arXiv：cs.AI· Weihang Su, Jiacheng Kang, Jingyan Xu, Qingyao Ai, Jianming Long, Hanwen Zhang, Bangde Du, Xinyuan Cao, Min Zhang, Yiqun Liu·· 5 小时前
ReGrad：以可检索梯度实现无累积权重漂移的持续后训练
Retrievable Gradients: Continual Post-Training Without Cumulative Weight Drift
AI 导读
研究者提出 ReGrad（Retrievable Gradients），将梯度视为可检索的知识单元：离线预计算文档专属梯度并存入索引化的 Gradient Bank，推理时仅检索与查询相关的梯度做临时权重适配，从而避免持续后训练中的权重累积漂移。
正文
Abstract:Continual post-training enables models to absorb emerging knowledge after deployment, but repeatedly updating shared parameters can accumulate weight drift, potentially causing catastrophic forgetting and degrading general capabilities. Retrieval-augmented generation avoids such parameter drift, yet often lacks the depth of parametric knowledge integration. In this paper, we propose ReGrad (Retrievable Gradients), a new paradigm that treats gradients as retrievable units of knowledge. ReGrad pre-computes document-specific gradients offline, stores them in an indexed Gradient Bank, and retrieves only query-relevant gradients at inference time for temporary weight adaptation. However, raw language-modeling gradients are optimized for token-level document reconstruction rather than for query-driven knowledge use. We therefore introduce a bi-level meta-learning objective that reshapes document-derived gradients into generalizable adaptation signals for downstream tasks. Experiments across general and domain-specific settings show that \textsc{ReGrad} outperforms CPT and RAG baselines, enabling scalable and reversible parametric knowledge injection without accumulating weight drift.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR); Machine Learning (cs.LG) |
| Cite as: | arXiv:2606.15734 [cs.CL] |
| (or arXiv:2606.15734v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2606.15734 arXiv-issued DOI via DataCite |
Submission history
From: Weihang Su [view email]
[v1]
Sun, 14 Jun 2026 10:31:36 UTC (294 KB)
[v2]
Sun, 11 Oct 2026 15:04:13 UTC (294 KB)
来源：arXiv：cs.AI · arxiv.org