原文出处：Prompt Applications 原作者：DAIR.AI · Elvis Saravia · 许可证：MIT License 中文译本由诸葛AI学院整理，仅供学习参考，版权归原作者与 DAIR.AI 所有。

这篇指南介绍提示工程（prompt engineering）里一些偏进阶、也挺有意思的用法，教你用提示词（prompt）去完成更复杂的实际任务。

本节仍在大量更新中。

内容目录： - 生成数据（Generating Data） - 程序辅助语言模型（Program-Aided Language Models，PAL） - Python 笔记本

生成数据

大语言模型（large language model，LLM）的文字生成能力很强。用对提示策略，你可以引导模型给出更好、更稳定、更符合事实的回答。LLM 还特别擅长生成数据，做各类实验正好用得上。比如你想给一个情感分类器快速造一批样本，可以这样下提示：

提示词： Produce 10 exemplars for sentiment analysis. Examples are categorized as either positive or negative. Produce 2 negative examples and 8 positive examples. Use this format for the examples: Q: <sentence> A: <sentiment>

输出： ``` Q: I just got the best news ever! A: Positive

Q: We just got a raise at work! A: Positive

Q: I'm so proud of what I accomplished today. A: Positive

Q: I'm having the best day ever! A: Positive

Q: I'm really looking forward to the weekend. A: Positive

Q: I just got the best present ever! A: Positive

Q: I'm so happy right now. A: Positive

Q: I'm so blessed to have such an amazing family. A: Positive

Q: The weather outside is so gloomy. A: Negative

Q: I just got some terrible news. A: Negative ```

要求 2 负 8 正，模型照单执行，格式也严格按 "Q: / A:" 来。这招很实用，指南后面另一节的测试就用到了这里生成的数据。

再看一个例子，展示怎么用 LLM 做数据标注和增广：

提示词： produce 3 wine reviews and label taste, flavor, aroma related token; present the result as a json file, in addition add the coordinate of each term for NER task

输出： [ { "review": "This Chardonnay has a rich, creamy texture with notes of vanilla and buttery oak. The taste is smooth and well-balanced, with a hint of tropical fruit on the finish.", "tokens": [ {"text": "rich", "label": "flavor", "start": 5, "end": 9}, {"text": "creamy", "label": "texture", "start": 10, "end": 16}, {"text": "vanilla", "label": "flavor", "start": 28, "end": 35}, {"text": "buttery", "label": "flavor", "start": 40, "end": 47}, {"text": "oak", "label": "aroma", "start": 48, "end": 51}, {"text": "smooth", "label": "taste", "start": 61, "end": 67}, {"text": "well-balanced", "label": "taste", "start": 68, "end": 81}, {"text": "hint", "label": "flavor", "start": 89, "end": 93}, {"text": "tropical fruit", "label": "flavor", "start": 98, "end": 112}, {"text": "finish", "label": "taste", "start": 116, "end": 122} ] }, { "review": "This Cabernet Sauvignon has a bold, full-bodied flavor with notes of black cherry and dark chocolate. The aroma is rich and complex, with hints of tobacco and leather.", "tokens": [ {"text": "bold", "label": "flavor", "start": 5, "end": 9}, {"text": "full-bodied", "label": "texture", "start": 10, "end": 21}, {"text": "black cherry", "label": "flavor", "start": 30, "end": 42}, {"text": "dark chocolate", "label": "flavor", "start": 47, "end": 60}, {"text": "rich", "label": "aroma", "start": 69, "end": 73}, {"text": "complex", "label": "aroma", "start": 78, "end": 85}, {"text": "hints", "label": "aroma", "start": 90, "end": 96}, {"text": "tobacco", "label": "aroma", "start": 101, "end": 108}, {"text": "leather", "label": "aroma", "start": 113, "end": 120} ] }, { "review": "This Riesling has a crisp, refreshing taste with notes of green apple and honey. The aroma is floral and fruity, with a hint of citrus.", "tokens": [ {"text": "crisp", "label": "texture", "start": 5, "end": 10}, {"text": "refreshing", "label": "texture", "start": 12, "end": 22}, {"text": "green apple", "label": "flavor", "start": 31, "end": 42}, ] }

模型生成 3 条酒评，把每个描述词标上类别（taste 口感、flavor 风味、aroma 香气、texture 质地），还给出每个词在句中的起止位置（start 和 end 是字符下标），正好是命名实体识别（named entity recognition，NER）任务需要的标注格式。原文的输出到第三条就中断了，这里原样保留。

