API_KEY\"] = \"your-api-key\"\n",
    "\n",
    "client = OpenAI(\n",
    "    api_key=os.getenv(\"DEEPSEEK_API_KEY\"),\n",
    "    base_url=\"https://api.deepseek.com\",\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8811ccbe-cfd1-45e4-be3f-c658e69bd6fa",
   "metadata": {},
   "source": [
    "### 单轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "5fa2d7da-cce2-404a-a603-f68ed45c1c47",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:38.834040Z",
     "iopub.status.busy": "2026-07-06T15:27:38.833891Z",
     "iopub.status.idle": "2026-07-06T15:27:40.383365Z",
     "shell.execute_reply": "2026-07-06T15:27:40.382615Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "你好！我是DeepSeek，由深度求索公司创造的AI助手。我在这里为你提供各种帮助，回答问题、处理信息、协助思考，或者就是和你聊聊天。无论你需要什么，尽管问我，我会尽力帮你！😊\n"
     ]
    }
   ],
   "source": [
    "# 单轮对话示例\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=[\n",
    "        {'role': 'system', 'content': 'You are a helpful assistant.'}, \n",
    "        {'role': 'user', 'content': '你是谁？'}\n",
    "    ]\n",
    ")\n",
    "\n",
    "# 打印模型消息\n",
    "print(completion.choices[0].message.content)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3170b31e-0cb5-4be5-b7f1-a8f2d0576e7f",
   "metadata": {},
   "source": [
    "### 两轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "8e22b864-d4c2-4815-a00e-f91c25dfa306",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:40.385683Z",
     "iopub.status.busy": "2026-07-06T15:27:40.385532Z",
     "iopub.status.idle": "2026-07-06T15:27:44.704538Z",
     "shell.execute_reply": "2026-07-06T15:27:44.703668Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 这个问题其实是个常见的“陷阱题”，很容易被人误解。\n",
      "\n",
      "**如果单纯比较数值大小：**  \n",
      "9.11 和 9.8，9.11 更大。因为 9.11 = 9 + 0.11，而 9.8 = 9 + 0.8，但注意 0.8 = 0.80，而 0.11 < 0.80，所以实际上 **9.8 > 9.11**。\n",
      "\n",
      "所以正确答案是：**9.8 更大**。\n",
      "\n",
      "但很多人会误以为 9.11 更大，因为觉得“11 > 8”，其实这是忽略了小数点后的位数不同。比较小数时，要从左到右逐位比较：  \n",
      "- 个位：9 = 9  \n",
      "- 十分位：1 < 8  \n",
      "所以 9.8 更大。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 我刚才针对的是 **9.11 和 9.8** 这两个数值进行了比较。\n",
      "\n",
      "所以回答是：**9.8 更大**，因为 9.8 的十分位是 8，而 9.11 的十分位是 1，所以 9.8 > 9.11。\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "# 第一轮对话\n",
    "messages.append({'role': 'user', 'content': '9.11 和 9.8 哪个更大？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=messages\n",
    ")\n",
    "content = completion.choices[0].message.content\n",
    "print(\"AI:\", content)\n",
    "\n",
    "# 将第一轮模型回复加入历史记录\n",
    "messages.append({'role': 'assistant', 'content': content})\n",
    "\n",
    "# 第二轮对话\n",
    "messages.append({'role': 'user', 'content': '刚刚针对哪两个数值进行了比较？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=messages\n",
    ")\n",
    "content = completion.choices[0].message.content\n",
    "print(\"AI:\", content)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c3ee8c47-a782-4e91-875e-a99f7e32d6f2",
   "metadata": {},
   "source": [
    "### 多轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "78a21713-5256-4e5a-9c4a-60f2c172ca75",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:44.707741Z",
     "iopub.status.busy": "2026-07-06T15:27:44.707551Z",
     "iopub.status.idle": "2026-07-06T15:27:46.008286Z",
     "shell.execute_reply": "2026-07-06T15:27:46.007611Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话，输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 2\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 12\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "print(\"开始对话，输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    # 获取用户输入\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:  # 退出条件\n",
    "        break\n",
    "        \n",
    "    # 记录用户输入\n",
    "    messages.append({'role': 'user', 'content': user_input})\n",
    "    \n",
    "    # 调用 API 获取模型回复\n",
    "    completion = client.chat.completions.create(\n",
    "        model=\"deepseek-v4-flash\",\n",
    "        extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "        messages=messages\n",
    "    )\n",
    "    content = completion.choices[0].message.content\n",
    "    print(\"AI:\", content)\n",
    "    \n",
    "    # 记录模型回复\n",
    "    messages.append({'role': 'assistant', 'content': content})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eefec12c-7140-49bf-9a1c-05b080c7f1d1",
   "metadata": {},
   "source": [
    "### 封装对话逻辑"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "12de2c9c-58e8-43df-b671-ef0cd6491cd6",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:46.010260Z",
     "iopub.status.busy": "2026-07-06T15:27:46.010121Z",
     "iopub.status.idle": "2026-07-06T15:27:47.526071Z",
     "shell.execute_reply": "2026-07-06T15:27:47.525437Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话，输入 'exit' 或 'quit' 结束对话。\n",
      "You: 1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 2\n",
      "You: 再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 12\n",
      "You: exit\n"
     ]
    }
   ],
   "source": [
    "class ChatSession:\n",
    "    def __init__(self, client, system_message=\"You are a helpful assistant.\"):\n",
    "        \"\"\"\n",
    "        参数:\n",
    "            client (openai.OpenAI): OpenAI 客户端实例\n",
    "            system_message (str): 系统消息，用于设定对话背景\n",
    "        \"\"\"\n",
    "        self.client = client\n",
    "        self.messages = [{'role': 'system', 'content': system_message}]\n",
    "\n",
    "    def append_message(self, role, content):\n",
    "        \"\"\"\n",
    "        添加一条对话消息\n",
    "        参数:\n",
    "            role (str): 消息角色，为 'user' 或 'assistant'\n",
    "            content (str): 消息内容\n",
    "        \"\"\"\n",
    "        self.messages.append({'role': role, 'content': content})\n",
    "\n",
    "    def get_response(self, user_input):\n",
    "        \"\"\"\n",
    "        添加用户消息，调用 API 获取回复，并将回复加入历史记录\n",
    "        参数:\n",
    "            user_input (str): 用户输入的消息\n",
    "        返回:\n",
    "            content (str): 模型的回复内容\n",
    "        \"\"\"\n",
    "        self.append_message('user', user_input)\n",
    "        completion = self.client.chat.completions.create(\n",
    "            model=\"deepseek-v4-flash\",\n",
    "            extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "            messages=self.messages\n",
    "        )\n",
    "        content = completion.choices[0].message.content\n",
    "        self.append_message('assistant', content)\n",
    "        return content\n",
    "\n",
    "\n",
    "# 初始化对话会话，采用默认的系统消息\n",
    "session = ChatSession(client)\n",
    "\n",
    "print(\"开始对话，输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You: \")\n",
    "    if user_input.lower() in ['exit', 'quit']:\n",
    "        break\n",
    "    reply = session.get_response(user_input)\n",
    "    print(\"AI:\", reply)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d0a7d31b-0fb6-4415-8011-d1ded410af78",
   "metadata": {},
   "source": [
    "## 思考模式"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e498e26f-60bf-41bd-9974-601b64bd0961",
   "metadata": {},
   "source": [
    "### 单轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "52223508-99a5-455b-afe2-f02f216f1820",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:47.527876Z",
     "iopub.status.busy": "2026-07-06T15:27:47.527762Z",
     "iopub.status.idle": "2026-07-06T15:27:50.159000Z",
     "shell.execute_reply": "2026-07-06T15:27:50.158389Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 模型推理过程 =====\n",
      "嗯，用户问“你是谁”，这是一个非常基础的自我介绍问题。用户可能是第一次接触我，或者想确认我的身份和功能。我需要清晰、简洁地说明我是谁、能做什么，同时保持友好和热情。\n",
      "\n",
      "想到了直接说明我是DeepSeek，由深度求索公司创造。然后简要概括我的核心特点：免费、纯文本助手、支持文件处理、长上下文等。最后以开放式的提问结尾，邀请用户提出具体需求。\n",
      "\n",
      "用自然的语气组织这些信息，避免过于技术化或冗长，让用户快速了解我的能力。\n",
      "\n",
      "===== 模型回复 =====\n",
      "你好！我是DeepSeek，由深度求索公司创造的AI助手。我是一个纯文本模型，可以帮你解答问题、处理信息、进行对话交流等。我支持阅读链接、上传文件（图像、txt、pdf、ppt、word、excel等），还能联网搜索（需要手动开启）。我的上下文容量高达1M，可以一次性处理像《三体》三部曲那样体量的书籍。最棒的是，我完全免费使用！有什么我可以帮你的吗？😊\n",
      "\n"
     ]
    }
   ],
   "source": [
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    messages=[\n",
    "        {'role': 'system', 'content': 'You are a helpful assistant.'}, \n",
    "        {'role': 'user', 'content': '你是谁？'}\n",
    "    ]\n",
    ")\n",
    "\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 模型推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\n{content}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d4ac0f2b-d933-433b-856c-5c74d98ea995",
   "metadata": {},
   "source": [
    "### 两轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "10b95b32-1e4e-4abc-b93a-b385e2e86fde",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:50.160757Z",
     "iopub.status.busy": "2026-07-06T15:27:50.160639Z",
     "iopub.status.idle": "2026-07-06T15:27:53.343223Z",
     "shell.execute_reply": "2026-07-06T15:27:53.342819Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 第一轮推理过程 =====\n",
      "我们被问到：\"9.11 和 9.8 哪个更大？\" 这是一个简单的问题，但要注意可能有陷阱。通常，比较小数时，9.11和9.8，9.8更大，因为9.8 = 9.80，而9.11 < 9.80。所以答案是9.8更大。但有时人们会误解为版本号或类似的东西，但这里显然是数字。所以回答：9.8更大。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 9.8 更大，因为 9.8 相当于 9.80，而 9.11 小于 9.80。\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 第二轮推理过程 =====\n",
      "我们需要理解用户的问题。用户问“刚刚针对哪两个数值进行了比较？”指的是上一轮对话中，我比较了9.11和9.8，并指出9.8更大。所以答案是：9.11和9.8。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 刚刚是针对 **9.11** 和 **9.8** 这两个数值进行了比较。\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "# 第一轮对话\n",
    "messages.append({'role': 'user', 'content': '9.11 和 9.8 哪个更大？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    messages=messages\n",
    ")\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 第一轮推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\nAI: {content}\\n\")\n",
