SOURCE-LINKED INTELLIGENCE

LILA: Calibration-Free Structured Pruning of Large Language Models via Latent Spectral Geometry

arXiv · AI, language, vision and robotics · article · Sep 10, 2026 · UTC

Structured pruning of large language models (LLMs) offers hardware-efficient compression, yet existing methods require calibration data, gradient computation, or large auxiliary policy networks at pruning time. LILA (\emph{Latent-Informed Layer Analysis}) scores neuron importance via the Kolmogorov--Smirnov (KS) distance between empirical singular value distributions of the full and neuron-ablated feed-forward network (FFN) weight matrix, providing a closed-form spectral rule requiring no training, calibration data, or auxiliary network. Without any fine-tuning, LILA surpasses PruneNet (45M-pa

Read original source ↗ Open in workspace

recordType

paper

region

Global