{ "cells": [ { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "# 第7章：构建聊天应用\n", "## OpenAI API 快速入门\n", "\n", "本笔记本改编自包含访问 [Azure OpenAI](notebook-azure-openai.ipynb) 服务笔记本的 [Azure OpenAI 示例仓库](https://github.com/Azure/azure-openai-samples?WT.mc_id=academic-105485-koreyst)。\n", "\n", "Python OpenAI API 也适用于 Azure OpenAI 模型，只需做一些修改。了解更多差异，请访问：[如何使用 Python 在 OpenAI 和 Azure OpenAI 端点之间切换](https://learn.microsoft.com/azure/ai-services/openai/how-to/switching-endpoints?WT.mc_id=academic-109527-jasmineg)\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "# 概览 \n", "“大型语言模型是将文本映射到文本的函数。给定一段输入文本，大型语言模型会尝试预测接下来将出现的文本”(1)。本“快速入门”笔记本将向用户介绍高级的LLM概念、开始使用AML所需的核心包、提示设计的简单入门，以及几个不同用例的简短示例。 \n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "## 目录 \n", "\n", "[概述](#overview) \n", "[如何使用OpenAI服务](#how-to-use-openai-service) \n", "[1. 创建您的OpenAI服务](#1.-creating-your-openai-service) \n", "[2. 安装](#2.-installation) \n", "[3. 凭证](#3.-credentials) \n", "\n", "[用例](#use-cases) \n", "[1. 总结文本](#1.-summarize-text) \n", "[2. 分类文本](#2.-classify-text) \n", "[3. 生成新产品名称](#3.-generate-new-product-names) \n", "[4. 微调分类器](#4.fine-tune-a-classifier) \n", "\n", "[参考资料](#references)\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "### 构建你的第一个提示 \n", "这个简短的练习将为提交提示给 OpenAI 模型完成一个简单任务“摘要”提供基本介绍。\n", "\n", "\n", "步骤： \n", "1. 在你的 Python 环境中安装 OpenAI 库 \n", "2. 加载标准辅助库并设置你为创建的 OpenAI 服务准备的典型 OpenAI 安全凭证 \n", "3. 为你的任务选择一个模型 \n", "4. 为模型创建一个简单的提示 \n", "5. 向模型 API 提交你的请求！\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "### 1. 安装 OpenAI\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "gather": { "logged": 1674254990318 }, "jupyter": { "outputs_hidden": true, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "%pip install openai python-dotenv" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "### 2. 导入辅助库并实例化凭证\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "gather": { "logged": 1674829434433 }, "jupyter": { "outputs_hidden": false, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "import os\n", "from openai import OpenAI\n", "from dotenv import load_dotenv\n", "\n", "load_dotenv()\n", "\n", "API_KEY = os.getenv(\"OPENAI_API_KEY\",\"\")\n", "assert API_KEY, \"ERROR: OpenAI Key is missing\"\n", "\n", "client = OpenAI(\n", " api_key=API_KEY\n", " )\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "### 3. 找到合适的模型 \n", "像 GPT-4o 和 GPT-4o mini 这样的模型可以理解和生成自然语言，并且在 OpenAI 平台上提供，具有不同的性能和速度，适用于不同的任务。\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "gather": { "logged": 1674742720788 }, "jupyter": { "outputs_hidden": true, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "# Select a general purpose chat model\n", "model = \"gpt-4o-mini\"\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "## 4. 提示设计 \n", "\n", "“大型语言模型的魔力在于，通过在大量文本上训练以最小化这种预测错误，模型最终学会了对这些预测有用的概念。例如，它们学会了如下概念”(1):\n", "\n", "* 如何拼写\n", "* 语法如何运作\n", "* 如何改写\n", "* 如何回答问题\n", "* 如何进行对话\n", "* 如何用多种语言写作\n", "* 如何编程\n", "* 等等\n", "\n", "#### 如何控制大型语言模型 \n", "“在所有输入给大型语言模型的内容中，最有影响力的远远是文本提示”(1)。\n", "\n", "大型语言模型可以通过几种方式被提示来生成输出：\n", "\n", "指令：告诉模型你想要什么\n", "补全：引导模型完成你想要的开头部分\n", "演示：向模型展示你想要什么，包括：\n", "在提示中给出几个示例\n", "在微调训练数据集中给出成百上千个示例”\n", "\n", "\n", "\n", "#### 创建提示的三个基本准则：\n", "\n", "展示与说明。通过指令、示例或两者结合清楚表达你的需求。如果你想让模型按字母顺序排列一个项目列表，或按情感对段落进行分类，就明确告诉它这是你的需求。\n", "\n", "提供高质量数据。如果你在尝试构建分类器或让模型遵循某种模式，确保有足够的示例。务必校对你的示例——模型通常足够聪明，能够识别基本拼写错误并给出回应，但它也可能认为这是故意的，进而影响回答结果。\n", "\n", "**检查你的设置。