    "\n",
    "# 只将最终的回复（content）加入对话历史\n",
    "messages.append({'role': 'assistant', 'content': content})\n",
    "\n",
    "# 第二轮对话\n",
    "messages.append({'role': 'user', 'content': '刚刚针对哪两个数值进行了比较？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    messages=messages\n",
    ")\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 第二轮推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\nAI: {content}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "38e1a19a-f138-4e88-b7ec-348d570c7096",
   "metadata": {},
   "source": [
    "### 多轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "12d97e64-1841-4633-b828-f432ab33d3ec",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:53.345238Z",
     "iopub.status.busy": "2026-07-06T15:27:53.345116Z",
     "iopub.status.idle": "2026-07-06T15:27:55.228947Z",
     "shell.execute_reply": "2026-07-06T15:27:55.228144Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们要求回答1+1等于几，只回答数字。所以答案是2。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 2\n",
      "\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们被问到：\"再加10呢？只回答数字。\" 之前的提问是\"1+1等于几？\" 回答\"2\"。然后\"再加10呢？\" 意思是2+10=12。只回答数字。所以答案应该是12。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 12\n",
      "\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "print(\"开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:\n",
    "        break\n",
    "\n",
    "    # 1. 记录用户输入\n",
    "    messages.append({'role': 'user', 'content': user_input})\n",
    "\n",
    "    # 2. 调用 API\n",
    "    completion = client.chat.completions.create(\n",
    "        model=\"deepseek-v4-flash\",\n",
    "        messages=messages\n",
    "    )\n",
    "    reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "    content = completion.choices[0].message.content\n",
    "\n",
    "    # 3. 输出推理思考过程与最终回复\n",
    "    print(f\"===== 推理过程 =====\\n{reasoning_content}\\n\")\n",
    "    print(f\"===== 模型回复 =====\\nAI: {content}\\n\")\n",
    "\n",
    "    # 4. 只将模型的最终回复写入到历史消息中\n",
    "    messages.append({'role': 'assistant', 'content': content})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2d12ec5b-de8c-472b-88af-7f81760c5288",
   "metadata": {},
   "source": [
    "### 封装对话逻辑"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "fd05574d-eb19-4f4a-bf79-2aaf3d77658a",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:55.231136Z",
     "iopub.status.busy": "2026-07-06T15:27:55.231004Z",
     "iopub.status.idle": "2026-07-06T15:27:57.125659Z",
     "shell.execute_reply": "2026-07-06T15:27:57.124901Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们要求只回答数字，所以直接输出2。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 2\n",
      "\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们被问到：\"再加10呢？只回答数字。\" 之前的对话是：用户问\"1+1等于几？只回答数字。\" 我们回答了\"2\"。现在用户说\"再加10呢？\" 意思是基于之前的结果2，再加10，即2+10=12。只回答数字。所以回答12。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 12\n",
      "\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "class ReasonerSession:\n",
    "    def __init__(self, client, system_message=\"You are a helpful assistant.\"):\n",
    "        \"\"\"\n",
    "        参数:\n",
    "            client (openai.OpenAI): OpenAI 客户端实例\n",
    "            system_message (str): 系统消息，用于设定对话背景\n",
    "        \"\"\"\n",
    "        self.client = client\n",
    "        self.messages = [{'role': 'system', 'content': system_message}]\n",
    "\n",
    "    def append_message(self, role, content):\n",
    "        \"\"\"\n",
    "        添加一条对话消息\n",
    "        参数:\n",
    "            role (str): 消息角色，为 'user' 或 'assistant'\n",
    "            content (str): 消息内容\n",
    "        \"\"\"\n",
    "        self.messages.append({'role': role, 'content': content})\n",
    "\n",
    "    def get_response(self, user_input):\n",
    "        \"\"\"\n",
    "        添加用户消息，调用 API 获取思维链和回复，并将回复加入历史记录\n",
    "        参数:\n",
    "            user_input (str): 用户输入的消息\n",
    "        返回:\n",
    "            (reasoning_content, content) (tuple): 模型的推理过程和回复\n",
    "        \"\"\"\n",
    "        # 1. 记录用户输入\n",
    "        self.append_message('user', user_input)\n",
    "\n",
    "        # 2. 调用 API\n",
    "        completion = self.client.chat.completions.create(\n",
    "            model=\"deepseek-v4-flash\",\n",
    "            messages=self.messages\n",
    "        )\n",
    "\n",
    "        # 3. 获取推理思考过程和最终回复\n",
    "        reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "        content = completion.choices[0].message.content\n",
    "\n",
    "        # 4. 只将模型的最终回复加入到历史消息\n",
    "        self.append_message('assistant', content)\n",
    "\n",
    "        return reasoning_content, content\n",
    "\n",
    "\n",
    "# 初始化推理会话，采用默认的系统消息\n",
    "session = ReasonerSession(client)\n",
    "\n",
    "print(\"开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:  # 退出条件\n",
    "        break\n",
    "\n",
    "    reasoning, reply = session.get_response(user_input)\n",
    "    print(f\"===== 推理过程 =====\\n{reasoning}\\n\")\n",
    "    print(f\"===== 模型回复 =====\\nAI: {reply}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "502678c3-6645-49b6-a707-6a29f45c3895",
   "metadata": {
    "jp-MarkdownHeadingCollapsed": true
   },
   "source": [
    "# 硅基流动"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ff6d162c-a955-4e61-a60b-3d673c6dd871",
   "metadata": {},
   "source": [
    "## 非思考模式"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "8b444291-851d-40c6-86f0-241d4aec1d4d",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:57.127991Z",
     "iopub.status.busy": "2026-07-06T15:27:57.127866Z",
     "iopub.status.idle": "2026-07-06T15:27:57.188367Z",
     "shell.execute_reply": "2026-07-06T15:27:57.187306Z"
    }
   },
   "outputs": [],
   "source": [
    "from openai import OpenAI\n",
    "import os\n",
    "\n",
    "# 临时环境变量配置\n",
    "os.environ[\"SILICONFLOW_API_KEY\"] = \"your-api-key\"\n",
    "\n",
    "client = OpenAI(\n",
    "    api_key=os.getenv(\"SILICONFLOW_API_KEY\"),\n",
    "    base_url=\"https://api.siliconflow.cn/v1\",\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ea40b539-89f2-4120-be55-d02eadb6610a",
   "metadata": {},
   "source": [
    "### 单轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "7a960861-bd20-461a-85b0-1e873d7bafee",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:27:57.190773Z",
     "iopub.status.busy": "2026-07-06T15:27:57.190591Z",
     "iopub.status.idle": "2026-07-06T15:28:00.096788Z",
     "shell.execute_reply": "2026-07-06T15:28:00.095779Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "你好！我是DeepSeek，由深度求索公司创造的AI助手。我是一个纯文本模型，可以帮你解答问题、处理信息、进行对话等。虽然我不支持多模态识别（比如直接识别图片内容），但我可以读取链接，也能处理你上传的文件（图像、txt、pdf、ppt、word、excel等），从中提取文字信息。而且我完全免费，上下文长度可达1M，能一次性处理像《三体》三部曲那样体量的书籍。\n",
      "\n",
      "有什么我可以帮你的吗？😊\n"
     ]
    }
   ],
   "source": [
    "# 单轮对话示例\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-ai/DeepSeek-V4-Flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=[\n",
    "        {'role': 'system', 'content': 'You are a helpful assistant.'}, \n",
    "        {'role': 'user', 'content': '你是谁？'}\n",
    "    ]\n",
    ")\n",
    "\n",
    "# 打印模型消息\n",
    "print(completion.choices[0].message.content)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "222db473-49b0-4f5e-bb74-1d53fa67e50c",
   "metadata": {},
   "source": [
    "### 两轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "b7dda8b6-da57-45ed-81ea-5b9829e3a56c",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:28:00.099309Z",
     "iopub.status.busy": "2026-07-06T15:28:00.098986Z",
     "iopub.status.idle": "2026-07-06T15:28:06.037900Z",
     "shell.execute_reply": "2026-07-06T15:28:06.037148Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 这个问题其实有点陷阱，需要看具体语境：\n",
      "\n",
      "1. **如果是比较小数（数学上）**：  \n",
      "   9.11 和 9.8 比较，**9.8 更大**。  \n",
      "   因为 9.8 = 9.80，而 9.11 < 9.80。\n",
      "\n",
      "2. **如果是比较版本号（比如软件版本）**：  \n",
      "   有些地方会把版本号按“数字部分逐段比较”，这时 9.11 可能被认为比 9.8 大（因为 11 > 8）。但这并不是标准数学比较。\n",
      "\n",
      "所以，**通常情况下（数学语境）答案是 9.8 更大**。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 刚刚比较的两个数值是 **9.11** 和 **9.8**。\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "# 第一轮对话\n",
    "messages.append({'role': 'user', 'content': '9.11 和 9.8 哪个更大？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-ai/DeepSeek-V4-Flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=messages\n",
    ")\n",
    "content = completion.choices[0].message.content\n",
    "print(\"AI:\", content)\n",
    "\n",
    "# 将第一轮模型回复加入历史记录\n",
    "messages.append({'role': 'assistant', 'content': content})\n",
    "\n",
    "# 第二轮对话\n",
    "messages.append({'role': 'user', 'content': '刚刚针对哪两个数值进行了比较？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-ai/DeepSeek-V4-Flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=messages\n",
    ")\n",
    "content = completion.choices[0].message.content\n",
    "print(\"AI:\", content)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a62411aa-c9af-4312-9164-59127b57b98f",
   "metadata": {},
   "source": [
    "### 多轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "7dcf81f2-506b-4e46-8338-77f2d6b91e6a",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:28:06.039425Z",
     "iopub.status.busy": "2026-07-06T15:28:06.039307Z",
     "iopub.status.idle": "2026-07-06T15:29:05.855460Z",
     "shell.execute_reply": "2026-07-06T15:29:05.853923Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话，输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 2\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 12\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "print(\"开始对话，输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    # 获取用户输入\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:  # 退出条件\n",
