The paper introduces Global Relative Kinetic Utility (Global RKU), a label‑free method for calibrating cross‑layer credit in global structured pruning of large language models. Global RKU estimates channel importance via a final‑hidden‑state activation‑gradient signal and applies block‑relative normalization to remove block‑common scale while preserving within‑block ordering, enabling a single‑stage static pruning topology. Experiments on Qwen‑2.5‑7B show significant performance gains at various sparsity levels, and ablation studies confirm the effectiveness of the relative‑normalization step.
arXiv:2609.18239v1 Announce Type: new Abstract: Structured pruning is commonly formulated as ranking individual channels, although channel responses can be complementary or cancel through downstream...
LILA (Latent-Informed Layer Analysis) introduces a calibration‑free method for structured pruning of large language models by scoring neuron importance using the Kolmogorov–Smirnov distance between singular value distributions of full and neuron‑ablated feed‑forward network weight matrices. The approach requires no training, calibration data, or auxiliary networks, and outperforms existing methods such as PruneNet and SliceGPT on LLaMA‑2‑7B and Phi‑2 at various sparsity levels. After a single epoch of LoRA fine‑tuning, LILA matches heavily calibrated baselines, and a Neural Tangent Kernel analysis provides theoretical support for its spectral importance criterion. Additionally, LILA can dynamically allocate sparsity budgets, achieving state‑of‑the‑art generative preservation and revealing architectural bottlenecks at higher compression.
arXiv:2607. 22587v1 Announce Type: new Abstract: Large language models (LLMs) achieve strong performance across diverse tasks but their deployment is constrained by the memory and compute cost of their parameters.
arXiv:2609.30738v1 Announce Type: new Abstract: KV cache eviction methods such as SnapKV and PyramidKV rank tokens solely by mean attention over a small observation window. We study a unified score,...
arXiv:2606. 03328v3 Announce Type: replace Abstract: Calibration data are often treated as a minor implementation detail in post-training LLM pruning because averaged evaluations suggest only modest effects.
arXiv:2607. 04302v1 Announce Type: cross Abstract: We present HiFA4, a post-training operator-level design that executes both QK^T and PV in FlashAttention as 4-bit HIF4 Cube GEMMs for LLM inference on Ascend NPUs, while maintaining the online softmax state in FP16.
arXiv:2608.23744v1 Announce Type: new Abstract: Split conformal prediction, not the pruning rule, supplies finite-sample marginal coverage once a pruned model is fixed independently of the conformal...
arXiv:2608.29765v1 Announce Type: new Abstract: Structured pruning uses surrogate objectives because direct task evaluation over every feasible mask is too expensive. Most evaluations report average...
arXiv:2606. 03328v1 Announce Type: cross Abstract: Post-training pruning compresses large language models to high sparsity using a small unlabelled calibration set, and recent work has concluded that the choice of calibration source has only modest impact on averaged post-pruning accuracy.
arXiv:2606. 01544v1 Announce Type: new Abstract: Deploying Large Language Models (LLMs) in practice incurs substantial memory and computational costs.
arXiv:2607. 17568v1 Announce Type: cross Abstract: Structured pruning compresses large language models (LLMs) by removing whole computational units, such as attention heads and feed-forward (FFN) channel groups.
Deploying Large Language Models (LLMs) in practice incurs substantial memory and computational costs. Post-training pruning (PTP) is an effective approach to reducing these costs by removing weights without additional training.