paper-with-me

홈 › Papers

Cross-Layer Subspace Coupling for LLM Compression: A Unifying Framework and Its Empirical Limits

2026-05-29 · Snigdha Chandan Khilar arxiv

Recent SVD based compression methods for large language models like SVD LLM and Basis Sharing can be unified under one optimization problem. While mathematical proofs and tests on Pythia models show this unified approach improves weight reconstruction error by up to 46% percent it fails in practical tasks. Downstream metrics like perplexity and accuracy severely degrade compared to standard per layer SVD LLM. The authors explain this failure mechanistically. Although the bundle method mathematically couples adjacent layers the transformer residual stream actually decouples them during forward passes. Thus per layer optimality matters more than joint cross layer optimization. The paper concludes that weight space reconstruction is a flawed objective for cross layer compression and future methods must focus on per layer activation reconstruction instead.

📄 PDF Abstract BibTeX arXiv:2605.30836

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Beyond Variance: Knowledge-Aware LLM Compression via Fisher-Aligned Subspace Diagnostics

2026-01-12 · Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma arxiv

Post-training activation compression is essential for deploying Large Language Models (LLMs) on resource-constrained hardware. However, standard methods like Singular Value Decomposition (SVD) are gradient-blind: they pr…

A General Error-Theoretical Analysis Framework for Constructing Compression Strategies

2025-02-19 · Boyang Zhang, Daning Cheng, Yunquan Zhang, Meiqi Tu 외

The exponential growth in parameter size and computational complexity of deep models poses significant challenges for efficient deployment. The core problem of existing compression methods is that different layers of the…

Quantization

MIC: Maximizing Informational Capacity in Adaptive Representations via Isotropic Subspace Alignment

2026-05-28 · Dang Nguyen Hong, Nhi Ngoc-Yen Nguyen, Huy-Hieu Pham arxiv

Although multi-scales representation learning enables elastic-dimension embeddings, nested subspaces often suffer from dimensional redundancy and spectral collapse. To address this, we introduce MIC, a framework that opt…

Representation Learning

Compress Then Adapt? No, Do It Together via Task-aware Union of Subspaces

2026-05-04 · Jingze Ge, Yun Liu, Xue Geng, Wanqi Dong 외 arxiv

Adapting large pretrained models to diverse tasks is now routine, yet the two dominant strategies of parameter-efficient fine-tuning (PEFT) and low-rank compression are typically composed in sequence. This decoupled prac…

parameter-efficient fine-tuning

COMPOT: Calibration-Optimized Matrix Procrustes Orthogonalization for Transformers Compression

2026-02-16 · Denis Makhov, Dmitriy Shopkhoev, Magauiya Zhussip, Ammar Ali 외 arxiv

Post-training compression of Transformer models commonly relies on truncated singular value decomposition (SVD). However, enforcing a single shared subspace can degrade accuracy even at moderate compression. Sparse dicti…