    "        break\n",
    "        \n",
    "    # 记录用户输入\n",
    "    messages.append({'role': 'user', 'content': user_input})\n",
    "    \n",
    "    # 调用 API 获取模型回复\n",
    "    completion = client.chat.completions.create(\n",
    "        model=\"deepseek-ai/DeepSeek-V4-Flash\",\n",
    "        extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "        messages=messages\n",
    "    )\n",
    "    content = completion.choices[0].message.content\n",
    "    print(\"AI:\", content)\n",
    "    \n",
    "    # 记录模型回复\n",
    "    messages.append({'role': 'assistant', 'content': content})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "df1715d9-bdb0-4af2-b9c4-d3292d76e323",
   "metadata": {},
   "source": [
    "### 封装对话逻辑"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "e5da4161-6b31-4c5a-b895-40dc05e8b3e1",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:05.858553Z",
     "iopub.status.busy": "2026-07-06T15:29:05.858378Z",
     "iopub.status.idle": "2026-07-06T15:29:11.749284Z",
     "shell.execute_reply": "2026-07-06T15:29:11.748561Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话，输入 'exit' 或 'quit' 结束对话。\n",
      "You: 1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 2\n",
      "You: 再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 12\n",
      "You: exit\n"
     ]
    }
   ],
   "source": [
    "class ChatSession:\n",
    "    def __init__(self, client, system_message=\"You are a helpful assistant.\"):\n",
    "        \"\"\"\n",
    "        参数:\n",
    "            client (openai.OpenAI): OpenAI 客户端实例\n",
    "            system_message (str): 系统消息，用于设定对话背景\n",
    "        \"\"\"\n",
    "        self.client = client\n",
    "        self.messages = [{'role': 'system', 'content': system_message}]\n",
    "\n",
    "    def append_message(self, role, content):\n",
    "        \"\"\"\n",
    "        添加一条对话消息\n",
    "        参数:\n",
    "            role (str): 消息角色，为 'user' 或 'assistant'\n",
    "            content (str): 消息内容\n",
    "        \"\"\"\n",
    "        self.messages.append({'role': role, 'content': content})\n",
    "\n",
    "    def get_response(self, user_input):\n",
    "        \"\"\"\n",
    "        添加用户消息，调用 API 获取回复，并将回复加入历史记录\n",
    "        参数:\n",
    "            user_input (str): 用户输入的消息\n",
    "        返回:\n",
    "            content (str): 模型的回复内容\n",
    "        \"\"\"\n",
    "        self.append_message('user', user_input)\n",
    "        completion = self.client.chat.completions.create(\n",
    "            model=\"deepseek-ai/DeepSeek-V4-Flash\",\n",
    "            extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "            messages=self.messages\n",
    "        )\n",
    "        content = completion.choices[0].message.content\n",
    "        self.append_message('assistant', content)\n",
    "        return content\n",
    "\n",
    "\n",
    "# 初始化对话会话，采用默认的系统消息\n",
    "session = ChatSession(client)\n",
    "\n",
    "print(\"开始对话，输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You: \")\n",
    "    if user_input.lower() in ['exit', 'quit']:\n",
    "        break\n",
    "    reply = session.get_response(user_input)\n",
    "    print(\"AI:\", reply)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fe72a506-7390-497d-b072-f3a913b1b27f",
   "metadata": {},
   "source": [
    "## 思考模式"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "70fd3357-93da-4520-bbfa-6f0c5f31c5ed",
   "metadata": {},
   "source": [
    "### 单轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "c48ceb51-dbd9-4804-a141-3f3fc57eea0e",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:11.751416Z",
     "iopub.status.busy": "2026-07-06T15:29:11.751280Z",
     "iopub.status.idle": "2026-07-06T15:29:21.125636Z",
     "shell.execute_reply": "2026-07-06T15:29:21.124677Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 模型推理过程 =====\n",
      "嗯，用户问“你是谁”，这是一个非常基础的自我介绍问题。用户可能是初次接触，想了解我的身份和功能。需要简洁清晰地说明我是AI助手，由深度求索公司创造，并说明我的主要能力和特点，比如支持多语言、长文本处理、文件上传等。最后可以表达友好和愿意提供帮助的态度。\n",
      "\n",
      "===== 模型回复 =====\n",
      "你好！我是DeepSeek，由深度求索公司创造的AI助手。我可以帮你解答问题、提供建议、进行对话，或者处理文本任务。我支持处理长文本（比如一次性读完《三体》三部曲）、上传图像、PDF、Word等文件，还能联网搜索（需要你手动开启）。有什么我可以帮你的吗？😊\n",
      "\n"
     ]
    }
   ],
   "source": [
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-ai/DeepSeek-V4-Flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"enabled\"}},  # 硅基流动上偶尔不默认思考，所以这里显式开启\n",
    "    messages=[\n",
    "        {'role': 'system', 'content': 'You are a helpful assistant.'}, \n",
    "        {'role': 'user', 'content': '你是谁？'}\n",
    "    ]\n",
    ")\n",
    "\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 模型推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\n{content}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d9d7b618-ce47-4f86-82b6-8156a1fd2168",
   "metadata": {},
   "source": [
    "### 两轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "dd82894a-739c-481c-835d-6bbefa6c1c77",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:21.128186Z",
     "iopub.status.busy": "2026-07-06T15:29:21.128049Z",
     "iopub.status.idle": "2026-07-06T15:29:31.241582Z",
     "shell.execute_reply": "2026-07-06T15:29:31.240736Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 第一轮推理过程 =====\n",
      "我们被问到：\"9.11 和 9.8 哪个更大？\" 这是一个简单的数字比较问题。9.11和9.8，比较小数部分：9.11的十分位是1，9.8的十分位是8，所以9.8更大。但要注意，有些人可能会误解为版本号或别的，但通常数学上9.8更大。所以答案：9.8更大。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 9.8 更大。因为比较小数时，先看整数部分相同，再看十分位，9.8 的十分位是 8，而 9.11 的十分位是 1，所以 9.8 > 9.11。\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 第二轮推理过程 =====\n",
      "我们刚刚比较的是9.11和9.8这两个数值。用户问“刚刚针对哪两个数值进行了比较？”，答案就是9.11和9.8。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 刚刚比较的是 **9.11** 和 **9.8** 这两个数值。\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "# 第一轮对话\n",
    "messages.append({'role': 'user', 'content': '9.11 和 9.8 哪个更大？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-ai/DeepSeek-V4-Flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"enabled\"}},  # 硅基流动上偶尔不默认思考，所以这里显式开启\n",
    "    messages=messages\n",
    ")\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 第一轮推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\nAI: {content}\\n\")\n",
    "\n",
    "# 只将最终的回复（content）加入对话历史\n",
    "messages.append({'role': 'assistant', 'content': content})\n",
    "\n",
    "# 第二轮对话\n",
    "messages.append({'role': 'user', 'content': '刚刚针对哪两个数值进行了比较？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-ai/DeepSeek-V4-Flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"enabled\"}},  # 硅基流动上偶尔不默认思考，所以这里显式开启\n",
    "    messages=messages\n",
    ")\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 第二轮推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\nAI: {content}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6b31a73d-9d50-4db2-9eaf-b1b518d0d106",
   "metadata": {},
   "source": [
    "### 多轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "efdb2cba-66b4-4207-99b8-339fdb435bd1",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:31.245129Z",
     "iopub.status.busy": "2026-07-06T15:29:31.244899Z",
     "iopub.status.idle": "2026-07-06T15:29:33.868569Z",
     "shell.execute_reply": "2026-07-06T15:29:33.867674Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们要求回答1+1等于几，只输出数字。答案是2。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 2\n",
      "\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们被问到：\"再加10呢？只回答数字。\" 前面的对话是：\"1+1等于几？只回答数字。\" 然后回答\"2\"。现在问\"再加10呢？\" 意思是2加10，结果是12。只回答数字。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 12\n",
      "\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "print(\"开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:\n",
    "        break\n",
    "\n",
    "    # 1. 记录用户输入\n",
    "    messages.append({'role': 'user', 'content': user_input})\n",
    "\n",
    "    # 2. 调用 API\n",
    "    completion = client.chat.completions.create(\n",
    "        model=\"deepseek-ai/DeepSeek-V4-Flash\",\n",
    "        extra_body={\"thinking\": {\"type\": \"enabled\"}},  # 硅基流动上偶尔不默认思考，所以这里显式开启\n",
    "        messages=messages\n",
    "    )\n",
    "    reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "    content = completion.choices[0].message.content\n",
    "\n",
    "    # 3. 输出推理思考过程与最终回复\n",
    "    print(f\"===== 推理过程 =====\\n{reasoning_content}\\n\")\n",
    "    print(f\"===== 模型回复 =====\\nAI: {content}\\n\")\n",
    "\n",
    "    # 4. 只将模型的最终回复写入到历史消息中\n",
    "    messages.append({'role': 'assistant', 'content': content})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7c8607c0-7697-4031-80e6-bd062c85cd56",
   "metadata": {},
   "source": [
    "### 封装对话逻辑"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "a1cbfef1-a108-48d6-80a0-6209e1cd60be",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:33.871026Z",
     "iopub.status.busy": "2026-07-06T15:29:33.870870Z",
     "iopub.status.idle": "2026-07-06T15:29:37.834668Z",
     "shell.execute_reply": "2026-07-06T15:29:37.833813Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们被问到：\"1+1等于几？只回答数字。\" 所以答案应该是2。只输出数字。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 2\n",
      "\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们被要求只回答数字。上一个问题是1+1等于几，回答2。然后问再加10呢？即2+10=12。只回答数字。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 12\n",
      "\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "class ReasonerSession:\n",
    "    def __init__(self, client, system_message=\"You are a helpful assistant.\"):\n",
    "        \"\"\"\n",
    "        参数:\n",
    "            client (openai.OpenAI): OpenAI 客户端实例\n",
    "            system_message (str): 系统消息，用于设定对话背景\n",
    "        \"\"\"\n",
    "        self.client = client\n",
