LILA: Calibration-Free Structured Pruning of Large Language Models via Latent Spectral Geometry
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) distan...
Sankar Behera, D. Singh, Anshika Agnihotri et al.
· 0 citations