AI RESEARCH
CRMA: A Spectrally-Bounded Backbone for Modular Continual Fine-Tuning of LLMs
arXiv CS.LG
•
ArXi:2606.00382v1 Announce Type: new Sequential fine-tuning of large language models forces a choice: let the shared substrate keep learning and accept catastrophic forgetting, or freeze it after task one and foreclose cross-task refinement. Per-task adapter methods (LoRAHub, AdapterFusion, PackNet, Progressive Networks) take the second path. We