5 papers · 1 filter
Normalized Low-Rank Adaptation
Jiale Kang, Ziyin Yue, Zheng Zhan +2
While low-rank adaptation (LoRA) is widely used for parameter-efficient model adaptation, how to regularize its training dynamics for stable and effective optimization remains unde…
Sample-Efficient Post-Training for LEGO Spatial-Physics Reasoning
Yuhuan Yuan, Zhouliang Yu, Minghao Liu +2
LLM-based LEGO assembly requires both semantic grounding and physical feasibility. In this paper, we identify a data-induced failure mode, physhack, in which generated assemblies s…
PEFT-Arena: Understanding Parameter-Efficient Finetuning from a Stability-Plasticity Perspective
Yangyi Huang, Ruotian Peng, Zeju Qiu +4
Parameter-efficient finetuning (PEFT) has become the standard approach for adapting large language models, yet evaluations largely emphasize downstream accuracy while overlooking t…
Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation
Kexuan Shi, Hanxuan Li, Zeju Qiu +3
We introduce Pion, a spectrum-preserving optimizer for large language model (LLM) training based on orthogonal equivalence transformation. Unlike additive optimizers such as Adam a…
Model Merging with Functional Dual Anchors
Kexuan Shi, Yandong Wen, Weiyang Liu
Model merging is an efficient post-training strategy for integrating knowledge from multiple finetuned checkpoints of a shared foundation model. Existing methods operate in the par…