Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Token Sequence Compression for Efficient Multimodal Computing

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read This paper argues that averaging visual-token clusters, not attention-based selection, is the most effective training-free way to compress visual tokens in vision-language models.

desk verdict A small empirical paper with a real negative result—saliency-agnostic token compression matches attention-based selection—but the SoTA overclaim and unmeasured compute overhead need fixing. read the letter →

arxiv 2504.17892 v1 pith:N3PFCFKY submitted 2025-04-24 cs.CV cs.AIcs.CL

Yasmine Omri , Parth Shroff , Thierry Tambe This is my paper · ORCID
classification cs.CVcs.AIcs.CL
keywords visualtokencompressionclusteraggregationattentionsaliencytraining-freeinferencevision-languagemodelsK-means++selectionmultimodalefficiency
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

The paper is trying to establish that the most effective way to compress images for vision-language models is not to decide which visual tokens matter, but to group similar tokens and average each group into one token. Its proposed method, Cluster & Aggregate, clusters visual token embeddings with K-means++ after the vision encoder and replaces each cluster with its mean embedding, requiring no fine-tuning and no importance scores. On LLaVA-1.5-7B/13B and VILA-8B, the paper reports that this method matches or outperforms earlier token-selection and merging algorithms, including attention-based ones, at the same 11% token retention. It also reports that attention-based saliency rankings are volatile across layers and barely change with the text prompt, which challenges the common assumption that attention reveals which visual tokens are semantically important. If these results hold, efficient multimodal inference would not need learned or attention-based importance scoring at all.

What carries the argument

The central object is the Cluster & Aggregate pipeline: after the vision encoder and the projector, visual token embeddings are grouped with K-means++ on embedding similarity, and all tokens in a cluster are replaced by one averaged embedding; the resulting aggregate embeddings are concatenated, in random order, and fed to the LLM alongside text tokens. The method works in the shared pre-LLM embedding space, requires no fine-tuning, and has no saliency score or per-layer pruning decisions. The paper contrasts this with its Algorithm 1 saliency score, a cross-modal attention measure that ranks visual tokens by the attention they receive from text tokens. That contrast carries the argument: if a mean of a cluster preserves enough information to match or beat attention-selected tokens, then attention-based importance is not doing useful work for compression.

What would settle it

Run LLaVA-1.5-7B end to end on the same GPU with the full 576-token baseline, Cluster & Aggregate at 64 tokens, and an attention-based method at 64 tokens, measuring wall-clock latency and energy including the K-means++ step; if Cluster & Aggregate is not faster than the baseline, or is slower than the attention-based method at matched accuracy, the paper's efficiency claim fails. A complementary test: on a fine-grained localization benchmark, if attention-selected tokens clearly beat cluster-averaged tokens at equal retention, the paper's conclusion that saliency is not predictive would fail.

Watch

Extended reading notes

Core claim

The paper's central claim, stated in its own terms, is that simple cluster-level token aggregation outperforms prior finetuning-free state-of-the-art methods for visual token selection and merging while being computationally simpler. In the paper's experiments, Cluster & Aggregate retains 64 of 576 visual tokens from LLaVA-1.5-7B and achieves higher average accuracy on eight visual question-answering benchmarks than the paper's saliency-based variants and than finetuning-free versions of VisionZip; the pattern persists at 128 and 192 tokens and on LLaVA-1.5-13B and VILA-8B. Alongside this, the paper documents two puzzling trends in attention-based saliency: the highlighted regions often do not coincide with intuitively informative image regions, and the saliency ranking of visual tokens changes very little when the text prompt changes. The authors take these observations as evidence that the visual encoding pipeline is highly redundant and that attention-based importance metrics are not reliable guides for selecting which visual tokens to keep.

Load-bearing premise

The paper's efficiency claim assumes that cutting the visual token count is a valid proxy for real computation savings; since it reports only LLM-level estimates and never measures the added cost of K-means++ clustering and aggregation, high clustering overhead would break the link between the accuracy-versus-retention tradeoff and actual speedup.