    "        self.messages = [{'role': 'system', 'content': system_message}]\n",
    "\n",
    "    def append_message(self, role, content):\n",
    "        \"\"\"\n",
    "        添加一条对话消息\n",
    "        参数:\n",
    "            role (str): 消息角色，为 'user' 或 'assistant'\n",
    "            content (str): 消息内容\n",
    "        \"\"\"\n",
    "        self.messages.append({'role': role, 'content': content})\n",
    "\n",
    "    def get_response(self, user_input):\n",
    "        \"\"\"\n",
    "        添加用户消息，调用 API 获取思维链和回复，并将回复加入历史记录\n",
    "        参数:\n",
    "            user_input (str): 用户输入的消息\n",
    "        返回:\n",
    "            (reasoning_content, content) (tuple): 模型的推理过程和回复\n",
    "        \"\"\"\n",
    "        # 1. 记录用户输入\n",
    "        self.append_message('user', user_input)\n",
    "\n",
    "        # 2. 调用 API\n",
    "        completion = self.client.chat.completions.create(\n",
    "            model=\"deepseek-ai/DeepSeek-V4-Flash\",\n",
    "            extra_body={\"thinking\": {\"type\": \"enabled\"}},  # 硅基流动上偶尔不默认思考，所以这里显式开启\n",
    "            messages=self.messages\n",
    "        )\n",
    "\n",
    "        # 3. 获取推理思考过程和最终回复\n",
    "        reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "        content = completion.choices[0].message.content\n",
    "\n",
    "        # 4. 只将模型的最终回复加入到历史消息\n",
    "        self.append_message('assistant', content)\n",
    "\n",
    "        return reasoning_content, content\n",
    "\n",
    "\n",
    "# 初始化推理会话，采用默认的系统消息\n",
    "session = ReasonerSession(client)\n",
    "\n",
    "print(\"开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:  # 退出条件\n",
    "        break\n",
    "\n",
    "    reasoning, reply = session.get_response(user_input)\n",
    "    print(f\"===== 推理过程 =====\\n{reasoning}\\n\")\n",
    "    print(f\"===== 模型回复 =====\\nAI: {reply}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "668a9b2c-ec58-40cc-96d1-600a984eb161",
   "metadata": {
    "jp-MarkdownHeadingCollapsed": true
   },
   "source": [
    "# 阿里云百炼"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9ba21368-4577-439e-a594-58a76226e6b6",
   "metadata": {},
   "source": [
    "## 非思考模式"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "dbfd8aea-b5a0-4664-9026-f5c8cffa0654",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:37.837745Z",
     "iopub.status.busy": "2026-07-06T15:29:37.837403Z",
     "iopub.status.idle": "2026-07-06T15:29:37.914573Z",
     "shell.execute_reply": "2026-07-06T15:29:37.913369Z"
    }
   },
   "outputs": [],
   "source": [
    "from openai import OpenAI\n",
    "import os\n",
    "\n",
    "# 临时环境变量配置\n",
    "os.environ[\"DASHSCOPE_API_KEY\"] = \"your-api-key\"\n",
    "\n",
    "client = OpenAI(\n",
    "    api_key=os.getenv(\"DASHSCOPE_API_KEY\"),\n",
    "    base_url=\"https://dashscope.aliyuncs.com/compatible-mode/v1\",\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "afb3c8ae-c9bd-4718-b7bd-4df6ca34f673",
   "metadata": {},
   "source": [
    "### 单轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "b28a6d91-4c59-4f13-8ca9-99feb95356c6",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:37.918848Z",
     "iopub.status.busy": "2026-07-06T15:29:37.918612Z",
     "iopub.status.idle": "2026-07-06T15:29:39.501104Z",
     "shell.execute_reply": "2026-07-06T15:29:39.500194Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "你好！我是DeepSeek，由深度求索公司创造的AI助手。我是一个纯文本模型，可以帮你解答问题、提供信息、进行对话交流等。我的知识截止日期是2025年5月，目前是DeepSeek最新版本模型。\n",
      "\n",
      "有什么我可以帮你的吗？😊\n"
     ]
    }
   ],
   "source": [
    "# 单轮对话示例\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=[\n",
    "        {'role': 'system', 'content': 'You are a helpful assistant.'}, \n",
    "        {'role': 'user', 'content': '你是谁？'}\n",
    "    ]\n",
    ")\n",
    "\n",
    "# 打印模型消息\n",
    "print(completion.choices[0].message.content)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1790ea0a-2ae6-41e6-bd99-d466cd63e9d5",
   "metadata": {},
   "source": [
    "### 两轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "387109f2-f41a-4642-8ef7-75e495c9bafb",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:39.503890Z",
     "iopub.status.busy": "2026-07-06T15:29:39.503704Z",
     "iopub.status.idle": "2026-07-06T15:29:43.543556Z",
     "shell.execute_reply": "2026-07-06T15:29:43.542905Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 这个问题其实有点陷阱，要看是在什么语境下。\n",
      "\n",
      "- 如果是**小数**比较，9.11 和 9.8，9.8 更大，因为 9.8 = 9.80，而 9.11 < 9.80。\n",
      "- 如果是**版本号**（比如软件版本），9.11 通常被认为是比 9.8 更新的版本，所以更大。\n",
      "- 如果是**数字字符串**按字典序比较，则 \"9.11\" 小于 \"9.8\"，因为字符 '1' 小于 '8'。\n",
      "\n",
      "不过最常见的数学意义上，**9.8 更大**。如果你问的是别的情况，可以再说明一下。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 刚刚比较的两个数值是 **9.11** 和 **9.8**。\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "# 第一轮对话\n",
    "messages.append({'role': 'user', 'content': '9.11 和 9.8 哪个更大？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=messages\n",
    ")\n",
    "content = completion.choices[0].message.content\n",
    "print(\"AI:\", content)\n",
    "\n",
    "# 将第一轮模型回复加入历史记录\n",
    "messages.append({'role': 'assistant', 'content': content})\n",
    "\n",
    "# 第二轮对话\n",
    "messages.append({'role': 'user', 'content': '刚刚针对哪两个数值进行了比较？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=messages\n",
    ")\n",
    "content = completion.choices[0].message.content\n",
    "print(\"AI:\", content)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "be181f77-afb2-4400-b705-4e3b448d262a",
   "metadata": {},
   "source": [
    "### 多轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "77519a99-4203-41ab-a7a8-35510c50fdc4",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:43.545430Z",
     "iopub.status.busy": "2026-07-06T15:29:43.545327Z",
     "iopub.status.idle": "2026-07-06T15:29:45.648519Z",
     "shell.execute_reply": "2026-07-06T15:29:45.647598Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话，输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 2\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 12\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "print(\"开始对话，输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    # 获取用户输入\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:  # 退出条件\n",
    "        break\n",
    "        \n",
    "    # 记录用户输入\n",
    "    messages.append({'role': 'user', 'content': user_input})\n",
    "    \n",
    "    # 调用 API 获取模型回复\n",
    "    completion = client.chat.completions.create(\n",
    "        model=\"deepseek-v4-flash\",\n",
    "        extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "        messages=messages\n",
    "    )\n",
    "    content = completion.choices[0].message.content\n",
    "    print(\"AI:\", content)\n",
    "    \n",
    "    # 记录模型回复\n",
    "    messages.append({'role': 'assistant', 'content': content})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7f387a5a-fe7d-472c-91a1-6a06ecc56cae",
   "metadata": {},
   "source": [
    "### 封装对话逻辑"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "3d5ae085-3a1f-4160-9156-bb13a536bc66",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:45.650825Z",
     "iopub.status.busy": "2026-07-06T15:29:45.650703Z",
     "iopub.status.idle": "2026-07-06T15:29:47.908515Z",
     "shell.execute_reply": "2026-07-06T15:29:47.907486Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话，输入 'exit' 或 'quit' 结束对话。\n",
      "You: 1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 2\n",
      "You: 再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 12\n",
      "You: exit\n"
     ]
    }
   ],
   "source": [
    "class ChatSession:\n",
    "    def __init__(self, client, system_message=\"You are a helpful assistant.\"):\n",
    "        \"\"\"\n",
    "        参数:\n",
    "            client (openai.OpenAI): OpenAI 客户端实例\n",
    "            system_message (str): 系统消息，用于设定对话背景\n",
    "        \"\"\"\n",
    "        self.client = client\n",
    "        self.messages = [{'role': 'system', 'content': system_message}]\n",
    "\n",
    "    def append_message(self, role, content):\n",
    "        \"\"\"\n",
    "        添加一条对话消息\n",
    "        参数:\n",
    "            role (str): 消息角色，为 'user' 或 'assistant'\n",
    "            content (str): 消息内容\n",
    "        \"\"\"\n",
    "        self.messages.append({'role': role, 'content': content})\n",
    "\n",
    "    def get_response(self, user_input):\n",
    "        \"\"\"\n",
    "        添加用户消息，调用 API 获取回复，并将回复加入历史记录\n",
    "        参数:\n",
    "            user_input (str): 用户输入的消息\n",
    "        返回:\n",
    "            content (str): 模型的回复内容\n",
    "        \"\"\"\n",
    "        self.append_message('user', user_input)\n",
    "        completion = self.client.chat.completions.create(\n",
    "            model=\"deepseek-v4-flash\",\n",
    "            extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "            messages=self.messages\n",
    "        )\n",
    "        content = completion.choices[0].message.content\n",
    "        self.append_message('assistant', content)\n",
    "        return content\n",
    "\n",
    "\n",
    "# 初始化对话会话，采用默认的系统消息\n",
    "session = ChatSession(client)\n",
    "\n",
    "print(\"开始对话，输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You: \")\n",
    "    if user_input.lower() in ['exit', 'quit']:\n",
    "        break\n",
    "    reply = session.get_response(user_input)\n",
    "    print(\"AI:\", reply)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f0039960-8b0c-453d-963d-ae4f09a6f903",
   "metadata": {},
   "source": [
    "## 思考模式"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "021ca2d2-21f6-45e5-ae8e-f037906201be",
   "metadata": {},
   "source": [
