LLM Pruning

LLM: Large Language Model

Ask about this topic

Overview

Large language model pruning studies how to remove weights, neurons, attention heads, layers, or experts while retaining useful behavior. Methods range from magnitude and activation-based scoring to gradient, curvature, causal, and representation-based criteria, often using calibration data to estimate which components matter; some also recover performance through fine-tuning or distillation. Structured pruning can reduce computation more directly, while unstructured sparsity often needs specialized hardware and kernels to deliver real speedups. A central challenge is that the best pruning decisions depend on the model, task, calibration data, and deployment setting, and preserving benchmark accuracy may not preserve generation quality, fairness, or reliability.

Latest papers

CardsList

178 papers32 papers in the last four weeks, up 191% on the four weeks before. 0.2% of all new papers.

May 13, 2026

Rethinking Layer Relevance in Large Language Models Beyond Cosine Similarity

Large language models (LLMs) have revolutionized natural language processing. Understanding their internal mechanisms is crucial for developing more interpretable and optimized architectures. Mechanistic interpretability has led to the development of various methods for assessing layer relevance, with cosine similarity being a widely used tool in the field. On this work, we demonstrate that cosine similarity is a poor proxy for the actual performance degradation caused by layer removal. Our theoretical analysis shows that a layer can exhibit an arbitrarily low cosine similarity score while still being crucial to the model's performance. On the other hand, empirical evidence from a range of LLMs confirms that the correlation between cosine similarity and actual performance degradation is often weak or moderate, leading to misleading interpretations of a transformer's internal mechanisms. We propose a more robust metric for assessing layer relevance: the actual drop in model accuracy resulting from the removal of a layer. Even though it is a computationally costly metric, this approach offers a more accurate picture of layer importance, allowing for more informed pruning strategies and lightweight models. Our findings have significant implications for the development of interpretable LLMs and highlight the need to move beyond cosine similarity in assessing layer relevance.

Cristian Hinostroza, Rodrigo Toro Icarte, Christ Devia +4

1Pontificia Universidad Cat´olica de Chile · 2National Center for Artificial Intelligence (CENIA) · 3Universidad de Chile

LLM PruningLLM InterpretabilityNeural Network Interpretability
AskPDF
May 13, 2026

STOP: Structured On-Policy Pruning of Long-Form Reasoning in Low-Data Regimes

Long chain-of-thought (Long CoT) reasoning improves performance on multi-step problems, but it also induces overthinking. This inefficiency is especially problematic in low-data fine-tuning regimes, where real applications adapt reasoning models with limited supervision and cannot rely on large-scale teacher distillation or heavy test-time control. To address this, we propose STOP (Structured On-policy Pruning), an on-policy algorithm for analyzing and pruning long-form reasoning traces. STOP constructs self-distilled traces from the model. Then it maps each trace into a structured reasoning interface through node segmentation, taxonomy annotation, and reasoning-tree construction. On top of this interface, we introduce ECN (Earliest Correct Node), which retains the shortest prefix ending at the earliest node. Experiments on DeepSeek-R1-Distill-Qwen-7B and DeepSeek-R1-Distill-LLaMA-3-8B across GSM8K, Math 500, and AIME 2024 show that STOP reduces generated tokens by 19.4% to 42.4% while largely preserving accuracy in low-data fine-tuning. Beyond efficiency, our analyses show that STOP induces much smaller distributional shift than teacher-guided pruning, improves the structural efficiency of generated reasoning, and reallocates reasoning effort away from redundant verification and backtracking toward more productive exploration.

Chenjun Xu, Zhennan Zhou, Zhan Su +3

University of Washington · University of Montreal

LLM PruningLLM Fine-TuningEfficient Language Model ReasoningLLM Inference AccelerationStructured ReasoningLarge Reasoning Models+1
AskPDF
May 12, 2026

Layer-wise Representation Dynamics: An Empirical Investigation Across Embedders and Base LLMs

Hidden states change substantially across the layers of modern language models, but most layer-wise analyses focus on one aspect of that change. We propose Layer-wise Representation Dynamics (LRD), a framework with three layer-wise measurement families: Frenet (Grassmann speed and curvature) for global subspace motion, Neighborhood Retention Score (NRS) for local nearest-neighbor retention, and Graph Filtration Mutual Information (GFMI) for alignment with the final layer. Applying LRD to 31 models (encoder-based and decoder-based embedders, plus base LLMs) on 30 MTEB tasks reveals architectural and task-level differences that are not apparent from final-layer representations alone. We then use LRD for two applications: label-free model selection and inference-time layer pruning. For selection, all three model-level scores correlate positively with downstream MTEB performance, with end-to-end subspace displacement (d_{0,L}) the strongest, and the same direction holds on a smaller base-LLM MMLU panel. For pruning, GFMI is the only measurement-guided rule that beats Random at the 15% and 20% budgets and has the best median change at every budget. Frenet is effective only at the lightest budget, while NRS does not transfer from model selection to pruning. These results show that layer-wise structure provides signal for both interpretation and deployment decisions.

