perspectiva aiqbenchMachine learning

This paper introduces MGE-PO, a policy optimization method that integrates generation entropy into credit assignment for mathematical reasoning tasks. By distinguishing between low-entropy successes and high-entropy failures, the approach achieves significant performance gains on standard benchmarks without requiring model architecture changes. This work advances the efficiency of training large language models to solve complex reasoning problems through better uncertainty-aware reward shaping.

Sursă care acoperăarXiv

Ce este aceastaLectura unui model asupra articolului și a acoperirii conexe, scrisă pentru aiqbench. Nu sunt cuvintele publicației.

MGE-PO: Entropy-Aware Credit Assignment for Mathematical Reasoning

arXiv19 octombrie 20261 min de cititArticolul și perspectiva sunt în engleză.

arXiv:2610.23245v1 Announce Type: new Abstract: GRPO assigns the same group-relative advantage to equally rewarded responses within a question, leaving their generation uncertainty unused. We introduce MGE-PO, centered on four-quadrant (4Q) advantage shaping that combines verified correctness with rollout generation entropy. Correctness determines the sign of an additional correction, while entropy controls its magnitude: lower-entropy successes receive larger bonuses and higher-entropy failures receive larger penalties. An online entropy-based question curriculum supports this credit rule using cached rollout measurements. At the final checkpoint after eight training epochs on Qwen3-8B-Base, adding 4Q across three policy optimization settings yields absolute gains of 3.23-3.69% in six-benchmark macro avg@32 and 4.24-11.06% in macro pass@32. On Qwen2.5-7B-Base, the complete configuration raises four-benchmark macro avg@32 from 41.5% to 45.4%.

Citește articolul complet pe arXiv

Citește pe arXiv ↗Întreabă despre acest articol