** temperature 和 top_p 设置控制模型生成回应的确定性。如果你希望得到唯一正确答案的回应，应将这些参数设置较低。如果你想要更多样化的回答，可以设置得较高。人们在这些设置上最常犯的错误是以为它们是“聪明”或“创造力”的控制开关。\n", "\n", "\n", "来源：https://learn.microsoft.com/azure/ai-services/openai/overview\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "### 5. 提交！\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "gather": { "logged": 1674494935186 }, "jupyter": { "outputs_hidden": false, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "# Create your first prompt\n", "text_prompt = \"Should oxford commas always be used?\"\n", "\n", "response = client.responses.create(\n", " model=model,\n", " input = [{\"role\":\"system\", \"content\":\"You are a helpful assistant.\"},\n", " {\"role\":\"user\",\"content\":text_prompt},],\n", " store=False,)\n", "\n", "response.output_text\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "### 重复相同的调用，结果有何不同？ \n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "gather": { "logged": 1674494940872 }, "jupyter": { "outputs_hidden": false, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "\n", "response = client.responses.create(\n", " model=model,\n", " input = [{\"role\":\"system\", \"content\":\"You are a helpful assistant.\"},\n", " {\"role\":\"user\",\"content\":text_prompt},],\n", " store=False,)\n", "\n", "response.output_text\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "## 总结文本 \n", "#### 挑战 \n", "通过在文本段落末尾添加“tl;dr:”来总结文本。注意模型如何理解执行多种任务而无需额外指令。您可以尝试比tl;dr更具描述性的提示词，以修改模型的行为并自定义您收到的摘要(3)。 \n", "\n", "最近的研究表明，通过在大规模文本语料库上进行预训练，随后在特定任务上进行微调，可以在许多NLP任务和基准测试上取得显著提升。虽然模型架构通常是任务无关的，但这种方法仍然需要成千上万个特定任务的微调数据集。相比之下，人类通常只需通过少量示例或简单指令即可完成新的语言任务——这是当前的NLP系统仍普遍难以做到的。本文展示了扩大语言模型规模大大提高了任务无关的少样本性能，有时甚至能够达到以往微调方法的竞争水平。 \n", "\n", "\n", "\n", "tl;dr \n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "# 多种用例练习 \n", "1. 文本总结 \n", "2. 文本分类 \n", "3. 生成新产品名称\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "gather": { "logged": 1674495198534 }, "jupyter": { "outputs_hidden": false, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "prompt = \"Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in architecture, this method still requires task-specific fine-tuning datasets of thousands or tens of thousands of examples. By contrast, humans can generally perform a new language task from only a few examples or from simple instructions - something that current NLP systems still largely struggle to do. Here we show that scaling up language models greatly improves task-agnostic, few-shot performance, sometimes even reaching competitiveness with prior state-of-the-art fine-tuning approaches.\\n\\nTl;dr\"\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "gather": { "logged": 1674495201868 }, "jupyter": { "outputs_hidden": false, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "#Setting a few additional, typical parameters during API Call\n", "\n", "response = client.responses.create(\n", " model=model,\n", " input = [{\"role\":\"system\", \"content\":\"You are a helpful assistant.\"},\n", " {\"role\":\"user\",\"content\":prompt},],\n", " store=False,)\n", "\n", "response.output_text\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "## 分类文本 \n", "#### 挑战 \n", "将项目分类到推理时提供的类别中。在下面的示例中，我们在提示中同时提供了类别和要分类的文本(*playground_reference)。 \n", "\n", "客户询问：您好，我笔记本键盘上的一个键最近坏了，我需要更换一个：\n", "\n", "分类类别：\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "gather": { "logged": 1674499424645 }, "jupyter": { "outputs_hidden": false, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "prompt = \"Classify the following inquiry into one of the following: categories: [Pricing, Hardware Support, Software Support]\\n\\ninquiry: Hello, one of the keys on my laptop keyboard broke recently and I'll need a replacement:\\n\\nClassified category:\"\n", "print(prompt)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "gather": { "logged": 1674499378518 }, "jupyter": { "outputs_hidden": false, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "#Setting a few additional, typical parameters during API Call\n", "\n", "response = client.responses.create(\n", " model=model,\n", " input = [{\"role\":\"system\", \"content\":\"You are a helpful assistant.\"},\n", " {\"role\":\"user\",\"content\":prompt},],\n", " store=False,)\n", "\n", "response.output_text\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "## 生成新产品名称\n", "#### 挑战\n", "根据示例词创建产品名称。这里我们在提示中包含了我们将要为其生成名称的产品信息。我们还提供了类似示例以展示我们希望获得的模式。我们还将温度值设置较高，以增加随机性和更具创新性的响应。\n", "\n", "产品描述：家用奶昔机\n", "种子词：快速、健康、紧凑。\n", "产品名称：HomeShaker、Fit Shaker、QuickShake、Shake Maker\n", "\n", "产品描述：一双可以适合任何脚型的鞋子。\n", "种子词：适应性强、合脚、全适合。\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "gather": { "logged": 1674257087279 }, "jupyter": { "outputs_hidden": false, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "prompt = \"Product description: A home milkshake maker\\nSeed words: fast, healthy, compact.\\nProduct names: HomeShaker, Fit Shaker, QuickShake, Shake Maker\\n\\nProduct description: A pair of shoes that can fit any foot size.\\nSeed words: adaptable, fit, omni-fit.\"\n", "\n", "print(prompt)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "jupyter": { "outputs_hidden": false, "source_hidden": false }, "nteract": { "transient": { "deleting": false } } }, "outputs": [], "source": [ "#Setting a few additional, typical parameters during API Call\n", "\n", "response = client.responses.create(\n", " model=model,\n", " input = [{\"role\":\"system\", \"content\":\"You are a helpful assistant.\"},\n", " {\"role\":\"user\",\"content\":prompt}],\n", " store=False,)\n", "\n", "response.output_text\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "# 参考资料 \n", "- [Openai Cookbook](https://github.com/openai/openai-cookbook?WT.mc_id=academic-105485-koreyst) \n", "- [Microsoft Foundry 门户](https://ai.azure.com?WT.mc_id=academic-105485-koreyst) \n", "- [微调 GPT 模型以分类文本的最佳实践](https://platform.openai.com/docs/guides/fine-tuning?WT.mc_id=academic-105485-koreyst)\n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "# 需要更多帮助 \n", "[OpenAI Commercialization Team](AzureOpenAITeam@microsoft.com) \n" ] }, { "attachments": {}, "cell_type": "markdown", "metadata": { "nteract": { "transient": { "deleting": false } } }, "source": [ "# 贡献者\n", "* Louis Li \n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "---\n\n\n**免责声明**：\n本文件由 AI 翻译服务 [Co-op Translator](https://github.com/Azure/co-op-translator) 翻译完成。尽管我们力求准确，但请注意，自动翻译可能包含错误或不准确之处。原始语言版文件应视为权威来源。对于重要信息，建议使用专业人工翻译。我们对因使用本翻译而产生的任何误解或误释不承担责任。\n\n" ] } ], "metadata": { "kernel_info": { "name": "python310-sdkv2" }, "kernelspec": { "display_name": "base", "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": "3.10.13" }, "microsoft": { "host": { "AzureML": { "notebookHasBeenCompleted": true } } }, "nteract": { "version": "nteract-front-end@1.0.0" } }, "nbformat": 4, "nbformat_minor": 2 }