Expand description
Optimization decision engine for LLM Auto-Optimizer
This crate provides the decision-making logic for optimizing LLM configurations, including A/B testing, Thompson Sampling, statistical significance testing, contextual bandits for reinforcement learning, Pareto optimization, adaptive parameter tuning, drift & anomaly detection, and a comprehensive model registry for all major LLM providers.
Re-exports§
pub use ab_testing::ABTestEngine;pub use thompson_sampling::ThompsonSampling;pub use statistical::StatisticalTest;pub use statistical::ZTest;pub use variant_generator::VariantGenerator;pub use experiment_manager::ExperimentManager;pub use contextual_bandit::LinUCB;pub use contextual_bandit::ContextualThompson;pub use context::RequestContext;pub use context::OutputLengthCategory;pub use reward::RewardCalculator;pub use reward::RewardWeights;pub use reward::UserFeedback;pub use reward::ResponseMetrics;pub use reinforcement_feedback::ReinforcementEngine;pub use reinforcement_feedback::BanditAlgorithm;pub use reinforcement_feedback::VariantStats;pub use pareto::ModelCandidate;pub use pareto::Objectives;pub use pareto::ObjectiveWeights;pub use pareto::ParetoFrontier;pub use pareto::CostCalculator;pub use pareto::QualityMetrics;pub use model_registry::ModelRegistry;pub use model_registry::ModelDefinition;pub use model_registry::ModelPricing;pub use model_registry::ModelPerformance;pub use model_registry::ModelCapabilities;pub use model_registry::Provider;pub use model_registry::ModelTier;pub use adaptive_params::AdaptiveParameterTuner;pub use adaptive_params::ParameterConfig;pub use adaptive_params::ParameterRange;pub use adaptive_params::ParameterStats;pub use parameter_search::GridSearch;pub use parameter_search::GridSearchConfig;pub use parameter_search::RandomSearch;pub use parameter_search::LatinHypercubeSampling;pub use parameter_search::ParameterSearchManager;pub use parameter_search::SearchStrategy;pub use parameter_optimizer::ParameterOptimizer;pub use parameter_optimizer::OptimizationPolicy;pub use parameter_optimizer::OptimizationMode;pub use drift_detection::DriftStatus;pub use drift_detection::DriftAlgorithm;pub use drift_detection::ADWIN;pub use drift_detection::PageHinkley;pub use drift_detection::CUSUM;pub use drift_detection::StatisticalDriftDetector;pub use anomaly_detection::AnomalyResult;pub use anomaly_detection::ZScoreDetector;pub use anomaly_detection::IQRDetector;pub use anomaly_detection::MADDetector;pub use anomaly_detection::MahalanobisDetector;pub use threshold_monitor::ThresholdMonitoringSystem;pub use threshold_monitor::ThresholdConfig;pub use threshold_monitor::Alert;pub use threshold_monitor::AlertType;pub use threshold_monitor::AlertSeverity;pub use errors::DecisionError;pub use errors::Result;
Modules§
- ab_
testing - A/B testing engine with Thompson Sampling
- adaptive_
params - Adaptive Parameter Tuning for LLM configurations
- anomaly_
detection - Anomaly Detection
- context
- Context feature extraction for contextual bandits
- contextual_
bandit - Contextual bandit algorithms for adaptive model selection
- drift_
detection - Drift Detection
- errors
- Error types for decision engine
- experiment_
manager - Experiment lifecycle management
- model_
registry - LLM Model Registry and Catalog
- parameter_
optimizer - Parameter Optimizer
- parameter_
search - Parameter Search Strategies
- pareto
- Pareto optimization for multi-objective decision making
- reinforcement_
feedback - Reinforcement Feedback Engine
- reward
- Reward signal calculation for reinforcement learning
- statistical
- Statistical significance testing for A/B experiments
- thompson_
sampling - Thompson Sampling implementation for multi-armed bandit optimization
- threshold_
monitor - Threshold-Based Monitoring
- variant_
generator - Variant generation strategies for A/B testing