作者位次 #2/3ACL 2026Poster· Natural Language Processing
SED - SFT : Selectively Encouraging Diversity in Supervised Fine-Tuning
Yijie Chen · Yijin Liu · Fandong Meng
摘要
Supervised Fine-Tuning (SFT) followed by Reinforcement Learning (RL) has emerged as the standard post-training paradigm for large language models (LLMs). However, the conventional SFT process, driven by Cross-Entropy (CE) loss, often induces mode collapse, where models over-concentrate on specifi...