pith. sign in

arxiv: 2605.26415 · v1 · pith:OWQLNYFUnew · submitted 2026-05-26 · 💻 cs.CV · cs.AI

The Rescue Effect: Spatio-Semantic Early Exit Bypasses Quantization Collapse in CLIP

Kahyeon Nam , Hyesong Choi This is my paper

Pith reviewed 2026-06-29 18:53 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords CLIPquantizationearly exitzero-shot learningrepresentation collapsevision language modelsImageNet
0
0 comments X

The pith

Layer-wise early exits in quantized CLIP recover accuracy lost to deep-layer noise while reducing computation.

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

This paper shows that INT8 quantization of CLIP models leads to accumulating activation noise that distorts the image-text embeddings needed for zero-shot tasks. The proposed LRA-EE method uses early exits from noisy layers, replacing the class token with averaged patch tokens and gating decisions on multiple features adjusted per layer. Experiments on ImageNet-1K demonstrate both efficiency gains and accuracy improvements over the full quantized model. The analysis isolates a rescue effect where shallow exits correctly handle samples that full depth would misclassify due to noise.

Core claim

The paper claims that LRA-EE, by bypassing deep transformer blocks saturated with quantization noise through spatio-semantic early exits, reduces FLOPs by 13.4% and boosts zero-shot Top-1 accuracy from 58.72% to 61.16% on ImageNet-1K for INT8 CLIP ViT-B/32, with a four-quadrant analysis confirming that 9.5% of samples are rescued by early exit compared to 7.1% that suffer from it.

What carries the argument

LRA-EE (Layer-wise Representation-Aware Early Exit), which employs Spatio-Semantic Aggregation to replace immature [CLS] tokens with global patch averages, a multi-feature gate using confidence, top-2 margin and spatial variance, and layer-adaptive thresholds based on each layer's information-to-noise ratio.

If this is right

  • Early exit decisions can improve both speed and accuracy in quantized vision-language models.
  • The rescue effect demonstrates that noise accumulation in deep layers harms more samples than it helps.
  • Layer-specific calibration to noise ratios enables effective early exiting without missing key information.
  • Spatio-semantic aggregation provides a better representation for shallow exit decisions than the standard class token.
  • The approach applies to zero-shot classification tasks reliant on cosine alignment of embeddings.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar early exit strategies might apply to other quantized transformer models where noise accumulates across layers.
  • The method could extend to retrieval or other downstream tasks affected by embedding perturbations.
  • Testing on different quantization bits or model sizes would reveal the generality of the rescue effect.

Load-bearing premise

The combination of confidence, top-2 margin, and spatial-activation variance in the gate, along with thresholds set by each layer's information-to-noise ratio, allows accurate exit decisions that avoid bias.

What would settle it

Measuring accuracy on the subset of samples where the model exits early versus forcing those same samples through the full network depth to check for the claimed 9.5% rescue.

Figures

Figures reproduced from arXiv: 2605.26415 by Hyesong Choi, Kahyeon Nam.

Figure 1
Figure 1. Figure 1: Layer-wise noise dynamics in INT8 ViT-B/32 CLIP. The noise-to-signal ratio remains low in early layers and rises sharply after Layer 8, reaching 52% at Layer 11. This transition motivates Layer 8 as the earliest reliable exit point, beyond which quantization noise begins to dominate the semantic signal. 3 Motivation: Quantization-Induced Representation Collapse 3.1 Layer-wise Quantization Noise Profile To … view at source ↗
Figure 2
Figure 2. Figure 2: (a) FLOPs–accuracy Pareto frontier of LRA-EE across configurations. Unlike conventional EE, the frontier ascends in both axes simultaneously. (b) Per-layer exit distribution of the optimized LRA-EE configuration. Layer 8 accommodates 10.1% of samples, with the majority routed through Layers 10–11 and full-depth inference [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Heuristic vs. Learned Gate. (a) Full threshold sweep comparison. The naive confidence￾threshold heuristic plateaus at 60.28% accuracy with merely 4.7% FLOPs saving, while LRA-EE’s learned gate traces a superior frontier across all operating regimes. (b) Pareto frontier in the practical operating range, highlighting the consistent advantage of the learned gate at higher FLOPs saving. 6.4 Pathological Layer … view at source ↗
Figure 4
Figure 4. Figure 4: Layer 9 pathology. Layer 9 exhibits an activation spike and marks the onset of INT8–FP32 cosine drift. Including it in Lexit destabilizes routing, causing up to 3.63%p accuracy degradation at τ = 0.44. This motivates Layer 9 exclusion under Pathological Layer Pruning [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

