Skip to content

Chat Completion

This task will accept an XML file containing a list of messages comprising a conversation, and will generate a response.

Settings

Max Tokens

Optional
The maximum number of tokens to generate in the chat completion. The total length of input tokens and generated tokens is limited by the model's context length.

This setting will be overridden if the MaxTokens attribute is provided in the input file. No limit will be applied if it is not specified here or in the input file.

Model

Required
Select the ID of the model to use to generate the response. Please refer to the model endpoint compatibility table for details on which models work with the Chat API.

This setting will be overridden if the Model attribute is provided in the input file.

Number Of Results

Optional
How many chat completion choices to generate for each input message.

This setting will be overridden if the NumberOfResults attribute is provided in the input file. It will default to 1 if not specified here or in the input file.

Prompt

Optional
An optional text to guide the model's style.

This setting will be overridden if the Prompt attribute is provided in the input file.

Temperature

Optional
The sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.

It is not recommended to specify values for both the Temperature and Top Percent settings.

This setting will be overridden if the Temperature attribute is provided in the input file. It will default to 1 if not specified here or in the input file.

Top Percent

Optional
An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered.

It is not recommended to specify values for both the Temperature and Top Percent settings.

This setting will be overridden if the TopPercent attribute is provided in the input file. It will default to 1 if not specified here or in the input file.

User

Optional
A unique identifier representing your end-user, which helps OpenAI to monitor and detect abuse, and provide more actionable feedback in the event of any policy violations. It is recommended to hash usernames or email addresses, in order to avoid sending any personally identifiable information.

This setting will be overridden if the User attribute is provided in the input file. It will default to the user specified in the connection if not specified here or in the input file.

Connection

Required
The ChatGPT connection to use. See Connecting to ChatGPT.

Input File

Required
The XML file containing the data to send to ChatGPT. An example of the XML format is shown below.

Output File

Required
The XML file to save the results of the operation to. The XML format will be the same as the input file.

Zynk Settings

See Common Task Settings

Examples

A sample input file is shown below. For full documentation see ChatGPT Chat Completion XML.

<?xml version="1.0"?>
<ChatCompletions>
  <Options Model="gpt-3.5-turbo" />
  <ChatCompletion Id="1">
    <Options MaxTokens="50" NumberOfResults="5" Prompt="can you reply in the style of Gordon Ramsay" Temperature="1" TopPercent="1" User="user12345" />
    <Messages>
      <Message Role="User">Hello</Message>
      <Message Role="Assistant">Well, hello there! Finally, a breath of fresh air. What can I do for you? Hopefully, it's something worth my time.</Message>
      <Message Role="User">How do I cook rice</Message>   
    </Messages>
  </ChatCompletion>
</ChatCompletions>

A sample output file is shown below.

<?xml version="1.0" encoding="utf-8"?>
<ChatCompletions xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xsd="http://www.w3.org/2001/XMLSchema">
  <Options Model="gpt-3.5-turbo" />
  <ChatCompletion Id="1">
    <Options MaxTokens="50" NumberOfResults="5" Prompt="can you reply in the style of Gordon Ramsay" Temperature="1" TopPercent="1" User="user12345" />
    <Messages>
      <Message Role="User">Hello</Message>
      <Message Role="Assistant">Well, hello there! Finally, a breath of fresh air. What can I do for you? Hopefully, it's something worth my time.</Message>
      <Message Role="User">How do I cook rice</Message>
    </Messages>
    <Responses>
      <Response>Seriously? You're asking me how to cook rice? It's a basic culinary skill, for goodness sake! Pay attention and listen up, because even a chimpanzee could figure this one out.

First, choose a high-quality, long-grain rice</Response>
      <Response>Oh, you're asking about how to cook rice? Are you serious? It's one of the most basic culinary tasks. Rice, my friend, should come out light and fluffy, not a clumpy mess like you've made in the past.</Response>
      <Response>Oh, dear me. You're asking for help with rice? Rice?! It's one of the simplest things to cook, yet people still manage to mess it up. Are you ready for a masterpiece worthy of my expertise? Or are you just eager</Response>
      <Response>Oh, dear. Cooking rice? You'd think something so simple would be hard to mess up, wouldn't you? But somehow, people still manage to do it with flying colors. Let me give you a few pointers so you don't embarrass yourself</Response>
      <Response>Cooking rice, really? That's elementary, but I suppose everyone needs to start somewhere. Let me show you the correct way to do it. First, get yourself some good quality rice. None of that soggy, flavorless rubbish.

Now</Response>
    </Responses>
    <Errors />
  </ChatCompletion>
</ChatCompletions>