Outcome-Based RL Provably Leads Transformers to Reason, but Only With the Right Data 文章

ArXiv CS.AI2026-06-04NEWSen作者: Yuval Ran-Milo, Yotam Alexander, Shahar Mendel, Nadav Cohen
大语言模型
查看原文 →
关系图谱

Outcome-Based RL Provably Leads Transformers to Reason, but Only With the Right Data · 相关技术

相关技术

single-layer TransformersPolicy Gradienttransformerschain-of-thoughtreinforcement learning