CLOVER: Cross-Layer Orthogonal Vectors Pruning and Fine-Tuning

• URL: http://arxiv.org/abs/2411.17426v3
• Date: Fri, 31 Jan 2025 14:13:49 GMT
• Title: CLOVER: Cross-Layer Orthogonal Vectors Pruning and Fine-Tuning
• Abstract summary: CLOVER (Cross-Layer Orthogonal Vectors) is a novel approach that treats pairs of attention layers as a set of low-rank decompositions.<n>We apply CLOVER to various models, including GPT-2 XL, DeepSeek-V2-Lite, Whisper-Large-v3, Stable Diffusion XL, and LLaMA-3.2-11B-Vision.
• Score: 24.697865155104644
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Decoder-only models generate tokens autoregressively by caching key/value vectors, but as the cache grows, inference becomes memory-bound. To address this issue, we introduce CLOVER (Cross-Layer Orthogonal Vectors), a novel approach that treats pairs of attention layers as a set of low-rank decompositions. CLOVER applies Singular Value Decomposition (SVD) to the \( Q \)-\( K \) and \( V \)-\( O \) pairs within each attention head. The resulting singular values can either guide pruning or serve as trainable parameters for efficient fine-tuning of all orthogonal vectors. After pruning or fine-tuning, these values are reintegrated into the model without increasing its parameter count. We apply CLOVER to various models, including GPT-2 XL, DeepSeek-V2-Lite, Whisper-Large-v3, Stable Diffusion XL, and LLaMA-3.2-11B-Vision. Our results demonstrate that CLOVER significantly improves pruning efficiency. For instance, the perplexity of pruning 70\% of the \( Q \)-\( K \) pairs in GPT-2 XL is similar to that of pruning just 8\% with vanilla methods. Fine-tuning the singular values further results in a full-rank update, outperforming state-of-the-art methods (LoRA, DoRA, HiRA, and PiSSA) by 7.6\%, 5.5\%, 3.8\%, and 0.7\%, respectively, on eight commonsense tasks for LLaMA-2 7B.

Related papers

• SemKV: Semantic Mixed-Precision KV Cache Quantization Guided by the Quality Cliff for Long-Context LLM Inference [3.2938220347051175]
  We show that uniform KV quantization on a fractional-bit grid does not degrade gracefully.<n>We show that replacing the affine base with a distortion-optimized quantizer (TurboQuant-MSE) prunes the cliff in every protocol tested.
  arXiv  Detail & Related papers  (2026-08-28T22:19:47Z)
• CARVE: Content-Aware Recurrent with Value Efficiency for Chunk-Parallel Linear Attention [0.0]
  Recurrent models must forget in order to remember, yet the state of the art decides what to erase without consulting what is stored.<n>This memory-blind gating is one of three coupled defects in the leading delta-rule architecture (GDN-2)<n>We introduce CARVE, which resolves all three problems through one principle: erase only on the key axis.
  arXiv  Detail & Related papers  (2026-06-25T16:16:51Z)
• TriSP: Tri-Signal Structured Pruning for Large Language Models [0.0]
  Large language models (LLMs) achieve strong performance across diverse tasks but their deployment is constrained by the memory and compute cost of their parameters.<n> Structured pruning addresses this by removing entire structures such as attention heads and Multi-Layer Perceptron (MLP) neurons to produce smaller dense models that run efficiently on standard hardware.<n>We propose TriSP (Tri-Signal Structured Pruning), an importance metric that combines weight magnitude scaled by activation norm with first-order gradient sensitivity via a geometric mean.
  arXiv  Detail & Related papers  (2026-06-09T17:02:14Z)
• LimiX-2M: Mitigating Low-Rank Collapse and Attention Bottlenecks in Tabular Foundation Models [56.999481798138625]
  LimiX-2M is a 2M- parameter model that outperforms larger TabPFN-v2 and TabICL baselines on widely used benchmarks.<n>We present a unified tokenize-and-route framework for strong Tabular foundation models (TFMs)<n>Results highlight value-aware tokenization and readout-aligned routing as key levers for improving the accuracy--efficiency trade-off in TFMs.
