FineTuningBase<T, TInput, TOutput>

Base Classes · AiDotNet.FineTuning

Base class for all fine-tuning methods.

Implements: IFineTuning<T, TInput, TOutput>, IModelShape

Known derived types: ConstitutionalAIFineTuning<T, TInput, TOutput>, ContrastivePreferenceOptimization<T, TInput, TOutput>, DirectPreferenceOptimization<T, TInput, TOutput>, GroupRelativePolicyOptimization<T, TInput, TOutput>, IdentityPreferenceOptimization<T, TInput, TOutput>, KahnemanTverskyOptimization<T, TInput, TOutput>, OddsRatioPreferenceOptimization<T, TInput, TOutput>, PairwiseRankingOptimization<T, TInput, TOutput>, RankResponsesHumanFeedback<T, TInput, TOutput>, ReinforcementLearningHumanFeedback<T, TInput, TOutput>, RobustDirectPreferenceOptimization<T, TInput, TOutput>, SelfPlayFineTuning<T, TInput, TOutput>, SimplePreferenceOptimization<T, TInput, TOutput>, StatisticalRejectionSampling<T, TInput, TOutput>, SupervisedFineTuning<T, TInput, TOutput>

For Beginners

This is the foundation that all specific fine-tuning methods (like DPO, RLHF, SimPO) build upon. It provides common functionality so each method only needs to implement its unique training logic.

How It Works

This abstract class provides common functionality shared by all fine-tuning implementations, including serialization, option management, and utility methods.

Properties

Property Summary
FineTuningCategory Category { get; }
bool RequiresReferenceModel { get; }
bool RequiresRewardModel { get; }
bool SupportsPEFT { get; }
string MethodName { get; }

Methods

Method Summary
DynamicShapeInfo GetDynamicShapeInfo()
FineTuningOptions<T> GetOptions()
Task<FineTuningMetrics<T>> EvaluateAsync(IFullModel<T, TInput, TOutput> model, FineTuningData<T, TInput, TOutput> evaluationData, CancellationToken cancellationToken)
Task<IFullModel<T, TInput, TOutput>> FineTuneAsync(IFullModel<T, TInput, TOutput> baseModel, FineTuningData<T, TInput, TOutput> trainingData, CancellationToken cancellationToken)
byte[] Serialize()
int[] GetInputShape()
int[] GetOutputShape()
void Deserialize(byte[] data)
void LoadModel(string filePath)
void Reset()
void SaveModel(string filePath)