Using Functions in Generated Speech Nodes

This feature allows Large Language Models (LLMs), such as GPTs, to use user-defined functions. In broader terms, it provides the capability for GPTs to utilize a set of tools to solve the problem at hand. This is particularly useful when you need the LLM to carry out specific, well-defined tasks within your application.

One key area where this functionality is leveraged is within the Generated Speech Node. However, using functions in this context can be a bit tricky and requires some basic knowledge of Kotlin programming. Below you will find a detailed example and an explanation of how to implement and use these functions.

Example Code

To apply user-defined functions to a generated speech node, simply use the function.

Here's a code snippet.

data class GetCurrentWeatherInput(
    @Description("The city and state, e.g. San Francisco, CA")
    val location: String
)

val myFunctions = listOf(
    llmFunction(
        name = "get_current_weather",
        description = "Get the current weather in a given location"
    ) { input: GetCurrentWeatherInput ->
        "Weather in ${input.location} is 20 degrees."
    },
)

Let's break down the example code step-by-step:

Data Class Definition

data class GetCurrentWeatherInput(
    @Description("The city and state, e.g. San Francisco, CA")
    val location: String
)

This represents input for one function.

Explanation:

  • data class GetCurrentWeatherInput: This defines a Kotlin data class named GetCurrentWeatherInput. You can name it anything you like, for consistency, we suffix with Input.

  • @Description("The city and state, e.g. San Francisco, CA"): This annotation provides a description for the location parameter. It helps document the purpose of the field and what kind of data it expects.

  • val location: String: This defines a single property, location, of type String. It signifies the city and state for which the weather is to be fetched. These data types are also supported, such as String, numbers (Integer, Float, Double) and Boolean.

List of Functions

val myFunctions = listOf(
    llmFunction(
        name = "get_current_weather",
        description = "Get the current weather in a given location"
    ) { input: GetCurrentWeatherInput ->
        "Weather in ${input.location} is 20 degrees."
    }
)

Explanation:

  • val myFunctions: Declares a variable myFunctions which contains a list of functions.

  • listOf(...): Creates a list with the specified elements—in this case, a single function created by the llmFunction method. You can add as many functions as needed.

  • llmFunction(...) { ... }: Defines a function that the LLM can call. It specifies a unique name and description for the function and provides its implementation.

Inside the llmFunction call:

  • name = "get_current_weather": Assigns the name get_current_weather to the function. This is the name used to invoke the function via the LLM.

  • description = "Get the current weather in a given location": Provides a description that explains the purpose of the function.

  • Function Body: Defined inside the curly braces {...}, it is a lambda expression that has input and output types. In this example, the input type is GetCurrentWeatherInput and the output is a simple String.

PreviousModeration APINextDetect sentiment

Last updated