Himani S. Deshpande

Step-Level Gradient Masking for GRPO: Selective Optimization of Reasoning Trajectories

Group Relative Policy Optimization (GRPO) has recently emerged as an effective algorithm for Reinforcement Learning from Verifiable Rewards (RLVR), allowing for improvements in logical reasoning, more specifically in mathematical reasoning in large language models (LLMs) without supervised reasoning traces. It does, ho...