LoRAWeaver selects LoRA adapters without retraining LoRAWeaver, described in arXiv paper 2611.17340, introduces training-free adapter routing for LoRA modules that matches query representations to task prototypes using cosine similarity, so new task-specific adapters can be added without retraining the router or encoder. The approach reaches performance close to trained retrieval encoders while eliminating retraining when adapters are added, cutting the cost of dynamic multi-task serving and inference-pipeline rebuilds as deployments scale to dozens of tasks. arXiv https://arxiv.org/abs/2611.17340 LoRAWeaver selects LoRA adapters without retraining Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated. LoRAWeaver enables training-free adapter routing for LoRA modules, letting you add new task-specific adapters without retraining the router or encoder. This cuts the cost and complexity of dynamic multi-task serving—no more rebuilding inference pipelines when new adapters arrive. For production, it means faster iteration on specialized models and lower overhead when scaling to dozens of tasks. LoRAWeaver enables training-free router selection for LoRA adapters by matching query representations with task prototypes using cosine similarity, achieving performance close to trained retrieval encoders. This eliminates the need for retraining when new adapters are added, making it practical for dynamic environments where tasks and adapters evolve frequently.