Editorial extensions

If this is right

  • The paper's experiments imply that attention-based importance scoring is not necessary: saliency-agnostic aggregation matches or exceeds it at equal token retention.
  • Performance depends more on how many tokens remain than on how they are chosen; the clustering variants show no statistically significant differences once retention is controlled.
  • Because Cluster & Aggregate is training-free and runs after the projector, it can be applied to any existing LLaVA-style autoregressive VLM without fine-tuning.
  • In some benchmarks, such as VizWiz, compressed token sequences yield higher accuracy than the full 576-token baseline, indicating redundancy and noise in the uncompressed encoding.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Not in the paper, but a direct corollary of its retention-rate finding: replacing K-means++ with a cheaper clustering routine should preserve most of the accuracy, since the paper's own results show performance tracks the number of retained tokens more than the clustering method.
  • If the reported prompt-invariance of saliency generalizes, prompt-conditioned compression could be simplified to content-based compression computed once per image, with little accuracy loss.
  • A direct wall-clock and energy comparison that includes clustering overhead would settle whether the accuracy-versus-retention tradeoff becomes a real speedup; the paper explicitly leaves that measurement for future work.
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

4 major / 6 minor

Summary. The paper studies compression of visual token sequences in autoregressive vision-language models. It proposes and evaluates several training-free methods: a cross-modal attention saliency score, cluster-based variants of that score, and a purely importance-agnostic "Cluster & Aggregate" method that clusters visual embeddings with K-means++ and averages each cluster into a single token. Experiments are run on LLaVA-1.5-7B/13B and VILA-8B across seven to eight VQA/hallucination benchmarks, with comparisons to FastV, SparseVLM, VisionZip, ToMe, and random/spatial sampling. The authors conclude that simple cluster aggregation outperforms all previous finetuning-free selection/merging methods and that attention-based saliency is volatile and largely prompt-insensitive.

Significance. If the empirical patterns hold, the paper makes a useful contribution by challenging the assumption that attention-based importance scoring is necessary for visual token selection: it shows that random/spatial sampling and simple aggregation are competitive, and it offers a concrete, falsifiable observation about saliency being prompt-insensitive. The comparative benchmark across three model families is a strength, and the qualitative heatmap analysis helps the reader interpret the surprising trends. However, the strongest claims are not backed by the reported data: VisionZip wins on several benchmarks in Tables 3 and 4, and no end-to-end compute or clustering overhead is measured, so the "computationally simpler" claim is not established. The paper provides no code, seeds, or significance tests, so reproducibility and statistical support are limited. The observations are nevertheless valuable as a stress-test of current token-reduction assumptions.

major comments (4)
  1. [Abstract; §3.2 (Fig. 6); Conclusion; Tables 3–4] The claim that Cluster & Aggregate "outperforms all previous finetuning-free SoTA" is not supported by the data in Tables 3 and 4. For LLaVA-1.5-7B at 128 retained tokens, VisionZip is higher on TextVQA (56.8 vs 55.25), MME (1761.7 vs 1729.42), and MMVet (32.6 vs 30.9); at 192 tokens VisionZip is higher on TextVQA, MME, and GQA (1782.6 vs 1744.44 on MME, 59.3 vs 59.16 on GQA). Similar mixes occur for LLaVA-1.5-13B, where VisionZip wins SQA, TextVQA, MMBench, and MMVet at 64 tokens. The paper needs a more precise statement, such as reporting per-benchmark wins and an appropriate aggregate statistic, rather than "outperforms all."
  2. [§2 (Fig. 2); Conclusion] The efficiency claim ("computationally simpler") is not established because no end-to-end compute is measured. Fig. 2 shows only LLM-level estimates from LLMViewer, which exclude the early LLM layers where FastV and SparseVLM retain all tokens, and no measurement of K-means++/aggregation overhead is reported. The Conclusion explicitly states "We also aim to quantify the computational overhead and net gains," acknowledging this gap. Without latency, FLOPs, or memory measurements, token-count retention is not a valid proxy for compute across methods that prune at different stages.
  3. [§3.1 (Table 2)] The statement that "our empirical evidence indicates no statistically significant performance differential among variants when controlling for retention percentage" is unsupported: no error bars, number of seeds, or significance tests are reported anywhere in the paper. Since Table 2 shows differences of up to several points on TextVQA (52.97 vs 48.61 between Variant 3 with k=40 and Variant 1) and large drops on VizWiz (53.99 vs 49.58), either the statistical evidence should be provided or the sentence should be removed or explicitly weakened to a qualitative observation.
  4. [§3.2 (Table 5)] The VILA-8B results in Table 5 do not support the sentence that Cluster & Aggregate compression variants "achieve superior performance compared to these baselines." Relative to the uncompressed VILA-8B baseline, Cluster & Aggregate loses substantially on SQA (77.94 vs 81.95), TextVQA (56.38 vs 68.54), MME (1566.2 vs 1744.41), MMBench (55.84 vs 66.83), and MMVet (32.2 vs 40.5), while improving only on VizWiz (64.14 vs 63.24) and GQA (44.02 vs 30.41). The sentence should clarify that the comparison is against the random/spatial compression baselines, not against the uncompressed model.
minor comments (6)
  1. [Table 5] The header row lists nine metric columns (including VizWiz) but each data row contains only eight values; the baseline row appears to be missing a VizWiz value or the column positions are shifted.
  2. [§3.2, Fig. 6] The text says results are reported for "64 retained tokens" in Fig. 6, while the caption says 11% of visual tokens; since 11% of 576 is 63.4, please state the exact token count used and keep the two statements consistent.
  3. [§3.1] The word "intutions" should be "intuitions"; also, "variants" vs "variant's" are used inconsistently in the same paragraph.
  4. [Algorithm 1] The notation "kvi·Qtext" is unclear because a single key vector is multiplied by a text query matrix; please specify the shapes and the softmax axis explicitly to match the described computation.
  5. [§3.2] The claim that "random insertion is a simpler method and proved to yield comparable results" is not supported by any table or figure comparing random-order and position-ordered aggregation; either add the comparison or mark this as an unreported observation.
  6. [Fig. 5b] The caption says "layer-wise visualization of saliency heatmaps" but does not indicate which layers are shown or how heatmaps are normalized; please expand the caption so the reader can interpret the volatility claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: benchmark study with externally evaluated methods; no fitted parameter is renamed as a prediction and no equation reduces to its own input.

