Initial commit

crates/training/rtx-rl/src/rlhf/mod.rs

@@ -0,0 +1,11 @@
+//! RLHF (Reinforcement Learning from Human Feedback) implementation
+
+mod ppo_trainer;
+mod preference_dataset;
+mod reward_model;
+mod rlhf_trainer;
+
+pub use ppo_trainer::{PPOConfig, PPOMetrics, PPOTrainer};
+pub use preference_dataset::{PreferenceDataset, PreferencePair};
+pub use reward_model::{RewardModel, RewardModelConfig};
+pub use rlhf_trainer::{RLHFConfig, RLHFTrainer, RewardMetrics};