FederatedRLHFOptions

Options & Configuration · AiDotNet.FederatedLearning.Alignment

Configuration options for Federated RLHF.

Constructors

Constructor Summary
FederatedRLHFOptions()

Properties

Property Summary
bool AdapterOnlyRewardModel { get; set; } Gets or sets whether to use adapter-only reward models.
double ClipRange { get; set; } Gets or sets the PPO clip range.
double KLCoefficient { get; set; } Gets or sets the KL divergence penalty coefficient (beta).
double RewardModelLearningRate { get; set; } Gets or sets the reward model learning rate.
int PPOEpochs { get; set; } Gets or sets the number of PPO epochs per round.