Struct Parameters

pub struct Parameters {

    pub temperature: Option<f32>,
    pub top_p: Option<f32>,
    pub top_k: Option<u32>,
    pub frequency_penalty: Option<f32>,
    pub presence_penalty: Option<f32>,
    pub repetition_penalty: Option<f32>,
    pub min_p: Option<f32>,
    pub top_a: Option<f32>,
    pub seed: Option<u32>,
    pub max_tokens: Option<u32>,
    pub logit_bias: Option<Vec<(String, f32)>>,
    pub logprobs: Option<bool>,
    pub top_logprobs: Option<u8>,
    pub stop: Option<Vec<String>>,
    pub tools: Tools,
    pub tool_choice: Option<Vec<String>>,
}

Parameters for configuring the behavior of a language model.

This struct contains various parameters that can be used to control how a language model generates responses. All parameters are optional and use the builder pattern for easy configuration.

Examples

use ai_types::llm::model::Parameters;

let params = Parameters::default()
    .temperature(0.7)
    .top_p(0.9)
    .max_tokens(1000)
    .seed(42);

Fields

temperature: Option<f32>

Sampling temperature.

Controls randomness in generation. Higher values (e.g., 1.0) make output more random, lower values (e.g., 0.1) make it more deterministic.

top_p: Option<f32>

Nucleus sampling probability.

Only consider tokens with cumulative probability up to this value. Typical values are between 0.9 and 1.0.

top_k: Option<u32>

Top-k sampling parameter.

Only consider the k most likely tokens at each step.

frequency_penalty: Option<f32>

Frequency penalty to reduce repetition.

Positive values penalize tokens that have already appeared.

presence_penalty: Option<f32>

Presence penalty to encourage new tokens.

Positive values encourage the model to talk about new topics.

repetition_penalty: Option<f32>

Repetition penalty to penalize repeated tokens.

Values > 1.0 discourage repetition, values < 1.0 encourage it.

min_p: Option<f32>

Minimum probability for nucleus sampling.

Alternative to top_p that sets a minimum threshold for token probabilities.

top_a: Option<f32>

Top-a sampling parameter.

Adaptive sampling that adjusts the number of considered tokens.

seed: Option<u32>

Random seed for reproducibility.

Use the same seed to get deterministic outputs.

max_tokens: Option<u32>

Maximum number of tokens to generate.

Limits the length of the generated response.

logit_bias: Option<Vec<(String, f32)>>

Biases for specific logits.

Each tuple contains a token string and its bias value.

logprobs: Option<bool>

Whether to return log probabilities.

When true, the model returns probability information for tokens.

top_logprobs: Option<u8>

Number of top log probabilities to return.

Only used when logprobs is true.

stop: Option<Vec<String>>

Stop sequences to end generation.

Generation stops when any of these strings are encountered.

tools: Tools

Tools available to the model.

Defines what external functions the model can call.

tool_choice: Option<Vec<String>>

Tool choices available to the model.

Specifies which tools the model is allowed to use.

Implementations

impl Parameters

pub fn temperature(self, value: f32) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn top_p(self, value: f32) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn top_k(self, value: u32) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn frequency_penalty(self, value: f32) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn presence_penalty(self, value: f32) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn repetition_penalty(self, value: f32) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn min_p(self, value: f32) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn top_a(self, value: f32) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn seed(self, value: u32) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn max_tokens(self, value: u32) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn logit_bias(self, value: Vec<(String, f32)>) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn logprobs(self, value: bool) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn top_logprobs(self, value: u8) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

pub fn stop(self, value: Vec<String>) -> Self

Sets the parameter value using a builder pattern.

Arguments

• value - The value to set for this parameter

Trait Implementations

impl Debug for Parameters

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for Parameters

fn default() -> Parameters

Returns the “default value” for a type.