Getting Started

One of the key component of any LLM and DotnetPrompt is a PromptTemplate.

This article aims to clarify the concept of prompt templates and how they can assist you in producing high-quality inputs for your models. You will understand how to create prompt templates that are customized to your specific requirements and how to use few-shot learning to generate additional examples.

What is a prompt template?

A prompt template is a pre-defined structure that can be filled in with specific information to generate a prompt. The beauty of prompt templates lies in their ability to be highly reproducible. By providing a standardized format for prompts, prompt templates can help ensure consistency in the quality of inputs that are fed into the model.

At its core, a prompt template consists of a text string or "the template," which can take in parameters from the end user to generate a prompt. These parameters can include language model instructions, few-shot examples to improve the model's response, or specific questions for the model to answer.

To help you better understand prompt templates, we've included a code snippet below that showcases a basic prompt template. With the ability to customize this template with specific parameters, you can generate a wide range of different prompts to suit your needs and tasks.

I want you to act as a naming consultant for new companies.
Here are some examples of good company names:
- search engine, Google
- social media, Facebook
- video sharing, YouTube
The name should be short, catchy and easy to remember.
What is a good name for a company that makes {product}?

Input Variables

Input variables are the variables that are used to fill in the template string. In the example above, the input variable is a {product}.

Given an input variables, the PromptTemplate can generate a prompt by filling in the template string with input values. For example, if the input value is mobile phone, the template string can be formatted by IPromptTemplate.Format method to generate the following prompt:

I want you to act as a naming consultant for new companies.
Here are some examples of good company names:
- search engine, Google
- social media, Facebook
- video sharing, YouTube
The name should be short, catchy and easy to remember." +
What is a good name for a company that makes mobile phone?

Create a Prompt Template

You can create prompts using the PromptTemplate class. Prompt templates can take any number of input variables, and can be formatted with input values to generate a prompt.

var template = "I want you to act as a naming consultant for new companies.\n\n" +
               "Here are some examples of good company names:\n\n" +
               "- search engine, Google\n" +
               "- social media, Facebook\n" +
               "- video sharing, YouTube\n\n" +
               "The name should be short, catchy and easy to remember.\n\n" +
               "What is a good name for a company that makes {product}?\n";

var prompt = new PromptTemplate(
    template: template,
    inputVariables: new[] { "product" });

When you want to fill template with values you need to use Dictionary<string, string> where keys should be the same as your input variables and values could be any valid string that need to be fill in template.

var values = new Dictionary<string, string>()
{
    { "product", "toy car" }
};

var finalPrompt = prompt.Format(values);

You could read more about prompt template and how to use it here.

Note

Currently, the template should be formatted as a C# formatted string. In the future, we will add more templating languages such as Mustache.

Pass few shot examples to a prompt template

The FewShotPromptTemplate class combines prefixes, examples, and suffixes into a prompt that is suitable for LLMs. This allows the user to create prompts with a few examples that can significantly enhance the model's accuracy.

For a complete list of examples and use cases for few-shot learning, you can refer to this page. Here's an example to demonstrate how FewShotPromptTemplate works:

var prefix = new PromptTemplate("I want you to act as a naming consultant for new companies.\n" +
                                "Here are some examples of good company names:");
var example = new PromptTemplate("- {product}, {company}");
var suffix = new PromptTemplate("The name should be short, catchy and easy to remember.\n" +
                                "What is a good name for a company that makes {product}?\n");

var examples = new List<IDictionary<string, string>>()
{
    new Dictionary<string, string>()
        { { "product", "search engine" }, { "company", "Google" } },
    new Dictionary<string, string>()
        { { "product", "social media" }, { "company", "Facebook" } },
    new Dictionary<string, string>()
        { { "product", "video sharing" }, { "company", "YouTube" } },
};

var prompt = new FewShotPromptTemplate(prefix, example, suffix, examples)
{
    ExampleSeparator = "\n"
};

var values = new Dictionary<string, string>()
{
    { "product", "toy cars" }
};

Console.WriteLine(prompt.Format(values));

I want you to act as a naming consultant for new companies.
Here are some examples of good company names:
- search engine, Google
- social media, Facebook
- video sharing, YouTube
The name should be short, catchy and easy to remember.
What is a good name for a company that makes toy cars?

Select examples for a prompt template

If you have a large number of examples, you can use the implementation of IExampleSelector to select a subset of examples that will be most informative for the Language Model. This will help you generate a prompt that is more likely to generate a good response.

There are a lot of different approaches to select examples, but currently only one is supported: LengthBasedExampleSelector which try to fit your examples into maximum allowed size of input for a model input size.