Jingzhou Jiang, Yi Yang, Kar Yan Tam

The Hong Kong University of Science and Technology

Text EmbeddingsLLM PruningRepresentation LearningLLM InterpretabilityText Embedding ModelsRepresentation Geometry in Language Models
AskPDF
May 11, 2026

Compute Where it Counts: Self Optimizing Language Models

Efficient LLM inference research has largely focused on reducing the cost of each decoding step (e.g., using quantization, pruning, or sparse attention), typically applying a uniform computation budget to every generated token. In practice, token difficulty varies widely, so static compression can over-compute on easy steps and under-compute on hard ones. We study dynamic budget allocation for autoregressive decoding: learning how much computation to spend per token from within a single model. Self-Optimizing Language Models (SOL) pair a frozen LLM with a lightweight policy network that reads the LLM hidden state and selects a discrete efficiency action at each decode step. Actions can jointly control (i) token-level attention sparsity, (ii) structured activation pruning in the MLP, and (iii) activation quantization bit-width, while leaving the base model weights unchanged. We train the policy with group-relative policy optimization on teacher-forced episodes: the token sequence is fixed, while we sample multiple compute schedules (i.e., "counterfactual" schedules that vary only the efficiency actions for the same token path) and compare their likelihoods under the same supervision. Our reward trades off language-model quality against soft penalties that encourage episode-average budget usage to match a requested target. Across model variants and compute regimes, SOL improves quality at matched budget over static allocation and strong random schedule search, offering a complementary axis for inference-efficiency optimization. SOL discovers a better quality-efficiency pareto-front across all our experiments and improves MMLU accuracy by up to 7.3% over uniform budget allocation strategies.

Yash Akhauri, Mohamed S. Abdelfattah

Cornell University

LLM Inference EfficiencyRL for Language ModelsLLM PruningLLM Inference SchedulingInference-Time Compute Allocation
AskPDF
May 10, 2026

A Game Theoretic Free Energy Analysis of Higher Order Synergy in Attention Heads of Large Language Models

Large language models rely on multihead attention, but interactions among heads remain poorly understood. We apply the Game Theoretic Free Energy Principle (GTFEP): a framework casting multiagent systems as distributed variational inference to analyze attention heads as bounded rational agents. According to GTFEP, each head minimizes its variational free energy, and collective behavior follows a Gibbs distribution over coalition structures whose energy is decomposed into Harsanyi dividends. Using a tractable approximation (uniform prior, deterministic dynamics), coalition free energy reduces to joint Shannon entropy of discretized head outputs (argmax key index). Pairwise dividends become mutual information (nonnegative), while triple dividends correspond to interaction information and can be negative. On BERT, GPT2, and Llama with GSM8K, triple dividends are consistently negative, revealing higher order redundancy. The Nash FEP correspondence guarantees that stationary points of collective free energy are epsilon Nash equilibria; thus, heads with negligible contribution can be pruned with minimal performance loss. Pruning heads with low marginal contribution reduces computational cost with minimal performance loss: for example, pruning 20% of heads in GPT2 reduces FLOPs by 18%, increases throughput by 22%, and raises perplexity only modestly (from 28.4 to 33.4 on GSM8K). Our work shows GTFEP provides a principled foundation for analyzing and optimizing transformer architectures.

Djamel Bouchaffra

DAVID Lab, University of Paris-Saclay, UVSQ Campus, 78035 Versailles, France

LLM PruningAttention Head AnalysisSelf-AttentionFree Energy PrincipleNeural Network InterpretabilityMulti-Head Attention
AskPDF
May 9, 2026

Relative Kinetic Utility: Calibrating Cross-Layer Credit for Global Structured LLM Pruning

Global structured pruning requires channels from different layers to compete under a shared sparsity budget, raising two coupled challenges: identifying which channels should be retained and making their scores comparable across layers. Raw channel scores can contain block-common scale that leaves within-block ordering unchanged but distorts model-wide competition. Our experiment indicates that similar layer-wise allocations can retain substantially different FFN channels, so layer allocation alone does not determine channel identity. Motivated by this separation, we introduce Global Relative Kinetic Utility (Global RKU), a label-free criterion that separates channel importance estimation from cross-layer comparison. Global RKU measures channel participation using a final-hidden-state activation-gradient signal, then applies block-relative normalization to mitigate block-common scale while preserving within-block ordering, requires only unlabeled calibration inputs, and produces a static pruning topology in a single calibration stage. Under questions-only calibration on Qwen-2.5-7B, RKU-GISP Mean3 margins are -0.98, +3.79, and +8.61 points at 30%, 40%, and 50% sparsity, respectively (average +3.81). Additional Qwen evaluations cover non-mathematical reasoning, recovery, held-out transfer, and physical deployment. Separately, replacing Wiki16K with questions-only Q16K improves RKU's Mean3 at every tested sparsity on Qwen, Llama, and Gemma. Our ablation study shows relative-normalization gains of 14.42 and 5.53 Mean3 points at 40% and 50% sparsity, respectively; the common-seed audit is positive in all 27 seed-task comparisons.

Tianhao Qian, Guilin Qi, Jiayu Chen

School of Mathematics, Southeast University Nanjing 210096, China

LLM PruningStructured SparsityChannel PruningStructured PruningNeural Network Pruning
AskPDF
May 9, 2026

ProxyKV: Cross-Model Proxy Pruning for Efficient Long-Context LLM Inference

Efficient long-context inference in Large Language Models (LLMs) is severely constrained by the Key-Value (KV) cache memory wall, yet existing pruning methods force a choice between low-latency heuristics that sacrifice precision and high-precision reconstruction methods that incur prohibitive prefilling overhead. To bridge this scoring-cost--accuracy gap, we propose ProxyKV, a cross-model proxy pruning framework that offloads importance scoring to a lightweight intra-family Small-Model Proxy executed asynchronously to the Large-Model Target. To bridge the architectural gap between heterogeneous models, we design the HybridAxialMapper, which disentangles temporal feature extraction from cross-head alignment, together with a Multi-Granularity Hybrid Loss that shifts the learning objective from rigid regression to relative ranking consistency. Across the Llama-3.1, Qwen-2.5, and Qwen-3 families spanning targets from 7B up to 32B parameters on LongBench, SCBench, and RULER, ProxyKV matches KVZip on aggregate (recovering \sim$$98.7\% of its mean accuracy) while delivering up to a 3.21×3.21\times prefilling speedup on Llama-3.1-8B (dual-GPU; \sim$$1.5\times shared single-GPU) and sustaining the speedup at contexts up to 170k tokens on Qwen-2.5-7B.

