# Improving MLLM Training Efficiency via Stage-Aware Sparsity

arXiv: https://arxiv.org/abs/2509.18150 · HTML: https://mlsanity.com/paper/2509.18150/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Visual token pruning and efficiency, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 24 of its 30 references that are in the map.
- First posted on arXiv 16 Sep 2025.

## Placed in

- Multimodal LLMs / vision-language models / [Visual token pruning and efficiency](https://mlsanity.com/mllm/visual-token-pruning-and-efficiency/)

## Builds on (most cited first, 24 of 24)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://mlsanity.com/paper/2301.12597/) (2023)
- [Flamingo: a Visual Language Model for Few-Shot Learning](https://mlsanity.com/paper/2204.14198/) (2022)
- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond](https://mlsanity.com/paper/2308.12966/) (2023)
- [Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering](https://mlsanity.com/paper/2209.09513/) (2022)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [GLM: General Language Model Pretraining with Autoregressive Blank Infilling](https://mlsanity.com/paper/2103.10360/) (2021)
- [Deep Networks with Stochastic Depth](https://mlsanity.com/paper/1603.09382/) (2016)
- [An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models](https://mlsanity.com/paper/2403.06764/) (2024)
- [Reducing Transformer Depth on Demand with Structured Dropout](https://mlsanity.com/paper/1909.11556/) (2019)
- [SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference](https://mlsanity.com/paper/2410.04417/) (2024)
- [MoE-LLaVA: Mixture of Experts for Large Vision-Language Models](https://mlsanity.com/paper/2401.15947/) (2024)
- [PyramidDrop: Accelerating Your Large Vision-Language Models via Pyramid Visual Redundancy Reduction](https://mlsanity.com/paper/2410.17247/) (2024)
- [PuMer: Pruning and Merging Tokens for Efficient Vision Language Models](https://mlsanity.com/paper/2305.17530/) (2023)
- [Staged Training for Transformer Language Models](https://mlsanity.com/paper/2203.06211/) (2022)
- [Efficient Large Multi-modal Models via Visual Context Compression](https://mlsanity.com/paper/2406.20092/) (2024)
- [No Train No Gain: Revisiting Efficient Training Algorithms For Transformer-based Language Models](https://mlsanity.com/paper/2307.06440/) (2023)
- [Variation-aware Vision Token Dropping for Faster Large Vision-Language Models](https://mlsanity.com/paper/2509.01552/) (2025)
- [Mipha: A Comprehensive Overhaul of Multimodal Assistant with Small Language Models](https://mlsanity.com/paper/2403.06199/) (2024)
- [Leveraging Visual Tokens for Extended Text Contexts in Multi-Modal Learning](https://mlsanity.com/paper/2406.02547/) (2024)
- [Dynamic Token Reduction during Generation for Vision Language Models](https://mlsanity.com/paper/2501.14204/) (2025)
- [Exploring Activation Patterns of Parameters in Language Models](https://mlsanity.com/paper/2405.17799/) (2024)

## Abstract (arXiv)

Multimodal Large Language Models (MLLMs) have demonstrated outstanding performance across a variety of domains. However, training MLLMs is often inefficient, as much of the computation is redundant due to the long input sequences from multimodal data and underutilized inter-layer operations. Notably, such redundancy is not static but varies across different stages of training. Building on this observation, we shift the focus to the training process itself and propose a training-efficient framework based on sparse representations, termed the Sparse Training Scheme (STS). Instead of applying a uniform sparsity strategy, STS adopts a stage-aware design that adapts to different sources of redundancy during training. Specifically, the framework consists of two complementary components: the Visual Token Compressor, which reduces the information load by compressing visual tokens during modality alignment, and the Layer Dynamic Skipper, which mitigates computational overhead by dynamically skipping unnecessary layers during instruction tuning. Our approach is broadly applicable to diverse MLLM architectures and has been extensively evaluated on multiple benchmarks, demonstrating its effectiveness and efficiency.