full rationale

This paper is an empirical benchmarking and qualitative study, not a derivation. It defines several token-selection and aggregation schemes (Basic Saliency, Cluster & Saliency variants, Random Sampling, Spatial Sampling, Cluster & Aggregate), evaluates them on fixed public benchmarks against prior methods, and reports accuracy at user-chosen retention levels and cluster counts. There is no fitted parameter that is later relabeled as a prediction, no equation is defined in terms of the quantity it claims to derive, and no load-bearing claim rests on a self-citation: the reference list contains no prior work by the authors, and the central comparison is made against externally reported or reproduced baselines. The paper's strongest claim, that cluster-level aggregation is 'computationally simpler', is not established because the work never measures end-to-end latency, FLOPs, or the overhead of K-means++ clustering; the conclusion explicitly concedes 'We also aim to quantify the computational overhead and net gains.' That concession is a genuine limitation and a threat to the efficiency claim, but it is not circularity: under-measured overhead and token-count-as-proxy reasoning are empirical validity concerns, not an input-output equivalence. Similarly, the observation that different methods prune at different layers means retained-token count is not a perfect compute-equivalence metric across methods, but this is a comparison-design issue rather than a self-referential reduction. No step of the paper's argument derives X from a definition of X or from the very data it claims to predict, so the circularity score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on empirical benchmark measurements and a few hand-chosen parameters, rather than new theoretical entities. The most important unstated premises are the semantic validity of embedding-space clusters and the use of token count as a proxy for real compute cost.

free parameters (3)
  • Number of clusters k = k=20 and k=40 reported for cluster variants; not reported for the main Cluster & Aggregate result
    The compression rate and aggregation granularity depend on k, but the paper does not tune it systematically or report the value used in the headline Figure 6 result.
  • Retention percentage x = 11%, and fixed token counts 64, 128, 192
    The retention level is adopted from prior work, and the paper itself notes performance correlates strongly with the proportion of retained tokens, so this choice heavily influences the conclusions.
  • Lambda scale lambda = 1 and 1.5
    User-input scale for the Dynamic cluster variant; not central to the main Cluster & Aggregate method, but affects the comparison table.
assumptions (5)
  • domain assumption K-means++ clusters in the shared embedding space correspond to semantically meaningful image regions.
    The Cluster & Aggregate method averages tokens within each cluster; if clusters are arbitrary, the aggregation could destroy information. This assumption is invoked in Section 3.2.
  • domain assumption Averaged token embeddings plus random ordering preserve enough information for downstream VLM benchmarks.
    The method's core operation is replacing many tokens with one mean vector; the paper provides empirical support but no theoretical or diagnostic guarantee. Section 3.2 states random insertion yields comparable results.
  • domain assumption Benchmark accuracy at a fixed token retention rate is a valid proxy for the efficiency-accuracy tradeoff.
    The paper motivates the work by compute savings but never measures actual latency, FLOPs, memory, or the overhead of clustering; Figure 2 uses an external estimator and the conclusion defers overhead quantification.
  • domain assumption First-layer cross-modal attention is representative for saliency analysis.
    Algorithm 1 computes saliency from the first layer's attention matrices, while Trend 1 in Section 3.1 notes that saliency rankings are volatile and do not converge across layers, which undermines this assumption.
  • standard math K-means++ and softmax attention are standard and correct as used.
    The paper relies on these standard algorithms without proof; they are not part of the claimed contribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Token Sequence Compression for Efficient Multimodal Computing." pith.science (2026). https://pith.science/paper/N3PFCFKY

