1
From Parameters to Data: A Task-Parameter-Guided Fine-Tuning Pipeline for Efficient LLM Alignment
提出强映射假设，揭示数据选择与参数高效微调的内在耦合，为LLM对齐提供新思路。
arXiv:2605.21558v1 Announce Type: cross Abstract: Adapting Large Language Models (LLMs) to specialized domains typically incurs high data and computat…