TFGN：无需任务和回放的连续预训练，避免灾难性遗忘的LLM规模
TFGN: Task-Free, Replay-Free Continual Pre-Training Without Catastrophic Forgetting at LLM Scale
AI总结：
TFGN通过输入条件化、参数高效更新，在无需回放或任务标签的情况下，实现了在异构文本领域连续预训练，避免灾难性遗忘，展示了跨领域正向迁移和高梯度分离。
摘要：
在异构文本领域上持续预训练大规模语言模型，无需回放或任务标签，一直是LLM规模下的未解决架构问题。现有方法依赖于回放缓冲区、任务标识符、扩展性差的正则化惩罚或句子分类级评估。我们引入TFGN，一种用于Transformer语言模型的架构叠加，产生输入条件化的、参数高效的更新，同时保持Transformer其余部分不变。在六个异构文本领域（Prose、Python、Math、Biomedical、Chinese、JavaScript）上，每个阶段10亿个token，三个模型规模（约398M、约739M、约9B）和两种模式（从头开始和Retrofit），TFGN在LLaMA 3.1 8B Retrofit上实现了反向迁移-0.007，在HellaSwag保留0.506/0.504/0.510，并在领域对之间实现了>=99.59%的L2正交梯度分离，无需回放、无需任务ID、无需Fisher惩罚。相同的矩阵显示了正向跨领域迁移：在LLaMA-8B Retrofit上，被排除的JavaScript PPL下降了26.8%，在GPT-2 Medium From-Scratch上下降了62.0%，纯从Python训练获得。两个扩展在相同基质上进一步解决了开放问题。一个闭环元控制层（扩展A）在~398M模型上减少了81%的遗忘，映射到Dupoux等人（arXiv:2603.15381）的System A和System M角色。一个操作级计划向量（扩展B）在30个源到目标对上以99.96%的余弦保真度重塑前向传递行为。架构洞察是一种读/写分解：前向传递完全密集，而跨领域参数更新结构化，使得先前领域子空间不被写入。据我们所知，TFGN是第一个同时关闭LLM规模灾难性遗忘、实现闭环自主学习元控制器，并携带操作级潜在计划的架构。
英文摘要：
Continually pre-training a large language model on heterogeneous text domains, without replay or task labels, has remained an unsolved architectural problem at LLM scale. Existing methods rely on replay buffers, task identifiers, regularization penalties that scale poorly, or sentence-classification-scale evaluation. We introduce TFGN, an architectural overlay for transformer language models that produces input-conditioned, parameter-efficient updates while leaving the rest of the transformer unchanged. On six heterogeneous text domains (Prose, Python, Math, Biomedical, Chinese, JavaScript) at 1B tokens per phase across three model scales (~398M, ~739M, ~9B) and two regimes (From-Scratch and Retrofit), TFGN achieves backward transfer of -0.007 at LLaMA 3.1 8B Retrofit, HellaSwag retention 0.506/0.504/0.510, and >=99.59% L2-orthogonal gradient separation between domain pairs - with no replay, no task IDs, no Fisher penalty. The same matrices show positive cross-domain forward transfer: held-out JavaScript PPL drops 26.8% at LLaMA-8B Retrofit and 62.0% at GPT-2 Medium From-Scratch purely from Python training. Two extensions on the same substrate close further open problems. A closed-loop meta-control layer (Extension A) reduces forgetting by an additional 81% at ~398M, mapping onto the System A and System M roles of Dupoux et al. (arXiv:2603.15381). An operator-level plan vector (Extension B) reshapes forward-pass behavior at 99.96% cosine fidelity over 30 source->target pairs. The architectural insight is a Read/Write decomposition: the forward pass is fully dense, while cross-domain parameter updates are structured so prior-domain subspaces are not written to. To our knowledge, TFGN is the first architecture that simultaneously closes catastrophic forgetting at LLM scale, realizes a closed-loop autonomous-learning meta-controller, and carries an operator-level latent planner.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