Junjie Li, Jiong Lou, Jie Li

Shanghai Jiao Tong University

LLM PruningKV CachingLong-Context Language Model InferenceKV-Cache EvictionLLM Inference AccelerationKV-Cache Pruning
AskPDF
May 9, 2026

SlimQwen: Exploring the Pruning and Distillation in Large MoE Model Pre-training

Structured pruning and knowledge distillation (KD) are typical techniques for compressing large language models, but it remains unclear how they should be applied at pretraining scale, especially to recent mixture-of-experts (MoE) models. In this work, we systematically study MoE compression in large-scale pretraining, focusing on three key questions: whether pruning provides a better initialization than training from scratch, how expert compression choices affect the final model after continued training, and which training strategy is most effective. We have the following findings: First, across depth, width, and expert compression, pruning a pretrained MoE consistently outperforms training the target architecture from scratch under the same training budget. Second, different one-shot expert compression methods converge to similar final performance after large-scale continual pretraining. Motivated by this, we introduce a simple partial-preservation expert merging strategy that improves downstream performance across most benchmarks. Third, combining KD with the language modeling loss outperforms KD alone, particularly on knowledge-intensive tasks. We further propose multi-token prediction (MTP) distillation, which yields consistent gains. Finally, given the same training tokens, progressive pruning schedules outperform one-shot compression, suggesting that gradual architecture transitions lead to better optimization trajectories. Putting it all together, we compress Qwen3-Next-80A3B to a 23A2B model that retains competitive performance. These results offer practical guidance for efficient MoE compression at scale.

Shengkun Tang, Zekun Wang, Bo Zheng +7

Qwen Team, Alibaba Inc. · MBZUAI · KAUST

Mixture-of-Experts PruningLLM PruningMixture-of-Experts Language ModelsContinued PretrainingKnowledge DistillationNeural Network Compression
AskPDF
May 8, 2026

Understanding Performance Collapse in Layer-Pruned Large Language Models via Decision Representation Transitions

Layer pruning efficiently reduces Large Language Model (LLM) computational costs but often triggers sudden performance collapse. Existing representation-based analyses struggle to explain this mechanism. We propose studying pruning through decision representation. Focusing on multiple-choice tasks, we introduce two metrics, Decision Margin and Option Frequency, and an Iterative Pruning method to analyze layer-wise decision dynamics. Our findings reveal a sharp decision transition that partitions the network into two stages: a Silent Phase, where the model cannot yet predict the correct answer, and a Decisive Phase, where the correct prediction emerges. We also find that pruning the Decisive Phase has minimal impact, whereas pruning the Silent Phase triggers immediate performance collapse, highlighting its extreme sensitivity to structural changes. Therefore, we conclude that pruning-induced collapse stems from disrupting the Silent Phase, which prevents the critical decision transition from occurring.

Boyu Shi, Chang Liu, ChuanBao Gao +2

School of Computer Science and Engineering, Southeast University, Nanjing, China · Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China

LLM PruningLLM InterpretabilityLanguage ModelingNeural Network Pruning
AskPDF
May 7, 2026

SparseForge: Efficient Semi-Structured LLM Sparsification via Annealing of Hessian-Guided Soft-Mask

Semi-structured sparsity provides a practical path to accelerate large language models (LLMs) with native hardware support, but post-training semi-structured pruning often suffers from substantial quality degradation due to strong structural coupling. Existing methods rely on large-scale sparse retraining to recover accuracy, resulting in high computational cost. We propose SparseForge, a post-training framework that improves recovery efficiency by directly optimizing the sparsity mask rather than scaling up retraining tokens. SparseForge combines Hessian-aware importance estimation with progressive annealing of soft masks into hardware-executable structured sparsity, enabling stable and efficient sparse recovery. On LLaMA-2-7B under 2:4 sparsity, SparseForge achieves 57.27% average zero-shot accuracy with only 5B\textbf{5B} retraining tokens, surpassing the dense model's 56.43% accuracy and approaching the 57.52% result of a state-of-the-art method using 40B\textbf{40B} tokens. Such improvements on the accuracy-efficiency trade-off from SparseForge are shown to be consistent across model families.

Liu Hanzuo, Chaofan Lin, Weixuan Sun +4

†Tsinghua University · ‡Tencent

LLM PruningSparse RecoveryStructured SparsityNeural Network Pruning
AskPDF
May 2, 2026

Importance-Guided Basis Selection for Low-Rank Decomposition of Large Language Models

Low-rank decomposition is a compelling approach for compressing large language models, but its effectiveness hinges on selecting which singular-vector bases to retain for a target task. Existing methods such as Basel adapt singular-value coefficients on downstream data and prune bases with small re-learned magnitudes, a heuristic that can be misaligned with task performance because it ignores the local geometry of the loss landscape. We present Basis Selection with Importance (BSI), a principled low-rank compression framework that ranks and prunes bases by directly estimating the expected loss increase incurred when each basis is removed. BSI derives a derivative-based importance score from a second-order Taylor expansion of the task loss with respect to singular values, combining first-order sensitivity and second-order curvature to quantify pruning impact. To make this criterion practical for LLMs, we develop an efficient Hessian-diagonal estimator by adapting the Hutchinson randomized-probing method to loss curvature with symmetric parameter perturbations. We provide a comprehensive theoretical analysis, including loss-increase bounds under basis pruning, explicit propagation of Hessian-diagonal estimation error into these bounds, variance characterization tied to the Hessian spectrum, high-probability sample-complexity guarantees for achieving a target estimation accuracy, and guidance on perturbation intensity. Extensive experiments on mathematical reasoning benchmarks demonstrate that BSI consistently outperforms state-of-the-art low-rank decomposition baselines, with especially strong improvements under deep compression.

