InfoSFT: Learn More and Forget Less with Information-Aware Token Weighting
InfoSFT：信息感知的令牌加权，学得更多，忘得更少
🧭 Training ★ 深度解读
University of Pennsylvania · University of Southern California
提出基于模型置信度的令牌加权规则，为中等置信度令牌分配最高权重，提升泛化并减轻灾难性遗忘
#Proximal Policy Optimization
1 篇