MGRPO adds a second GRPO layer trained on successful self-corrections of the model's own first answers, and reports higher math benchmark accuracy than standard GRPO.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Multi-Layer GRPO: Enhancing Reasoning and Self-Correction in Large Language Models
MGRPO adds a second GRPO layer trained on successful self-corrections of the model's own first answers, and reports higher math benchmark accuracy than standard GRPO.