Daniel Agyei Asante, Ernie Chang, Yang Li

Iowa State University · Meta

LLM PruningLLM CompressionLow-Rank Matrix DecompositionLow-Rank CompressionLow-Rank ApproximationNeural Network Pruning
AskPDF
May 2, 2026

Weight Pruning Amplifies Bias: A Multi-Method Study of Compressed LLMs for Edge AI

Weight pruning is widely advocated for deploying Large Language Models on resource-constrained IoT and edge devices, yet its impact on model fairness remains poorly understood. We conduct a controlled empirical study of three instruction-tuned models (Gemma-2-9b-it, Mistral-7B-Instruct-v0.3, Phi-3.5-mini-instruct) across three pruning methods (Random, Magnitude, Wanda) at four sparsity levels (10-70%) on 12,148 BBQ bias benchmark items with 5 random seeds, totaling 2,368,860 inference records. Our results reveal a Smart Pruning Paradox: activation-aware pruning (Wanda) preserves perplexity nearly perfectly (just 3.5% increase at 50% sparsity for Mistral-7B), yet produces the highest bias amplification, with Stereotype Reliance Score increasing 83.7% and 47-59% of previously unbiased items developing new stereotypical behaviors at 70% sparsity. Random pruning destroys language capability entirely (perplexity exceeding 10410^4 and reaching 10810^8) but produces only random-chance bias. We further show that unstructured pruning provides zero storage savings and zero inference latency reduction on real edge hardware, undermining the primary motivation for its use in IoT deployment. Of 180 dense-vs-pruned comparisons, 141 (78.3%) are significant (p<0.05p < 0.05) with mean ∣h∣=0.305|h| = 0.305. Published quantization studies report up to 21% of responses flipping between biased and unbiased states; our pruning results show transition rates nearly three times higher (47-59%), suggesting pruning poses a categorically greater risk to alignment than quantization. These findings demonstrate that perplexity-based evaluation provides false assurance of behavioral equivalence, and that IoT deployment pipelines require bias-aware validation before deploying pruned models at the edge.

Plawan Kumar Rath, Rahul Maliakkal

Meta

LLM PruningOn-Device Language Model InferenceSocial Bias in Language ModelsLanguage Model Bias Evaluation
AskPDF
Apr 29, 2026

Exploring the Limits of Pruning: Task-Specific Neurons, Model Collapse, and Recovery in Task-Specific Large Language Models

Neuron pruning is widely used to reduce the computational cost and parameter footprint of large language models, yet it remains unclear whether neurons in task-specific models contribute uniformly to task performance. In this work, we provide empirical evidence for the existence and importance of task-specific neurons through a systematic pruning study on language models specialized for mathematical reasoning and code generation. We introduce an activation-based selectivity metric to identify neurons with low contribution to the target task and prune them while preserving target-task accuracy, and compare selective pruning with random pruning. Selective pruning consistently outperforms random pruning, indicating that activation-based selectivity provides a systematic advantage over random pruning. Reverse pruning experiments further show that removing a small subset of highly task-specific neurons (~10%) causes complete performance collapse, suggesting that there exist task specific neurons and critical task information is concentrated in a small portion of the network. In contrast, selective pruning of less critical neurons (~30% - ~35%) reduces accuracy but still preserves significant performance. We also observed consistent reductions in parameters and runtime VRAM usage, along with improved inference throughput as pruning increases. Experiments on both 1.5B and 7B models reveal a robustness threshold around 15-20% pruning, beyond which accuracy loss and generation failures increase sharply. Fine-tuning substantially recovers performance across pruning levels, particularly for aggressively pruned models. These findings provide empirical evidence of neuron specialization in task-specific language models and offer insights into pruning robustness, model redundancy, and post-pruning recoverability.

M. K. Khalidi Siam, Md. Tausif-Ul-Islam, Md. Reshad Romim Khan +5

1BRAC University

LLM PruningLanguage ModelingNeural Network Pruning
AskPDF
Apr 28, 2026

Revisiting the Effectiveness of LLM Pruning for Test-Time Scaling

Large Language Models (LLMs) now exhibit remarkable reasoning capabilities through test-time compute scaling (TTS), with impressive performance across math and coding benchmarks. In parallel, research in model compression has developed pruning methods that seek to remove redundant/detrimental parameters without sacrificing task performance. The intersection of these two research advancements lays the foundation for our work. Specific to reasoning LLMs, prior work has shown that structured pruning (methods which remove entire set of layer blocks), significantly degrades TTS reasoning performance. However, in this work, we revisit this assumption and investigate whether unstructured pruning (methods that carefully remove only certain redundant/detrimental weights) exhibits similar limitations. Surprisingly, our extensive experiments across four reasoning benchmarks on two reasoning LLMs: s1.1-7B and Qwen3-8B, consistently show that unstructured pruning augments TTS performance compared to structured pruning, and at times can even outperform the unpruned full-weight LLMs. Furthermore, we also empirically study the impact of different layer-wise sparsity allocation strategies, which are an important parametric choice for instantiating these unstructured methods. These findings challenge the conventional notion that pruning always reduces TTS performance and in fact, suggest that carefully undertaken pruning can retain TTS effectiveness.

