Imagen de Portada

Predicting LLM Compression Degradation from Spectral Statistics

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autor principal:	Xu, Mingxue
Formato:	Preprint
Publicado:	2026
Materias:	Machine Learning
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Geometry is All You Need: A Unified Taxonomy of Matrix and Tensor Factorization for Compression of Generative Language Models
por: Xu, Mingxue, et al.
Publicado: (2024)

TensorSLM: Energy-efficient Embedding Compression of Sub-billion Parameter Language Models on Low-end Devices
por: Xu, Mingxue, et al.
Publicado: (2025)

TensorGPT: Efficient Compression of Large Language Models based on Tensor-Train Decomposition
por: Xu, Mingxue, et al.
Publicado: (2023)

How can representation dimension dominate structurally pruned LLMs?
por: Xu, Mingxue, et al.
Publicado: (2025)

The Lifecycle of the Spectral Edge: From Gradient Learning to Weight-Decay Compression
por: Xu, Yongzhong
Publicado: (2026)

Enhancing Spectral Graph Neural Networks with LLM-Predicted Homophily
por: Lu, Kangkang, et al.
Publicado: (2025)

On-Policy Consistency Training Improves LLM Safety with Minimal Capability Degradation
por: Han, Andy, et al.
Publicado: (2026)

Statistical Analysis of Policy Space Compression Problem
por: Molaei, Majid, et al.
Publicado: (2024)

Dobi-SVD: Differentiable SVD for LLM Compression and Some New Perspectives
por: Wang, Qinsi, et al.
Publicado: (2025)

A Generalized Spectral Framework to Expain Neural Scaling and Compression Dynamics
por: Zhang, Yizhou
Publicado: (2025)

Do More Predictions Improve Statistical Inference? Filtered Prediction-Powered Inference
por: Xu, Shirong, et al.
Publicado: (2026)

Lossless Compression for LLM Tensor Incremental Snapshots
por: Waddington, Daniel, et al.
Publicado: (2025)

Can Compressed LLMs Truly Act? An Empirical Evaluation of Agentic Capabilities in LLM Compression
por: Dong, Peijie, et al.
Publicado: (2025)

Task-Driven Kernel Flows: Label Rank Compression and Laplacian Spectral Filtering
por: Li, Hongxi, et al.
Publicado: (2026)

Self-Indexing KVCache: Predicting Sparse Attention from Compressed Keys
por: Yang, Xu, et al.
Publicado: (2026)

ProCut: LLM Prompt Compression via Attribution Estimation
por: Xu, Zhentao, et al.
Publicado: (2025)

Efficient Evaluation of LLM Performance with Statistical Guarantees
por: Wu, Skyler, et al.
Publicado: (2026)

Leech Lattice Vector Quantization for Efficient LLM Compression
por: van der Ouderaa, Tycho F. A., et al.
Publicado: (2026)

Efficient Multi-Cohort Inference for Long-Term Effects and Lifetime Value in A/B Testing with User Learning
por: Simionato, Dario, et al.
Publicado: (2026)

Generating Statistical Charts with Validation-Driven LLM Workflows
por: Poličar, Pavlin G., et al.
Publicado: (2026)

Statistically Accurate and Robust Generative Prediction of Rock Discontinuities with A Tabular Foundation Model
por: Meng, Han, et al.
Publicado: (2025)

When Less is More: The LLM Scaling Paradox in Context Compression
por: Guo, Ruishan, et al.
Publicado: (2026)

ProjQ: Project-and-Quantize for Adapter-Aware LLM Compression
por: Yu, Wneya, et al.
Publicado: (2026)

The Spectral Lifecycle of Transformer Training: Transient Compression Waves, Persistent Spectral Gradients, and the Q/K--V Asymmetry
por: Liu, Yi
Publicado: (2026)

Controlled LLM Training on Spectral Sphere
por: Xie, Tian, et al.
Publicado: (2026)

Spectral Lens: Activation and Gradient Spectra as Diagnostics of LLM Optimization
por: Liu, Andy Zeyi, et al.
Publicado: (2026)

Kernelized Advantage Estimation: From Nonparametric Statistics to LLM Reasoning
por: Gong, Shijin, et al.
Publicado: (2026)

Spectral Journey: How Transformers Predict the Shortest Path
por: Cohen, Andrew, et al.
Publicado: (2025)

Wine Characterisation with Spectral Information and Predictive Artificial Intelligence
por: Yao, Jianping, et al.
Publicado: (2025)

Compression Aware Certified Training
por: Xu, Changming, et al.
Publicado: (2025)

Huff-LLM: End-to-End Lossless Compression for Efficient LLM Inference
por: Yubeaton, Patrick, et al.
Publicado: (2025)

The Significance of Latent Data Divergence in Predicting System Degradation
por: Fernandes, Miguel, et al.
Publicado: (2024)

NuMuon: Nuclear-Norm-Constrained Muon for Compressible LLM Training
por: Dolatabadi, Hadi Mohaghegh, et al.
Publicado: (2026)

PV-Tuning: Beyond Straight-Through Estimation for Extreme LLM Compression
por: Malinovskii, Vladimir, et al.
Publicado: (2024)

Compress the Context, Keep the Commitments: A Formal Framework for Verifiable LLM Context Compression
por: Trukhina, Natalia, et al.
Publicado: (2026)

General Machine Learning Models for Interpreting and Predicting Efficiency Degradation in Organic Solar Cells
por: Valiente, David, et al.
Publicado: (2024)

Spectral and Spatial Graph Learning for Multispectral Solar Image Compression
por: Siwakoti, Prasiddha, et al.
Publicado: (2025)

Zero Sum SVD: Balancing Loss Sensitivity for Low Rank LLM Compression
por: Abbasi, Ali, et al.
Publicado: (2026)

Elastic-dLLM: Position Preserving Context Compression and Augmentation of Diffusion LLMs
por: Wu, Junyi, et al.
Publicado: (2026)

IO-SVD: Input-Output Whitened SVD for Adaptive-Rank LLM Compression
por: Abbasi, Ali, et al.
Publicado: (2026)