sparse-critical-rlvr-token-analysis

Analyzes how Reinforcement Learning from Verification Rewards (RLVR) improves reasoning by examining token-level probability distributions. Finds that >83% of token positions exhibit near-zero divergence—RL operates through sparse, targeted refinements. Cross-sampling experiments show 1.5-7.8% RL-selected tokens recover full gains, while reverting 5-10% of …

Independent GitHub snapshot · Automated classification · Original documentation governs