baidu-ai-search

{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# AppBuilder Assistant SDK FunctionCall\n", "\n", "您可基于 Assistants SDK，可通过全代码形式创建和调试专属智能体，实现FunctionCall等功能。\n", "\n", "以下通过一个端到端的示例，介绍如何使用Assistants SDK创建智能体，并调用AppBuilder SDK内置的工具组件。\n", "\n", "该应用的场景是动物识别，通过Assistant SDK创建智能体，并调用AppBuilder SDK内置的动物识别组件，实现识别动物的功能。\n", "\n", "Assistants API/SDK 正在内测中，敬请期待公测版本。" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 一、环境准备\n", "\n", "首先需要安装AppBuilder-SDK代码库，若已在开发环境安装，则可跳过此步。\n", "\n", "**注意：**: appbuilder-sdk 的python版本要求 3.9+，安装的SDK version >= 0.7.0" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "!python3 -m pip install appbuilder-sdk" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 二、代码实现\n", "\n", "### 2.1 创建Assitant & Thread & Message\n", "\n", "我们此处调用 `appbuilder.AnimalRecognition()`组件，需要Assitant识别的图片为一只大熊猫，示例图如下：\n", "\n", "\n", "<img src=\"https://bj.bcebos.com/v1/appbuilder/animal_recognize_test.png?authorization=bce-auth-v1%2FALTAKGa8m4qCUasgoljdEDAzLm%2F2024-01-24T12%3A19%3A16Z%2F-1%2Fhost%2F411bad53034fa8f9c6edbe5c4909d76ecf6fad6862cf937c03f8c5260d51c6ae\" alt=\"drawing\" width=\"1000\"/>" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import os\n", "import appbuilder\n", "\n", "# 配置你的密钥，主要在这之前需要首先申请Assistant API的内测资格\n", "os.environ[\"APPBUILDER_TOKEN\"] = \"your_appbuilder_token\"\n", "\n", "# 创建 Assitant\n", "assistant = appbuilder.assistant.assistants.create(\n", " name=\"test_function\",\n", " description=\"你是一个热心的朋友\",\n", " instructions=\"请用友善的语气回答问题\",\n", " tools=[\n", " {'type': 'function', 'function': appbuilder.AnimalRecognition().manifests[0]}\n", " ]\n", ")\n", "\n", "# 创建 Thread（即会话：conversation）\n", "thread = appbuilder.assistant.threads.create()\n", "\n", "image_url = \"https://bj.bcebos.com/v1/appbuilder/animal_recognize_test.png?\" \\\n", " \"authorization=bce-auth-v1%2FALTAKGa8m4qCUasgoljdEDAzLm%2F2024-01-24T\" \\\n", " \"12%3A19%3A16Z%2F-1%2Fhost%2F411bad53034fa8f9c6edbe5c4909d76ecf6fad68\" \\\n", " \"62cf937c03f8c5260d51c6ae\"\n", "\n", "origin_query = \"我有一张图片，url是: {}, 麻烦帮我看看这是什么动物\".format(image_url)\n", "\n", "# 在Thread中添加一条Message，内容为原始的query\n", "appbuilder.assistant.threads.messages.create(\n", " thread_id=thread.id,\n", " content=origin_query,\n", ")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 2.2 第一次运行，触发函数调用\n", "\n", "我们向Assistant询问 `我有一张图片，url是: X, 麻烦帮我看看这是什么动物`, Assistant会调用我们预先配置好的函数，并返回调用参数，我们收到调用触发的信号后，在本地调用函数" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 触发第一次run，并收到函数本地调用的信号\n", "run_result = appbuilder.assistant.threads.runs.run(\n", " thread_id=thread.id,\n", " assistant_id=assistant.id,\n", ")\n", "\n", "print(\"\\nFirst run result: {}\\n\".format(run_result.model_dump_json(indent=4)))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "运行结果为：\n", "\n", "```bash\n", "First run result: {\n", " \"id\": \"run_089cffe9a82944bd9cdb87a8a0d8b54f\",\n", " \"object\": \"run.result\",\n", " \"assistant_id\": \"asst_922cacd503354349a1a0a5afb0be0408\",\n", " \"thread_id\": \"thread_523a4aacc6a845a0b8252009af5771ea\",\n", " \"model\": \"\",\n", " \"instructions\": \"\",\n", " \"thought_instructions\": \"\",\n", " \"chat_instructions\": \"\",\n", " \"tools\": null,\n", " \"file_ids\": null,\n", " \"status\": \"requires_action\",\n", " \"required_action\": {\n", " \"type\": \"submit_tool_outputs\",\n", " \"submit_tool_outputs\": {\n", " \"tool_calls\": [\n", " {\n", " \"id\": \"call-thread_523a4aacc6a845a0b8252009