聊聊如何设计一个优秀的提示应用程序？

在本指南中，我们将介绍一些高级和有趣的方法，我们可以使用提示工程来执行有用和更高级的任务。

请注意，此部分正在大量开发中。主题：

生成数据
程序辅助语言模型
Python 笔记本

一、生成数据

LLM 具有强大的文本生成能力。使用有效的提示策略可以引导模型产生更好、更一致和更符合事实的响应。LLM 对于生成对运行各种实验有用的数据也特别有用。例如，我们可以使用它为情感分类器生成快速样本，如下所示：

迅速的：

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></sentiment></sentence>

输出：

Q: 我刚得到了有史以来最好的消息！
A: 积极的
 
Q: 我们刚刚在工作中得到了加薪！
A: 积极的
 
Q: 我为今天所取得的成就感到非常自豪。
A: 积极的
 
Q: 我度过了有史以来最好的一天！
A: 积极的
 
Q: 我非常期待周末的到来。
A: 积极的
 
Q: 我刚收到了有史以来最好的礼物！
A: 积极的
 
Q: 我现在非常开心。
A: 积极的
 
Q: 我很幸运有这么一支了不起的家庭。
A: 积极的
 
Q: 外面的天气很阴暗。
A: 消极的
 
Q: 我刚得到了一些可怕的消息。
A: 消极的

这非常有用。我们将此示例用于指南另一部分中的不同测试。

二、PAL（程序辅助语言模型）

Gao 等人 (2022)提出了一种使用 LLM 阅读自然语言问题并生成程序作为中间推理步骤的方法。创造的程序辅助语言模型 (PAL) 与思维链提示的不同之处在于，它不是使用自由格式的文本来获得解决方案，而是将解决方案步骤卸载到编程运行时，例如 Python 解释器。

我们一起聊聊提示应用程序

让我们看一个使用 LangChain 和 OpenAI GPT-3 的例子。我们有兴趣开发一个简单的应用程序，该应用程序能够解释所提出的问题并通过利用 Python 解释器提供答案。

具体来说，我们有兴趣创建一个函数，允许使用 LLM 来回答需要理解日期的问题。我们将为 LLM 提供提示，其中包括从此处采用的一些范例。

这些是我们需要的导入：

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

让我们首先配置一些东西：

1	`load_dotenv()`

API configuration

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

for LangChain

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

设置模型实例：

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

设置提示+问题：

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'