@misc{pith2026250417892,
  author       = {Pith},
  title        = {Pith review of: Token Sequence Compression for Efficient Multimodal Computing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N3PFCFKY}},
  note         = {Machine review of arXiv:2504.17892}
}
read the original abstract

The exponential growth of Large Multimodal Models (LMMs) has driven advancements in cross-modal reasoning but at significant computational costs. In this work, we focus on visual language models. We highlight the redundancy and inefficiency in current vision encoders, and seek to construct an adaptive compression method for multimodal data. In this work, we characterize a panoply of visual token selection and merging approaches through both benchmarking and qualitative analysis. In particular, we demonstrate that simple cluster-level token aggregation outperforms prior state-of-the-art works in token selection and merging, including merging at the vision encoder level and attention-based approaches. We underline the redundancy in current vision encoders, and shed light on several puzzling trends regarding principles of visual token selection through cross-modal attention visualizations. This work is a first effort towards more effective encoding and processing of high-dimensional data, and paves the way for more scalable and sustainable multimodal systems.

Figures

Figures reproduced from arXiv: 2504.17892 by the authors.

Figure 1
Figure 1. Architecture of auto-regressive visual language models. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Pipeline for dynamic training-free visual token sequence [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 2
Figure 2. Computational savings estimates at the LLM level from [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Basic cross modality saliency visualization [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]
Figure 5
Figure 5. Figure 5: (a) Performance of LLaVA1.5-7B with 11% of the [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Comparison of importance-agnostic selection methods [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. p-MoD: Building Mixture-of-Depths MLLMs via Progressive Ratio Decay

    cs.CV 2024-12 conditional novelty 6.0 of 10

    A mixture-of-depths strategy with a progressively decaying token retention ratio lets MLLMs keep full benchmark accuracy while cutting inference compute and memory by nearly half.

Reference graph

Works this paper leans on

25 extracted references · 18 canonical work pages · cited by 1 Pith paper

  1. [1]

    A multimodal architecture for ai agents, 2023

    Adept AI. A multimodal architecture for ai agents, 2023. https://www.adept.ai/blog/fuyu-8b. 1, 2

    work page 2023

  2. [2]

    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. In Proceedings of the Inter- national Conference on Learning Representations (ICLR) ,

    work page

  3. [3]

    An image is worth 1/2 tokens after layer 2: Plug-and-play inference acceleration for large vision-language models

    Liang Chen, Haozhe Zhao, Tianyu Liu, Shuai Bai, Junyang Lin, Chang Zhou, and Baobao Chang. An image is worth 1/2 tokens after layer 2: Plug-and-play inference acceleration for large vision-language models. 2024. 2

    work page 2024

  4. [4]

    Sun, Xing

    Chaoyou Fu, Peixian Chen, Yunhang Shen, Yulei Qin, Mengdan Zhang, Xu Lin, Jinrui Yang, Xiawu Zheng, Ke Li, and et al. Sun, Xing. Mme: A comprehensive evalua- tion benchmark for multimodal large language models.arXiv preprint, arXiv:2306.13394, 2023. 3

    arXiv 2023

  5. [5]

    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. In Conference on Computer Vision and Pattern Recognition (CVPR), 2019. 3

    work page 2019

  6. [6]

    Perceiver: Gen- eral perception with iterative attention

    Andrew Jaegle, Felix Gimeno, Andrew Brock, Andrew Zis- serman, Oriol Vinyals, and Joao Carreira. Perceiver: Gen- eral perception with iterative attention. arXiv preprint arXiv:2103.03206, 2021. 1

    arXiv 2021

  7. [7]

    Towards efficient visual-language alignment of the q-former for visual reason- ing tasks

    Sungkyung Kim, Adam Lee, Junyoung Park, Andrew Chung, Jusang Oh, and Jay-Yoon Lee. Towards efficient visual-language alignment of the q-former for visual reason- ing tasks. 2024. 1

    work page 2024

  8. [8]