    "### 单轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "1fe0cd25-e290-4f23-b99b-e98f391091f5",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:47.910598Z",
     "iopub.status.busy": "2026-07-06T15:29:47.910437Z",
     "iopub.status.idle": "2026-07-06T15:29:49.841649Z",
     "shell.execute_reply": "2026-07-06T15:29:49.840822Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 模型推理过程 =====\n",
      "我们被问到\"你是谁？\"，这是一个简单的问题。作为AI助手，我需要明确自己的身份。用户可能想知道我的名称、创建者或功能。通常我会回答：我是DeepSeek，由深度求索公司创造的AI助手。我会用友好的语气回答。\n",
      "\n",
      "===== 模型回复 =====\n",
      "你好！我是DeepSeek，由深度求索公司创造的AI助手。我在这里帮助你解答问题、提供信息，以及协助你完成各种任务。无论是学习、工作还是日常生活中的疑问，都可以向我提问，我会尽力提供准确、有用的回答。有什么我可以帮你的吗？😊\n",
      "\n"
     ]
    }
   ],
   "source": [
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    messages=[\n",
    "        {'role': 'system', 'content': 'You are a helpful assistant.'}, \n",
    "        {'role': 'user', 'content': '你是谁？'}\n",
    "    ]\n",
    ")\n",
    "\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 模型推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\n{content}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e89e32e5-2fbc-4172-b519-ff0dd50c5ca2",
   "metadata": {},
   "source": [
    "### 两轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "a472056d-100c-41b2-87a3-11b48429a2fb",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:49.844231Z",
     "iopub.status.busy": "2026-07-06T15:29:49.844067Z",
     "iopub.status.idle": "2026-07-06T15:29:53.661655Z",
     "shell.execute_reply": "2026-07-06T15:29:53.660865Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 第一轮推理过程 =====\n",
      "我们比较两个小数：9.11 和 9.8。直观上，9.8 = 9.80，而9.11 < 9.80，所以9.8更大。但有人可能会误解为9.11更大，因为11>8，但实际上是小数比较，先比较整数部分相同，再比较十分位，9.8的十分位是8，9.11的十分位是1，所以9.8大。因此答案是9.8。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 9.8 更大。因为比较小数时，先看整数部分（都是9），然后看十分位，9.8的十分位是8，而9.11的十分位是1，所以9.8 > 9.11。\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 第二轮推理过程 =====\n",
      "我们刚刚比较了9.11和9.8。用户问“刚刚针对哪两个数值进行了比较？”，所以回答应该是9.11和9.8。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 我们刚刚比较的是 9.11 和 9.8 这两个数值。\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "# 第一轮对话\n",
    "messages.append({'role': 'user', 'content': '9.11 和 9.8 哪个更大？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    messages=messages\n",
    ")\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 第一轮推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\nAI: {content}\\n\")\n",
    "\n",
    "# 只将最终的回复（content）加入对话历史\n",
    "messages.append({'role': 'assistant', 'content': content})\n",
    "\n",
    "# 第二轮对话\n",
    "messages.append({'role': 'user', 'content': '刚刚针对哪两个数值进行了比较？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    messages=messages\n",
    ")\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 第二轮推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\nAI: {content}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bd05db35-06b8-4671-b632-dc84f3faa471",
   "metadata": {},
   "source": [
    "### 多轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "8622d231-dae3-45e8-baae-6d2c78780a76",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:53.665912Z",
     "iopub.status.busy": "2026-07-06T15:29:53.665593Z",
     "iopub.status.idle": "2026-07-06T15:29:58.450893Z",
     "shell.execute_reply": "2026-07-06T15:29:58.448326Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们只需要回答数字，所以答案是2。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 2\n",
      "\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "用户问了1+1等于几，我只回答了数字2。现在用户继续问再加10呢，并且要求只回答数字。这是在前一个结果的基础上加10，也就是2+10=12。按照要求，只输出数字，不输出任何其他内容。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 12\n",
      "\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "print(\"开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:\n",
    "        break\n",
    "\n",
    "    # 1. 记录用户输入\n",
    "    messages.append({'role': 'user', 'content': user_input})\n",
    "\n",
    "    # 2. 调用 API\n",
    "    completion = client.chat.completions.create(\n",
    "        model=\"deepseek-v4-flash\",\n",
    "        messages=messages\n",
    "    )\n",
    "    reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "    content = completion.choices[0].message.content\n",
    "\n",
    "    # 3. 输出推理思考过程与最终回复\n",
    "    print(f\"===== 推理过程 =====\\n{reasoning_content}\\n\")\n",
    "    print(f\"===== 模型回复 =====\\nAI: {content}\\n\")\n",
    "\n",
    "    # 4. 只将模型的最终回复写入到历史消息中\n",
    "    messages.append({'role': 'assistant', 'content': content})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9c4eca43-71aa-4226-a6c7-eb414aebe0f2",
   "metadata": {},
   "source": [
    "### 封装对话逻辑"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "866b4c99-7d4d-412c-9d1b-0dd12d7310e9",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:29:58.453894Z",
     "iopub.status.busy": "2026-07-06T15:29:58.453650Z",
     "iopub.status.idle": "2026-07-06T15:30:00.390011Z",
     "shell.execute_reply": "2026-07-06T15:30:00.388659Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们要求回答1+1等于几，只回答数字。所以答案是2。\n",
      "\n",
      "===== AI回复 =====\n",
      "AI: 2\n",
      "\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "用户问“再加10”，在之前回答1+1=2的基础上，加10就是12。只输出数字。\n",
      "\n",
      "===== AI回复 =====\n",
      "AI: 12\n",
      "\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "class ReasonerSession:\n",
    "    def __init__(self, client, system_message=\"You are a helpful assistant.\"):\n",
    "        \"\"\"\n",
    "        参数:\n",
    "            client (openai.OpenAI): OpenAI 客户端实例\n",
    "            system_message (str): 系统消息，用于设定对话背景\n",
    "        \"\"\"\n",
    "        self.client = client\n",
    "        self.messages = [{'role': 'system', 'content': system_message}]\n",
    "\n",
    "    def append_message(self, role, content):\n",
    "        \"\"\"\n",
    "        添加一条对话消息\n",
    "        参数:\n",
    "            role (str): 消息角色，为 'user' 或 'assistant'\n",
    "            content (str): 消息内容\n",
    "        \"\"\"\n",
    "        self.messages.append({'role': role, 'content': content})\n",
    "\n",
    "    def get_response(self, user_input):\n",
    "        \"\"\"\n",
    "        添加用户消息，调用 API 获取思维链和回复，并将回复加入历史记录\n",
    "        参数:\n",
    "            user_input (str): 用户输入的消息\n",
    "        返回:\n",
    "            (reasoning_content, content) (tuple): 模型的推理过程和回复\n",
    "        \"\"\"\n",
    "        # 1. 记录用户输入\n",
    "        self.append_message('user', user_input)\n",
    "\n",
    "        # 2. 调用 API\n",
    "        completion = self.client.chat.completions.create(\n",
    "            model=\"deepseek-v4-flash\",\n",
    "            messages=self.messages\n",
    "        )\n",
    "\n",
    "        # 3. 获取推理思考过程和最终回复\n",
    "        reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "        content = completion.choices[0].message.content\n",
    "\n",
    "        # 4. 只将模型的最终回复加入到历史消息\n",
    "        self.append_message('assistant', content)\n",
    "\n",
    "        return reasoning_content, content\n",
    "\n",
    "\n",
    "# 初始化推理会话，采用默认的系统消息\n",
    "session = ReasonerSession(client)\n",
    "\n",
    "print(\"开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:  # 退出条件\n",
    "        break\n",
    "\n",
    "    reasoning, reply = session.get_response(user_input)\n",
    "    print(f\"===== 推理过程 =====\\n{reasoning}\\n\")\n",
    "    print(f\"===== AI回复 =====\\nAI: {reply}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c196c658-9054-4aaf-89ec-aa715bd8f1e3",
   "metadata": {
    "jp-MarkdownHeadingCollapsed": true
   },
   "source": [
    "# 百度智能云"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c2eadd99-8d5f-4e9b-9763-e7a2fb672d19",
   "metadata": {},
   "source": [
    "## 非思考模式"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "4a28226b-1a13-422d-9c50-cddfd7447bb9",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:00.393490Z",
     "iopub.status.busy": "2026-07-06T15:30:00.393311Z",
     "iopub.status.idle": "2026-07-06T15:30:00.467884Z",
     "shell.execute_reply": "2026-07-06T15:30:00.466562Z"
    }
   },
   "outputs": [],
   "source": [
    "from openai import OpenAI\n",
    "import os\n",
    "\n",
    "# 临时环境变量配置\n",
    "os.environ[\"BAIDU_API_KEY\"] = \"your-api-key\"\n",
    "\n",
    "client = OpenAI(\n",
    "    api_key=os.getenv(\"BAIDU_API_KEY\"),\n",
    "    base_url=\"https://qianfan.baidubce.com/v2\",\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0b35207b-cb22-445f-9b7e-3b26d09b5f07",
   "metadata": {},
   "source": [
    "### 单轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "44ac6e82-7408-41b2-b7a1-28698b536de3",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:00.471959Z",
     "iopub.status.busy": "2026-07-06T15:30:00.471576Z",
     "iopub.status.idle": "2026-07-06T15:30:01.618225Z",
     "shell.execute_reply": "2026-07-06T15:30:01.617335Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "你好！我是 DeepSeek，一个由深度求索公司开发的 AI 助手。我可以帮你回答问题、提供信息、进行对话，或者协助处理各种任务。有什么我可以帮你的吗？😊\n"
     ]
    }
   ],
   "source": [
    "# 单轮对话示例\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=[\n",
    "        {'role': 'system', 'content': 'You are a helpful assistant.'}, \n",
    "        {'role': 'user', 'content': '你是谁？'}\n",
    "    ]\n",
    ")\n",
    "\n",
    "# 打印模型消息\n",
    "print(completion.choices[0].message.content)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "becbede1-d20c-4779-aeba-32a8704a4e62",
   "metadata": {},
   "source": [
    "### 两轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "4c0ee0cb-cd8a-4268-9d59-7c0a8f006948",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:01.621247Z",
     "iopub.status.busy": "2026-07-06T15:30:01.620933Z",
     "iopub.status.idle": "2026-07-06T15:30:04.108798Z",
