Model Options

The PromptOptions class configures Azure OpenAI model parameters without exposing the Azure SDK. Use it to control temperature, token limits, sampling, and penalties across all prompt methods.

Basic Usage

using Prompt;

var opts = new PromptOptions
{
    Temperature = 0.3f,
    MaxTokens = 2000,
    TopP = 0.9f
};

// With Main.GetResponseAsync
string? response = await Main.GetResponseAsync(
    "Explain recursion", options: opts);

// With Conversation
var conv = new Conversation("You are a tutor.", opts);

// With PromptChain
var chain = new PromptChain()
    .WithOptions(opts)
    .AddStep("step1", template, "output");

Parameters

Parameter	Range	Default	Effect
`Temperature`	0.0–2.0	0.7	Controls randomness. Lower = more focused and deterministic. Higher = more creative and varied.
`MaxTokens`	≥ 1	800	Maximum tokens in the response.
`TopP`	0.0–1.0	0.95	Nucleus sampling — the model considers tokens within the top probability mass. Alternative to Temperature.
`FrequencyPenalty`	-2.0–2.0	0.0	Penalizes tokens based on how often they appear. Positive values reduce repetition.
`PresencePenalty`	-2.0–2.0	0.0	Penalizes tokens based on whether they've appeared at all. Positive values encourage topic diversity.

Parameter Guidelines

Temperature vs TopP: Generally, adjust one or the other — not both. Temperature controls the shape of the probability distribution; TopP truncates it.

Temperature 0.0–0.2: Deterministic, factual answers. Best for code, math, data extraction.
Temperature 0.5–0.7: Balanced. Good default for most tasks.
Temperature 0.8–1.2: Creative, varied outputs. Good for writing, brainstorming.

Factory Presets

PromptOptions includes static factory methods for common use cases:

Code Generation

var opts = PromptOptions.ForCodeGeneration();
// Temperature: 0.1, MaxTokens: 4000, TopP: 0.95
// Low temperature for precise, deterministic code

Creative Writing

var opts = PromptOptions.ForCreativeWriting();
// Temperature: 0.9, MaxTokens: 2000, TopP: 0.9
// High temperature for varied, imaginative output

Data Extraction

var opts = PromptOptions.ForDataExtraction();
// Temperature: 0.0, MaxTokens: 2000, TopP: 1.0
// Zero temperature for maximum determinism

Summarization

var opts = PromptOptions.ForSummarization();
// Temperature: 0.3, MaxTokens: 1000, TopP: 0.9
// Moderate temperature for faithful but readable summaries

Custom Presets

Build your own presets:

public static class MyPresets
{
    // Chatbot — warm and conversational
    public static PromptOptions Chatbot => new()
    {
        Temperature = 0.8f,
        MaxTokens = 500,
        TopP = 0.9f,
        PresencePenalty = 0.6f  // encourage varied responses
    };

    // Classification — deterministic labeling
    public static PromptOptions Classification => new()
    {
        Temperature = 0.0f,
        MaxTokens = 50,
        TopP = 1.0f
    };

    // Long-form — detailed technical writing
    public static PromptOptions LongForm => new()
    {
        Temperature = 0.5f,
        MaxTokens = 4000,
        TopP = 0.95f,
        FrequencyPenalty = 0.3f  // reduce repetition in long outputs
    };
}

JSON Serialization

PromptOptions properties are annotated with [JsonPropertyName] for serialization:

var opts = new PromptOptions { Temperature = 0.3f, MaxTokens = 2000 };
string json = JsonSerializer.Serialize(opts);
// {"temperature":0.3,"maxTokens":2000,"topP":0.95,"frequencyPenalty":0,"presencePenalty":0}

var restored = JsonSerializer.Deserialize<PromptOptions>(json);

Validation

All setters validate ranges and throw ArgumentOutOfRangeException for invalid values:

var opts = new PromptOptions();

opts.Temperature = 3.0f;    // throws: must be 0.0–2.0
opts.MaxTokens = 0;         // throws: must be ≥ 1
opts.TopP = -0.5f;          // throws: must be 0.0–1.0
opts.FrequencyPenalty = 5f;  // throws: must be -2.0–2.0

Table of Contents