{"total":27,"items":[{"citing_arxiv_id":"2606.31383","ref_index":27,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"MS-Resampler: Multi-Scope Visual Resampling for Efficient Multimodal LLMs","primary_cat":"cs.CV","submitted_at":"2026-06-30T09:11:50+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"MS-Resampler deploys multiple scope-specific resamplers with explicit spatial priors and adaptive fusion to outperform single-scope global cross-attention in MLLMs on ten benchmarks with minimal added cost.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.29350","ref_index":36,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Fast Enough to Act: Spatio-Temporal Visual Token Merging for Low-Latency Robotic VLMs and VLAs","primary_cat":"cs.CV","submitted_at":"2026-06-28T11:42:55+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"ST-Merge is a plug-and-play spatio-temporal token merging method that delivers 2x speedup on VLMs and 8.3x on a VLA at high resolution with minimal accuracy loss via 3D coordinate matching and positional correction.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.22352","ref_index":35,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"On the Sparsity-Storage-Accuracy Tradeoff in Parsimoniously Activated Dictionary Learning","primary_cat":"cs.LG","submitted_at":"2026-06-21T06:20:55+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"PADL is shown equivalent to MAP estimation under a structured generative model, yielding generalization guarantees, an analytical sparsity-storage-accuracy tradeoff, and a tuning-free algorithm tested on visual benchmarks and vision-language model acceleration.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.09131","ref_index":16,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Late-Layer Fusion is Enough: Dual-Path Vision Token Routing for Multimodal Large Language Models under Visual Saturation","primary_cat":"cs.AI","submitted_at":"2026-06-08T07:28:14+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"DPVR-LF routes saturated vision tokens into a one-layer side branch after layer 4, runs text-only processing through layers 5-17, and performs late fusion at the final layer to reduce visual computation while preserving multimodal performance.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.01503","ref_index":30,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"On the Limits of Token Reduction for Efficient Unified Vision Language Training","primary_cat":"cs.CV","submitted_at":"2026-05-31T23:59:12+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Analysis of unified VLM training reveals asymmetric image token dependence between understanding and generation, leading to synergy loss when applying task-specific token reduction in joint optimization.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.00535","ref_index":145,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"DREAM-S: Speculative Decoding with Searchable Drafting and Target-Aware Refinement for Multimodal Generation","primary_cat":"cs.LG","submitted_at":"2026-05-30T05:05:24+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"DREAM-S combines neural architecture search, target-aware supernet training, and attention-entropy-guided distillation to accelerate speculative decoding in VLMs, reporting up to 3.85x speedup over standard methods.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.25343","ref_index":205,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Toward Native Multimodal Modeling: A Roadmap","primary_cat":"cs.CV","submitted_at":"2026-05-25T01:57:43+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":3.0,"formal_verification":"none","one_line_summary":"A roadmap that defines architectural nativity for multimodal models and categorizes them into Multi-to-Text, Multi-to-Target, and Multi-to-Multi types while outlining an industrial pipeline toward unified transformer-based native multimodal modeling.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"idea appears in production-oriented models such as MiniCPM-V 4.5 and Gemma3 [ 201], where image and video features are summarized into compact visual sequences before being passed to the language backbone. More adaptive methods further observe that most visual tokens are redundant for a given query. VisionZip [202], SparseVLM [203], FitPrune [204], and LLaV A-PruMerge [205] select, prune, recycle, or merge visual tokens according to information density, attention behavior, or similarity structure, while trainable methods such as VisionSelector [206] and LaCo [207] move compression into the learned visual pathway itself. The intuition is that multimodal reasoning rarely requires preserving every patch with equal fidelity: global semantics, task-relevant regions, and fine-grained details should"},{"citing_arxiv_id":"2605.24785","ref_index":13,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"PANDO: Efficient Multimodal AI Agents via Online Skill Distillation","primary_cat":"cs.AI","submitted_at":"2026-05-24T00:07:25+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"PANDO introduces an online skill-distillation method with a structured library, reflection, demotion, routing, compression, and cache-aware prompting that reaches 58.3% success on 910 VisualWebArena tasks using 58-61% fewer tokens than prior methods.