  arXiv  Detail & Related papers  (2026-06-03T06:07:33Z)
• Rethinking Attention Output Projection: Structured Hadamard Transforms for Efficient Transformers [0.0]
  We propose replacing the dense output projection in multi-head attention with a fixed, parameter-free Walsh Hadamard Transform.<n>Across different model sizes, we demonstrate that this structured substitution maintains comparable or slightly superior downstream task performance.<n>We observe that structured Hadamard-based models exhibit a steeper validation loss curve relative to training FLOPs, suggesting more favorable compute utilization during training.
  arXiv  Detail & Related papers  (2026-03-09T13:05:08Z)
• LINA: Linear Autoregressive Image Generative Models with Continuous Tokens [56.80443965097921]
  Autoregressive models with continuous tokens form a promising paradigm for visual generation, especially for text-to-image (T2I) synthesis.<n>We study how to design compute-efficient linear attention within this framework.<n>We present LINA, a simple and compute-efficient T2I model built entirely on linear attention, capable of generating high-fidelity 1024x1024 images from user instructions.
  arXiv  Detail & Related papers  (2026-01-30T06:44:33Z)
• ARMOR: High-Performance Semi-Structured Pruning via Adaptive Matrix Factorization [99.96330641363396]
  ARMOR: (Adaptive Representation with Matrix-factORization) is a novel one-shot post-training pruning algorithm.<n>Instead of directly pruning weights, ARMOR factorizes each weight matrix into a 2:4 sparse core wrapped by two low-overhead, block diagonal matrices.<n>We demonstrate ARMOR consistently and significantly outperforms state-of-the-art 2:4 pruning methods across a wide range of downstream tasks and perplexity evaluations.
  arXiv  Detail & Related papers  (2025-10-07T02:39:20Z)
• High-Layer Attention Pruning with Rescaling [14.141903038286362]
  Pruning is a highly effective approach for compressing large language models (LLMs)<n>We propose a novel pruning algorithm that strategically prunes attention heads in the model's higher layers.<n>We conduct comprehensive experiments on a wide range of LLMs, including LLaMA3.1-8B, Mistral-7B-v0.3, Qwen2-7B, and Gemma2-9B.
  arXiv  Detail & Related papers  (2025-07-02T17:15:05Z)
• A Simple Linear Patch Revives Layer-Pruned Large Language Models [58.056251480151104]
  Layer pruning has emerged as a widely used technique for compressing large language models (LLMs)<n>textscLinearPatch fuses two operations into one matrix multiply at the pruning interface.<n>The patch can be further refined with 5K unlabeled samples via memory-efficient offline distillation, pushing the retention to 95.16% within only 30 minutes on a single GPU.
  arXiv  Detail & Related papers  (2025-05-30T15:06:08Z)
• Generalized Kullback-Leibler Divergence Loss [105.66549870868971]
  We prove that the Kullback-Leibler (KL) Divergence loss is equivalent to the Decoupled Kullback-Leibler (DKL) Divergence loss. Thanks to the decoupled structure of DKL loss, we have identified two areas for improvement.
  arXiv  Detail & Related papers  (2025-03-11T04:43:33Z)
• Align Attention Heads Before Merging Them: An Effective Way for Converting MHA to GQA [8.305827430948654]
  We propose a cost-effective method for converting MHA into GQA with any compression ratio of KV heads.<n>We employ $mathitL$ regularization to prune redundant parameters.<n> Experimental results show that our strategy can compress up to 87.5% KV heads of LLaMA2-7B model and 75% KV heads of Sheared-LLaMA-1.3B with acceptable performance degradation.
  arXiv  Detail & Related papers  (2024-12-30T03:05:45Z)
• Efficient Adaptation of Pre-trained Vision Transformer via Householder Transformation [53.88562288388169]
  A common strategy for. Efficient Fine-Tuning (PEFT) of pre-trained Vision Transformers (ViTs) involves adapting the model to downstream tasks. We propose a novel PEFT approach inspired by Singular Value Decomposition (SVD) for representing the adaptation matrix. SVD decomposes a matrix into the product of a left unitary matrix, a diagonal matrix of scaling values, and a right unitary matrix.
  arXiv  Detail & Related papers  (2024-10-30T12:08:30Z)
• EoRA: Training-free Compensation for Compressed LLM with Eigenspace Low-Rank Approximation [79.56709262189953]
  EoRA consistently outperforms previous methods in compensating errors for compressed LLaMA2/3 models on various tasks. EoRA offers a scalable, training-free solution to compensate for compression errors.
