python - OpenGPT API 输出中的 `index`

Question

下面是代码

import os
import openai


openai.api_key = "..."


response = openai.Completion.create(
  model="text-davinci-003",     
  prompt="I am happy!",
  temperature=0, #creativity
  max_tokens=10,
  top_p=1,
  frequency_penalty=0.0,
  presence_penalty=0.0,
  suffix='I am even more happy!'
)

print(response)

输出如下

{
  "choices": [
    {
      "finish_reason": "length",
      "index": 0,
      "logprobs": null,
      "text": "\n\nI am happy because I am surrounded by"
    }
  ],
  "created": 1674640360,
  "id": "cmpl-6cWkK124234ho8C2134afasdasdnwDKLUMP",
  "model": "text-davinci-003",
  "object": "text_completion",
  "usage": {
    "completion_tokens": 10,
    "prompt_tokens": 10,
    "total_tokens": 20
  }
}

下面输出中的index代表什么？

Answer 1

索引是指回复的提示的索引。

由于 prompt 可以是字符串或数组，documentation将其称为:

prompt ^{string or array}

The prompt(s) to generate completions for, encoded as a string, array of strings, array of tokens, or array of token arrays.

---

例如，考虑调用 /completions带有 2 个提示的端点:

{
    "model": "text-davinci-003",
    "prompt": [ 
      "How should I",                            // [ 0 ]
      "Should I"                                 // [ 1 ] 
    ],
    "max_tokens": 7,
    "temperature": 0
}

OpenAI 将回答这两个提示，index 将引用 prompt 中原始索引的回复:

{
    "choices": [
        {
            "text": " go about doing this?\n\n",
            "index": 0,                              // Answers prompt[0]
            "logprobs": null,
            "finish_reason": "length"
        },
        {
            "text": " use a VPN on my iPhone?",
            "index": 1,                              // Answers prompt[1]
            "logprobs": null,
            "finish_reason": "length"
        }
    ]
}

---

在您的示例中，只有 1 个 prompt，因此 API 只会在 choices 数组中发送 1 个对象，因此 index将始终为 0。