Sparse Training Scheme for Multimodal LLM

• URL: http://arxiv.org/abs/2509.18150v1
• Date: Tue, 16 Sep 2025 11:33:20 GMT
• Title: Sparse Training Scheme for Multimodal LLM
• Authors: Kean Shi, Liang Chen, Haozhe Zhao, Baobao Chang,
• Abstract summary: Multimodal Large Language Models (MLLMs) have demonstrated outstanding performance across a variety of domains.<n>We propose a novel training-efficient framework based on sparse representations, termed the Sparse Training Scheme (STS)<n>This scheme consists of two key components: the Visual Token, which reduces the information load by compressing visual tokens, and the Layer Dynamic Skipper, which mitigates the computational overhead by skipping unnecessary layers in the language model during both forward and backward passes.
• Score: 26.81140959413325
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Multimodal Large Language Models (MLLMs) have demonstrated outstanding performance across a variety of domains. However, training MLLMs is often inefficient due to the significantly longer input sequences introduced by multimodal data and the low utilization of inter-layer computations. To address this challenge, we shift the focus to the training process itself and propose a novel training-efficient framework based on sparse representations, termed the Sparse Training Scheme (STS). This scheme consists of two key components: the Visual Token Compressor, which reduces the information load by compressing visual tokens, and the Layer Dynamic Skipper, which mitigates the computational overhead by dynamically skipping unnecessary layers in the language model during both forward and backward passes. Our approach is broadly applicable to diverse MLLM architectures and has been extensively evaluated on multiple benchmarks, demonstrating its effectiveness and efficiency.

