AI RESEARCH
W2T: LoRA Weights Already Know What They Can Do
arXiv CS.LG
•
ArXi:2603.15990v1 Announce Type: new Each LoRA checkpoint compactly s task-specific updates in low-rank weight matrices, offering an efficient way to adapt large language models to new tasks and domains. In principle, these weights already encode what the adapter does and how well it performs. In this paper, we ask whether this information can be read directly from the weights, without running the base model or accessing