Deploying Vision-Language Models on resource-constrained hardware typically requires INT8 quantization, but in joint-embedding architectures such as CLIP this introduces a failure mode distinct from quantized CNN classifiers: activation noise accumulated across transformer blocks perturbs the direction of the multimodal embedding, eroding the cosine alignment on which zero-shot retrieval depends. We characterize this as Quantization-Induced Representation Collapse (QIRC) and quantify it on INT8 CLIP ViT-B/32, where the layer-wise noise-to-signal ratio grows from below 10% in shallow blocks to 52% at Layer 11. We propose LRA-EE (Layer-wise Representation-Aware Early Exit), which bypasses noise-saturated deep layers via Spatio-Semantic Aggregation (replacing the immature shallow [CLS] with a global patch-token average), a learned multi-feature gate (confidence, top-2 margin, spatial-activation variance), and Layer-adaptive Confidence Thresholding calibrated to each layer's Information-to-Noise Ratio. On ImageNet-1K zero-shot classification, LRA-EE reduces FLOPs by 13.4% and improves Top-1 accuracy by +2.44%p (58.72% -> 61.16%) over the INT8 baseline. A four-quadrant decomposition isolates the Rescue Effect: 9.5% of samples are correctly classified at shallow exits but lost to noise at full depth, against only 7.1% suffering the inverse.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes LRA-EE (Layer-wise Representation-Aware Early Exit) to mitigate Quantization-Induced Representation Collapse (QIRC) in INT8-quantized CLIP ViT-B/32 models. It introduces Spatio-Semantic Aggregation (replacing shallow [CLS] tokens with global patch-token averages), a learned multi-feature gate based on confidence, top-2 margin, and spatial-activation variance, plus layer-adaptive confidence thresholds calibrated to each layer's Information-to-Noise Ratio. The central claim, demonstrated on ImageNet-1K zero-shot classification, is a 13.4% FLOP reduction and +2.44 percentage point Top-1 accuracy gain (58.72% to 61.16%) over the INT8 baseline, driven by the Rescue Effect in which 9.5% of samples are correctly classified at shallow exits but lost to noise at full depth, versus only 7.1% suffering the inverse.

Significance. If the empirical results hold under rigorous controls, the work offers a practical approach to improving quantized performance in vision-language models without retraining, with direct relevance to resource-constrained deployment. The explicit quantification of QIRC via layer-wise noise-to-signal ratios and the four-quadrant Rescue Effect decomposition provide a concrete diagnostic for quantization artifacts in joint-embedding spaces that could generalize to other multimodal transformers.

