ReasoningConfig

Options & Configuration · AiDotNet.Reasoning.Models

Configuration options for reasoning strategies that control how problems are solved.

For Beginners

Think of this class as a control panel with knobs and switches that adjust how the AI thinks about problems. Just like you might adjust the difficulty level in a video game, these settings let you control things like:

  • How many steps the AI should take when thinking
  • How thoroughly it should explore different solution paths
  • Whether it should verify its work
  • How much computing power to use

Different problems might need different settings. A simple math problem might only need a few steps, while a complex reasoning task might benefit from exploring many different approaches.

How It Works

Example Usage:

Constructors

Constructor Summary
ReasoningConfig()

Properties

Property Summary
bool EnableContradictionDetection { get; set; } Whether to enable contradiction detection across reasoning steps.
bool EnableDiversitySampling { get; set; } Whether to enable diversity sampling for exploring varied reasoning paths.
bool EnableExternalVerification { get; set; } Whether to enable external tool verification (calculators, code execution, etc.).
bool EnableSelfRefinement { get; set; } Whether to enable self-refinement when verification fails.
bool EnableTestTimeCompute { get; set; } Whether to enable test-time compute scaling (adaptive computation based on problem difficulty).
bool EnableVerification { get; set; } Whether to enable step-by-step verification with critic models.
double ComputeScalingFactor { get; set; } Multiplier for compute resources based on estimated problem difficulty (1.0 = baseline, 2.0 = double).
double Temperature { get; set; } Temperature for sampling diverse reasoning paths (0.0 = deterministic, 1.0+ = creative).
double VerificationThreshold { get; set; } Minimum verification score to accept a reasoning step (0.0 to 1.0).
int BeamWidth { get; set; } Beam width for beam search algorithms.
int BranchingFactor { get; set; } Number of alternative thoughts to generate at each step (for Tree-of-Thoughts).
int ExplorationDepth { get; set; } Maximum depth for tree-based reasoning strategies (Tree-of-Thoughts, MCTS).
int MaxReasoningTimeSeconds { get; set; } Maximum total reasoning time in seconds (0 = no limit).
int MaxRefinementAttempts { get; set; } Maximum number of refinement attempts per step.
int MaxSteps { get; set; } Maximum number of reasoning steps to generate.
int NumSamples { get; set; } Number of independent reasoning attempts for self-consistency (majority voting).