IAiModelTrainingCore<T, TInput, TOutput>

Interfaces · AiDotNet.Configuration

Component that owns the core training configuration for an AI model build: the model itself, optimizer, regularization, fitness calculator, fit detector, training pipeline, training monitor, checkpoint manager, and memory management. Extracted from AiModelBuilder as slice 2 of the audit-2026-05 phase-2a DI refactor (see docs/internal/audit-2026-05-phase2a-aimodelbuilder-refactor.md).

Known implementations: AiModelTrainingCore<T, TInput, TOutput>

How It Works

This concern is the dependency root for slices 3 / 4 / 6 / 7 / 9 / 10 / 11 per the migration plan: cross-validation, compliance evaluation, workflow orchestration (FL / distributed), advanced learning, storage, observability, and agent / export each consume the trained model or the optimizer in some way, so they all wait for this component to land before they can migrate.

Properties

Property Summary
ICheckpointManager<T, TInput, TOutput> CheckpointManager { get; } The configured checkpoint manager, or null if not configured (no checkpointing).
IFitDetector<T, TInput, TOutput> FitDetector { get; } The configured fit detector (over- / under-fitting diagnostic), or null if not configured.
IFitnessCalculator<T, TInput, TOutput> FitnessCalculator { get; } The configured fitness calculator, or null if not configured.
IFullModel<T, TInput, TOutput> Model { get; } The configured model, or null if ConfigureModel hasn’t been called.
IOptimizer<T, TInput, TOutput> Optimizer { get; } The configured optimizer, or null if ConfigureOptimizer hasn’t been called.
IRegularization<T, TInput, TOutput> Regularization { get; } The configured regularization strategy, or null if not configured.
ITrainingMonitor<T> TrainingMonitor { get; } The configured training monitor, or null if not configured (no monitoring callbacks).
TrainingMemoryConfig MemoryConfig { get; } The configured training-memory configuration (gradient checkpointing, activation pooling, sharding), or null for default settings.
TrainingPipelineConfiguration<T, TInput, TOutput> TrainingPipelineConfiguration { get; } The configured training-pipeline configuration (custom stage definitions), or null for the default linear training pipeline.

Methods

Method Summary
void ConfigureCheckpointManager(ICheckpointManager<T, TInput, TOutput> manager) Sets the checkpoint manager.
void ConfigureFitDetector(IFitDetector<T, TInput, TOutput> detector) Sets the fit detector.
void ConfigureFitnessCalculator(IFitnessCalculator<T, TInput, TOutput> calculator) Sets the fitness calculator.
void ConfigureMemoryManagement(TrainingMemoryConfig configuration) Sets the training-memory configuration.
void ConfigureModel(IFullModel<T, TInput, TOutput> model) Sets the model.
void ConfigureOptimizer(IOptimizer<T, TInput, TOutput> optimizationAlgorithm) Sets the optimizer.
void ConfigureRegularization(IRegularization<T, TInput, TOutput> regularization) Sets the regularization strategy.
void ConfigureTrainingMonitor(ITrainingMonitor<T> monitor) Sets the training monitor.
void ConfigureTrainingPipeline(TrainingPipelineConfiguration<T, TInput, TOutput> configuration) Sets the training-pipeline configuration.