     "shell.execute_reply": "2026-07-06T15:30:04.107750Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 9.11 大于 9.8。比较时先看整数部分（都是9），然后比较小数部分：9.11 的小数部分为 0.11，而 9.8 的小数部分为 0.80，因此 9.11 < 9.8？  \n",
      "实际上，9.8 等于 9.80，0.80 大于 0.11，所以 **9.8 更大**。抱歉之前的错误。正确答案是 9.8 更大。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 刚刚针对 **9.11** 和 **9.8** 这两个数值进行了比较。\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "# 第一轮对话\n",
    "messages.append({'role': 'user', 'content': '9.11 和 9.8 哪个更大？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=messages\n",
    ")\n",
    "content = completion.choices[0].message.content\n",
    "print(\"AI:\", content)\n",
    "\n",
    "# 将第一轮模型回复加入历史记录\n",
    "messages.append({'role': 'assistant', 'content': content})\n",
    "\n",
    "# 第二轮对话\n",
    "messages.append({'role': 'user', 'content': '刚刚针对哪两个数值进行了比较？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=messages\n",
    ")\n",
    "content = completion.choices[0].message.content\n",
    "print(\"AI:\", content)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "69a9ea1b-d3a1-47ea-8663-adf063d14a85",
   "metadata": {},
   "source": [
    "### 多轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "2adeea2e-34ca-4d6c-b336-6953a2a792d1",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:04.110453Z",
     "iopub.status.busy": "2026-07-06T15:30:04.110328Z",
     "iopub.status.idle": "2026-07-06T15:30:05.569523Z",
     "shell.execute_reply": "2026-07-06T15:30:05.569005Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话，输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 2\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 12\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "print(\"开始对话，输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    # 获取用户输入\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:  # 退出条件\n",
    "        break\n",
    "        \n",
    "    # 记录用户输入\n",
    "    messages.append({'role': 'user', 'content': user_input})\n",
    "    \n",
    "    # 调用 API 获取模型回复\n",
    "    completion = client.chat.completions.create(\n",
    "        model=\"deepseek-v4-flash\",\n",
    "        extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "        messages=messages\n",
    "    )\n",
    "    content = completion.choices[0].message.content\n",
    "    print(\"AI:\", content)\n",
    "    \n",
    "    # 记录模型回复\n",
    "    messages.append({'role': 'assistant', 'content': content})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c443c69b-e9cb-4a73-b7db-8c075806f2e1",
   "metadata": {},
   "source": [
    "### 封装对话逻辑"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "fbf27e5a-51fb-41f0-ae74-efe01bb46cb5",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:05.571365Z",
     "iopub.status.busy": "2026-07-06T15:30:05.571233Z",
     "iopub.status.idle": "2026-07-06T15:30:06.511882Z",
     "shell.execute_reply": "2026-07-06T15:30:06.511090Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话，输入 'exit' 或 'quit' 结束对话。\n",
      "You: 1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 2\n",
      "You: 再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 12\n",
      "You: exit\n"
     ]
    }
   ],
   "source": [
    "class ChatSession:\n",
    "    def __init__(self, client, system_message=\"You are a helpful assistant.\"):\n",
    "        \"\"\"\n",
    "        参数:\n",
    "            client (openai.OpenAI): OpenAI 客户端实例\n",
    "            system_message (str): 系统消息，用于设定对话背景\n",
    "        \"\"\"\n",
    "        self.client = client\n",
    "        self.messages = [{'role': 'system', 'content': system_message}]\n",
    "\n",
    "    def append_message(self, role, content):\n",
    "        \"\"\"\n",
    "        添加一条对话消息\n",
    "        参数:\n",
    "            role (str): 消息角色，为 'user' 或 'assistant'\n",
    "            content (str): 消息内容\n",
    "        \"\"\"\n",
    "        self.messages.append({'role': role, 'content': content})\n",
    "\n",
    "    def get_response(self, user_input):\n",
    "        \"\"\"\n",
    "        添加用户消息，调用 API 获取回复，并将回复加入历史记录\n",
    "        参数:\n",
    "            user_input (str): 用户输入的消息\n",
    "        返回:\n",
    "            content (str): 模型的回复内容\n",
    "        \"\"\"\n",
    "        self.append_message('user', user_input)\n",
    "        completion = self.client.chat.completions.create(\n",
    "            model=\"deepseek-v4-flash\",\n",
    "            extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "            messages=self.messages\n",
    "        )\n",
    "        content = completion.choices[0].message.content\n",
    "        self.append_message('assistant', content)\n",
    "        return content\n",
    "\n",
    "\n",
    "# 初始化对话会话，采用默认的系统消息\n",
    "session = ChatSession(client)\n",
    "\n",
    "print(\"开始对话，输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You: \")\n",
    "    if user_input.lower() in ['exit', 'quit']:\n",
    "        break\n",
    "    reply = session.get_response(user_input)\n",
    "    print(\"AI:\", reply)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1b9e5536-06f8-4701-bc03-c7a5ec13728a",
   "metadata": {},
   "source": [
    "## 思考模式"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "54a64dd4-30cd-4666-ad7c-d0cb012ef154",
   "metadata": {},
   "source": [
    "### 单轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "68262aa1-14a4-4013-9461-d2fc2ce7cff2",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:06.513797Z",
     "iopub.status.busy": "2026-07-06T15:30:06.513649Z",
     "iopub.status.idle": "2026-07-06T15:30:09.069676Z",
     "shell.execute_reply": "2026-07-06T15:30:09.068657Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 模型推理过程 =====\n",
      "嗯，用户问“你是谁？”，这是一个非常基础的自我介绍问题。用户可能是初次接触，想确认我的身份和功能。我需要简洁清晰地说明自己的名称、创造者、核心能力和特点。\n",
      "\n",
      "想到了直接告诉用户我是DeepSeek，由深度求索公司创造，并列举主要功能如对话、文件处理、联网搜索等，同时说明免费和知识截止日期。这样用户能快速了解我能做什么和不能做什么。另外可以加上友好结尾，邀请用户提问。\n",
      "\n",
      "===== 模型回复 =====\n",
      "你好！我是 DeepSeek，由深度求索公司创造的 AI 助手。我可以帮你回答问题、提供信息、创作内容，或者只是聊聊天。我支持处理文本、阅读链接和上传的文件（如图像、PDF、Word 等），还能进行联网搜索（需要你手动开启）。目前我是完全免费的，知识截止于 2025 年 5 月。有什么我可以帮你的吗？😊\n",
      "\n"
     ]
    }
   ],
   "source": [
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    messages=[\n",
    "        {'role': 'system', 'content': 'You are a helpful assistant.'}, \n",
    "        {'role': 'user', 'content': '你是谁？'}\n",
    "    ]\n",
    ")\n",
    "\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 模型推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\n{content}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8d49695d-a3fa-40ea-a95d-ec5a7b5a56c2",
   "metadata": {},
   "source": [
    "### 两轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "c88b932a-430e-4e0b-a8e7-798810a149a2",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:09.072236Z",
     "iopub.status.busy": "2026-07-06T15:30:09.072091Z",
     "iopub.status.idle": "2026-07-06T15:30:12.029845Z",
     "shell.execute_reply": "2026-07-06T15:30:12.028314Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 第一轮推理过程 =====\n",
      "我们比较两个数字9.11和9.8。注意这是十进制数，不是版本号。9.8 = 9.80，而9.11 = 9.11，显然9.8更大，因为十分位8大于1。所以9.8 > 9.11。答案：9.8更大。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 9.8 更大。因为比较小数时，先比较整数部分，两者都是9，然后比较十分位，8大于1，所以9.8 > 9.11。\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 第二轮推理过程 =====\n",
      "我们刚刚比较的是9.11和9.8。用户问“刚刚针对哪两个数值进行了比较？”，所以回答9.11和9.8即可。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 刚刚比较的是 9.11 和 9.8 这两个数值。\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "# 第一轮对话\n",
    "messages.append({'role': 'user', 'content': '9.11 和 9.8 哪个更大？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    messages=messages\n",
    ")\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 第一轮推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\nAI: {content}\\n\")\n",
    "\n",
    "# 只将最终的回复（content）加入对话历史\n",
    "messages.append({'role': 'assistant', 'content': content})\n",
    "\n",
    "# 第二轮对话\n",
    "messages.append({'role': 'user', 'content': '刚刚针对哪两个数值进行了比较？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash\",\n",
    "    messages=messages\n",
    ")\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 第二轮推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\nAI: {content}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2cf8d1e8-45ef-484a-bb9f-bdc794183482",
   "metadata": {},
   "source": [
    "### 多轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "5af4cce2-b5db-4138-989b-53ebceee6902",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:12.033810Z",
     "iopub.status.busy": "2026-07-06T15:30:12.033440Z",
     "iopub.status.idle": "2026-07-06T15:30:13.615950Z",
     "shell.execute_reply": "2026-07-06T15:30:13.614922Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们要求只回答数字，所以答案是2。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 2\n",
      "\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们被要求只回答数字。第一问已经回答了2，第二问是“再加10呢”，即2+10=12。所以回答12。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 12\n",
      "\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "print(\"开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:\n",
    "        break\n",
    "\n",
    "    # 1. 记录用户输入\n",
    "    messages.append({'role': 'user', 'content': user_input})\n",
    "\n",
    "    # 2. 调用 API\n",
    "    completion = client.chat.completions.create(\n",
    "        model=\"deepseek-v4-flash\",\n",
    "        messages=messages\n",
    "    )\n",
    "    reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "    content = completion.choices[0].message.content\n",
    "\n",
    "    # 3. 输出推理思考过程与最终回复\n",
    "    print(f\"===== 推理过程 =====\\n{reasoning_content}\\n\")\n",
    "    print(f\"===== 模型回复 =====\\nAI: {content}\\n\")\n",
    "\n",