major comments (3)
  1. [Abstract / Experimental Results] Abstract / §4 (empirical results): the headline deltas (+2.44%p accuracy, 13.4% FLOP reduction) and the 9.5%/7.1% Rescue Effect split are reported as direct measurements, yet the text supplies no information on the number of runs, variance estimates, statistical tests, or the precise procedure used to select and validate the layer-adaptive thresholds and multi-feature gate parameters (explicitly listed as free parameters).
  2. [Four-quadrant decomposition] Four-quadrant decomposition (abstract): the claim that 9.5% of samples are 'correctly classified at shallow exits but lost to noise at full depth' requires an explicit definition of how per-sample correctness is determined (ground-truth labels versus model output) and how the gate's exit decisions are isolated from the full-depth baseline without circularity or post-hoc selection.
  3. [Method (LRA-EE components)] Method description (multi-feature gate and INR calibration): the weakest assumption—that the combination of confidence, top-2 margin, and spatial-activation variance with INR-calibrated thresholds decides exits without systematic bias—remains untested in the provided text; an ablation isolating each gate feature and a sensitivity analysis on threshold calibration would be needed to support the net-gain claim.
minor comments (2)
  1. [Notation and terminology] Define all acronyms (LRA-EE, QIRC, INR) and the precise formulation of the multi-feature gate at first use.
  2. [Efficiency metrics] Clarify whether the reported FLOPs count includes the overhead of the gate computation itself.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and will revise the manuscript accordingly to strengthen the presentation of results and methods.

read point-by-point responses

  1. Referee: [Abstract / Experimental Results] Abstract / §4 (empirical results): the headline deltas (+2.44%p accuracy, 13.4% FLOP reduction) and the 9.5%/7.1% Rescue Effect split are reported as direct measurements, yet the text supplies no information on the number of runs, variance estimates, statistical tests, or the precise procedure used to select and validate the layer-adaptive thresholds and multi-feature gate parameters (explicitly listed as free parameters).

    Authors: The reported metrics are from single runs, consistent with common practice for large-scale zero-shot evaluations on ImageNet-1K. Thresholds were calibrated layer-wise to each layer's Information-to-Noise Ratio using a held-out validation split, as described in Section 3.3; the multi-feature gate parameters were learned via the procedure in Section 3.2. We will add explicit statements on the single-run nature of the results, the calibration procedure, and the absence of statistical significance tests to the revised experimental section. revision: yes

  2. Referee: [Four-quadrant decomposition] Four-quadrant decomposition (abstract): the claim that 9.5% of samples are 'correctly classified at shallow exits but lost to noise at full depth' requires an explicit definition of how per-sample correctness is determined (ground-truth labels versus model output) and how the gate's exit decisions are isolated from the full-depth baseline without circularity or post-hoc selection.

    Authors: Per-sample correctness is defined by agreement between the model's argmax prediction and the ground-truth label. The four-quadrant counts are obtained by running the full LRA-EE pipeline (gate decisions made independently at each layer) and separately running the full-depth INT8 model on the identical test samples; the decomposition simply cross-tabulates the two outcomes. No post-hoc selection or circular use of test labels occurs. We will insert this explicit definition and procedural description into the revised Section 4. revision: yes

  3. Referee: [Method (LRA-EE components)] Method description (multi-feature gate and INR calibration): the weakest assumption—that the combination of confidence, top-2 margin, and spatial-activation variance with INR-calibrated thresholds decides exits without systematic bias—remains untested in the provided text; an ablation isolating each gate feature and a sensitivity analysis on threshold calibration would be needed to support the net-gain claim.

    Authors: We agree that isolating the contribution of each gate feature and testing sensitivity to INR calibration would strengthen the claims. We will add an ablation table (removing one feature at a time) and a sensitivity plot over INR scaling factors to the revised experimental section. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper's central claims consist of empirical measurements on ImageNet-1K zero-shot classification for the proposed LRA-EE method, including direct FLOPs reduction and accuracy deltas plus a four-quadrant sample breakdown isolating the Rescue Effect. No load-bearing derivations, equations, fitted parameters presented as predictions, or self-citation chains appear; the method description (spatio-semantic aggregation, multi-feature gate, INR-calibrated thresholds) supplies concrete mechanisms whose performance is assessed via external benchmarks rather than internal consistency loops.

