{"slug": "loraweaver-directional-adapter-routing-in-lora-representation-space", "title": "LoRAWeaver: Directional Adapter Routing in LoRA Representation Space", "summary": "A training-free adapter-routing method called LoRAWeaver selects task-specific LoRA adapters by matching adapter-conditioned query representations against task prototypes using cosine similarity, avoiding the retraining that existing routing and encoder-based methods require when new adapters are added. In experiments with LLaMA-2-7B and LLaMA-2-13B on a 48-task FLAN-v2 benchmark and on out-of-distribution tasks, LoRAWeaver outperformed training-free baselines in the cross-task setting and approached a trained retrieval encoder without learning a router, with an ablation confirming the benefit of directional alignment in the adapter-output space. The work, posted as arXiv:2611.17340v1, targets the growing pool of task-specific LoRA adapters that complicates adapter selection during inference.", "body_md": "arXiv:2611.17340v1 Announce Type: new \nAbstract: Large language models are expensive to update through full retraining. Parameter-efficient fine-tuning reduces the cost of adaptation, but repeated adaptation creates a growing pool of task-specific LoRA adapters. This pool complicates adapter selection during inference. Existing routing and encoder-based methods often require retraining when new adapters are added, which limits their use in dynamic settings. This paper treats adapter routing as a representation alignment problem in the adapter-output space. The proposed training-free method selects adapters by matching adapter-conditioned query representations against task prototypes using cosine similarity. Experiments with LLaMA-2-7B and LLaMA-2-13B on a 48-task FLAN-v2 benchmark and on out-of-distribution tasks show that the method outperforms training-free baselines in the cross-task setting and approaches a trained retrieval encoder without learning a router. An ablation confirms the benefit of directional alignment in the adapter-output space.", "url": "https://wpnews.pro/news/loraweaver-directional-adapter-routing-in-lora-representation-space", "canonical_source": "https://arxiv.org/abs/2611.17340", "published_at": "2026-11-16 05:00:00+00:00", "updated_at": "2026-11-16 05:19:33.376928+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "ai-research", "natural-language-processing"], "entities": ["LoRAWeaver", "LLaMA-2-7B", "LLaMA-2-13B", "FLAN-v2", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/loraweaver-directional-adapter-routing-in-lora-representation-space", "markdown": "https://wpnews.pro/news/loraweaver-directional-adapter-routing-in-lora-representation-space.md", "text": "https://wpnews.pro/news/loraweaver-directional-adapter-routing-in-lora-representation-space.txt", "jsonld": "https://wpnews.pro/news/loraweaver-directional-adapter-routing-in-lora-representation-space.jsonld"}}