  arXiv  Detail & Related papers  (2024-10-28T17:59:03Z)
• Memory-Efficient LLM Training with Online Subspace Descent [8.393403749426097]
  We propose Online Subspace Descent, a new family of subspace descent without singular value decomposition (SVD) Online Subspace Descent is flexible and introduces only minimum overhead to training. We show that for the task of pretraining LLaMA models ranging from 60M to 7B parameters on the C4 dataset, Online Subspace Descent achieves lower perplexity and better downstream tasks performance.
  arXiv  Detail & Related papers  (2024-08-23T05:54:53Z)
• LoRA$^2$ : Multi-Scale Low-Rank Approximations for Fine-Tuning Large Language Models [3.7049613588433497]
  Low-Rank Adaptation (LoRA) significantly reduces the number of trainable parameters for fine-tuning. We extend the LoRA to multiple scales, dubbed as LoRA$2$.
  arXiv  Detail & Related papers  (2024-08-13T12:31:30Z)
• ADMM Based Semi-Structured Pattern Pruning Framework For Transformer [4.02487511510606]
  This paper introduces Alternating Direction Method of Multipliers(ADMM) based pattern pruning framework to reshape the distribution of activation map. We conduct extensive experiments on classification tasks over GLUE dataset. We achieve 50% percent compression ratio while maintaining overall score 80.1 on GLUE dataset.
  arXiv  Detail & Related papers  (2024-07-11T09:35:08Z)
• DELLA-Merging: Reducing Interference in Model Merging through Magnitude-Based Sampling [24.270321913746233]
  We propose a new model merging technique, Drop and rEscaLe via sampLing with mAgnitude (DELLA-Merging), that employs a novel pruning technique, MAGPRUNE. MAGPRUNE first ranks the parameters in order of their magnitude and assigns higher dropout probabilities (p) to parameters with lower ranks corresponding to lower magnitudes.
  arXiv  Detail & Related papers  (2024-06-17T15:02:45Z)
• MiLoRA: Harnessing Minor Singular Components for Parameter-Efficient LLM Finetuning [16.67302585857681]
  We propose MiLoRA, a simple yet effective LLM finetuning approach that only updates the minor singular components of the weight matrix. It is observed that the minor matrix corresponds to the noisy or long-tail information, while the principal matrix contains important knowledge. During finetuning, MiLoRA makes the most use of the less-optimized subspace for learning the labeled dataset.
  arXiv  Detail & Related papers  (2024-06-13T12:30:02Z)
• AffineQuant: Affine Transformation Quantization for Large Language Models [58.45460102764]
  Post-Training Quantization (PTQ) has emerged as a subject of considerable interest due to its compression efficiency and cost-effectiveness in the context of training. Existing PTQ methods for Large-scale Language Models (LLMs) limit the optimization scope to scaling transformations between pre- and post-quantization weights. In this paper, we advocate for the direct optimization using equivalent Affine transformations in PTQ (AffineQuant)
  arXiv  Detail & Related papers  (2024-03-19T08:40:21Z)
• ConvLoRA and AdaBN based Domain Adaptation via Self-Training [4.006331916849688]
  We propose Convolutional Low-Rank Adaptation (ConvLoRA) for multi-target domain adaptation. ConvLoRA freezes pre-trained model weights, adds trainable low-rank decomposition matrices to convolutional layers, and backpropagates the gradient. Our method has fewer trainable parameters and performs better or on-par with large independent fine-tuned networks.
  arXiv  Detail & Related papers  (2024-02-07T15:43:50Z)
• LQ-LoRA: Low-rank Plus Quantized Matrix Decomposition for Efficient Language Model Finetuning [66.85589263870702]
  Our approach uses an iterative algorithm to decompose each pretrained matrix into a high-precision low-rank component and a memory-efficient quantized component. Experiments on finetuning RoBERTa and LLaMA-2 demonstrate that our low-rank plus quantized matrix decomposition approach (LQ-LoRA) outperforms strong QLoRA and GPTQ-LoRA baselines.
  arXiv  Detail & Related papers  (2023-11-20T18:57:41Z)
• CLIP-AD: A Language-Guided Staged Dual-Path Model for Zero-shot Anomaly Detection [49.510604614688745]
  We propose a framework called CLIP-AD to leverage the zero-shot capabilities of the large vision-language model CLIP. We note opposite predictions and irrelevant highlights in the direct computation of the anomaly maps.