    "    # 4. 只将模型的最终回复写入到历史消息中\n",
    "    messages.append({'role': 'assistant', 'content': content})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f16b814d-ffbe-4c69-8fad-bd2915cdf2af",
   "metadata": {},
   "source": [
    "### 封装对话逻辑"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "6c60fc72-f2f7-4b8e-a11c-510868225aac",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:13.618875Z",
     "iopub.status.busy": "2026-07-06T15:30:13.618561Z",
     "iopub.status.idle": "2026-07-06T15:30:15.276837Z",
     "shell.execute_reply": "2026-07-06T15:30:15.275975Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们要求只回答数字，所以输出就是2。\n",
      "\n",
      "===== AI回复 =====\n",
      "AI: 2\n",
      "\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们被问到：\"再加10呢？只回答数字。\" 前一个问题：\"1+1等于几？只回答数字。\" 回答是2。然后\"再加10呢？\" 意思是在2的基础上加10，所以是12。只回答数字。\n",
      "\n",
      "===== AI回复 =====\n",
      "AI: 12\n",
      "\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "class ReasonerSession:\n",
    "    def __init__(self, client, system_message=\"You are a helpful assistant.\"):\n",
    "        \"\"\"\n",
    "        参数:\n",
    "            client (openai.OpenAI): OpenAI 客户端实例\n",
    "            system_message (str): 系统消息，用于设定对话背景\n",
    "        \"\"\"\n",
    "        self.client = client\n",
    "        self.messages = [{'role': 'system', 'content': system_message}]\n",
    "\n",
    "    def append_message(self, role, content):\n",
    "        \"\"\"\n",
    "        添加一条对话消息\n",
    "        参数:\n",
    "            role (str): 消息角色，为 'user' 或 'assistant'\n",
    "            content (str): 消息内容\n",
    "        \"\"\"\n",
    "        self.messages.append({'role': role, 'content': content})\n",
    "\n",
    "    def get_response(self, user_input):\n",
    "        \"\"\"\n",
    "        添加用户消息，调用 API 获取思维链和回复，并将回复加入历史记录\n",
    "        参数:\n",
    "            user_input (str): 用户输入的消息\n",
    "        返回:\n",
    "            (reasoning_content, content) (tuple): 模型的推理过程和回复\n",
    "        \"\"\"\n",
    "        # 1. 记录用户输入\n",
    "        self.append_message('user', user_input)\n",
    "\n",
    "        # 2. 调用 API\n",
    "        completion = self.client.chat.completions.create(\n",
    "            model=\"deepseek-v4-flash\",\n",
    "            messages=self.messages\n",
    "        )\n",
    "\n",
    "        # 3. 获取推理思考过程和最终回复\n",
    "        reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "        content = completion.choices[0].message.content\n",
    "\n",
    "        # 4. 只将模型的最终回复加入到历史消息\n",
    "        self.append_message('assistant', content)\n",
    "\n",
    "        return reasoning_content, content\n",
    "\n",
    "\n",
    "# 初始化推理会话，采用默认的系统消息\n",
    "session = ReasonerSession(client)\n",
    "\n",
    "print(\"开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:  # 退出条件\n",
    "        break\n",
    "\n",
    "    reasoning, reply = session.get_response(user_input)\n",
    "    print(f\"===== 推理过程 =====\\n{reasoning}\\n\")\n",
    "    print(f\"===== AI回复 =====\\nAI: {reply}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3e531cf4-a20e-4647-9bc1-0389b6caaab2",
   "metadata": {},
   "source": [
    "# 字节火山引擎"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ffde8cc5-4f58-4ad8-8e02-59ad0b084c66",
   "metadata": {},
   "source": [
    "## 非思考模式"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "4d59241c-2c82-4dd3-9b9d-0c1ea3ebb1b0",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:15.279192Z",
     "iopub.status.busy": "2026-07-06T15:30:15.278941Z",
     "iopub.status.idle": "2026-07-06T15:30:15.351346Z",
     "shell.execute_reply": "2026-07-06T15:30:15.350512Z"
    }
   },
   "outputs": [],
   "source": [
    "from openai import OpenAI\n",
    "import os\n",
    "\n",
    "# 临时环境变量配置\n",
    "os.environ[\"ARK_API_KEY\"] = \"your-api-key\"\n",
    "\n",
    "client = OpenAI(\n",
    "    api_key=os.getenv(\"ARK_API_KEY\"),\n",
    "    base_url=\"https://ark.cn-beijing.volces.com/api/v3\",\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "79b317e8-c526-49c1-bca8-7fa18cadd1d7",
   "metadata": {},
   "source": [
    "### 单轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "a1b0169f-23d9-4d18-81f5-687c1031ab4e",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:15.353693Z",
     "iopub.status.busy": "2026-07-06T15:30:15.353565Z",
     "iopub.status.idle": "2026-07-06T15:30:17.740264Z",
     "shell.execute_reply": "2026-07-06T15:30:17.739557Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "你好！我是DeepSeek，由深度求索公司创造的AI助手。😊\n",
      "\n",
      "我是一个纯文本模型，可以帮你回答各种问题、进行对话、处理文字信息等。我的知识截止日期是2025年5月，拥有1M的上下文窗口，能够一次性处理像《三体》三部曲那样体量的超长文本。\n",
      "\n",
      "我支持文件上传（图像、txt、pdf、ppt、word、excel等），可以从中读取文字信息进行处理。我还支持联网搜索功能（需要你在Web/App端手动点开联网搜索按键），以及阅读链接内容。\n",
      "\n",
      "最重要的是，我完全免费使用！你可以在官方应用商店下载App，App端还支持语音输入功能。\n",
      "\n",
      "有什么我可以帮你的吗？😊\n"
     ]
    }
   ],
   "source": [
    "# 单轮对话示例\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash-260425\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=[\n",
    "        {'role': 'system', 'content': 'You are a helpful assistant.'}, \n",
    "        {'role': 'user', 'content': '你是谁？'}\n",
    "    ]\n",
    ")\n",
    "\n",
    "# 打印模型消息\n",
    "print(completion.choices[0].message.content)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a75f4439-1aff-4a42-ac40-b4da3af4ead5",
   "metadata": {},
   "source": [
    "### 两轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "9c5d8560-2d45-431a-9eff-dc86c49c9fed",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:17.743445Z",
     "iopub.status.busy": "2026-07-06T15:30:17.743257Z",
     "iopub.status.idle": "2026-07-06T15:30:21.300065Z",
     "shell.execute_reply": "2026-07-06T15:30:21.299091Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 这个问题其实是一个常见的小陷阱。从数字大小来看，**9.11 比 9.8 小**，而不是更大。\n",
      "\n",
      "解释一下：\n",
      "- 比较小数时，先比较整数部分：都是 9。\n",
      "- 然后比较小数部分：9.11 的小数部分是 0.11，9.8 的小数部分是 0.8。\n",
      "- 0.11 < 0.8，所以 **9.8 更大**。\n",
      "\n",
      "很多人会误以为 9.11 更大，是因为把 9.11 看成了“9 点 11”，而 9.8 看成了“9 点 8”，但按十进制小数比较，9.8 = 9.80，确实大于 9.11。\n",
      "\n",
      "所以答案是：**9.8 更大**。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 刚刚比较的两个数值是 **9.11** 和 **9.8**。\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "# 第一轮对话\n",
    "messages.append({'role': 'user', 'content': '9.11 和 9.8 哪个更大？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash-260425\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=messages\n",
    ")\n",
    "content = completion.choices[0].message.content\n",
    "print(\"AI:\", content)\n",
    "\n",
    "# 将第一轮模型回复加入历史记录\n",
    "messages.append({'role': 'assistant', 'content': content})\n",
    "\n",
    "# 第二轮对话\n",
    "messages.append({'role': 'user', 'content': '刚刚针对哪两个数值进行了比较？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash-260425\",\n",
    "    extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "    messages=messages\n",
    ")\n",
    "content = completion.choices[0].message.content\n",
    "print(\"AI:\", content)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "03a01d98-2b66-4452-af45-96a69803273b",
   "metadata": {},
   "source": [
    "### 多轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "6a330243-c55f-4c3d-9090-50f4279c30c3",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:21.302023Z",
     "iopub.status.busy": "2026-07-06T15:30:21.301904Z",
     "iopub.status.idle": "2026-07-06T15:30:22.567706Z",
     "shell.execute_reply": "2026-07-06T15:30:22.567214Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话，输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 2\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 12\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "print(\"开始对话，输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    # 获取用户输入\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:  # 退出条件\n",
    "        break\n",
    "        \n",
    "    # 记录用户输入\n",
    "    messages.append({'role': 'user', 'content': user_input})\n",
    "    \n",
    "    # 调用 API 获取模型回复\n",
    "    completion = client.chat.completions.create(\n",
    "        model=\"deepseek-v4-flash-260425\",\n",
    "        extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "        messages=messages\n",
    "    )\n",
    "    content = completion.choices[0].message.content\n",
    "    print(\"AI:\", content)\n",
    "    \n",
    "    # 记录模型回复\n",
    "    messages.append({'role': 'assistant', 'content': content})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "20ade4bd-bd3e-4c7a-8ac8-b35394b6fc40",
   "metadata": {},
   "source": [
    "### 封装对话逻辑"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "d984c929-e834-4fdc-8747-5f444abca036",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:22.569563Z",
     "iopub.status.busy": "2026-07-06T15:30:22.569449Z",
     "iopub.status.idle": "2026-07-06T15:30:23.957792Z",
     "shell.execute_reply": "2026-07-06T15:30:23.956862Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话，输入 'exit' 或 'quit' 结束对话。\n",
      "You: 1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 2\n",
      "You: 再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AI: 12\n",
      "You: exit\n"
     ]
    }
   ],
   "source": [
    "class ChatSession:\n",
    "    def __init__(self, client, system_message=\"You are a helpful assistant.\"):\n",
    "        \"\"\"\n",
    "        参数:\n",
    "            client (openai.OpenAI): OpenAI 客户端实例\n",
    "            system_message (str): 系统消息，用于设定对话背景\n",
    "        \"\"\"\n",
    "        self.client = client\n",
    "        self.messages = [{'role': 'system', 'content': system_message}]\n",
    "\n",
    "    def append_message(self, role, content):\n",
    "        \"\"\"\n",