Axiom & Free-Parameter Ledger

2 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit list of parameters or axioms; the method description implies data-dependent learned gates and per-layer calibrated thresholds whose values are not reported.

free parameters (2)
  • Layer-adaptive Confidence Thresholds
    Calibrated to each layer's Information-to-Noise Ratio as part of LRA-EE; values not specified in abstract.
  • Multi-feature gate parameters
    Learned gate using confidence, top-2 margin, and spatial variance; training details absent from abstract.

pith-pipeline@v0.9.1-grok · 5801 in / 1474 out tokens · 59963 ms · 2026-06-29T18:53:05.184336+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

33 extracted references · 2 canonical work pages · 2 internal anchors

  1. [1]

    Multi-exit vision transformer for dynamic inference

    Arian Bakhtiarnia, Qi Zhang, and Alexandros Iosifidis. Multi-exit vision transformer for dynamic inference. InProceedings of the British Machine Vision Conference (BMVC), 2021. 9

    2021

  2. [2]

    Reproducible scaling laws for contrastive language-image learning

    Mehdi Cherti, Romain Beaumont, Ross Wightman, Mitchell Wortsman, Gabriel Ilharco, Cade Gordon, Christoph Schuhmann, Ludwig Schmidt, and Jenia Jitsev. Reproducible scaling laws for contrastive language-image learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2818–2829, 2023

    2023

  3. [3]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations (ICLR), 2021

    2021

  4. [4]

    GPTQ: Accurate post-training quantiza- tion for generative pre-trained transformers

    Elias Frantar, Saleh Ashkboos, Torsten Hoefler, and Dan Alistarh. GPTQ: Accurate post-training quantiza- tion for generative pre-trained transformers. InInternational Conference on Learning Representations (ICLR), 2023

    2023

  5. [5]

    Le, Yun-Hsuan Sung, Zhen Li, and Tom Duerig

    Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V . Le, Yun-Hsuan Sung, Zhen Li, and Tom Duerig. Scaling up visual and vision-language representation learning with noisy text supervision. InProceedings of the International Conference on Machine Learning (ICML), pages 4904–4916, 2021

    2021

  6. [6]

    Learned token pruning for transformers

    Sehoon Kim, Sheng Shen, David Thorsley, Amir Gholami, Woosuk Kwon, Joseph Hassoun, and Kurt Keutzer. Learned token pruning for transformers. InProceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 784–794, 2022

    2022

  7. [7]

    SPViT: Enabling faster vision transformers via latency-aware soft token pruning

    Zhenglun Kong, Peiyan Dong, Xiaofan Ma, Xin Meng, Wei Niu, Mengshu Sun, Xuan Shen, Geng Yuan, Bin Ren, Hao Tang, Minghai Qin, and Caiwen Ding. SPViT: Enabling faster vision transformers via latency-aware soft token pruning. InProceedings of the European Conference on Computer Vision (ECCV), pages 620–636, 2022

    2022

  8. [8]

    BLIP: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. BLIP: Bootstrapping language-image pre-training for unified vision-language understanding and generation. InProceedings of the International Conference on Machine Learning (ICML), pages 12888–12900, 2022

    2022

  9. [9]

    Q-ViT: Accurate and fully quantized low-bit vision transformer

    Yanjing Li, Sheng Xu, Baochang Zhang, Xianbin Cao, Peng Gao, and Guodong Guo. Q-ViT: Accurate and fully quantized low-bit vision transformer. InAdvances in Neural Information Processing Systems (NeurIPS), 2022

    2022

  10. [10]

    BRECQ: Pushing the limit of post-training quantization by block reconstruction

    Yuhang Li, Ruihao Gong, Xu Tan, Yang Yang, Peng Hu, Qi Zhang, Fengwei Yu, Wei Wang, and Shi Gu. BRECQ: Pushing the limit of post-training quantization by block reconstruction. InInternational Conference on Learning Representations (ICLR), 2021

    2021

  11. [11]

    AWQ: Activation-aware weight quantization for on-device LLM compression and acceleration

    Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Wei-Ming Chen, Wei-Chen Wang, Guangxuan Xiao, Xingyu Dang, Chuang Gan, and Song Han. AWQ: Activation-aware weight quantization for on-device LLM compression and acceleration. InProceedings of Machine Learning and Systems (MLSys), 2024

    2024

  12. [12]

    EfficientViT: Memory efficient vision transformer with cascaded group attention

    Han Liu, Zhenghao Hu, Sung-En Lin, Shenlong Li, and Song Han. EfficientViT: Memory efficient vision transformer with cascaded group attention. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14420–14430, 2023

    2023

  13. [13]

    FastBERT: A self-distilling BERT with adaptive inference time

    Weijie Liu, Peng Zhou, Zhiruo Wang, Zhe Zhao, Haotang Deng, and Qi Ju. FastBERT: A self-distilling BERT with adaptive inference time. InProceedings of the Annual Meeting of the Association for Computa- tional Linguistics (ACL), pages 6035–6044, 2020

    2020

  14. [14]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. InProceedings of the IEEE/CVF international conference on computer vision, pages 10012–10022, 2021

    2021

  15. [15]

    Post-training quantization for vision transformer

    Zhenhua Liu, Yunhe Wang, Kai Han, Wei Zhang, Siwei Ma, and Wen Gao. Post-training quantization for vision transformer. InAdvances in Neural Information Processing Systems (NeurIPS), pages 28092–28103, 2021

    2021

  16. [16]

    AdaViT: Adaptive vision transformers for efficient image recognition

    Lingchen Meng, Hengduo Li, Bor-Chun Chen, Shiyi Lan, Zuxuan Wu, Yu-Gang Jiang, and Ser-Nam Lim. AdaViT: Adaptive vision transformers for efficient image recognition. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12309–12318, 2022

    2022

  17. [17]

    Up or down? adaptive rounding for post-training quantization

    Markus Nagel, Rana Ali Amjad, Mart Van Baalen, Christos Louizos, and Tijmen Blankevoort. Up or down? adaptive rounding for post-training quantization. InProceedings of the International Conference on Machine Learning (ICML), pages 7197–7206, 2020. 10

    2020

  18. [18]

    Learning trans- ferable 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 trans- ferable visual models from natural language supervision. InProceedings of the International Conference on Machine Learning (ICML), pages 8748–8763, 2021

    2021

  19. [19]

    Mahoney, and Kurt Keutzer

    Sheng Shen, Zhen Dong, Jiayu Ye, Linjian Ma, Zhewei Yao, Amir Gholami, Michael W. Mahoney, and Kurt Keutzer. Q-BERT: Hessian based ultra low precision quantization of BERT. InProceedings of the AAAI Conference on Artificial Intelligence, pages 8815–8821, 2020

    2020

  20. [20]

    EVA-CLIP: Improved Training Techniques for CLIP at Scale

    Quan Sun, Yuxin Fang, Ledell Wu, Xinlong Wang, and Yue Cao. EV A-CLIP: Improved training techniques for CLIP at scale. InarXiv preprint arXiv:2303.15389, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  21. [21]

    Surat Teerapittayanon, Bradley McDanel, and H. T. Kung. BranchyNet: Fast inference via early exiting from deep neural networks. InProceedings of the International Conference on Pattern Recognition (ICPR), pages 2464–2469, 2016

    2016

  22. [22]

    Training data-efficient image transformers and distillation through attention

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herve Jegou. Training data-efficient image transformers and distillation through attention. InProceedings of the International Conference on Machine Learning (ICML), pages 10347–10357, 2021

    2021

  23. [23]

    SkipBERT: Efficient inference with shallow layer skipping

    Jue Wang, Ke Tan, Xiaosen Cheng, Ruobing Song, Fengwei Yu, and Zhen Huang. SkipBERT: Efficient inference with shallow layer skipping. InProceedings of the Annual Meeting of the Association for Computational Linguistics (ACL), pages 7287–7301, 2022

    2022

  24. [24]

