Inworld LLM Module

Overview > Inworld LLM Module Class: InworldLLMModule | Inherits from: InworldFrameworkModule Module for Large Language Model (LLM) integration in the Inworld framework. Provides text generation capabilities using both remote and local AI models.

​

Properties

• Provider
• ModelName
• ModelPath
• MaxToken
• MaxPromptLength
• Repetition
• Temperature
• TopP
• Frequency
• Presence

​

Methods

• GenerateText
• GenerateTextAsync
• SetupTextGenerationConfig

​

Reference

​

Provider

Gets or sets the AI model provider (e.g., OpenAI, Anthropic). Used when connecting to remote LLM services.

​

Returns

Type: string Description: The provider name as a string.

---

​

ModelName

Gets or sets the specific model name to use for text generation. This should correspond to models available from the configured provider.

​

Returns

Type: string Description: The model name as a string.

---

​

ModelPath

Gets or sets the file path to a local LLM model. If not set, defaults to the framework’s default LLM model path.

​

Returns

Type: string Description: The full file path to the local model.

---

​

MaxToken

Gets or sets the maximum number of tokens the model can generate in a single response. Value is automatically clamped between 1 and 2500.

​

Returns

Type: int Description: The maximum token count (1-2500).

---

​

MaxPromptLength

Gets or sets the maximum length of the input prompt in tokens. Value is automatically clamped between 1 and 2500.

​

Returns

Type: int Description: The maximum prompt length (1-2500).

---

​

Repetition

Gets or sets the repetition penalty factor. Higher values reduce the likelihood of repeating the same content. Value is automatically clamped between 0 and 1.

​

Returns

Type: float Description: The repetition penalty (0.0-1.0).

---

​

Temperature

Gets or sets the temperature for text generation. Higher values produce more creative/random output, lower values are more focused. Value is automatically clamped between 0 and 1.

​

Returns

Type: float Description: The temperature value (0.0-1.0).

---

​

TopP

Gets or sets the top-p (nucleus sampling) parameter. Controls diversity by only considering tokens with cumulative probability up to this value. Value is automatically clamped between 0 and 1.

​

Returns

Type: float Description: The top-p value (0.0-1.0).

---

​

Frequency

Gets or sets the frequency penalty factor. Reduces the likelihood of repeating frequently used tokens. Value is automatically clamped between 0 and 1.

​

Returns

Type: float Description: The frequency penalty (0.0-1.0).

---

​

Presence

Gets or sets the presence penalty factor. Encourages the model to talk about new topics by penalizing tokens that have already appeared. Value is automatically clamped between 0 and 1.

​

Returns

Type: float Description: The presence penalty (0.0-1.0).

---

​

GenerateText

Generates text synchronously using the configured LLM. This method blocks until the generation is complete.

​

Parameters

​

Returns

Type: string Description: The generated text response, or empty string if generation fails.

---

​

GenerateTextAsync

Generates text asynchronously using the configured LLM. This method performs generation on a background thread for improved performance.

​

Parameters

​

Returns

Type: Awaitable<string> Description: A task that completes with the generated text response, or empty string if generation fails.

---

​

SetupTextGenerationConfig

Sets up and returns a text generation configuration instance based on current module settings. This method configures all text generation parameters including token limits, penalties, and sampling parameters.

​

Returns

Type: TextGenerationConfig Description: A configuration instance for text generation with current module settings.

---

​

Serialized Fields

The following fields are configurable in the Unity Inspector:

​

Remote Configuration

• m_Provider (string) - The AI model provider (e.g., OpenAI, Anthropic)
• m_ModelName (string) - The specific model name to use for text generation

​

Local Configuration

• m_ModelPath (string) - The file path to a local LLM model

​

Text Configuration

• m_MaxToken (int) - Maximum number of tokens to generate (default: 100, range: 1-2500)
• m_MaxPromptLength (int) - Maximum length of input prompt in tokens (default: 1000, range: 1-2500)
• m_TopP (float) - Top-p (nucleus sampling) parameter (default: 0.95, range: 0-1)
• m_Temperature (float) - Temperature for text generation (default: 0.5, range: 0-1)

​

Penalty Configuration

• m_Repetition (float) - Repetition penalty factor (default: 1.0, range: 0-1)
• m_Frequency (float) - Frequency penalty factor (default: 0.0, range: 0-1)
• m_Presence (float) - Presence penalty factor (default: 0.0, range: 0-1)

​

Configuration Management

​

Remote Configuration

When using remote models, the module creates an LLMRemoteConfig with:

• Provider name (if specified)
• Model name (if specified)
• API key from framework utilities

​

Local Configuration

When using local models, the module creates an LLMLocalConfig with:

• Model path (resolved from StreamingAssets folder)
• Device configuration from framework utilities

​

Text Generation Configuration

The SetupTextGenerationConfig() method configures:

• Maximum token count
• Repetition penalty
• Top-p and temperature (with GPT-5 compatibility handling)
• Maximum prompt length
• Frequency and presence penalties

​

Important Notes

​

GPT-5 Compatibility

Special handling is implemented for GPT-5 models where TopP and Temperature parameters are not applied.

---