Class ZCL_LLM_OPTIONS

AI Generated documentation.

Overview

The ZCL_LLM_OPTIONS class manages configuration parameters for Large Language Model interactions. It provides methods to set and validate various LLM parameters such as temperature, top-p sampling, penalties, and custom parameters. The class implements the ZIF_LLM_OPTIONS interface.

Key public methods:

• SET_TEMPERATURE: Controls randomness (0.0-2.0)
• SET_TOP_P: Sets nucleus sampling parameter (0.0-1.0)
• SET_TOP_K: Sets top-k sampling parameter (min 1)
• SET_SEED: Sets random seed (min 0)
• SET_FREQUENCY_PENALTY/SET_PRESENCE_PENALTY: Control repetition (-2.0 to 2.0)
• SET_MIN_P: Sets minimum probability threshold (0.0-1.0)
• SET_TOP_A: Sets top-a sampling parameter (0.0-1.0)
• SET_MAX_TOKENS: Sets maximum output tokens (min 0)
• SET_CUSTOM_PARAMETERS: Allows setting arbitrary key-value parameters
• GET_PARAMTERS: Retrieves all stored parameters

Dependencies

• ZCX_LLM_VALIDATION: Exception class for parameter validation
• ZLLM_KEYVALUES: Table type for storing key-value parameters

Details

The class follows a parameter validation and storage pattern:

1. Each parameter setting method first validates the input
2. If valid, the parameter is stored in the internal parameter table
3. Custom parameters can override standard parameters

flowchart TD
    A[Parameter Input] --> B{Validate Range}
    B -->|Valid| C[Store Parameter]
    B -->|Invalid| D[Raise Exception]
    C --> E[Internal Parameter Table]
    F[Custom Parameters] --> G{Exists?}
    G -->|Yes| H[Update Value]
    G -->|No| I[Insert New]
    H --> E
    I --> E

The class uses two validation helper methods:

• VALIDATE_RANGE_FLOAT: For decimal parameters with min/max bounds
• VALIDATE_RANGE_INT: For integer parameters with min/optional max bounds

All parameters are stored as strings in the internal table, with numeric values being converted during storage. This allows for a unified storage mechanism while maintaining type safety through the public interface.