Class SingleTurnConversation

Namespace: LMKit.TextGeneration

Assembly: LM-Kit.NET.dll

A class designed for handling single-turn question answering.
Unlike a multi-turn conversation service, it does not preserve context between questions and answers.

public sealed class SingleTurnConversation : IConversation, ITextGenerationSettings

Inheritance: object

SingleTurnConversation

Implements: IConversation

ITextGenerationSettings

Inherited Members: object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

object.ToString()

Examples

Example: Basic single-turn question answering

using LMKit.Model;
using LMKit.TextGeneration;
using System;

// Load the language model
LM model = LM.LoadFromModelID("llama-3.2-1b");

// Create the single-turn conversation
SingleTurnConversation conversation = new SingleTurnConversation(model);

// Submit a question
var result = conversation.Submit("What is the capital of France?");

Console.WriteLine(result.Completion);
Console.WriteLine($"Tokens generated: {result.CompletionTokens}");

Example: Custom system prompt and sampling

using LMKit.Model;
using LMKit.TextGeneration;
using LMKit.TextGeneration.Sampling;
using System;

LM model = LM.LoadFromModelID("llama-3.2-1b");
SingleTurnConversation conversation = new SingleTurnConversation(model);

// Customize the system prompt
conversation.SystemPrompt = "You are a helpful coding assistant. Provide concise code examples.";

// Configure sampling for more deterministic output
conversation.SamplingMode = new TopPSampling { TopP = 0.9f, Temperature = 0.7f };

// Limit completion length
conversation.MaximumCompletionTokens = 500;

var result = conversation.Submit("Write a function to calculate factorial in Python.");
Console.WriteLine(result.Completion);

Remarks

Use this class when you need independent question-answer interactions where each request is processed without memory of previous exchanges. This is ideal for:

Simple Q&A systems
Text completion tasks
One-off queries that don't require conversation history
Parallel processing of independent requests

Key Features

Customizable system prompt via SystemPrompt
Token sampling control via SamplingMode
Repetition penalty configuration via RepetitionPenalty
Grammar-constrained generation via Grammar
Stop sequences for controlled output termination
Multimodal input support (text and images)

For conversations that need to maintain context across multiple exchanges, use MultiTurnConversation instead.

Constructors

SingleTurnConversation(LM): Creates an instance of the SingleTurnConversation class.

SingleTurnConversation(LM, ITextGenerationSettings): Initializes a new SingleTurnConversation instance using the specified language model and text-generation settings. This constructor configures the conversation for single-turn interactions, applying any provided ITextGenerationSettings.

Properties

Grammar: Gets or sets the Grammar object used to enforce grammatical rules during text generation. This ensures controlled and structured output from the model.

InferencePolicies: Gets the InferencePolicies instance that defines how inference is conducted for this single-turn conversation. This includes settings such as how to handle context overflow and input length overflow during generation.

LogitBias: A LogitBias object designed to adjust the likelihood of particular tokens (or text chunks) appearing during text completion.

MaximumCompletionTokens: Defines the maximum number of tokens (text chunks) permitted for text completion or generation.

MaximumContextLength: Gets or sets the maximum context length for inputs to the model.

MaximumInputTokens: Specifies the maximum number of tokens allowed as input to the model.

Model: Gets the Model instance associated with this object.

ReasoningLevel

Controls how (and whether) intermediate "reasoning"/"thinking" content is produced and/or exposed.

Use None to fully disable reasoning. Higher levels hint the model to allocate more budget to chain-of-thought style tokens when the model supports it. Support depends on model and template capabilities.

Typical semantics:

Level	Intended behavior
None	No reasoning tokens requested or exposed.
Low	Minimal reasoning; terse scratch space when helpful.
Medium	Balanced reasoning (default if enabled).
High	Maximize reasoning depth; may trade off speed.

RepetitionPenalty: A RepetitionPenalty object specifying the rules for repetition penalties applied during text completion.

SamplingMode: A TokenSampling object specifying the sampling strategy followed during text completion.

StopSequences: Specifies a set of sequences for which the API will stop generating additional tokens (or text chunks). The resultant text completion will exclude any occurrence of the specified stop sequences.

SystemPrompt: Specifies the system prompt that is applied to the model before forwarding the user's request.

Methods

Submit(Message, CancellationToken): Prompts the model with a structured request encapsulated in a ChatHistory.Message object.

Submit(string, CancellationToken): Submits a request to the model for text generation.

SubmitAsync(Message, CancellationToken): Prompts the model with a structured request encapsulated in a ChatHistory.Message object (asynchronously).

SubmitAsync(string, CancellationToken): Submits a request to the model for text generation.

Events

AfterTextCompletion: This event is triggered following the execution of a text completion.

AfterTokenSampling: This event is triggered just after the generation of a token.
The provided AfterTokenSamplingEventArgs argument enables detailed modifications to be made to the token selection process.

BeforeTokenSampling: This event is triggered just prior to the generation of a token.
The provided BeforeTokenSamplingEventArgs argument allows for precise adjustments to the token sampling process.