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.22158","ref_index":8,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"ST-SimDiff: Balancing Spatiotemporal Similarity and Difference for Efficient Video Understanding with MLLMs","primary_cat":"cs.AI","submitted_at":"2026-05-21T08:27:15+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"ST-SimDiff is a training-free method using a spatio-temporal graph and dual similarity-difference selection to compress video tokens for MLLMs while retaining static and dynamic content.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.17447","ref_index":30,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"FastOCR: Dynamic Visual Fixation via KV Cache Pruning for Efficient Document Parsing","primary_cat":"cs.CV","submitted_at":"2026-05-17T13:39:47+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"FastOCR dynamically selects a small subset of visual tokens per decoding step using focal-guided pruning and cross-step reuse, retaining 98% accuracy on Qwen2.5-VL while attending to only 5% of tokens and cutting attention latency by 3x.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.15621","ref_index":29,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"LRCP: Low-Rank Compressibility Guided Visual Token Pruning for Efficient LVLMs","primary_cat":"cs.CV","submitted_at":"2026-05-15T05:09:15+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"LRCP prunes visual tokens in LVLMs by scoring projection residuals onto a PCA-estimated low-rank subspace, achieving 88.9% image token reduction with 94.7% performance retention and 87.5% video reduction with 97.8% accuracy retention.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.13375","ref_index":16,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"GRIP-VLM: Group-Relative Importance Pruning for Efficient Vision-Language Models","primary_cat":"cs.CV","submitted_at":"2026-05-13T11:32:03+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"GRIP-VLM applies group-relative policy optimization via reinforcement learning to prune visual tokens in VLMs, yielding up to 15% inference speedup at matched accuracy over prior methods.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.09429","ref_index":25,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Evading Visual Aphasia: Contrastive Adaptive Semantic Token Pruning for Vision-Language Models","primary_cat":"cs.CV","submitted_at":"2026-05-10T09:07:04+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"COAST prunes 77.8% of visual tokens in LVLMs with a 2.15x speedup while keeping 98.64% of original performance by adaptively routing semantic and spatial context via contrastive scores.","context_count":1,"top_context_role":"baseline","top_context_polarity":"baseline","context_text":"This cost has motivated visual token reduction methods [ 19, 13, 20, 21, 22, 23, 14, 24] that compress the visual sequence before or during language decoding. Existing approaches [13, 23, 14, 22] typically rely on early or layer-static importance estimates. Text-aware methods such as FastV [13] prune tokens according to shallow-layer text-to-image attention, while spatial reduction methods such as LLaV A-PruMerge [25] merge patches based on local visual similarity. These strategies are effective for reducing computation, but they implicitly assume that token importance can be determined from an early, scalar signal. We find that this assumption is unreliable for compositional vision-language reasoning [26, 27]. Tokens that receive low attention in shallow layers may later become important for resolving"},{"citing_arxiv_id":"2605.08329","ref_index":48,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"An Efficient Token Compression Framework for Visual Object Tracking","primary_cat":"cs.CV","submitted_at":"2026-05-08T17:26:24+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"ETCTrack compresses template tokens by 60% in visual trackers via an adaptive compressor and hierarchical interaction, cutting MACs 21.4% with 0.4% accuracy drop on seven benchmarks.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Advances in Neu- ral Information Processing Systems, 37:130797-130818, 2024. 6 [47] Hamid Rezatofighi, Nathan Tsoi, JunYoung Gwak, Amir Sadeghian, Ian Reid, and Silvio Savarese. Generalized in- tersection over union: A metric and a loss for bounding box regression. In2019 IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), 2019. 5 [48] Yuzhang Shang, Mu Cai, Bingxin Xu, Yong Jae Lee, and Yan Yan. Llava-prumerge: Adaptive token reduction for efficient large multimodal models.arXiv preprint arXiv:2403.15388, 2024. 