Ocean Monjur, Shahriar Kabir Nahin, Anshuman Chhabra

Bellini College of AI, Cybersecurity, and Computing University of South Florida Tampa, FL, USA

LLM PruningTest-Time ScalingNeural Network Pruning
AskPDF
Apr 27, 2026

Rethinking Layer Redundancy: Calibration Matters More Than Search in LLM Depth Pruning

Depth pruning improves the inference efficiency of large language models by removing Transformer blocks. Prior work typically treats layer redundancy as an inherent structural property of pretrained networks, emphasizing importance criteria and search algorithms to identify removable layers. In this study, we empirically investigate depth pruning from a functional perspective. Evaluating representative LLM families across diverse calibration configurations and multiple search algorithms, we show that different configurations produce different pruning patterns. Furthermore, under a fixed calibration configuration, complex search algorithms yield marginal performance improvements over simple one-shot methods, converging to similar pruned subsets. Overall, our results suggest that the calibration configuration plays a substantially larger role than the choice of search algorithm in shaping pruning patterns and calibration perplexity, while contributing comparably to variance in downstream reasoning accuracy. This indicates that future pruning efforts may benefit from prioritizing the calibration configuration over search complexity.

Minkyu Kim, Vincent-Daniel Yun, Youngrae Kim +3

Neural Superintelligence Lab, MODULABS, Republic of Korea · University of Southern California, United States · Boston University, United States +2

LLM PruningLLM Inference AccelerationNeural Network Pruning
AskPDF
Apr 27, 2026

DepthKV: Layer-Dependent KV Cache Pruning for Long-Context LLM Inference

Long-context reasoning is a critical capability of large language models (LLMs), enabling applications such as long-document understanding, summarization, and code generation. However, efficient autoregressive inference relies on the key-value (KV) cache, whose memory footprint grows linearly with sequence length, leading to a major memory bottleneck. To mitigate this overhead, KV cache pruning methods discard cached tokens with low attention scores during inference. Most existing methods apply a uniform pruning ratio across layers, implicitly assuming that all layers contribute equally to overall model performance. We show that this assumption is suboptimal, as layers differ significantly in their sensitivity to pruning. We propose DepthKV, a layer-dependent pruning framework that allocates a fixed global KV budget across layers based on their sensitivity, rather than using a uniform allocation. Across multiple models and tasks, DepthKV consistently outperforms uniform pruning at the same global pruning ratio, demonstrating more effective utilization of the KV cache budget through layer-dependent allocation.

Zahra Dehghanighobadi, Asja Fischer

1Ruhr University Bochum · 2UAR Research Center for Trustworthy Data Science and Security

LLM PruningKV CachingLong-Context Language Model InferenceKV-Cache ManagementMemory-Efficient InferenceKV-Cache Pruning
AskPDF
Apr 27, 2026

Pruning via Causal Attribution Preserves Reasoning Performance in Large Language Models

Large language models (LLMs) excel at multi-step reasoning but incur substantial inference cost. We introduce Causal Attribution Pruning (CAP), a training-free method that identifies critical attention heads by measuring their causal impact on reasoning tasks and uses these head-level scores to guide fine-grained weight pruning. For each attention head, CAP estimates the expected performance degradation when the head is masked during forward passes on a small calibration set of reasoning problems. These causal scores are then converted into weight-level importance values for the corresponding projection matrices. Unlike magnitude-only or activation-based criteria, CAP's interventional measurement directly captures each head's functional contribution, yielding relative accuracy gains of up to 61% over Wanda on ARC-Challenge at 20% sparsity. We evaluate CAP on GSM8K, StrategyQA, and ARC-Challenge using Llama-3-8B-Instruct and Mistral-7B-Instruct at 10%, 20%, and 50% sparsity. At moderate sparsity (10-20%), CAP improves over Wanda in most model-benchmark configurations. with especially large gains on ARC-Challenge for Llama-3. Our results suggest that attention-head-level causal attribution can better preserve reasoning performance on downstream benchmarks than correlational pruning criteria at equivalent sparsity, while remaining limited by coarse MLP attribution at 50% sparsity.

Amogh Sheth, Andrew Lin, Biruk Assefa +2

Edison Academy Magnet School · Massachusetts Institute of Technology · State University of New York College at Plattsburgh +2

LLM PruningSelf-AttentionLLM Inference AccelerationAttention MechanismsAttention Head PruningCausal Interventions in Language Models+2
AskPDF
Apr 26, 2026

Supernodes and Halos: Loss-Critical Hubs in LLM Feed-Forward Layers

We study the organization of channel-level importance in transformer feed-forward networks (FFNs). Using a Fisher-style loss proxy (LP) based on activation-gradient second moments, we show that loss sensitivity is concentrated in a small set of channels within each layer. In Llama-3.1-8B, the top 1% of channels per layer accounts for a median of 58.7% of LP mass, with a range of 33.0% to 86.1%. We call these loss-critical channels supernodes. Although FFN layers also contain strong activation outliers, LP-defined supernodes overlap only weakly with activation-defined outliers and are not explained by activation power or weight norms alone. Around this core, we find a weaker but consistent halo structure: some non-supernode channels share the supernodes' write support and show stronger redundancy with the protected core. We use one-shot structured FFN pruning as a diagnostic test of this organization. At 50% FFN sparsity, baselines that prune many supernodes degrade sharply, whereas our SCAR variants explicitly protect the supernode core; the strongest variant, SCAR-Prot, reaches perplexity 54.8 compared with 989.2 for Wanda-channel. The LP-concentration pattern appears across Mistral-7B, Llama-2-7B, and Qwen2-7B, remains visible in targeted Llama-3.1-70B experiments, and increases during OLMo-2-7B pretraining. These results suggest that LLM FFNs develop a small learned core of loss-critical channels, and that preserving this core is important for reliable structured pruning.

