paper
Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories
paperactiveprovisional
language-models-need-sleep-learning-to-self-modify-and-consolidate-memories-f19c9e75·1 events·first seen 14d agoAliases: Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories
Co-occurring entities
More like this (12)
A sleep-like consolidation mechanism for LLMsSleep Consolidation MechanismAttention Amnesia in Hybrid LLMs: When CoT Fine-Tuning Breaks Long-Range Recall, and How to Fix ItRecalling Too Well: Sycophancy Evaluation and Mitigation in Memory-Augmented ModelsSupervised Memory TrainingLanguage Modeling LossLearning from the Self-future: On-policy Self-distillation for dLLMsLanguage Model FinetuningModeling Complex Behaviors: Multi-Personality Composition and Dynamic Switching in Vision-Language ModelsFlashbackCL: Mitigating Temporal Forgetting in Federated LearningWhen Good Verifiers Go Bad: Self-Improving VLMs Can Regress on New TasksLanguage Model Safety Monitor
Recent events (1)
Sleep paradigm for LLMs enables continual learning and memory consolidation via distillation and RL
A new arXiv preprint proposes a 'Sleep' paradigm for language models that enables continual learning by consolidating short-term in-context memories into long-term parameters. The framework has two stages: Knowledge Seeding (distilling a smaller model's memories into a larger network via on-policy distillation combined with RL-based imitation learning) and Dreaming (self-improvement via RL-generated synthetic curricula without human supervision). Experiments cover long-horizon tasks, continual learning, knowledge incorporation, and few-shot generalization, addressing a known weakness of current LLMs in retaining temporal knowledge across contexts.