arXiv：cs.AI· Jeesu Jung, Hwan Chang, Juseon Do, Jeonghwan Choi, Jinho Choo, Sungwoo Nam, S. K. Hong, Hwanjun Song·· 5 小时前
Ready2Blend：从自然语言指令到可组合对齐提示词
Ready2Blend: From Natural-Language Instructions to Composable Alignment Prompts
AI 导读
Ready2Blend 通过 AlignFormer 将每条自然语言需求映射为存入模块化 prompt bank 的定长对齐提示词，主干与已有提示词保持冻结，并可借助可组合性正则化在推理时混合与重加权。
正文
Abstract:Continual alignment requires LLMs to adapt to new requirements without forgetting previously acquired behaviors. Natural-language instructions are flexible and composable but offer only indirect control, whereas post-training provides stronger adaptation at the cost of repeated parameter updates. We introduce Ready2Blend, which combines the flexibility of natural language with learned alignment. AlignFormer maps each requirement to a fixed-length alignment prompt stored in a modular prompt bank, while the backbone and prior prompts remain frozen. Composability regularization transfers the semantic geometry of textual requirements into prompt space, enabling inference-time blending and reweighting. Across two practical continual alignment settings, Ready2Blend is the only frozen-backbone method that matches post-training-based alignment methods, reaching $93.1$-$98.5\%$ of a joint-training reference with competitive retention, while requiring only a few prompt tokens and up to $4.3\times$ less training time. Its modular design further enables weighted personalization and order-free composition without retraining. Code will be released upon acceptance.
| Comments: | 24 pages |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.39365 [cs.CL] |
| (or arXiv:2609.39365v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2609.39365 arXiv-issued DOI via DataCite |
Submission history
From: Juseon Do [view email]
[v1]
Wed, 30 Sep 2026 09:21:38 UTC (586 KB)
[v2]
Mon, 12 Oct 2026 05:01:31 UTC (586 KB)
来源：arXiv：cs.AI · arxiv.org