Skip to content

Class ZCL_LLM_OPTIONS

AI Generated documentation.

Overview

The zcl_llm_options class implements configuration management for Large Language Model (LLM) parameters. It provides methods to set hyperparameters (e.g., temperature, top_p, seed) with validation, stores them as key-value pairs, and allows retrieval via get_parameters().

Key Public Methods:

  • Setters:
  • Numerical parameters: set_temperature, set_top_p, set_top_k, set_seed, etc., with boundary validation (e.g., 0.0-2.0 for temperature).
  • set_custom_parameters: Adds or updates arbitrary parameters.
  • Validation: All setters enforce valid ranges, raising zcx_llm_validation for invalid inputs.
  • Data Management: get_parameters returns the current configuration as a key-value table.

Dependencies

  • Exceptions: Relies on zcx_llm_validation for error handling.
  • Data Structure: Uses internal table type zllm_keyvalues for parameter storage.

Details

flowchart LR  
  A[Public Set Method] --> B{Validate Input}  
  B -->|Valid| C[Update Parameter]  
  B -->|Invalid| D[Raise zcx_llm_validation]  
  C --> E[int_parameters Table]

Key Implementation Aspects:

  1. Validation Logic:
  2. validate_range_float/validate_range_int enforce min/max boundaries.

  3. Distinct error messages for out-of-range values (e.g., "Value 3.0 is above maximum 2.0").

  4. Parameter Storage:

  5. int_parameters table ensures unique keys via DELETE + INSERT in set_parameter.

  6. Values are stored as strings (e.g., |{ temperature }|), requiring parsing during retrieval.

  7. Custom Parameters:

  8. set_custom_parameters merges input parameters, overwriting existing keys.
  9. Supports extensibility for non-standard LLM parameters.

Test-Critical Behaviors (from test summary):

  • Boundary checks (e.g., 0.0/2.0 for temperature).
  • Parameter overwrite integrity (last-set value persists).
  • Exception propagation for invalid inputs.