Efficient Token Pruning for LLaDA-V

• URL: http://arxiv.org/abs/2601.20168v1
• Date: Wed, 28 Jan 2026 02:03:03 GMT
• Title: Efficient Token Pruning for LLaDA-V
• Authors: Zhewen Wan, Tianchen Song, Chen Lin, Zhiyong Zhao, Xianpeng Lang,
• Abstract summary: Diffusion-based large multimodal models, such as LLaDA-V, have demonstrated impressive capabilities in vision-language understanding and generation.<n>We conduct an in-depth attention analysis and reveal that, unlike autoregressive decoders, LLaDA-V aggregates cross-modal information predominantly in middle-to-late layers.<n>Motivated by this observation, we propose a structured token pruning strategy inspired by FastV.
• Score: 5.790228283932133
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Diffusion-based large multimodal models, such as LLaDA-V, have demonstrated impressive capabilities in vision-language understanding and generation. However, their bidirectional attention mechanism and diffusion-style iterative denoising paradigm introduce significant computational overhead, as visual tokens are repeatedly processed across all layers and denoising steps. In this work, we conduct an in-depth attention analysis and reveal that, unlike autoregressive decoders, LLaDA-V aggregates cross-modal information predominantly in middle-to-late layers, leading to delayed semantic alignment. Motivated by this observation, we propose a structured token pruning strategy inspired by FastV, selectively removing a proportion of visual tokens at designated layers to reduce FLOPs while preserving critical semantic information. To the best of our knowledge, this is the first work to investigate structured token pruning in diffusion-based large multimodal models. Unlike FastV, which focuses on shallow-layer pruning, our method targets the middle-to-late layers of the first denoising step to align with LLaDA-V's delayed attention aggregation to maintain output quality, and the first-step pruning strategy reduces the computation across all subsequent steps. Our framework provides an empirical basis for efficient LLaDA-V inference and highlights the potential of vision-aware pruning in diffusion-based multimodal models. Across multiple benchmarks, our best configuration reduces computational cost by up to 65% while preserving an average of 95% task performance.

Related papers

• A Comprehensive Study on Visual Token Redundancy for Discrete Diffusion-based Multimodal Large Language Models [85.30893355216486]
  We study how visual token redundancy evolves with different dMLLM architectures and tasks.<n>Our study reveals that visual redundancy emerges only in from-scratch dMLLMs while handling long-answer tasks.<n>Layer-skipping is promising for accelerating AR-to-diffusion dMLLMs, whereas progressive or late-step pruning is more effective for from-scratch dMLLMs.
  arXiv  Detail & Related papers  (2025-11-19T04:13:36Z)
• IAM: Efficient Inference through Attention Mapping between Different-scale LLMs [74.81417160018856]
  IAM framework achieves dual benefits of accelerated attention computation and reduced KV cache usage.<n>We show that IAM can accelerate prefill by 15% and reduce KV cache usage by 22.1% without appreciably sacrificing performance.
  arXiv  Detail & Related papers  (2025-07-16T06:39:11Z)
• MoCa: Modality-aware Continual Pre-training Makes Better Bidirectional Multimodal Embeddings [75.0617088717528]
  MoCa is a framework for transforming pre-trained VLM backbones into effective bidirectional embedding models.<n>MoCa consistently improves performance across MMEB and ViDoRe-v2 benchmarks, achieving new state-of-the-art results.
  arXiv  Detail & Related papers  (2025-06-29T06:41:00Z)
• Think Twice, Act Once: Token-Aware Compression and Action Reuse for Efficient Inference in Vision-Language-Action Models [30.7855782696894]
  Vision-Language-Action (VLA) models have emerged as a powerful paradigm for general-purpose robot control through natural language instructions.<n>We propose FlashVLA, the first training-free and plug-and-play acceleration framework that enables action reuse in VLA models.
  arXiv  Detail & Related papers  (2025-05-27T13:47:18Z)
• A Stitch in Time Saves Nine: Small VLM is a Precise Guidance for Accelerating Large VLMs [65.00970402080351]
  A promising approach to accelerating large vision-language models (VLMs) is using partial information, such as attention maps from specific layers, to assess token importance and prune less essential tokens.<n>Our study reveals three key insights: (i) Partial attention information is insufficient for accurately identifying critical visual tokens, resulting in suboptimal performance, especially at low token retention ratios; (ii) Global attention information, such as the attention map aggregated across all layers, more effectively preserves essential tokens and maintains comparable performance under aggressive pruning; and (iii) The global attention map aggregated from a small VLM closely resembles that of a large VLM,
  arXiv  Detail & Related papers  (2024-12-04T13:56:44Z)
• ZipVL: Efficient Large Vision-Language Models with Dynamic Token Sparsification [29.163757099307553]
  The efficiency of large vision-language models (LVLMs) is constrained by the computational bottleneck of the attention mechanism during the prefill phase.<n>We present ZipVL, an efficient inference framework designed for LVLMs through a dynamic ratio allocation strategy of important tokens.
  arXiv  Detail & Related papers  (2024-10-11T07:24:21Z)
• Universal Pooling Method of Multi-layer Features from Pretrained Models for Speaker Verification [7.005068872406135]
  Recent advancements in automatic speaker verification (ASV) studies have been achieved by leveraging large-scale pretrained networks. We present a novel approach for exploiting the multilayered nature of pretrained models for ASV. We show how the proposed interlayer processing aids in maximizing the advantage of utilizing pretrained models.
  arXiv  Detail & Related papers  (2024-09-12T05:55:32Z)
• An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models [65.37846460916042]
  We find out that the attention computation over visual tokens is of extreme inefficiency in the deep layers of popular LVLMs. We introduce FastV, a versatile plug-and-play method designed to optimize computational efficiency.
  arXiv  Detail & Related papers  (2024-03-11T14:35:32Z)
• Perceiver-VL: Efficient Vision-and-Language Modeling with Iterative Latent Attention [100.81495948184649]
  We present Perceiver-VL, a vision-and-language framework that efficiently handles high-dimensional multimodal inputs such as long videos and text. Our framework scales with linear complexity, in contrast to the quadratic complexity of self-attention used in many state-of-the-art transformer-based models.
  arXiv  Detail & Related papers  (2022-11-21T18:22:39Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.