analiza aiqbench
Structural Anchor Pruning offers a significant breakthrough for deploying Vision-Language Models in resource-constrained environments by achieving up to 90% compression without requiring query-dependent training. The method successfully isolates stable visual structures within model backbones, allowing for aggressive token pruning while maintaining high retrieval accuracy on challenging benchmarks like ViDoRe v2. This approach resolves previous limitations where heuristic layer choices led to sharp performance degradation under heavy compression.
Structural Anchor Pruning: Training-Free Multi-Vector Compression for Visual Document Retrieval
arXiv:2601.20107v3 Announce Type: replace-cross Abstract: Recent Vision-Language Models (e.g., ColPali) enable fine-grained Visual Document Retrieval (VDR) but incur prohibitive multi-vector index storage overhead. Existing training-free pruning methods either rely on heuristic layer choices or degrade sharply under aggressive compression, leading prior work to argue that effective high-compression pruning requires query-dependent training. We challenge this view with Structural Anchor Pruning (SAP), a self-calibrating, training-free, query-agnostic index-time framework combining (i) Score Retention (SR), a white-box per-layer compression diagnostic; (ii) SR-guided window selection, which automatically locates the structural pruning region of any backbone with no per-model hyperparameters; and (iii) a visual in-degree centrality scorer that identifies anchor patches within that window. On ViDoRe v1/v2 across three architectures spanning 18, 28, and 36 backbone layers, SAP retains 93--96\% of NDCG@5 on v1 and 88--90\% on the harder v2 while pruning 90\% of visual tokens; at 20$\times$ compression it retains 85--90\% and 76--79\% respectively. Our layer-resolved SR analysis reveals an Alignment-Aggregation Divergence: visual structure is preserved as a stable ``Structural Plateau'' within the backbone, while the final layers reshape it into a sparse, query-aligned form unsuitable for pruning. Probing the pre-retrieval base backbones shows that contrastive fine-tuning sharpens this boundary three- to eight-fold, explaining why final-layer methods fail.
Abstract:Recent Vision-Language Models (e.g., ColPali) enable fine-grained Visual Document Retrieval (VDR) but incur prohibitive multi-vector index storage overhead. Existing training-free pruning methods either rely on heuristic layer choices or degrade sharply under aggressive compression, leading prior work to argue that effective high-compression pruning requires query-dependent training. We challenge this view with Structural Anchor Pruning (SAP), a self-calibrating, training-free, query-agnostic index-time framework combining (i) Score Retention (SR), a white-box per-layer compression diagnostic; (ii) SR-guided window selection, which automatically locates the structural pruning region of any backbone with no per-model hyperparameters; and (iii) a visual in-degree centrality scorer that identifies anchor patches within that window. On ViDoRe v1/v2 across three architectures spanning 18, 28, and 36 backbone layers, SAP retains 93--96\% of NDCG@5 on v1 and 88--90\% on the harder v2 while pruning 90\% of visual tokens; at 20$\times$ compression it retains 85--90\% and 76--79\% respectively. Our layer-resolved SR analysis reveals an Alignment-Aggregation Divergence: visual structure is preserved as a stable ``Structural Plateau'' within the backbone, while the final layers reshape it into a sparse, query-aligned form unsuitable for pruning. Probing the pre-retrieval base backbones shows that contrastive fine-tuning sharpens this boundary three- to eight-fold, explaining why final-layer methods fail.
Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL); Information Retrieval (cs.IR)
arXiv:2601.20107 [cs.CV]
(or arXiv:2601.20107v3 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2601.20107 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 300 cuvinte
Subiecte conexe
Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
Hugging Face · 26 aug. 2026
Presentation: From S3 to GPU in One Copy: Rethinking Data Loading for ML Training
InfoQ · 4 sept. 2026
Task-Aware Structured Memory for Dynamic Multi-modal In-Context Learning
arXiv · 11 iun. 2026
Quadrillion Param Costs: KV Cache, Context Length, Frontier Margins
LessWrong · 27 iul. 2026