GRPO, Dr. GRPO, and DAPO Are Three Operations on One Number: The Group-Standard-Deviation Identity

GRPO, Dr. GRPO, and DAPO Are Three Operations on One Number: The Group-Standard-Deviation Identity

Three seemingly distinct training methods for language models are shown to be variations of a single approach based on standard deviation adjustment, with the disagreement among sa…

Hugging Face · Daily Papers ·Yong Yi Bay, Kathleen A. Yearick · ·▲ 6 upvotes

Este artigo está em destaque na seleção diária de papers do Hugging Face, curada pela comunidade de pesquisa em IA.

Autores: Yong Yi Bay, Kathleen A. Yearick

  • 6 upvotes da comunidade
  • Temas: standard deviation, prompt's sampled answers, automatic checker, Group Relative Policy Optimization, GRPO, GRPO Done Right

Resumo

Resumo original (em inglês), extraído do paper:

Three seemingly distinct training methods for language models are shown to be variations of a single approach based on standard deviation adjustment, with the disagreement among sampled answers determining learning effectiveness and update magnitude.

Onde ler

compartilhar: