Pith. sign in

REVIEW 2 major objections 4 minor 37 references

TORINO: Token Reduction via Interpretable Concept Overlap in Vision-Language Models

T0 review · 2 major / 4 minor · reviewed 2026-07-11 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Visual tokens can be reduced by grouping patches that share the same sparse-autoencoder concepts, adapting compression to each image without fine-tuning the VLM.

desk verdict Solid training-free token reduction that uses SAE concept overlap for content-adaptive prune/merge; moderate-regime gains look real, but rest on a CLS-trained dictionary whose patch transfer is only partially checked. read the letter →

arxiv 2607.04593 v1 pith:GZRUNGX7 submitted 2026-07-06 cs.CV cs.AI

Riccardo Renzulli , Gabriele Spadaro , Shruthi Gowda , Alaa Eddine Mazouz , Van-Tam Nguyen This is my paper · ORCID
classification cs.CVcs.AI
keywords tokenreductionvision-languagemodelssparseautoencodersconceptoverlapvisualtokensadaptivecompressionplug-and-play
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Vision-language models are slowed by the hundreds of visual patch tokens they feed into the language model. Prior reduction methods score tokens by attention or raw similarity and never name what each patch means. TORINO projects every patch into a frozen sparse autoencoder concept space, links patches that share active latents, and then prunes or merges inside each concept group. The number of groups is set by the image itself, so plain scenes collapse more aggressively than busy ones. Across nine multimodal benchmarks the method keeps nearly full accuracy at moderate compression rates while remaining a pure inference-time plug-in.

What carries the argument

Concept overlap: two tokens are joined if they share at least δ of their top-k active SAE features; connected components of that graph become the groups that are then pruned or merged to a single primary token.

What would settle it

Run the identical top-k overlap pipeline on raw patch embeddings instead of SAE latents at a matched average budget of roughly 47 tokens; if the reported multi-point accuracy gap vanishes, the SAE is not supplying the claimed semantic basis.

Watch

Extended reading notes

Core claim

Measuring agreement among active SAE latents produces semantically coherent groups of visual tokens; reducing each group to one representative (by peak activation or log-rescaled merge) removes redundancy more accurately than attention- or similarity-based baselines in the moderate-reduction regime, and the retained token count emerges automatically from image content rather than a fixed budget.

Load-bearing premise

The method assumes that a sparse autoencoder trained only on class-token activations still supplies faithful monosemantic concepts for grouping the encoder’s spatial patch tokens.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. TORINO is a training-free, plug-and-play visual token reduction method for VLMs that projects patch embeddings (CLIP ViT-L/14@336, block 22) through a frozen Matryoshka BatchTopK SAE into a sparse concept space, builds a token graph by δ-overlap of top-k active latents, and reduces each connected component by either peak-activation pruning (TORINO-P) or log-size-rescaled merging (TORINO-M). A fixed-budget variant adds truncation of the largest groups and padding by peak activation. On LLaVA-1.5-7B/13B across nine VLMEvalKit benchmarks the method reports content-adaptive retention (~217 / ~129 / ~47 tokens) with relative macro-scores of 98.7% (TORINO-M) and 97.1% (TORINO-P) at moderate compression, outperforming Random, FOLDER, PruneSID and PruMerge in that regime; an SAE-vs-raw ablation (−5.2 pp), expansion-factor Pareto, latency isolation and qualitative failure modes are also provided.

Significance. If the moderate-regime gains hold under broader SAE training, the work supplies a concrete instance of pragmatic interpretability: monosemantic SAE latents are used not only for analysis but as the decision basis for inference-time compression. The content-adaptive budget (G_dyn emerging from concept groups without an explicit complexity score), the unified prune/merge framework, the matched dynamic-vs-fixed comparison, and the public-dictionary reproducibility are genuine strengths. The contribution is incremental rather than foundational—limited to LLaVA + one CLIP SAE—but it is a clean, well-documented step that other VLM token-reduction lines can build on.

major comments (2)
  1. [Sec. 3.1, 4.1, 4.5; Tables 4, 10] Sec. 3.1, 4.1 and 4.5 (and Tables 4, 10): the central claim that concept-overlap grouping is superior rests on a Matryoshka BatchTopK SAE trained only on CLS activations at block 22. The paper’s own spatial-SAE ablation (two random patches) shows large, compression-widening drops (−2.7 / −7.7 / −19.6 pp for TORINO-M at B=192/128/64). Late-layer mixing is offered as justification, yet no monosemanticity scores, concept-coverage statistics or reconstruction metrics are reported for patch activations under the CLS SAE. Without that validation the 98.7 % / 97.1 % relative scores could be dictionary-specific rather than a general property of SAE concept overlap; either a full-patch SAE or direct patch-level monosemanticity numbers are needed to underwrite the claim.
  2. [Table 1, Sec. 4.2, Abstract] Table 1 and Sec. 4.2: at the most aggressive tier (~47 tokens, ↓92 %) PruneSID leads (92.8 % vs TORINO-P 91.1 %). The abstract and introduction frame TORINO as achieving “favorable efficiency-accuracy trade-offs” without regime qualification. The moderate-regime advantage is real and should be stated as the primary result; the extreme-compression ranking should be reported with equal prominence so that the contribution is not overstated.
minor comments (4)
  1. [Eq. (10), Sec. 3.3] Eq. (10): the log-size rescale 1+log|G_g| is taken from ToMe without an ablation against plain averaging or other sublinear factors; a one-line comparison would strengthen the design choice.
  2. [Fig. 3, Table 1] Fig. 3 and latency column of Table 1: no error bars or multi-seed statistics are given for either the per-image G_dyn distributions or the isolated reduction-module timings; even a short note on variance would help.
  3. [Sec. 4.1] Sec. 4.1: the three (k,δ) pairs are presented as inducing distinct average budgets, yet no sensitivity sweep around those discrete points is shown; a brief grid would clarify robustness.
  4. [Throughout] Typographical inconsistencies appear throughout (LLaV A / LLaVA, fac ¸ade, T´el´ecom); a single pass of copy-editing would remove them.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: empirical token-reduction method evaluated on external benchmarks with frozen public SAE; no claim reduces by construction to its inputs.

full rationale

TORINO is a training-free inference-time pipeline that encodes frozen CLIP patch tokens with a publicly released Matryoshka BatchTopK SAE (Pach et al., independent authors), groups by top-k concept overlap (k, δ), and prunes or merges within connected components. The claimed efficiency-accuracy trade-offs are measured on nine external VLMEvalKit benchmarks against independent baselines (Random, FOLDER, PruneSID, PruMerge) using a relative macro-average against a reproduced full-token LLaVA baseline. Hyperparameters (k, δ, ε, B) are swept and ablated; none is fitted to the target relative scores. The SAE dictionary is frozen and not re-trained on the evaluation data. No equation equates a reported accuracy gain to a quantity defined by the same data used to declare success; the content-adaptive G_dyn length emerges from the grouping graph rather than being reverse-engineered from benchmark labels. The CLS-to-patch transfer assumption is a validity risk (addressed by the paper’s own spatial-SAE ablation) but is not a circular step under the listed patterns. The derivation chain is therefore self-contained and non-circular.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The method rests on a frozen public SAE, standard VLM architecture assumptions, and a small set of hand-chosen grouping hyperparameters. No new physical entities are postulated. Free parameters are the discrete grouping knobs and the choice of SAE expansion/layer; axioms are domain assumptions about SAE monosemanticity transferring from CLS training to patch grouping.

free parameters (4)
  • grouping pair (k, δ)
    Hand-chosen discrete settings {(1,1),(2,2),(3,3)} that control group granularity and average compression; not learned from the evaluation targets but selected to induce three reduction tiers.
  • SAE expansion factor ε and active features K
    Primary results use ε=64, K=20 from the public Matryoshka BatchTopK SAE; ablations sweep ε but the reported best points depend on this choice.
  • extraction layer ℓ
    Fixed to CLIP ViT-L/14 block 22 (post-MLP residual) to match the SAE training layer and LLaVA projection input.
  • fixed budget B and log-size merge rescale
    B ∈ {192,128,64} for fair comparison; merge uses 1+log|G| factor following ToMe-style practice, a design choice affecting TORINO-M.
assumptions (4)
  • domain assumption Matryoshka BatchTopK SAE latents trained on CLS activations are sufficiently monosemantic and transferable to spatial patch tokens for semantic grouping.
    Invoked throughout Sec. 3.1–3.2 and 4.1; spatial-SAE ablation (Sec. 4.5, Table 4) shows the assumption is imperfect at high compression.
  • ad hoc to paper Tokens sharing ≥δ of their top-k active SAE concepts are semantically redundant enough that one representative (or average) preserves LLM-usable information.
    Core grouping rule Eq. (7)–(8); justified empirically, not derived from a formal information bound.
  • ad hoc to paper Peak activation a(i)=∥z_i∥_∞ is a valid importance score for selecting representatives and padding tokens.
    Eq. (6) and TORINO-P / padding rules; alternative importance signals are not compared inside the SAE space.
  • domain assumption Standard transformer attention quadratic cost and frozen LLaVA CLIP+LLM pipeline make visual token count the dominant inference bottleneck.
    Motivation in Sec. 1; standard in the token-reduction literature.
invented entities (1)
  • concept overlap graph H with δ-overlap edges on top-k SAE latents
    purpose: Defines content-adaptive groups whose connected components become the reduction units.
    New operational construct for this paper; independent evidence is only the empirical accuracy tables, not an external measurement of concept identity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TORINO: Token Reduction via Interpretable Concept Overlap in Vision-Language Models." pith.science (2026). https://pith.science/paper/GZRUNGX7

@misc{pith2026260704593,
  author       = {Pith},
  title        = {Pith review of: TORINO: Token Reduction via Interpretable Concept Overlap in Vision-Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GZRUNGX7}},
  note         = {Machine review of arXiv:2607.04593}
}
read the original abstract

Vision-Language Models (VLMs) have demonstrated impressive capabilities across different tasks, but their computational cost is dominated by the large number of visual tokens fed to the language model. Existing token reduction methods rely on attention-based scores or pairwise similarity, without an explicit semantic representation of each token. We introduce TORINO (TOken Reduction via Interpretable coNcept Overlap), a plug-and-play framework for adaptive visual token reduction in VLMs that requires no fine-tuning of the underlying model. TORINO leverages Sparse Autoencoders (SAEs) to project visual tokens into an interpretable latent space where token relationships can be analyzed through shared concept activations. Specifically, we define concept overlap as the degree of agreement between active SAE latents and use it to group tokens that share semantic content. Reduction within each group is then performed by either pruning or merging, providing a unified framework that preserves semantically important visual information while removing redundancy. Unlike fixed-budget approaches, TORINO dynamically adapts the reduction rate to input complexity, allowing different images to retain different numbers of tokens. Experiments across multiple vision-language benchmarks show that TORINO achieves favorable efficiency-accuracy trade-offs, reducing the number of visual tokens with minimal performance loss.

Figures

Figures reproduced from arXiv: 2607.04593 by the authors.

Figure 1
Figure 1. TORINO reduces visual tokens dynamically by group [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. TORINO pipeline. Visual patch tokens are extracted at layer [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Per-image distribution of removed tokens for TORINO in the dynamic setting, on the Matryoshka BatchTopK ε = 64 SAE, across the nine LLaVA-1.5-7B benchmarks. Each box summarises the distribution of per-image removal percentages within a benchmark: edges mark the 10th and 90th percentiles (covering 80% of images), the central line is the median, the dia￾mond is the mean, and whiskers are the minimum and maximum. boxes… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: shows visualization on four MME examples the base model LLaVA-1.5-7B without any reduction method and TORINO-P (ε = 64) across three (k, δ) configurations. In columns 1 and 2, the model answers correctly at all com￾pression tiers: the retained patches concentrate on th…
Figure 5
Figure 5. Figure 5: Pareto trade-off across SAE expansion factors ε. Each marker is one TORINO run; each curve connects three grouping configurations (k, δ) ∈ {(1, 1),(2, 2),(3, 3)}. is consistent with the monosemanticity argument: highly specialised features produce groups whose members …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 2 linked inside Pith

  1. [1]

    Flamingo: a visual language model for few-shot learning

    Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Men- sch, Katherine Millican, Malcolm Reynolds, Roman Ring, Eliza Rutherford, Serkan Cabi, Tengda Han, Zhitao Gong, Sina Samangooei, Marianne Monteiro, Jacob L Menick, Sebastian Borgeaud, Andy Brock, Aida Nematzadeh, Sa- hand Sharifzadeh, Mikoł aj B...

    2022

  2. [2]

    Divprune: Diversity-based visual token pruning for large multimodal models

    Saeed Ranjbar Alvar, Gursimran Singh, Mohammad Akbari, and Yong Zhang. Divprune: Diversity-based visual token pruning for large multimodal models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9392–9401, 2025. 2, 3

    2025

  3. [3]

    Hired: Attention-guided token dropping for efficient inference of high-resolution vision-language models

    Kazi Hasan Ibn Arif, JinYi Yoon, Dimitrios S Nikolopou- los, Hans Vandierendonck, Deepu John, and Bo Ji. Hired: Attention-guided token dropping for efficient inference of high-resolution vision-language models. InProceedings of the AAAI Conference on Artificial Intelligence, pages 1773– 1781, 2025. 2, 3

    2025

  4. [4]

    Qwen-vl: A versatile vision-language model for un- derstanding, localization, text reading, and beyond.arXiv preprint arXiv:2308.12966, 2023

    Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. Qwen-vl: A versatile vision-language model for un- derstanding, localization, text reading, and beyond.arXiv preprint arXiv:2308.12966, 2023. 1

    arXiv 2023

  5. [5]

    Mechanistic inter- pretability for AI safety - a review.Transactions on Machine Learning Research, 2024

    Leonard Bereska and Stratis Gavves. Mechanistic inter- pretability for AI safety - a review.Transactions on Machine Learning Research, 2024. Survey Certification, Expert Cer- tification. 2

    2024

  6. [6]

    Token merging: Your vit but faster

    Daniel Bolya, Cheng-Yang Fu, Xiaoliang Dai, Peizhao Zhang, Christoph Feichtenhofer, and Judy Hoffman. Token merging: Your vit but faster. InThe Eleventh International Conference on Learning Representations, 2023. 2, 3, 4

    2023

  7. [7]

    Towards Monosemanticity: Decomposing Language Models With Dictionary Learning,

    Trenton Bricken, Adly Templeton, Joshua Batson, Brian Chen, Adam Jermyn, Tom Conerly, Nicholas L Turner, Cem Anil, Carson Denison, Amanda Askell, Robert Lasenby, Yi- fan Wu, Shauna Kravec, Nicholas Schiefer, Tim Maxwell, Nicholas Joseph, Alex Tamkin, Karina Nguyen, Brayden McLean, Josiah E Burke, Tristan Hume, Shan Carter, Tom Henighan, and Chris Olah. To...

  8. [8]

    Batchtopk sparse autoencoders

    Bart Bussmann, Patrick Leask, and Neel Nanda. Batchtopk sparse autoencoders. InNeurIPS 2024 Workshop on Scien- tific Methods for Understanding Deep Learning, 2024. 2

    2024

Show all 37 references
  1. [9]

    Learning multi-level features with matryoshka sparse autoencoders

    Bart Bussmann, Noa Nabeshima, Adam Karvonen, and Neel Nanda. Learning multi-level features with matryoshka sparse autoencoders. InForty-second International Conference on Machine Learning, 2025. 2

    2025

  2. [10]

    SAEmne- sia: Erasing concepts in diffusion models with supervised sparse autoencoders

    Enrico Cassano, Riccardo Renzulli, Marco Nurisso, Mirko Zaffaroni, Alan Perotti, and Marco Grangetto. SAEmne- sia: Erasing concepts in diffusion models with supervised sparse autoencoders. InForty-third International Conference on Machine Learning, 2026. 3

    2026

  3. [11]

    SAeuron: Interpretable concept unlearning in diffusion models with sparse autoen- coders

    Bartosz Cywi ´nski and Kamil Deja. SAeuron: Interpretable concept unlearning in diffusion models with sparse autoen- coders. InForty-second International Conference on Ma- chine Learning, 2025. 3

    2025

  4. [12]

    InstructBLIP: Towards general-purpose vision-language models with instruction tuning

    Wenliang Dai, Junnan Li, Dongxu Li, Anthony Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, and Steven Hoi. InstructBLIP: Towards general-purpose vision-language models with instruction tuning. InThirty- seventh Conference on Neural Information Processing Sys- tems, 2023. 1

    2023

  5. [13]

    Vlmevalkit: An open-source toolkit for evaluating large multi-modality models

    Haodong Duan, Junming Yang, Yuxuan Qiao, Xinyu Fang, Lin Chen, Yuan Liu, Xiaoyi Dong, Yuhang Zang, Pan Zhang, Jiaqi Wang, et al. Vlmevalkit: An open-source toolkit for evaluating large multi-modality models. InProceedings of the 32nd ACM International Conference on Multimedia,...

    2024

  6. [14]

    Davis, Gaowen Liu, George K

    Nicholas John Eliopoulos, Purvish Jajal, James C. Davis, Gaowen Liu, George K. Thiravathukal, and Yung-Hsiang Lu. Pruning one more token is enough: Leveraging latency- workload non-linearities for vision transformers on the edge. In2025 IEEE/CVF Winter Conference on Applicatio...

    2025

  7. [15]

    Prune redundancy, pre- serve essence: Vision token compression in VLMs via syner- gistic importance-diversity

    Zhengyao Fang, Pengyuan Lyu, Chengquan Zhang, Guang- ming Lu, Jun Yu, and Wenjie Pei. Prune redundancy, pre- serve essence: Vision token compression in VLMs via syner- gistic importance-diversity. InThe Fourteenth International Conference on Learning Representations, 2026. 2, 3, 5

    2026

  8. [16]

    Mme: A compre- hensive evaluation benchmark for multimodal large language models.arXiv preprint arXiv:2306.13394, 2023

    Chaoyou Fu, Peixian Chen, Yunhang Shen, Yulei Qin, Mengdan Zhang, Xu Lin, Jinrui Yang, Xiawu Zheng, Ke Li, Xing Sun, Yunsheng Wu, and Rongrong Ji. Mme: A compre- hensive evaluation benchmark for multimodal large language models.arXiv preprint arXiv:2306.13394, 2023. 5 9

    2023 arXiv

  9. [17]

    Scaling and evaluating sparse autoencoders

    Leo Gao, Tom Dupre la Tour, Henk Tillman, Gabriel Goh, Rajan Troll, Alec Radford, Ilya Sutskever, Jan Leike, and Jeffrey Wu. Scaling and evaluating sparse autoencoders. In The Thirteenth International Conference on Learning Rep- resentations, 2025. 2

    2025

  10. [18]

    Vizwiz grand challenge: Answering visual questions from blind people

    Danna Gurari, Qing Li, Abigale J Stangl, Anhong Guo, Chi Lin, Kristen Grauman, Jiebo Luo, and Jeffrey P Bigham. Vizwiz grand challenge: Answering visual questions from blind people. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 3608–3617,

  11. [19]

    Gqa: A new dataset for real-world visual reasoning and compositional question answering

    Drew A Hudson and Christopher D Manning. Gqa: A new dataset for real-world visual reasoning and compositional question answering. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 6700–6709, 2019. 5

    2019

  12. [20]

    Evaluating object hallucination in large vision-language models

    Yifan Li, Yifan Du, Kun Zhou, Jinpeng Wang, Wayne Xin Zhao, and Ji-Rong Wen. Evaluating object hallucination in large vision-language models. InProceedings of the Confer- ence on Empirical Methods in Natural Language Processing, pages 292–305, 2023. 5

    2023

  13. [21]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. InAdvances in Neural Information Processing Systems, pages 34892–34916. Curran Associates, Inc., 2023. 1, 5

    2023

  14. [22]

    Mmbench: Is your multi-modal model an all-around player? InEuropean conference on computer vi- sion, pages 216–233

    Yuan Liu, Haodong Duan, Yuanhan Zhang, Bo Li, Songyang Zhang, Wangbo Zhao, Yike Yuan, Jiaqi Wang, Conghui He, Ziwei Liu, et al. Mmbench: Is your multi-modal model an all-around player? InEuropean conference on computer vi- sion, pages 216–233. Springer, 2024. 5

    2024

  15. [23]

    Learn to explain: Multimodal reasoning via thought chains for science question answering.Advances in Neural Information Processing Systems, 35:2507–2521,

    Pan Lu, Swaroop Mishra, Tanglin Xia, Liang Qiu, Kai-Wei Chang, Song-Chun Zhu, Oyvind Tafjord, Peter Clark, and Ashwin Kalyan. Learn to explain: Multimodal reasoning via thought chains for science question answering.Advances in Neural Information Processing Systems, 35:2507–2521,

  16. [24]

    ToMA: Token merge with attention for diffusion models

    Wenbo Lu, Shaoyi Zheng, Yuxuan Xia, and Shengjie Wang. ToMA: Token merge with attention for diffusion models. In Forty-second International Conference on Machine Learn- ing, 2025. 2

    2025

  17. [25]

    A pragmatic vision for inter- pretability

    Neel Nanda, Josh Engels, Arthur Conmy, Senthooran Ra- jamanoharan, Bilal Chughtai, Callum McDougall, J ´anos Kram´ar, and Lewis Smith. A pragmatic vision for inter- pretability. AI Alignment Forum, 2025. 2

    2025

  18. [26]

    Sparse autoencoders learn monosemantic features in vision-language models

    Mateusz Pach, Shyamgopal Karthik, Quentin Bouniot, Serge Belongie, and Zeynep Akata. Sparse autoencoders learn monosemantic features in vision-language models. InThe Thirty-ninth Annual Conference on Neural Information Pro- cessing Systems, 2026. 2, 3, 5

    2026

  19. [27]

    Save: Sparse autoencoder-driven visual information enhancement for mitigating object hallucination

    Sangha Park, Seungryong Yoo, Jisoo Mok, and Sungroh Yoon. Save: Sparse autoencoder-driven visual information enhancement for mitigating object hallucination. InProceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 7935–7944, 2026. 3

    2026

  20. [28]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. InInternational conference on machine learning, p...

    2021

  21. [29]

    Discover-then-name: Task-agnostic concept bottle- necks via automated concept discovery

    Sukrut Rao, Sweta Mahajan, Moritz B ¨ohle, and Bernt Schiele. Discover-then-name: Task-agnostic concept bottle- necks via automated concept discovery. InEuropean Con- ference on Computer Vision, pages 444–461. Springer, 2024. 3

    2024

  22. [30]

    Llava-prumerge: Adaptive token reduction for efficient large multimodal models

    Yuzhang Shang, Mu Cai, Bingxin Xu, Yong Jae Lee, and Yan Yan. Llava-prumerge: Adaptive token reduction for efficient large multimodal models. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 22857–22867, 2025. 2, 3, 5

    2025

  23. [31]

    VL-SAE: Interpreting and enhancing vision-language alignment with a unified concept set

    Shufan Shen, Junshu Sun, Qingming Huang, and Shuhui Wang. VL-SAE: Interpreting and enhancing vision-language alignment with a unified concept set. InThe Thirty-ninth An- nual Conference on Neural Information Processing Systems,

  24. [32]

    A survey on sparse autoencoders: Interpreting the internal mechanisms of large language models

    Dong Shu, Xuansheng Wu, Haiyan Zhao, Daking Rai, Ziyu Yao, Ninghao Liu, and Mengnan Du. A survey on sparse autoencoders: Interpreting the internal mechanisms of large language models. InFindings of the Association for Com- putational Linguistics: EMNLP 2025, pages 1690–1712, S...

    2025

  25. [33]

    Towards vqa models that can read

    Amanpreet Singh, Vivek Natarajan, Meet Shah, Yu Jiang, Xinlei Chen, Dhruv Batra, Devi Parikh, and Marcus Rohrbach. Towards vqa models that can read. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8317–8326, 2019. 5

    2019

  26. [34]

    Folder: Accelerating multi-modal large language models with en- hanced performance

    Haicheng Wang, Zhemeng Yu, Gabriele Spadaro, Chen Ju, Victor Qu ´etu, Shuai Xiao, and Enzo Tartaglione. Folder: Accelerating multi-modal large language models with en- hanced performance. InProceedings of the IEEE/CVF In- ternational Conference on Computer Vision (ICCV), pages...

    2025

  27. [35]

    Gtp-vit: Efficient vision trans- formers via graph-based token propagation

    Xuwei Xu, Sen Wang, Yudong Chen, Yanping Zheng, Zhewei Wei, and Jiajun Liu. Gtp-vit: Efficient vision trans- formers via graph-based token propagation. InProceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 86–95, 2024. 3

    2024

  28. [36]

    Visionzip: Longer is better but not necessary in vision language models

    Senqiao Yang, Yukang Chen, Zhuotao Tian, Chengyao Wang, Jingyao Li, Bei Yu, and Jiaya Jia. Visionzip: Longer is better but not necessary in vision language models. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 19792–19802, 2025. 3

    2025

  29. [37]

    Mm-vet: Evaluating large multimodal models for integrated capabilities

    Weihao Yu, Zhengyuan Yang, Linjie Li, Jianfeng Wang, Kevin Lin, Zicheng Liu, Xinchao Wang, and Lijuan Wang. Mm-vet: Evaluating large multimodal models for integrated capabilities. InInternational conference on machine learn- ing. PMLR, 2024. 5 10 A. Pseudocode for TORINO We pr...

    2024

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.