prompt-model/ai.koog.prompt.params/LLMParams

LLMParams

data class LLMParams(val temperature: Double? = null, val speculation: String? = null, val schema: LLMParams.Schema? = null, val toolChoice: LLMParams.ToolChoice? = null)(source)

Represents configuration parameters for controlling the behavior of a language model.

Constructors

LLMParams

constructor(temperature: Double? = null, speculation: String? = null, schema: LLMParams.Schema? = null, toolChoice: LLMParams.ToolChoice? = null)

Types

Schema

@Serializable

sealed interface Schema

Represents a generic schema for structured data, defining a common contract for schemas. This is a sealed interface, enabling a restrictive set of implementations.

ToolChoice

@Serializable

sealed class ToolChoice

Used to switch tool calling behavior of LLM

Properties

schema

val schema: LLMParams.Schema? = null

speculation

val speculation: String? = null

Reserved for speculative proposition of how result would look like, supported only by a number of models, but may greatly improve speed and accuracy of result. For example, in OpenAI that feature is called PredictedOutput

temperature

val temperature: Double? = null

A parameter to control the randomness in the output. Higher values encourage more diverse results, while lower values produce deterministically focused outputs. The value is optional and defaults to null.

toolChoice

val toolChoice: LLMParams.ToolChoice? = null

Used to switch tool calling behavior of LLM.

Functions

default

fun default(default: LLMParams): LLMParams

Combines the parameters of the current LLMParams instance with the provided default LLMParams to produce a new instance. Fields that are null in the current instance are replaced by the corresponding fields from the default instance.