From $\log π$ to $π$: Taming Divergence in Soft Clipping via Bilateral Decoupled Decay of Probability Gradient Weight

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Fu, Xiaoliang, Lin, Jiaye, Fang, Yangyi, Hu, Chaowen, Qin, Cong, Shao, Zekai, Zheng, Binbin, Pan, Lu, Zeng, Ke
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!