drpo-llm-rl - DRPO for stable LLM reinforcement learning
DRPO LLM RL Reinforcement Learning Policy Optimization
Updated 2026-09-15
Replaces hard trust-region masks with a smooth advantage-weighted quadratic regularizer to stabilize LLM reinforcement learning.
⚙ Assess current RL training setup ⚙ Design DRPO quadratic regularizer ⚙ Implement smooth policy regularization