Related papers

• Magic-MM-Embedding: Towards Visual-Token-Efficient Universal Multimodal Embedding with MLLMs [10.443777669301983]
  Multimodal Large Language Models (MLLMs) have shown immense promise in universal multimodal retrieval.<n>But their practical application is often hindered by the substantial computational cost incurred from processing a large number of tokens from visual inputs.<n>We propose Magic-MM-Embedding, a series of novel models that achieve both high efficiency and state-of-the-art performance in universal multimodal embedding.
  arXiv  Detail & Related papers  (2026-02-05T04:01:01Z)
• 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)
• HyperET: Efficient Training in Hyperbolic Space for Multi-modal Large Language Models [50.31704374968706]
  Multi-modal large language models (MLLMs) have emerged as a transformative approach for aligning visual and textual understanding.<n>They typically require extremely high computational resources for training to achieve cross-modal alignment at multi-granularity levels.<n>We argue that a key source of this inefficiency lies in the vision encoders they widely equip with, e.g., CLIP and SAM, which lack the alignment with language at multi-granularity levels.
  arXiv  Detail & Related papers  (2025-10-23T08:16:44Z)
• $\mathcal{V}isi\mathcal{P}runer$: Decoding Discontinuous Cross-Modal Dynamics for Efficient Multimodal LLMs [26.779915891040236]
  We propose emphVisiPruner, a training-free pruning framework that reduces up to 99% of vision-related attention computations and 53.9% of FLOPs on LLaVA-v1.5 7B.<n>Our insights further provide actionable guidelines for training efficient MLLMs by aligning model architecture with its intrinsic layer-wise processing dynamics.
  arXiv  Detail & Related papers  (2025-10-20T06:40:17Z)
• NaViL: Rethinking Scaling Properties of Native Multimodal Large Language Models under Data Constraints [100.02131897927484]
  This paper focuses on the native training of Multimodal Large Language Models (MLLMs) in an end-to-end manner.<n>We propose a native MLLM called NaViL, combined with a simple and cost-effective recipe.<n> Experimental results on 14 multimodal benchmarks confirm the competitive performance of NaViL against existing MLLMs.
  arXiv  Detail & Related papers  (2025-10-09T17:59:37Z)
• Discrete Tokenization for Multimodal LLMs: A Comprehensive Survey [69.45421620616486]
  This work presents the first structured taxonomy and analysis of discrete tokenization methods designed for large language models (LLMs)<n>We categorize 8 representative VQ variants that span classical and modern paradigms and analyze their algorithmic principles, training dynamics, and integration challenges with LLM pipelines.<n>We identify key challenges including codebook collapse, unstable gradient estimation, and modality-specific encoding constraints.
  arXiv  Detail & Related papers  (2025-07-21T10:52:14Z)
• PUMA: Layer-Pruned Language Model for Efficient Unified Multimodal Retrieval with Modality-Adaptive Learning [54.73049408950049]
  We propose a Layer-Pruned Language Model for Efficient Unified Multimodal Retrieval with Modality-Adaptive Learning.<n>Our approach improves unified multimodal retrieval from both structural and learning perspectives.
  arXiv  Detail & Related papers  (2025-07-10T16:47:25Z)
• Efficient Multi-modal Long Context Learning for Training-free Adaptation [96.21248144937627]
  This paper introduces Efficient Multi-Modal Long Context Learning (EMLoC)<n>It embeds demonstration examples directly into the model input.<n>It condenses long-context multimodal inputs into compact, task-specific memory representations.
  arXiv  Detail & Related papers  (2025-05-26T10:49:44Z)
• TAMP: Token-Adaptive Layerwise Pruning in Multimodal Large Language Models [23.916205754112774]
  Multimodal Large Language Models (MLLMs) have shown remarkable versatility in understanding diverse multimodal data and tasks.<n>We propose TAMP, a simple yet effective pruning framework tailored for MLLMs.<n>We validate our method on two state-of-the-art MLLMs: LLaVA-NeXT, designed for vision-language tasks, and VideoLLaMA2, capable of processing audio, visual, and language modalities.
  arXiv  Detail & Related papers  (2025-04-14T05:44:38Z)
• FOLDER: Accelerating Multi-modal Large Language Models with Enhanced Performance [9.782362715017596]
  We introduce FOLDER, a simple yet effective plug-and-play module designed to reduce the length of the visual token sequence.<n>We analyze the information loss introduced by different reduction strategies and develop FOLDER to preserve key information while removing visual redundancy.<n>FOLDER achieves comparable or even better performance than the original models, while dramatically reducing complexity by removing up to 70% of visual tokens.
  arXiv  Detail & Related papers  (2025-01-05T03:28:45Z)
• Mono-InternVL: Pushing the Boundaries of Monolithic Multimodal Large Language Models with Endogenous Visual Pre-training [48.455597568212944]
  We present Mono-InternVL, a novel monolithic MLLM that seamlessly integrates a set of visual experts via a multimodal mixture-of-experts structure.<n>In particular, EViP is designed as a progressive learning process for visual experts, which aims to fully exploit the visual knowledge from noisy data to high-quality data.
  arXiv  Detail & Related papers  (2024-10-10T17:59:22Z)
• EE-MLLM: A Data-Efficient and Compute-Efficient Multimodal Large Language Model [15.449472477182061]
  Current approaches for vision and language interaction fall into two categories: self-attention-based and cross-attention-based methods.<n>We modify the original self-attention mechanism in MLLM to a composite attention mechanism.<n>EE-MLLM significantly outperforms Flamingo with limited training data, and reduces the prefilling time to 79 ms on an H800 GPU.<n>We present a training-free variant named EE-MLLM-F, which reduces the computation cost of self-attention-based method without additional training.
  arXiv  Detail & Related papers  (2024-08-21T17:36:37Z)
• SoupLM: Model Integration in Large Language and Multi-Modal Models [51.12227693121004]
  Training large language models (LLMs) requires significant computing resources. Existing publicly available LLMs are typically pre-trained on diverse, privately curated datasets spanning various tasks.
  arXiv  Detail & Related papers  (2024-07-11T05:38:15Z)

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.