Real-time Self-Distillation Connects Short-Term and Long-Term Memory in LLMs
中文摘要
LLM实时自蒸馏技术，有效连接短期和长期记忆，提升模型性能。
English Summary
Real-time self-distillation enhances large language models by bridging short-term and long-term memory, improving overall performance and information retention.
原文节选
Self-distillation works https://arxiv.org/abs/2604.01193. Continue reading on Medium »