Audrey Cherilyn, Houman Safaai

Kempner Institute at Harvard University

Transformer InterpretabilityLLM PruningTransformer FFNsNeural Network Pruning
AskPDF
Apr 21, 2026

SimDiff: Depth Pruning via Similarity and Difference

Depth pruning improves the deployment efficiency of large language models (LLMs) by identifying and removing redundant layers. A widely accepted standard for this identification process is to measure the similarity between layers using cosine distance. However, we find that methods relying solely on this one-dimensional heuristic can exhibit unpredictable performance and even catastrophic collapse across different architectures. To address this issue, we propose SimDiff, a novel layer importance criterion that jointly evaluates layers from two orthogonal perspectives: representational similarity and transformation difference. The difference is quantified using two distinct metrics: MSSD, which is sensitive to outliers and identifies layers that make decisive corrections, and MASD, which robustly measures a layer's average contribution. Extensive experiments on multiple models ranging from 0.5B to 13B parameters demonstrate that SimDiff significantly outperforms state-of-the-art baselines across various pruning ratios. Notably, our method retains over 91% of LLaMA2-7B's performance at a 25% pruning ratio and achieves up to a 1.49x inference speedup when pruning 12 layers on LLaMA3.1-8B. We also show that pruned models can be effectively recovered with minimal fine-tuning.

Yuli Chen, Shuhao Zhang, Fanshen Meng +4

LLM PruningLLM Inference AccelerationNeural Network Pruning
AskPDF
Apr 21, 2026

GRASPrune: Global Gating for Budgeted Structured Pruning of Large Language Models

Large language models (LLMs) are expensive to serve because model parameters, attention computation, and KV caches impose substantial memory and latency costs. We present GRASPrune, a structured pruning framework applied after pretraining that jointly prunes FFN channels and KV head groups under a single global budget. Instead of learning importance scores without constraints and applying the budget only after training, GRASPrune learns lightweight gate scores with a projected straight-through estimator that enforces a hard mask satisfying the budget at every step while keeping the backbone weights frozen. After the mask is fixed, we calibrate scaling factors on the retained units to mitigate scale mismatch caused by pruning, and fold these factors into the pruned weights to obtain a smaller dense checkpoint with no extra parameters at inference. On LLaMA-2-7B, GRASPrune removes 50% of parameters and achieves 12.18 perplexity on WikiText-2 while maintaining competitive average zero-shot accuracy on five benchmarks, using four epochs on 512 unlabeled calibration sequences on a single NVIDIA A100 80GB GPU without any full model fine-tuning.

Ziyang Wang, Jiangfeng Xiao, Chuan Xiao +3

Beijing Institute of Technology, Zhuhai · Shenzhen University · Osaka University

LLM PruningLLM CompressionStructured Pruning
AskPDF
Apr 19, 2026

CRISP: Compressing Redundancy in Chain-of-Thought via Intrinsic Saliency Pruning

Long Chain-of-Thought (CoT) reasoning is pivotal for the success of recent reasoning models but suffers from high computational overhead and latency. While prior works attempt to compress CoT via external compressor, they often fail to align with the model's internal reasoning dynamics, resulting in the loss of critical logical steps. This paper presents Compressing Redundancy in Chain-of-Thought via Intrinsic Saliency Pruning (CRISP), a framework that compresses CoT by exploiting the model's intrinsic saliency. Our analysis reveals a distinct phenomenon: the reasoning termination token [object Object] acts as an information anchor, where its attention pattern effectively demarcates essential reasoning from redundancy. Based on this finding, we design a policy that utilizes these intrinsic attention signals to guide atomic compression operations. In contrast to coarse-grained pruning strategies, CRISP strategically distills the reasoning chain to maximize information density while preserving logical coherence. Empirical results across various backbone models and mathematical datasets demonstrate that CRISP achieves a 50-60% reduction in token count without compromising accuracy, effectively mitigating the efficiency bottleneck of long-context reasoning. We open-source our implementation to facilitate further research in efficient reasoning.

Yangsong Lan, Hongliang Dai, Piji Li

College of Artificial Intelligence, Nanjing University of Aeronautics and Astronautics, Nanjing, China · The Key Laboratory of Brain-Machine Intelligence Technology, Ministry of Education, Nanjing, China.

LLM PruningLLM CompressionLLM Inference AccelerationCoT Compression
AskPDF
Apr 18, 2026

Prune, Interpret, Evaluate: A Cross-Layer Transcoder-Native Framework for Efficient Circuit Discovery via Feature Attribution

Existing feature-interpretation pipelines typically operate on uniformly sampled units or exhaustive feature sets, incurring massive costs on units irrelevant to target behaviors. To address this, we introduce the first CLT-native end-to-end pruning framework, PIE, which pioneers the paradigm of pruning first and interpreting later. PIE connects Pruning, automatic Interpretation, and interpretation Evaluation, establishing a comprehensive benchmarking environment to systematically measure behavioral fidelity and downstream interpretability under pruning. Within this framework, we adapt strong relevance baselines and propose Feature Attribution Patching (FAP), a patch-grounded attribution method that scores CLT features by aggregating gradient-weighted write contributions. Furthermore, we introduce FAP-Synergy, a systematic synergy-aware reranking procedure. We evaluate pruning using KL-divergence behavior retention and assess interpretation quality with FADE-style metrics across IOI and Doc-String datasets. Across budget constraints of K in {50, 100, 200, 400, 800}, our rigorous benchmarking reveals distinct operational regimes: while base FAP and adapted baselines perform robustly at relaxed budgets, FAP-Synergy excels in highly constrained, strict-budget regimes. Crucially, we demonstrate a practical "Effective Budget" advantage: on the IOI task for both Llama-3.2-1B and Gemma-2-2B, FAP-Synergy at K=50 functionally matches the behavioral fidelity of baseline circuits at K=75. Because downstream evaluation costs scale linearly per feature, Synergy effectively grants the pipeline 25 "free" features, achieving K=75 fidelity while reducing interpretation costs by 33%.

