Class ChatHistory

Namespace: LMKit.TextGeneration.Chat

Assembly: LM-Kit.NET.dll

This class represents the chat history between a user and an assistant within a chat session.

public sealed class ChatHistory : ISerializableData

Inheritance: object

ChatHistory

Implements: ISerializableData

Inherited Members: object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

Examples

Example: Create and populate a chat history

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

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

// Create chat history
ChatHistory history = new ChatHistory(model);

// Add system message (optional, must be first)
history.AddMessage(AuthorRole.System, "You are a helpful assistant.");

// Add conversation turns
history.AddMessage(AuthorRole.User, "What is the capital of France?");
history.AddMessage(AuthorRole.Assistant, "The capital of France is Paris.");
history.AddMessage(AuthorRole.User, "What about Germany?");
history.AddMessage(AuthorRole.Assistant, "The capital of Germany is Berlin.");

// Display the conversation
Console.WriteLine($"Message count: {history.MessageCount}");
foreach (var message in history.Messages)
{
    Console.WriteLine($"[{message.AuthorRole}]: {message.Text}");
}

Example: Use chat history for fine-tuning

using LMKit.Model;
using LMKit.TextGeneration.Chat;
using LMKit.Finetuning;
using System;

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

// Create training data as chat history
ChatHistory trainingData = new ChatHistory(model);

// Add training examples
trainingData.AddMessage(AuthorRole.User, "Translate 'hello' to Spanish");
trainingData.AddMessage(AuthorRole.Assistant, "Hola");

trainingData.StartNewConversation(); // Separator between examples

trainingData.AddMessage(AuthorRole.User, "Translate 'goodbye' to Spanish");
trainingData.AddMessage(AuthorRole.Assistant, "Adios");

// Use for fine-tuning
using var finetuning = new LoraFinetuning(model);
int samples = finetuning.LoadTrainingDataFromChatHistory(trainingData);
Console.WriteLine($"Loaded {samples} training samples");

Example: Chat history with attachments

using LMKit.Model;
using LMKit.TextGeneration.Chat;
using LMKit.Data;
using System;

// Load a vision-enabled model
LM model = LM.LoadFromModelID("gemma-3-4b");

ChatHistory history = new ChatHistory(model);

// Create a message with an image attachment
var userMessage = new ChatHistory.Message(AuthorRole.User, "What do you see in this image?");
userMessage.AddAttachment(new Attachment("photo.jpg"));

history.AddMessage(userMessage);
history.AddMessage(AuthorRole.Assistant, "I see a beautiful sunset over the ocean.");

Console.WriteLine($"Messages: {history.MessageCount}");

Remarks

The ChatHistory class manages the sequence of messages exchanged between users and assistants. It is used internally by conversation classes like MultiTurnConversation and can also be used directly for fine-tuning or custom conversation management.

Key Features

Store and manage conversation messages with different author roles
Support for system, user, assistant, and developer messages
Attachment support for multimodal conversations
Serialization support for saving and loading chat history
Clone support for creating conversation branches
Automatic chat template formatting based on model

Author Roles

System - System instructions (must be first)
User - User messages
Assistant - Assistant responses
Developer - Developer instructions

Common Use Cases

Building training data for fine-tuning
Custom conversation management outside of MultiTurnConversation
Serializing conversations for later continuation
Creating conversation templates

Constructors

ChatHistory(LM): Create a new instance of the ChatHistory class.

Properties

AssistantMessageSuffix: Gets or sets the suffix to append after each assistant message.

AssistantPrefix: Specifies the assistant message prefix used to prompt the language model.

ConversationId: Gets the unique identifier for this conversation session.

MessageCount: Returns the number of messages in the history.

Messages: Specifies a list of all messages exchanged in the chat.

Model: Gets or sets the LM (Language Model) instance associated with this object. The model instance can be changed unless the object is marked as internal/sealed. Throws exceptions when an invalid model is provided or the instance is sealed.

PromptPrefix: Defines a text string that will be added at the start of the chat prompt.

SystemMessageSuffix: Gets or sets the suffix to append after each system message.

SystemPrefix: Gets or sets the system message prefix used to prompt the language model.

TokenCount: Indicates the total count of tokens needed by the associated language model to tokenize the content of this instance.

UserMessageCount: Returns the number of user messages in the history.

UserMessageSuffix: Gets or sets the suffix to append after each user message.

UserPrefix: Specifies the user message prefix used to prompt the language model.

Methods

AddMessage(AuthorRole, string): Adds a message to the history.

AddMessage(Message): Adds a message to the history.

ClearHistory(): Clears the history.

Clone(): Creates a deep copy of the current ChatHistory instance.

Deserialize(byte[], LM): Deserializes the given binary data into a ChatHistory instance using the specified model for context.
This method reconstructs a ChatHistory object from its binary form, applying the model to interpret or configure the data appropriately.

Deserialize(Stream, LM): Deserializes a ChatHistory from a stream using the provided model for context.
This method reconstructs a ChatHistory object from its binary form, applying the model to interpret or configure the data appropriately.

Deserialize(string, LM): Deserializes the binary data from the specified file path into a ChatHistory instance using the provided model for context. This method reconstructs a ChatHistory object from its binary form, applying the model to interpret or configure the data appropriately.

Serialize(): Serializes this instance into a binary format and returns the resulting data as a byte array.

Serialize(Stream): Serializes this instance into a binary format and writes it to the specified stream.

Serialize(string): Serializes this instance into a binary format and writes it to the specified file path. This method converts the current instance into its binary representation and writes the data to a file at the given path.

StartNewConversation(): Initiates the beginning of a new conversation.

ToString(): Overrides ToString() by directly invoking the ToText() method.

ToText(): Converts the complete chat history, including the formatting provided by the chat template, into a single string.