Primary sourceIT之家
View sourceEvent ·Single source
谷歌发布"睡眠框架":让大模型边用边学,不忘旧知识
Google Research 于 10 月 27 日发布持续学习研究成果"睡眠框架",论文题为《Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories》。该框架将模型运行分为"清醒"和"睡眠"两个阶段:清醒时把输入暂存于短期记忆,睡眠时通过参数扩展、混合专家记忆块和知识蒸馏,将上下文信息写入长期参数而不覆盖旧知识。"做梦"环节让模型生成合成训练样本,用强化学习练习推理。实验中模型可处理最长约 1000 万 token 的上下文,在未见语言的持续学习中性能下降接近零,并在 AIME 24/25 数学基准上取得更高准确率。
Current read
谷歌研究团队搞了个"睡眠框架",模仿人睡觉巩固记忆,让模型边用边学新知识还不忘旧的,AIME 数学成绩也涨了。
2 sources62° AI heatLast updated 2026/10/30 05:51:20
Evidence
2 sourcesRelated source 1shao__meng
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.