af5771ea-run_089cffe9a82944bd9cdb87a8a0d8b54f-step-1\",\n", " \"type\": \"function\",\n", " \"function\": {\n", " \"name\": \"animal_rec\",\n", " \"arguments\": \"{\\\"img_url\\\":\\\"https://bj.bcebos.com/v1/appbuilder/animal_recognize_test.png?authorization=bce-auth-v1%2FALTAKGa8m4qCUasgoljdEDAzLm%2F2024-01-24T12%3A19%3A16Z%2F-1%2Fhost%2F411bad53034fa8f9c6edbe5c4909d76ecf6fad6862cf937c03f8c5260d51c6ae\\\"}\",\n", " \"output\": \"\"\n", " }\n", " }\n", " ]\n", " }\n", " },\n", " \"last_error\": null,\n", " \"final_answer\": null,\n", " \"created_at\": 1713868218144,\n", " \"started_at\": 1713868218164,\n", " \"expired_at\": 0,\n", " \"cancelled_at\": 0,\n", " \"failed_at\": 0,\n", " \"completed_at\": 0\n", "}\n", "```" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 2.3 识别函数调用的触发信号，调用本地函数" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 触发函数调用的标志是： status == 'requires_action'\n", "assert run_result.status == 'requires_action'\n", "\n", "# 获取函数调用的信息\n", "tool_call = run_result.required_action.submit_tool_outputs.tool_calls[0]\n", "\n", "# 调用工具组件的 tool_eval 方法，获取函数本地的运行结果\n", "func_res = appbuilder.AnimalRecognition().tool_eval(\n", " name=\"animal_rec\",\n", " streaming=True,\n", " origin_query=origin_query,\n", " **eval(tool_call.function.arguments))\n", "\n", "func_message = \"\"\n", "for res in func_res:\n", " func_message += res\n", "print(\"\\nFunction result: {}\\n\".format(func_message))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "运行结果为：\n", "\n", "\n", "```bash\n", "Function result: 模型识别结果为：\n", "类别: 国宝大熊猫 置信度: 0.975161\n", "```" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 2.4 上传函数运行结果，继续触发运行，得到完整对话结果" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "run_result = appbuilder.assistant.threads.runs.run(\n", " thread_id=thread.id,\n", " assistant_id=assistant.id,\n", " tool_output={\n", " \"tool_call_id\":tool_call.id,\n", " \"output\": func_message,\n", " \"run_id\": run_result.id\n", " },\n", ")\n", "print(\"\\nFinal run result: {}\\n\".format(run_result.model_dump_json(indent=4)))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "运行结果为：\n", "\n", "```bash\n", "\n", "Final run result: {\n", " \"id\": \"run_089cffe9a82944bd9cdb87a8a0d8b54f\",\n", " \"object\": \"run.result\",\n", " \"assistant_id\": \"asst_922cacd503354349a1a0a5afb0be0408\",\n", " \"thread_id\": \"thread_523a4aacc6a845a0b8252009af5771ea\",\n", " \"model\": \"\",\n", " \"instructions\": \"\",\n", " \"thought_instructions\": \"\",\n", " \"chat_instructions\": \"\",\n", " \"tools\": null,\n", " \"file_ids\": null,\n", " \"status\": \"completed\",\n", " \"required_action\": null,\n", " \"last_error\": null,\n", " \"final_answer\": {\n", " \"type\": \"message\",\n", " \"message\": {\n", " \"message_id\": \"chatmsg_8d54f8c3c6714665baeda5eed013a5e9\",\n", " \"content\": {\n", " \"type\": \"text\",\n", " \"text\": {\n", " \"value\": \"根据您的图片，我识别出图片中的动物是**国宝大熊猫**，识别结果的置信度很高，为0.975161。如果您还有其他问题或需要识别其他图片，请随时告诉我。\",\n", " \"annotations\": null\n", " }\n", " }\n", " }\n", " },\n", " \"created_at\": 1713868218144,\n", " \"started_at\": 1713868239553,\n", " \"expired_at\": 0,\n", " \"cancelled_at\": 0,\n", " \"failed_at\": 0,\n", " \"completed_at\": 1713868251090\n", "}\n", "```" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 总结\n", "\n", "在本示例中，展示了如何使用Assistant SDK，并调用AppBuilder中的能力组件，原子化的进行Agent流程编排与交互" ] } ], "metadata": { "language_info": { "name": "python" } }, "nbformat": 4, "nbformat_minor": 2 }