SparseDecoding is a decoding‑aware pruning framework that improves large language model inference by aligning pruning with the activations seen during actual autoregressive generation. It constructs calibration matrices from dense‑model activations (excluding prefill) and introduces an optimized N:M sparse matrix‑vector kernel with bitmask indexing to accelerate decoding. Experiments on models such as Llama‑3.1‑8B, Llama‑3.3‑70B, and Qwen3‑14B/32B show consistent performance gains over standard fixed‑text calibration and up to 1.48× wall‑clock speedup on A100 GPUs.
arXiv:2607. 21985v1 Announce Type: cross Abstract: The increasing deployment of large language models (LLMs) has magnified the computational and memory bottlenecks of autoregressive decoding, where low compute intensity and bandwidth-bound kernels dominate inference cost.
arXiv:2606. 09080v1 Announce Type: new Abstract: Pruning has emerged as a dominant paradigm for accelerating large language model (LLM) inference, spanning a broad spectrum of methods that remove computation across tokens, layers, heads, dimensions, and attention patterns.
arXiv:2607. 28418v1 Announce Type: cross Abstract: Pruning is a promising approach for improving the efficiency of LLMs.
arXiv:2607. 08786v1 Announce Type: cross Abstract: With the growing deployment of large language models (LLMs), LLM inference cost has become a key challenge.
Sparse autoencoders (SAEs) are commonly used to interpret large language models, but their reliability after pruning is unclear. This study shows that pruning’s effect on an SAE is governed by perturbation energy, a covariance-weighted norm, and that magnitude pruning distorts the representation space by ignoring activation geometry. Activation-aware pruning methods such as Wanda and SparseGPT better preserve SAE behavior, and the authors find that middle layers are especially vulnerable, leading them to propose a layer‑wise sparsity allocation that reduces perplexity for a given sparsity level.
arXiv:2606. 10445v1 Announce Type: new Abstract: Semi-structured 2:4 sparsity is widely supported by modern accelerators, providing up to a 2x theoretical speedup.
Semi-structured 2:4 sparsity is widely supported by modern accelerators, providing up to a 2x theoretical speedup. However, its strict 50% sparsity constraint often causes non-negligible accuracy degradation under post-training pruning.
arXiv:2608. 01536v1 Announce Type: cross Abstract: Large Language Models (LLMs) increasingly rely on sparsity to reduce inference cost, but most prior work targets a single sparsity source-either weight or activation-and optimizes for batched multi-user inference.
arXiv:2504.04342v2 Announce Type: replace Abstract: Scaling up model parameters and training data consistently improves the performance of large language models (LLMs), but at the cost of rapidly gro...
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.
The paper introduces 3BASiL-TM, a one‑shot post‑training method for compressing large language models using a sparse plus low‑rank decomposition. It employs a novel 3‑block ADMM algorithm to minimize layer‑wise reconstruction error and a transformer‑matching refinement that jointly optimizes sparse and low‑rank components across layers. Experiments show a >30% reduction in WikiText2 perplexity gap for LLaMA‑8B under a (2:4 sparse + 64 LR) setting and a 2.5× faster compression runtime on an A100 GPU compared to state‑of‑the‑art methods.
The paper introduces GaugeLasso, a method that applies symmetric group‑lasso penalties to transformer channels during training, enabling entire tensor slices to be zeroed out while maintaining dense tensors for GPU efficiency. By calibrating channel penalties based on inference utility per compute, the network self‑organizes into depth‑dependent structural profiles that can be dramatically smaller than the original architecture, achieving up to 255‑fold compression on a polynomial division task and outperforming hand‑designed baselines on language modeling and autoencoding benchmarks. The approach also accelerates training and reveals over‑provisioned axes that guide subsequent design iterations.