Qinhao Chen, Linyang He, Nima Mesgarani

Columbia University

LLM PruningFeature AttributionCircuit DiscoveryMechanistic InterpretabilityExplainability Evaluation
AskPDF
Apr 17, 2026

Optimizing Korean-Centric LLMs via Token Pruning

This paper presents a systematic benchmark of state-of-the-art multilingual large language models (LLMs) adapted via token pruning - a compression technique that eliminates tokens and embedding parameters corresponding to languages irrelevant to the target application. Focusing on Korean-centric natural language processing (NLP) tasks, we evaluate architectures including Qwen3, Gemma-3, Llama-3, and Aya across three vocabulary configurations: Original, English-Korean (EnKo), and English-Korean-Chinese (EnKoZh). Performance is assessed using established benchmarks for general aptitude, cultural literacy, instruction following, and machine translation. Our findings indicate that token pruning significantly improves generation stability by eliminating language confusion, and in the case of machine translation, frequently enhances performance on Korean-specific tasks. While instruction-following capabilities display architecture-dependent variance linked to latent cross-lingual representations, the significant reduction in vocabulary size validates token pruning as a highly effective optimization strategy for memory-constrained, domain-specific deployments, despite modest gains in inference latency.

Hoyeol Kim, Hyeonwoo Kim

College of Computing · Independent Researcher

Multilingual Language Model EvaluationLLM PruningLLM CompressionLanguage Modeling
AskPDF
Apr 17, 2026

Pruning Unsafe Tickets: A Resource-Efficient Framework for Safer and More Robust LLMs

Machine learning models are increasingly deployed in real-world applications, but even aligned models such as Mistral and LLaVA still exhibit unsafe behaviors inherited from pre-training. Current alignment methods like SFT and RLHF primarily encourage models to generate preferred responses, but do not explicitly remove the unsafe subnetworks that trigger harmful outputs. In this work, we introduce a resource-efficient pruning framework that directly identifies and removes parameters associated with unsafe behaviors while preserving model utility. Our method employs a gradient-free attribution mechanism, requiring only modest GPU resources, and generalizes across architectures and quantized variants. Empirical evaluations on ML models show substantial reductions in unsafe generations and improved robustness against jailbreak attacks, with minimal utility loss. From the perspective of the Lottery Ticket Hypothesis, our results suggest that ML models contain "unsafe tickets" responsible for harmful behaviors, and pruning reveals "safety tickets" that maintain performance while aligning outputs. This provides a lightweight, post-hoc alignment strategy suitable for deployment in resource-constrained settings.

Wai Man Si, Mingjie Li, Michael Backes +1

CISPA Helmholtz Center for Information Security

LLM PruningLLM Safety AlignmentLLM SafetyNeural Network Pruning
AskPDF
Apr 14, 2026

MOONSHOT : A Framework for Multi-Objective Pruning of Vision and Large Language Models

Weight pruning is a common technique for compressing large neural networks. We focus on the challenging post-training one-shot setting, where a pre-trained model is compressed without any retraining. Existing one-shot pruning methods typically optimize a single objective, such as a layer-wise reconstruction loss or a second-order Taylor approximation of the training loss. We highlight that neither objective alone is consistently the most effective across architectures and sparsity levels. Motivated by this insight, we propose MOONSHOT, a general and flexible framework that extends any single-objective pruning method into a multi-objective formulation by jointly optimizing both the layer-wise reconstruction error and second-order Taylor approximation of the training loss. MOONSHOT acts as a wrapper around existing pruning algorithms. To enable this integration while maintaining scalability to billion-parameter models, we propose modeling decisions and introduce an efficient procedure for computing the inverse Hessian, preserving the efficiency of state-of-the-art one-shot pruners. When combined with state-of-the-art pruning methods on Llama-3.2 and Llama-2 models, MOONSHOT reduces C4 perplexity by up to 32.6% at 2:4 sparsity and improves zero-shot mean accuracy across seven classification benchmarks by up to 4.9 points. On Vision Transformers, it improves accuracy on ImageNet-1k by over 5 points at 70% sparsity, and on ResNet-50, it yields a 4-point gain at 90% sparsity.

Gabriel Afriat, Xiang Meng, Shibal Ibrahim +2

Operations Research Center Massachusetts Institute of Technology · Google · OpenAI +1

LLM PruningNeural Network PruningTransformer Pruning
AskPDF
Apr 12, 2026

MoEITS: A Green AI approach for simplifying MoE-LLMs

Large language models are transforming all areas of academia and industry, attracting the attention of researchers, professionals, and the general public. In the trek for more powerful architectures, Mixture-of-Experts, inspired by ensemble models, have emerged as one of the most effective ways to follow. However, this implies a high computational burden for both training and inference. To reduce the impact on computing and memory footprint as well as the energy consumption, simplification methods has arisen as very effective procedures. In this paper, an original algorithm, MoEITS, for MoE-LLMs simplification is presented. The algorithm is characterized by a refined simplicity, underpinned by standardized Information Theoretic frameworks. MoEITS is analyzed in depth from theoretical and practical points of view. Its computational complexity is studied. Its performance on the accuracy of the simplified LLMs and the reduction rate achieved is assessed through a thoroughly designed experimentation. This empirical evaluation includes a comparison with state-of-the-art MoE-LLM pruning methods applied on Mixtral 8×78\times7B, Qwen1.5-2.7B, and DeepSeek-V2-Lite. The extensive experimentation conducted demonstrates that MoEITS outperforms state-of-the-art techniques by generating models that are both effective across all benchmarks and computationally efficient. The code implementing the method will be available at https://github.com/luisbalru/MoEITS.