PAL（程序辅助语言模型）

Gao 等人（2022）提出了一种方法：让 LLM 读自然语言题目，把生成的程序代码当作中间推理步骤。这种方法被称为程序辅助语言模型（program-aided language models，PAL）。它和思维链（chain of thought，CoT）提示的区别在于：思维链靠自由文本推出答案，PAL 则把求解这一步交给 Python 解释器这类程序运行时去算。

下面用 LangChain 和 OpenAI GPT-3 举个例子。我们想做一个简单的应用：理解用户的问题，再借助 Python 解释器算出答案。

具体来说，我们想让 LLM 回答需要理解日期的问题。提示词里放了几个示例，取自 PAL 官方仓库。

先准备需要的 import：

python import openai from datetime import datetime from dateutil.relativedelta import relativedelta import os from langchain.llms import OpenAI from dotenv import load_dotenv

先做几项配置：

```python load_dotenv()

API 配置

openai.api_key = os.getenv("OPENAI_API_KEY")

供 LangChain 使用

os.environ["OPENAI_API_KEY"] = os.getenv("OPENAI_API_KEY") ```

创建模型实例：

python llm = OpenAI(model_name='text-davinci-003', temperature=0)

准备提示词和问题：

```python question = "Today is 27 February 2023. I was born exactly 25 years ago. What is the date I was born in MM/DD/YYYY?"

DATE_UNDERSTANDING_PROMPT = """

Q: 2015 is coming in 36 hours. What is the date one week from today in MM/DD/YYYY?

If 2015 is coming in 36 hours, then today is 36 hours before.

today = datetime(2015, 1, 1) - relativedelta(hours=36)

One week from today,

one_week_from_today = today + relativedelta(weeks=1)

The answer formatted with %m/%d/%Y is

one_week_from_today.strftime('%m/%d/%Y')

Q: The first day of 2019 is a Tuesday, and today is the first Monday of 2019. What is the date today in MM/DD/YYYY?

If the first day of 2019 is a Tuesday, and today is the first Monday of 2019, then today is 6 days later.

today = datetime(2019, 1, 1) + relativedelta(days=6)

The answer formatted with %m/%d/%Y is

today.strftime('%m/%d/%Y')

Q: The concert was scheduled to be on 06/01/1943, but was delayed by one day to today. What is the date 10 days ago in MM/DD/YYYY?

If the concert was scheduled to be on 06/01/1943, but was delayed by one day to today, then today is one day later.

today = datetime(1943, 6, 1) + relativedelta(days=1)

10 days ago,

ten_days_ago = today - relativedelta(days=10)

The answer formatted with %m/%d/%Y is

ten_days_ago.strftime('%m/%d/%Y')

Q: It is 4/19/1969 today. What is the date 24 hours later in MM/DD/YYYY?

It is 4/19/1969 today.

today = datetime(1969, 4, 19)

24 hours later,

later = today + relativedelta(hours=24)

The answer formatted with %m/%d/%Y is

today.strftime('%m/%d/%Y')

Q: Jane thought today is 3/11/2002, but today is in fact Mar 12, which is 1 day later. What is the date 24 hours later in MM/DD/YYYY?

If Jane thought today is 3/11/2002, but today is in fact Mar 12, then today is 3/1/2002.

today = datetime(2002, 3, 12)

24 hours later,

later = today + relativedelta(hours=24)

The answer formatted with %m/%d/%Y is

later.strftime('%m/%d/%Y')

Q: Jane was born on the last day of Feburary in 2001. Today is her 16-year-old birthday. What is the date yesterday in MM/DD/YYYY?

If Jane was born on the last day of Feburary in 2001 and today is her 16-year-old birthday, then today is 16 years later.

today = datetime(2001, 2, 28) + relativedelta(years=16)

Yesterday,

yesterday = today - relativedelta(days=1)

The answer formatted with %m/%d/%Y is

yesterday.strftime('%m/%d/%Y')

Q: {question}

""".strip() + '\n' ```

提示词里每个示例都先推理"今天到底是哪天"，再逐步写成日期计算的代码。模型照着这个套路，把我们的新问题也翻译成一段 Python。

让模型生成代码：

python llm_out = llm(DATE_UNDERSTANDING_PROMPT.format(question=question)) print(llm_out)

执行生成的代码，输出结果：

python exec(llm_out) print(born)

最终输出：02/27/1998。日期题不在模型内部计算，全部交给解释器，答案就不会在算术上出错。

Python 笔记本

更多示例筹备中。

原章节导航：上一章：进阶提示技巧 · 下一章：ChatGPT 提示