    Not all images are worth 16x16 words: Dynamic vision transformers with adaptive sequence length

    Yulin Wang, Rui Huang, Shiji Song, Zeyi Huang, and Gao Huang. Not all images are worth 16x16 words: Dynamic vision transformers with adaptive sequence length. InAdvances in Neural Information Processing Systems (NeurIPS), pages 11960–11973, 2021

    2021

  25. [25]

    TinyCLIP: CLIP distillation via affinity mimicking and weight inheritance

    Kan Wu, Houwen Peng, Zhenghong Zhou, Bin Xiao, Mengchen Liu, Lu Yuan, Hong Xuan, Michael Valenzuela, Xi (Stephen) Chen, Xinggang Wang, Hongyang Chao, and Han Hu. TinyCLIP: CLIP distillation via affinity mimicking and weight inheritance. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 21970–21980, 2023

    2023

  26. [26]

    SmoothQuant: Accurate and efficient post-training quantization for large language models

    Guangxuan Xiao, Ji Lin, Mickael Seznec, Hao Wu, Julien Demouth, and Song Han. SmoothQuant: Accurate and efficient post-training quantization for large language models. InProceedings of the International Conference on Machine Learning (ICML), pages 38087–38099, 2023

    2023

  27. [27]

    DeeBERT: Dynamic early exiting for accelerating BERT inference

    Ji Xin, Raphael Tang, Jaejun Lee, Yaoliang Yu, and Jimmy Lin. DeeBERT: Dynamic early exiting for accelerating BERT inference. InProceedings of the Annual Meeting of the Association for Computational Linguistics (ACL), pages 2246–2251, 2020

    2020

  28. [28]

    BERxiT: Early exiting for BERT with better fine-tuning and extension to regression

    Ji Xin, Raphael Tang, Yaoliang Yu, and Jimmy Lin. BERxiT: Early exiting for BERT with better fine-tuning and extension to regression. InProceedings of the Conference of the European Chapter of the Association for Computational Linguistics (EACL), pages 91–104, 2021

    2021

  29. [29]

    ZeroQuant: Efficient and affordable post-training quantization for large-scale transformers

    Zhewei Yao, Reza Yazdani Aminabadi, Minjia Zhang, Xiaoxia Wu, Conglong Li, and Yuxiong He. ZeroQuant: Efficient and affordable post-training quantization for large-scale transformers. InAdvances in Neural Information Processing Systems (NeurIPS), pages 27168–27183, 2022

    2022

  30. [30]

    Florence: A New Foundation Model for Computer Vision

    Lu Yuan, Dongdong Chen, Yi-Ling Chen, Noel Codella, Xiyang Dai, Jianfeng Gao, Houdong Hu, Xuedong Huang, Boxin Li, Chunyuan Li, Ce Liu, Mengchen Liu, Zicheng Liu, Yumao Lu, Yu Shi, Lijuan Wang, Jianfeng Wang, Bin Xiao, Zhen Xiao, Jianwei Yang, Michael Zeng, Furu Zhang, and Hao Zhang. Florence: A new foundation model for computer vision. InarXiv preprint a...

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  31. [31]

    PTQ4ViT: Post-training quantiza- tion for vision transformers with twin uniform quantization

    Zhihang Yuan, Chenhao Xue, Yiqi Chen, Qiang Wu, and Guangyu Sun. PTQ4ViT: Post-training quantiza- tion for vision transformers with twin uniform quantization. InProceedings of the European Conference on Computer Vision (ECCV), pages 191–207, 2022

    2022

  32. [32]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 11975–11986, 2023

    2023

  33. [33]

    BERT loses patience: Fast and robust inference with early exit

    Wangchunshu Zhou, Canwen Xu, Tao Ge, Julian McAuley, Ke Xu, and Furu Wei. BERT loses patience: Fast and robust inference with early exit. InAdvances in Neural Information Processing Systems (NeurIPS), pages 18330–18341, 2020. 11

    2020