  arXiv  Detail & Related papers  (2023-11-01T11:39:22Z)
• Attention Map Guided Transformer Pruning for Edge Device [98.42178656762114]
  Vision transformer (ViT) has achieved promising success in both holistic and occluded person re-identification (Re-ID) tasks. We propose a novel attention map guided (AMG) transformer pruning method, which removes both redundant tokens and heads. Comprehensive experiments on Occluded DukeMTMC and Market-1501 demonstrate the effectiveness of our proposals.
  arXiv  Detail & Related papers  (2023-04-04T01:51:53Z)
• Transformers meet Stochastic Block Models: Attention with Data-Adaptive Sparsity and Cost [53.746169882193456]
  Recent works have proposed various sparse attention modules to overcome the quadratic cost of self-attention. We propose a model that resolves both problems by endowing each attention head with a mixed-membership Block Model. Our model outperforms previous efficient variants as well as the original Transformer with full attention.
  arXiv  Detail & Related papers  (2022-10-27T15:30:52Z)
• On the Eigenvalues of Global Covariance Pooling for Fine-grained Visual Recognition [65.67315418971688]
  We show that truncating small eigenvalues of the Global Covariance Pooling (GCP) can attain smoother gradient. On fine-grained datasets, truncating the small eigenvalues would make the model fail to converge. Inspired by this observation, we propose a network branch dedicated to magnifying the importance of small eigenvalues.
  arXiv  Detail & Related papers  (2022-05-26T11:41:36Z)
• Basis Scaling and Double Pruning for Efficient Inference in Network-Based Transfer Learning [1.3467579878240454]
  We decompose a convolutional layer into two layers: a convolutional layer with the orthonormal basis vectors as the filters, and a "BasisScalingConv" layer which is responsible for rescaling the features. We can achieve pruning ratios up to 74.6% for CIFAR-10 and 98.9% for MNIST in model parameters.
  arXiv  Detail & Related papers  (2021-08-06T00:04:02Z)
• MLPruning: A Multilevel Structured Pruning Framework for Transformer-based Models [78.45898846056303]
  Pruning is an effective method to reduce the memory footprint and computational cost associated with large natural language processing models. We develop a novel MultiLevel structured Pruning framework, which uses three different levels of structured pruning: head pruning, row pruning, and block-wise sparse pruning.
  arXiv  Detail & Related papers  (2021-05-30T22:00:44Z)
• Solving weakly supervised regression problem using low-rank manifold regularization [77.34726150561087]
  We solve a weakly supervised regression problem. Under "weakly" we understand that for some training points the labels are known, for some unknown, and for others uncertain due to the presence of random noise or other reasons such as lack of resources. In the numerical section, we applied the suggested method to artificial and real datasets using Monte-Carlo modeling.
  arXiv  Detail & Related papers  (2021-04-13T23:21:01Z)
• Training Invertible Linear Layers through Rank-One Perturbations [0.0]
  This work presents a novel approach for training invertible linear layers. In lieu of directly optimizing the network parameters, we train rank-one perturbations and add them to the actual weight matrices infrequently. We show how such invertible blocks improve the mixing and thus normalizing the mode separation of the resulting flows.
  arXiv  Detail & Related papers  (2020-10-14T12:43:47Z)
• Joint Multi-Dimension Pruning via Numerical Gradient Update [120.59697866489668]
  We present joint multi-dimension pruning (abbreviated as JointPruning), an effective method of pruning a network on three crucial aspects: spatial, depth and channel simultaneously. We show that our method is optimized collaboratively across the three dimensions in a single end-to-end training and it is more efficient than the previous exhaustive methods.
  arXiv  Detail & Related papers  (2020-05-18T17:57:09Z)
• Controllable Orthogonalization in Training DNNs [96.1365404059924]
  Orthogonality is widely used for training deep neural networks (DNNs) due to its ability to maintain all singular values of the Jacobian close to 1. This paper proposes a computationally efficient and numerically stable orthogonalization method using Newton's iteration (ONI) We show that our method improves the performance of image classification networks by effectively controlling the orthogonality to provide an optimal tradeoff between optimization benefits and representational capacity reduction. We also show that ONI stabilizes the training of generative adversarial networks (GANs) by maintaining the Lipschitz continuity of a network, similar to spectral normalization (
  arXiv  Detail & Related papers  (2020-04-02T10:14:27Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.