    Llava-onevision: Easy visual task transfer

    Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Peiyuan Zhang, Yanwei Li, Zi- wei Liu, and Chunyuan Li. Llava-onevision: Easy visual task transfer. Arxiv e-prints, 2024. 1, 2

    work page 2024

Show all 25 references
  1. [9]

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In In- ternational conference on machine learning, 2023. 1

    2023

  2. [10]

    Evaluating object hallucina- tion in large vision-language models

    Yifan Li, Yifan Du, Kun Zhou, Jinpeng Wang, Wayne Xin Zhao, and Ji-Rong Wen. Evaluating object hallucina- tion in large vision-language models. arXiv preprint , arXiv:2305.10355, 2023. 3

    2023 arXiv

  3. [11]

    Vila: On pre-training for visual language models

    Ji Lin, Hongxu Yin, Wei Ping, Yao Lu, Pavlo Molchanov, Andrew Tao, Huizi Mao, Jan Kautz, Mohammad Shoeybi, and Song Han. Vila: On pre-training for visual language models. Arxiv e-prints, 2024. 1, 2

    2024

  4. [12]

    Improved baselines with visual instruction tuning

    Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. Improved baselines with visual instruction tuning. arXiv:2310.03744, 2023. 1

    2023 arXiv

  5. [13]

    Llava-next: Im- proved reasoning, ocr, and world knowledge

    Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. Llava-next: Im- proved reasoning, ocr, and world knowledge. 2024. 1, 2

    2024

  6. [14]

    Liu, Ziwei

    Yuan Liu, Haodong Duan, Yuanhan Zhang, Bo Li, Songyang Zhang, Wangbo Zhao, Yike Yuan, Jiaqi Wang, Conghui He, and et al. Liu, Ziwei. Mmbench: Is your multi-modal model an all-around player? arXiv preprint, arXiv:2307.06281,

    arXiv

  7. [15]

    Learn to explain: Multimodal reasoning via thought chains for science question answering

    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. In Advances in Neural Information Processing Systems (NeurIPS...

    2022

  8. [16]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. Arxiv e-prints,

  9. [17]

    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. Arxiv e-prints, 2024. 2, 3 6

    2024

  10. [18]

    Towards vqa models that can read

    Amanpreet Singh, Vivek Natarjan, Meet Shah, Yu Jiang, Xinlei Chen, Devi Parikh, and Marcus Rohrbach. Towards vqa models that can read. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 8317–8326, 2019. 3

    2019

  11. [19]

    Next-gpt: Any-to-any multimodal llm

    Shengqiong Wu, Hao Fei, Leigang Qu, Wei Ji, and Tat-Seng Chua. Next-gpt: Any-to-any multimodal llm. arXiv preprint, arXiv:2023, 2023. 1

    2023

  12. [20]

    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. Arxiv e-prints, 2024. 2

    2024

  13. [21]

    X-vila: Cross-modality align- ment for large language model

    Hanrong Ye, De-An Huang, Yao Lu, Zhiding Yu, Wei Ping, Andrew Tao, Jan Kautz, Song Han, Dan Xu, Pavlo Molchanov, and Hongxu Yin. X-vila: Cross-modality align- ment for large language model. arXiv preprint, arXiv:2024,

    2024

  14. [22]

    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. In International Conference on Machine Learn- ing (ICML). PMLR, 2024. 3

    2024

  15. [23]

    Llm inference unveiled: Survey and roofline model insights

    Zhihang Yuan, Yuzhang Shang, Yang Zhou, Zhen Dong, Chenhao Xue, Bingzhe Wu, Zhikai Li, Qingyi Gu, Yong Jae Lee, Yan Yan, et al. Llm inference unveiled: Survey and roofline model insights. arXiv preprint arXiv:2402.16363 ,

    arXiv

  16. [24]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. Arxiv e-prints, 2023. 1, 2

    2023

  17. [25]

    Sparsevlm: Vi- sual token sparsification for efficient vision-language model inference

    Yuan Zhang, Chun-Kai Fan, Junpeng Ma, Wenzhao Zheng, Tao Huang, Kuan Cheng, Denis Gudovskiy, Tomoyuki Okuno, Yohei Nakata, Kurt Keutzer, et al. Sparsevlm: Vi- sual token sparsification for efficient vision-language model inference. arXiv preprint arXiv:2410.04417, 2024. 2 7

    arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.