256-Token Soft Prompts Match Full-Parameter RL for Continual Post-Training

xwang_lk · x · 2026-11-19

Original post →

More from Models

Models channel →