openai-api - 如何记住来自不同 ChatCompletion 实例的消息

Question

我正在尝试使用 chatgpt api 运行一些命令，但它并没有运行所有内容，因此我创建了另一个 ChatCompletion 实例。如何让 chatgpt 记住我在先前实例中发送的提示？这是我的代码:

data = openai.ChatCompletion.create(
  model="gpt-3.5-turbo",
  messages=[
        {"role": "user", "content": "Generate a title"},
        {"role": "user", "content": "Generate 30 chapters"},
    ]
)

data2 = openai.ChatCompletion.create(
  model="gpt-3.5-turbo",
  messages=[
          {"role": "user", "content": "Generate chapter 1"},
])

Answer 1

为了能够使模型保持在上下文中，您必须将以前的对话与当前提示一起发送。

[注意:对话历史记录必须小于 token 限制才能预期模型的正确行为，为避免这种情况，您可以在第 n 个响应后总结对话或使用文本嵌入]

这里是一个示例代码:

import openai
import time

# Set your OpenAI API key
openai.api_key = "sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxx"

# Define the engine and the query
engine = "gpt-3.5-turbo"
query = "Generate a title"

# Start a new conversation with an empty list of messages
messages = []

# Call the Chat Completion API to generate a title
data = openai.ChatCompletion.create(
  engine=engine,
  query=query,
  messages=messages
)

# Print the generated title
title = data["message"]
print(title)

# Update the messages with the title
messages.append({"role": "assistant", "content": title})

# Define a function to generate chapters
def generate_chapters(n):
  # Loop for n times
  for i in range(n):
    # Generate a query for each chapter
    query = f"Generate chapter {i+1}"
    # Call the Chat Completion API to generate a chapter
    data = openai.ChatCompletion.create(
      engine=engine,
      query=query,
      messages=messages
    )
    # Print the generated chapter
    chapter = data["message"]
    print(chapter)
    # Update the messages with the chapter
    messages.append({"role": "assistant", "content": chapter})
    # Wait for 10 seconds before generating the next chapter
    time.sleep(10)

# Generate 30 chapters
generate_chapters(30)