Module sampling 

Sampling and generation parameters

Structs

MirostatParams

Mirostat sampling parameters

ResponseCompletionEnvelope

A lexical response envelope that can complete an otherwise pending response.

SamplingParams

Sampling parameters for generation

SamplingPresets

Sampling presets

SpecialTokens

Special tokens configuration

Enums

FinishReason

Reason for completion

Priority

Request priority levels

ResponseCompletionBoundary

Typed model-EOS boundary for product response completion.

ResponseFormat

Response format for structured output. Mirrors OpenAI’s response_format API — no proprietary extensions.

StructuredOutputStart

Typed activation boundary for constrained decoding.

Constants

DEFAULT_CHAT_REPETITION_PENALTY

Default repetition penalty for product chat entrypoints.