3 [49] Liangtao Shi, Bineng Zhong, Qihua Liang, Ning Li, Sheng- ping Zhang, and Xianxian Li. Explicit visual prompts for visual object tracking. InProceedings of the AAAI Confer-"},{"citing_arxiv_id":"2605.03351","ref_index":22,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"VLMaxxing through FrameMogging Training-Free Anti-Recomputation for Video Vision-Language Models","primary_cat":"cs.CV","submitted_at":"2026-05-05T04:13:32+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Training-free adaptive reuse of stable visual state in video VLMs reduces follow-up latency by 15-36x on Qwen2.5-VL while preserving correctness on VideoMME, with smaller first-query speedups via pruning.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"deployment question we care about here: how much redundancy a frozen stack can exploit before retraining or representation learning enters the picture. Another family reduces visual cost after tokens already exist. FastV [ 8], FastVID [ 9], FlashVID [11], VisionZip [ 40], FrameFusion [ 12], SparseVLM [ 31], VScan [ 43], STTM [ 34], SparseVILA [ 32], and LLaV A-PruMerge [22] prune, retrieve, or merge visual tokens using attention-derived, density, query-aware, or feature-similarity signals. ToMe [ 38] is the older training-free Vision Transformer (ViT) token-merging ancestor, and EvoPrune [ 7] is a late-breaking MLLM example of pruning inside the visual encoder itself. Those methods are adjacent but not identical to our setting."},{"citing_arxiv_id":"2605.01048","ref_index":87,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Compared to What? Baselines and Metrics for Counterfactual Prompting","primary_cat":"cs.CL","submitted_at":"2026-05-01T19:23:33+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Counterfactual prompting effects on LLMs are often indistinguishable from those caused by meaning-preserving paraphrases, causing most previously reported demographic sensitivities to disappear under proper statistical comparison.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.17087","ref_index":45,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"EvoComp: Learning Visual Token Compression for Multimodal Large Language Models via Semantic-Guided Evolutionary Labeling","primary_cat":"cs.CV","submitted_at":"2026-04-18T17:52:02+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"EvoComp compresses visual tokens in MLLMs by 3x while retaining 99.3% accuracy via an evolutionary labeling strategy that searches for low-loss, semantically diverse token subsets.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.12358","ref_index":36,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Why and When Visual Token Pruning Fails? A Study on Relevant Visual Information Shift in MLLMs Decoding","primary_cat":"cs.CV","submitted_at":"2026-04-14T06:48:31+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Visual token pruning in MLLMs fails on complex reasoning due to Relevant Visual Information Shift during decoding, but the DSTP framework fixes it training-free across models.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"training-based and training-free. Training-based methods (e.g., LLaVolta [8], Title Suppressed Due to Excessive Length 15 ZipR1 [7], VCM [31]) incorporate learnable modules to enforce sparsity but incur significant overhead. Consequently, training-free methods have gained prevalence, utilizing three primary strategies: (1) Vision-Encoder Based: LLaVA- PruMerge [36] and VisionZip [45] utilize encoder-side features like [CLS] attention or CLIP-generated scores. (2) LLM-Based: FastV [9], ZipVL [13], and PDrop [44] prune based on internal LLM attention, while DivPrune [1] treats pruning as a diversity maximization problem. (3) Cross-Modal: SparseVLM [53] and Spar- seVILA [17] leverage query-aware interactions to guide sparsification, with the"},{"citing_arxiv_id":"2604.11627","ref_index":68,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"POINTS-Long: Adaptive Dual-Mode Visual Reasoning in MLLMs","primary_cat":"cs.CV","submitted_at":"2026-04-13T15:38:22+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"POINTS-Long is a dual-mode multimodal large language model that uses dynamic visual token scaling to retain 97.7-99.7% accuracy on long-form tasks with 1/40 to 1/10th the tokens and supports streaming via detachable KV-cache.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"arXiv preprint arXiv:2405.08813, 2024. 1 [67] Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Worts- man, et al. Laion-5b: An open large-scale dataset for train- ing next generation image-text models.Advances in neural information processing systems, 35:25278-25294, 2022. 1 [68] Yuzhang Shang, Mu Cai, Bingxin Xu, Yong Jae Lee, and Yan Yan. Llava-prumerge: Adaptive token reduc- tion for efficient large multimodal models.arXiv preprint arXiv:2403.15388, 2024. 2 [69] Shuai Shao, Zeming Li, Tianyuan Zhang, Chao Peng, Gang Yu, Xiangyu Zhang, Jing Li, and Jian Sun. Objects365: A large-scale, high-quality dataset for object detection."