Luis Balderas, Miguel Lastra, José M. Benítez

Department of Computer Science and Artificial Intelligence, DiCITS, iMUDS, DaSCI, University of Granada, Granada, Spain 18071 · Department of Software Engineering, DiCITS, iMUDS, DaSCI, University of Granada, Granada, Spain 18071

LLM CompressionMixture-of-Experts PruningMixture-of-Experts Language ModelsLLM PruningEnergy-Efficient ML
AskPDF
Apr 8, 2026

NanoSpec: Accelerating Speculative Decoding using Minimalist In-Context Vocabularies

The massive vocabulary sizes of large language models, often exceeding 100k tokens, impose a computational bottleneck on the final linear projection layer during speculative decoding. Existing vocabulary pruning solutions rely on static or coarsely-grained sub-vocabularies that necessitate large active sizes (∼\sim30k) to maintain draft quality. We propose NanoSpec, a novel training-free approach that breaks this trade-off by dynamically constructing a minimalist, context-aware active vocabulary for each generation step. Leveraging the inherent temporal locality of language generation, NanoSpec achieves high coverage while slashing the average vocabulary size by over 40×40\times (to <<3k tokens) without requiring any auxiliary trained parameters. To realize the theoretical benefits of such high sparsity on modern hardware, we introduce a system-algorithm co-design that overcomes the inefficiencies of sparse memory access through asynchronous gathering and GPU-resident state management. As a complementary plug-and-play module, NanoSpec cuts draft time by an average of 51.6%, delivering a 1.171.17-1.29×1.29\times end-to-end speedup over the state-of-the-art speculative decoding methods EAGLE-2 and EAGLE-3 across 7 tasks and outperforming complex training-based pruning baselines.

Zhiyang Chen, Daliang Xu, Yinyuan Zhang +3

Institute for Artificial Intelligence, Peking University, Beijing, China · State Key Laboratory of Networking and Switching Technology; School of Computer Science, Beijing University of Posts and Telecommunications, Beijing, China · Key Laboratory of High Confidence Software Technologies (Peking University), Ministry of Education; School of Computer Science, Peking University, Beijing, China

Speculative DecodingLLM Inference AccelerationLLM Pruning
AskPDF
Apr 6, 2026

SLaB: Sparse-Lowrank-Binary Decomposition for Efficient Large Language Models

The rapid growth of large language models (LLMs) presents significant deployment challenges due to their massive computational and memory demands. While model compression, such as network pruning, offers potential solutions, most existing methods often fail to maintain good performance at high compression ratios. To address this, we propose SLaB, a novel framework that decomposes each linear layer weight into three complementary components: a sparse matrix, a low-rank matrix, and a binary matrix. SLaB eliminates the need for retraining and leverages activation-aware pruning scores to guide the decomposition process. Experiments on Llama-family models demonstrate that SLaB achieves state-of-the-art performance, reducing perplexity by up to 36% compared to existing methods at 50% compression and improving accuracy by up to 8.98% over the baseline on zero-shot tasks.

Ziwei Li, Yuang Ma, Yi Kang

School of Microelectronics, University of Science and Technology of China, Hefei, Anhui, China

Binary QuantizationLLM CompressionLLM PruningLow-Rank Compression
AskPDF
Mar 26, 2026

How Pruning Reshapes Features: Sparse Autoencoder Analysis of Weight-Pruned Language Models

Weight pruning is a standard technique for compressing large language models, yet its effect on learned internal representations remains poorly understood. We present the first systematic study of how unstructured pruning reshapes the feature geometry of language models, using Sparse Autoencoders (SAEs) as interpretability probes. Across three model families (Gemma 3 1B, Gemma 2 2B, Llama 3.2 1B), two pruning methods (magnitude and Wanda), and six sparsity levels (0--60%), we investigate five research questions spanning seed stability, feature survival, SAE transferability, feature fragility, and causal relevance. Our most striking finding is that rare SAE features--those with low firing rates--survive pruning far better than frequent ones, with within-condition Spearman correlations of rho = -1.0 in 11 of 17 experimental conditions. This counter-intuitive result suggests that pruning acts as implicit feature selection, preferentially destroying high-frequency generic features while preserving specialized rare ones. We further show that Wanda pruning preserves feature structure up to 3.7x better than magnitude pruning, that pre-trained SAEs remain viable on Wanda-pruned models up to 50% sparsity, and that geometric feature survival does not predict causal importance--a dissociation with implications for interpretability under compression.

Hector Borobia, Elies Seguí-Mas, Guillermina Tormo-Carbó

VRAIN – Valencian Research Institute for Artificial Intelligence, Universitat Politècnica de València, Valencia, Spain · Department of Economics and Social Sciences, Universitat Politècnica de València, Valencia, Spain · Department of Business Organisation, Universitat Politècnica de València, Valencia, Spain

LLM PruningRepresentation Geometry in Language ModelsLLM Compression
AskPDF
Mar 25, 2026

Demystifying When Pruning Works via Representation Hierarchies

Network pruning, which removes less important parameters or architectures, is often expected to improve efficiency while preserving performance. However, this expectation does not consistently hold across language tasks: pruned models can perform well on non-generative tasks but frequently fail in generative settings. To understand this discrepancy, we analyze network pruning from a representation-hierarchy perspective, decomposing the internal computation of language models into three sequential spaces: embedding (hidden representations), logit (pre-softmax outputs), and probability (post-softmax distributions). We find that representations in the embedding and logit spaces are largely robust to pruning-induced perturbations. However, the nonlinear transformation from logits to probabilities amplifies these deviations, which accumulate across time steps and lead to substantial degradation during generation. In contrast, the stability of the categorical-token probability subspace, together with the robustness of the embedding space, supports the effectiveness of pruning for non-generative tasks such as retrieval and multiple-choice selection. Our analysis disentangles the effects of pruning across tasks and provides practical guidance for its application. Code is available at https://github.com/CASE-Lab-UMD/Pruning-on-Representations

Shwai He, Guoheng Sun, Haichao Zhang +2

University of Maryland, College Park, USA · Northeastern University, USA

LLM PruningLanguage Modeling
AskPDF