PPL
Jiajun Chai
Papers (5)
Joint Training of Multi-Token Prediction in Reinforcement Learning via Optimal Coefficient CalibrationZipRL: Adaptive Multi-Turn Context Compression with Hindsight Response ReplayWhen Self-Belief Misleads: Active Label Acquisition for Reinforcement Learning with Verifiable RewardsAMR-SD: Asymmetric Meta-Reflective Self-Distillation for Token-Level Credit AssignmentImplicit Hierarchical GRPO: Decoupling Tool Invocation from Execution for Tool-Integrated Mathematical Reasoning