The paper introduces CHASE, a framework that leverages Geometric and Spectral Alignment (GSA) to inform practical model design. GSA characterizes trained networks via spectral concentration, channel alignment, support structure, and singular basis changes, and CHASE applies these insights to six applications including parameter‑efficient fine‑tuning, structured pruning, and model merging. New methods—CAGA, SAKV, and CAPS—use GSA to share attention heads, compress KV caches, and prune structured outputs, demonstrating improved performance over baselines.
arXiv:2607. 09757v1 Announce Type: cross Abstract: Low-Rank Adaptation (LoRA) has become a cornerstone of parameter-efficient fine-tuning (PEFT); however, the conventional practice of uniform rank assignment ignores the functional heterogeneity of neural layers.
arXiv:2602. 02599v4 Announce Type: replace Abstract: Long-context inference in large language models (LLMs) is bottlenecked by the memory and compute of the key-value (KV) cache.
arXiv:2607. 24331v1 Announce Type: new Abstract: As the inference phase of Large Language Models (LLMs) requires handling long context windows, the Key-Value (KV) cache initially appears to address this challenge but eventually becomes a significant bottleneck as the context window continues to grow.
arXiv:2610.28385v1 Announce Type: new Abstract: Head-dimension compression can reduce the memory required by the key-value (KV) cache, but existing methods often reconstruct full-dimensional represen...
The paper proposes a sensitivity‑aware residual‑stream pruning method for large language models that goes beyond minimizing activation reconstruction error. By using a second‑order approximation of output KL divergence, the authors derive a spectral upper bound that selects pruning subspaces based on both activation covariance and output sensitivity, enabling efficient eigendecomposition. Experiments on instruction‑tuned language models show that this approach consistently reduces calibration KL divergence, improves perplexity, and enhances downstream task performance across various compression levels.
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...
arXiv:2608. 06901v1 Announce Type: cross Abstract: Vision-language models (VLMs) have achieved remarkable generalization across diverse multimodal tasks through large-scale pre-training, yet their rapidly increasing computational and memory requirements pose significant challenges for deployment in constrained environments.
COEC (Calibrated Orthogonal-Equivalence Compensation) is a training‑free framework that improves structured pruning of large language models by applying alternating left and right orthogonal rotations to the retained weight matrix. The method optimizes the right rotation on a reduced Stiefel manifold, rescales singular values via generalized cross‑validation, tempers the calibration Gram matrix, and adds an alignment penalty to preserve geometric relations between attention projections. Experiments on Llama‑3, Llama‑3.1, and Qwen2.5 show that COEC consistently improves perplexity and zero‑shot accuracy across multiple sparsity levels, outperforming existing compensation techniques.
The paper introduces STD, a hierarchical token pruning framework for Large Vision‑Language Models that aligns pruning strategies with the functional roles of different network stages. By using high‑frequency spectral analysis in shallow layers, Gaussian‑smoothed attention in intermediate layers, and a stability‑adaptive trigger in deep layers, STD preserves essential visual information while aggressively reducing token counts. Experiments demonstrate that STD outperforms existing pruning methods, achieving up to 94.4% token reduction and a 3.9× speed‑up on LLaVA‑NeXT‑7B.
Position Anchor Tuning (PAT) is a parameter‑efficient fine‑tuning method for pre‑trained point cloud transformers that improves inference efficiency. PAT reduces the computational cost of multi‑head attention and feed‑forward networks by using token aggregation‑expansion pairs: a token aggregation module (TAM) selects representative tokens based on 3D position anchors, and a token expansion module (TEM) propagates the learned representations back to the full token set. Combined with base‑sharing low‑rank adaptation (BSLoRA) for the TAMs, PAT achieves performance comparable to state‑of‑the‑art methods while requiring fewer trainable parameters and lower computational overhead.
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. 28418v1 Announce Type: cross Abstract: Pruning is a promising approach for improving the efficiency of LLMs.