},{"citing_arxiv_id":"2604.09442","ref_index":34,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"UIPress: Bringing Optical Token Compression to UI-to-Code Generation","primary_cat":"cs.CL","submitted_at":"2026-04-10T15:58:31+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"UIPress is the first encoder-side learned optical compression method for UI-to-Code that compresses visual tokens to 256, outperforming the uncompressed baseline by 7.5% CLIP score and the best inference-time baseline by 4.6% while delivering 9.1x TTFT speedup.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"token overhead is therefore critical for practical deployment. A growing body of work addresses visual token reduction for VLMs [10, 19, 21, 24, 34, 46, 50, 61, 62, 65]. Inference-time meth- ods compress tokens without retraining: FastV [5] zeros out low- attention features after early layers; VisionZip [61] selects dominant tokens by L2 norm and merges the rest; LLaVA-PruMerge [34] uses CLS attention for pruning. These methods achieve impressive re- sults on VQA benchmarks-FlashVLM [43] reports \"beyond-lossless\" performance at 78% pruning on TextVQA and MMBench [25]. For UI2CODE specifically, EfficientUICoder [64] combines UI element detection with token selection, achieving 55-60% compression on LLaVA-v1.6 [18, 19, 24, 58]."},{"citing_arxiv_id":"2604.16462","ref_index":4,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"From Inheritance to Saturation: Disentangling the Evolution of Visual Redundancy for Architecture-Aware MLLM Inference Acceleration","primary_cat":"cs.CV","submitted_at":"2026-04-08T07:42:16+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"HalfV disentangles MLLM visual redundancy into universal IVR and architecture-dependent SSR via a three-stage lifecycle, delivering 4.1x FLOPs speedup with 96.8% performance retention on Qwen25-VL.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2603.07080","ref_index":27,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"VLN-Cache: Enabling Token Caching for VLN Models with Visual/Semantic Dynamics Awareness","primary_cat":"cs.RO","submitted_at":"2026-03-07T07:30:35+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"VLN-Cache delivers up to 1.52x faster inference in VLN models by using view-aligned remapping for geometric consistency and a task-relevance saliency filter to manage semantic changes during navigation.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2511.06754","ref_index":20,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"SlotVLA: Towards Modeling of Object-Relation Representations in Robotic Manipulation","primary_cat":"cs.RO","submitted_at":"2025-11-10T06:33:44+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"SlotVLA uses slot attention to model object-relation representations for multitask robotic manipulation, reducing visual tokens while achieving competitive generalization on the new LIBERO+ benchmark.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2508.06038","ref_index":14,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Fourier Compressor: Frequency-Domain Visual Token Compression for Vision-Language Models","primary_cat":"cs.CV","submitted_at":"2025-08-08T05:49:42+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Fourier Compressor uses FFT to remove frequency-domain redundancy from visual tokens in VLMs, retaining over 96% accuracy with up to 83.8% FLOP reduction.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2503.14075","ref_index":43,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Growing a Multi-head Twig via Distillation and Reinforcement Learning to Accelerate Large Vision-Language Models","primary_cat":"cs.CV","submitted_at":"2025-03-18T09:52:45+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"TwigVLM adds a twig module to VLMs for twig-guided token pruning and self-speculative decoding, retaining 96% performance after pruning 88.9% visual tokens and delivering 154% speedup on long responses for LLaVA-1.5-7B.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2410.17247","ref_index":42,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"PyramidDrop: Accelerating Your Large Vision-Language Models via Pyramid Visual Redundancy Reduction","primary_cat":"cs.CV","submitted_at":"2024-10-22T17:59:53+00:00","verdict":"ACCEPT","verdict_confidence":"HIGH","novelty_score":7.0,"formal_verification":"none","one_line_summary":"PyramidDrop accelerates LVLMs by staged, similarity-based dropping of visual tokens that become redundant in deeper layers, delivering 40% faster training and 55% lower inference cost with comparable accuracy.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2410.04417","ref_index":83,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference","primary_cat":"cs.CV","submitted_at":"2024-10-06T09:18:04+00:00","verdict":"ACCEPT","verdict_confidence":"MODERATE","novelty_score":6.0,"formal_verification":"none","one_line_summary":"SparseVLM uses text-guided attention to prune and recycle visual tokens in VLMs, delivering 54% FLOPs reduction and 37% lower latency with 97% accuracy retention on LLaVA.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null}],"limit":50,"offset":0}