    "        添加一条对话消息\n",
    "        参数:\n",
    "            role (str): 消息角色，为 'user' 或 'assistant'\n",
    "            content (str): 消息内容\n",
    "        \"\"\"\n",
    "        self.messages.append({'role': role, 'content': content})\n",
    "\n",
    "    def get_response(self, user_input):\n",
    "        \"\"\"\n",
    "        添加用户消息，调用 API 获取回复，并将回复加入历史记录\n",
    "        参数:\n",
    "            user_input (str): 用户输入的消息\n",
    "        返回:\n",
    "            content (str): 模型的回复内容\n",
    "        \"\"\"\n",
    "        self.append_message('user', user_input)\n",
    "        completion = self.client.chat.completions.create(\n",
    "            model=\"deepseek-v4-flash-260425\",\n",
    "            extra_body={\"thinking\": {\"type\": \"disabled\"}},  # 关闭思考\n",
    "            messages=self.messages\n",
    "        )\n",
    "        content = completion.choices[0].message.content\n",
    "        self.append_message('assistant', content)\n",
    "        return content\n",
    "\n",
    "\n",
    "# 初始化对话会话，采用默认的系统消息\n",
    "session = ChatSession(client)\n",
    "\n",
    "print(\"开始对话，输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You: \")\n",
    "    if user_input.lower() in ['exit', 'quit']:\n",
    "        break\n",
    "    reply = session.get_response(user_input)\n",
    "    print(\"AI:\", reply)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "92ce4689-2261-499c-8519-d27c3edcf1c5",
   "metadata": {},
   "source": [
    "## 思考模式"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1624fcf7-d91b-41c2-be58-8f9f51043a01",
   "metadata": {},
   "source": [
    "### 单轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "57e4f515-589a-44ba-b283-a3c920323926",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:23.959706Z",
     "iopub.status.busy": "2026-07-06T15:30:23.959572Z",
     "iopub.status.idle": "2026-07-06T15:30:25.495348Z",
     "shell.execute_reply": "2026-07-06T15:30:25.494469Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 模型推理过程 =====\n",
      "我们被问到：\"你是谁？\" 这是一个简单的自我介绍问题。作为AI助手，我应该友好地回应，说明自己的身份和功能。\n",
      "\n",
      "===== 模型回复 =====\n",
      "你好！我是 DeepSeek，一个由深度求索公司创造的 AI 助手。我可以帮你解答问题、提供信息、处理文本、分析数据等等。有什么我可以帮你的吗？😊\n",
      "\n"
     ]
    }
   ],
   "source": [
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash-260425\",\n",
    "    messages=[\n",
    "        {'role': 'system', 'content': 'You are a helpful assistant.'}, \n",
    "        {'role': 'user', 'content': '你是谁？'}\n",
    "    ]\n",
    ")\n",
    "\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 模型推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\n{content}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9fa07faa-a1b0-4f11-bcac-937b44e69d59",
   "metadata": {},
   "source": [
    "### 两轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "be56e2e6-5336-46c0-a48d-033f2afc9812",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:25.497162Z",
     "iopub.status.busy": "2026-07-06T15:30:25.497030Z",
     "iopub.status.idle": "2026-07-06T15:30:30.548948Z",
     "shell.execute_reply": "2026-07-06T15:30:30.547073Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 第一轮推理过程 =====\n",
      "我们被问到：\"9.11 和 9.8 哪个更大？\" 这是一个简单的问题，但需要注意数字的表示。9.11和9.8，通常比较小数时，9.8 = 9.80，而9.11 < 9.80，所以9.8更大。但有时人们会误解为版本号或类似的东西，但这里显然是十进制数。所以答案是9.8更大。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 9.8 更大，因为 9.8 相当于 9.80，而 9.11 小于 9.80。\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 第二轮推理过程 =====\n",
      "我们刚刚比较的是9.11和9.8。用户问“刚刚针对哪两个数值进行了比较？”，需要明确指出是9.11和9.8。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 刚刚比较的是 **9.11** 和 **9.8** 这两个数值。\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "# 第一轮对话\n",
    "messages.append({'role': 'user', 'content': '9.11 和 9.8 哪个更大？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash-260425\",\n",
    "    messages=messages\n",
    ")\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 第一轮推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\nAI: {content}\\n\")\n",
    "\n",
    "# 只将最终的回复（content）加入对话历史\n",
    "messages.append({'role': 'assistant', 'content': content})\n",
    "\n",
    "# 第二轮对话\n",
    "messages.append({'role': 'user', 'content': '刚刚针对哪两个数值进行了比较？'})\n",
    "completion = client.chat.completions.create(\n",
    "    model=\"deepseek-v4-flash-260425\",\n",
    "    messages=messages\n",
    ")\n",
    "# 获取推理思考过程和模型回复\n",
    "reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "content = completion.choices[0].message.content\n",
    "\n",
    "print(f\"===== 第二轮推理过程 =====\\n{reasoning_content}\\n\")\n",
    "print(f\"===== 模型回复 =====\\nAI: {content}\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0fef03e7-de0a-4c3d-9c49-eaed07d24a08",
   "metadata": {},
   "source": [
    "### 多轮对话"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "6b522d2f-f264-48dd-ae2d-8a7dc918e387",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:30.553337Z",
     "iopub.status.busy": "2026-07-06T15:30:30.553106Z",
     "iopub.status.idle": "2026-07-06T15:30:32.479946Z",
     "shell.execute_reply": "2026-07-06T15:30:32.478701Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们要求只回答数字，所以直接输出2。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 2.0\n",
      "\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们被问到：\"再加10呢？只回答数字。\" 前面有1+1等于2，再加10就是12。所以答案应该是12。\n",
      "\n",
      "===== 模型回复 =====\n",
      "AI: 12.0\n",
      "\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "# 初始化对话历史，包含系统消息\n",
    "messages = [{'role': 'system', 'content': 'You are a helpful assistant.'}]\n",
    "\n",
    "print(\"开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:\n",
    "        break\n",
    "\n",
    "    # 1. 记录用户输入\n",
    "    messages.append({'role': 'user', 'content': user_input})\n",
    "\n",
    "    # 2. 调用 API\n",
    "    completion = client.chat.completions.create(\n",
    "        model=\"deepseek-v4-flash-260425\",\n",
    "        messages=messages\n",
    "    )\n",
    "    reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "    content = completion.choices[0].message.content\n",
    "\n",
    "    # 3. 输出推理思考过程与最终回复\n",
    "    print(f\"===== 推理过程 =====\\n{reasoning_content}\\n\")\n",
    "    print(f\"===== 模型回复 =====\\nAI: {content}\\n\")\n",
    "\n",
    "    # 4. 只将模型的最终回复写入到历史消息中\n",
    "    messages.append({'role': 'assistant', 'content': content})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0577a773-f052-409b-bf36-e5733b40e4a2",
   "metadata": {},
   "source": [
    "### 封装对话逻辑"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "f1f23a67-d853-4d84-a8f0-51c83b0d1ceb",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-07-06T15:30:32.483978Z",
     "iopub.status.busy": "2026-07-06T15:30:32.483766Z",
     "iopub.status.idle": "2026-07-06T15:30:34.996790Z",
     "shell.execute_reply": "2026-07-06T15:30:34.994770Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\n",
      "You：1+1等于几？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们要求只回答数字，所以直接输出结果。1+1=2。\n",
      "\n",
      "===== AI回复 =====\n",
      "AI: 2.\n",
      "\n",
      "You：再加10呢？只回答数字。\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===== 推理过程 =====\n",
      "我们被问到：\"再加10呢？只回答数字。\" 前一个问题是\"1+1等于几？\" 我们回答2。然后\"再加10呢？\" 意思是2+10=12。只回答数字。\n",
      "\n",
      "===== AI回复 =====\n",
      "AI: 12.\n",
      "\n",
      "You：exit\n"
     ]
    }
   ],
   "source": [
    "class ReasonerSession:\n",
    "    def __init__(self, client, system_message=\"You are a helpful assistant.\"):\n",
    "        \"\"\"\n",
    "        参数:\n",
    "            client (openai.OpenAI): OpenAI 客户端实例\n",
    "            system_message (str): 系统消息，用于设定对话背景\n",
    "        \"\"\"\n",
    "        self.client = client\n",
    "        self.messages = [{'role': 'system', 'content': system_message}]\n",
    "\n",
    "    def append_message(self, role, content):\n",
    "        \"\"\"\n",
    "        添加一条对话消息\n",
    "        参数:\n",
    "            role (str): 消息角色，为 'user' 或 'assistant'\n",
    "            content (str): 消息内容\n",
    "        \"\"\"\n",
    "        self.messages.append({'role': role, 'content': content})\n",
    "\n",
    "    def get_response(self, user_input):\n",
    "        \"\"\"\n",
    "        添加用户消息，调用 API 获取思维链和回复，并将回复加入历史记录\n",
    "        参数:\n",
    "            user_input (str): 用户输入的消息\n",
    "        返回:\n",
    "            (reasoning_content, content) (tuple): 模型的推理过程和回复\n",
    "        \"\"\"\n",
    "        # 1. 记录用户输入\n",
    "        self.append_message('user', user_input)\n",
    "\n",
    "        # 2. 调用 API\n",
    "        completion = self.client.chat.completions.create(\n",
    "            model=\"deepseek-v4-flash-260425\",\n",
    "            messages=self.messages\n",
    "        )\n",
    "\n",
    "        # 3. 获取推理思考过程和最终回复\n",
    "        reasoning_content = getattr(completion.choices[0].message, 'reasoning_content', None)\n",
    "        content = completion.choices[0].message.content\n",
    "\n",
    "        # 4. 只将模型的最终回复加入到历史消息\n",
    "        self.append_message('assistant', content)\n",
    "\n",
    "        return reasoning_content, content\n",
    "\n",
    "\n",
    "# 初始化推理会话，采用默认的系统消息\n",
    "session = ReasonerSession(client)\n",
    "\n",
    "print(\"开始对话（Reasoner），输入 'exit' 或 'quit' 结束对话。\")\n",
    "while True:\n",
    "    user_input = input(\"You：\")\n",
    "    if user_input.lower() in ['exit', 'quit']:  # 退出条件\n",
    "        break\n",
    "\n",
    "    reasoning, reply = session.get_response(user_input)\n",
    "    print(f\"===== 推理过程 =====\\n{reasoning}\\n\")\n",
    "    print(f\"===== AI回复 =====\\nAI: {reply}\\n\")"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.12.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}