MentorAgent Docs
Menu

ReferencePackage: Novior.MentorAgent

MentorSamplingOptions class

The MentorSamplingOptions class of Novior.MentorAgent.Options: How a model samples its reply (MentorOptions.CoordinatorSampling). Each value left null is the…

View as Markdown
On this page

How a model samples its reply (MentorOptions.CoordinatorSampling). Each value left null is the provider's own; each one set goes into the request's ChatOptions as it is.

Namespace: Novior.MentorAgent.Options. Package: Novior.MentorAgent.

public sealed class MentorSamplingOptions

Terminal
dotnet add package Novior.MentorAgent

Stop sequences are left out on purpose: a sequence that also occurs inside a tool call's arguments or a citation would end the reply there, and the call would reach the application cut in half.

public MentorSamplingOptions()

public float? FrequencyPenalty { get; set; }

Above 0 discourages repeating the same tokens, the more they recur; below 0 encourages it. The range is the provider's.

public int? MaxOutputTokens { get; set; }

The longest reply, in tokens, of one request. At least 1: a reply cut short stops mid-sentence.

public float? PresencePenalty { get; set; }

Above 0 discourages any token already used, once is enough; below 0 encourages it. The range is the provider's.

public long? Seed { get; set; }

Asks the provider for repeatable sampling, where it supports it.

public float? Temperature { get; set; }

Randomness of the reply: 0 is the most predictable. Not negative.

public int? TopK { get; set; }

Only this many most likely tokens are drawn from. At least 1; not every provider takes it.

public float? TopP { get; set; }

Nucleus sampling: only the most likely tokens whose probabilities add up to this are drawn from. In (0, 1].