Struct SamplingParams 

pub struct SamplingParams {

    pub max_tokens: usize,
    pub temperature: f32,
    pub top_p: f32,
    pub top_k: Option<usize>,
    pub repetition_penalty: f32,
    pub presence_penalty: f32,
    pub frequency_penalty: f32,
    pub stop_sequences: Vec<String>,
    pub seed: Option<u64>,
    pub min_p: Option<f32>,
    pub tfs: Option<f32>,
    pub typical_p: Option<f32>,
    pub mirostat: Option<MirostatParams>,
    pub response_format: ResponseFormat,
    pub structured_output_start: StructuredOutputStart,
    pub response_completion_boundary: ResponseCompletionBoundary,
}

Sampling parameters for generation

Fields

max_tokens: usize

Maximum number of tokens to generate

temperature: f32

Temperature for randomness (0.0 = deterministic, higher = more random)

top_p: f32

Nucleus sampling probability threshold

top_k: Option<usize>

Top-k sampling - consider only top k tokens

repetition_penalty: f32

Repetition penalty to reduce repetitive text

presence_penalty: f32

Presence penalty for token diversity

frequency_penalty: f32

Frequency penalty based on token frequency

stop_sequences: Vec<String>

Stop sequences to end generation

seed: Option<u64>

Random seed for reproducible generation

min_p: Option<f32>

Minimum probability threshold for tokens

tfs: Option<f32>

Tail free sampling parameter

typical_p: Option<f32>

Typical sampling parameter

mirostat: Option<MirostatParams>

Mirostat sampling parameters

response_format: ResponseFormat

Response format constraint (JSON mode, schema-constrained, etc.)

structured_output_start: StructuredOutputStart

Point at which a structured-output grammar starts constraining model output. Thinking templates can open a reasoning block in the prompt; in that case the grammar activates only after the typed delimiter.

response_completion_boundary: ResponseCompletionBoundary

Product completion boundary that must be satisfied before model EOS tokens may terminate generation.

This is independent of structured-output grammar activation: ordinary text and structured responses need a complete reasoning-to-payload transition when the rendered prompt opens one. A complete lexical response envelope, such as a tool call, is an alternate terminal path.

Implementations

impl SamplingParams

pub fn greedy() -> Self

Create greedy sampling parameters (deterministic)

pub fn with_temperature(temperature: f32) -> Self

Create default sampling parameters with temperature

pub fn validate(&self) -> Result<()>

Validate sampling parameters

Trait Implementations

impl Clone for SamplingParams

fn clone(&self) -> SamplingParams

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for SamplingParams

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

Formats the value using the given formatter.

impl Default for SamplingParams

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for SamplingParams

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for SamplingParams

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.