# Growing a Multi-head Twig via Distillation and Reinforcement Learning to Accelerate Large Vision-Language Models

arXiv: https://arxiv.org/abs/2503.14075 · HTML: https://mlsanity.com/paper/2503.14075/

## Key facts

- Cited by 15 papers in the map (data through 14 Oct 2026).
- Placed in Visual token pruning and efficiency, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 32 of its 80 references that are in the map.
- First posted on arXiv 18 Mar 2025.

## Placed in

- Multimodal LLMs / vision-language models / [Visual token pruning and efficiency](https://mlsanity.com/mllm/visual-token-pruning-and-efficiency/)
- Efficient LLM inference and serving / [Visual token pruning in multimodal models](https://mlsanity.com/inference/visual-token-pruning-in-multimodal-models/)

## Builds on (most cited first, 30 of 32)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control](https://mlsanity.com/paper/2307.15818/) (2023)
- [OpenVLA: An Open-Source Vision-Language-Action Model](https://mlsanity.com/paper/2406.09246/) (2024)
- [Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond](https://mlsanity.com/paper/2308.12966/) (2023)
- [Evaluating Object Hallucination in Large Vision-Language Models](https://mlsanity.com/paper/2305.10355/) (2023)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilities](https://mlsanity.com/paper/2308.02490/) (2023)
- [Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads](https://mlsanity.com/paper/2401.10774/) (2024)
- [DeepSeek-V3 Technical Report](https://mlsanity.com/paper/2412.19437/) (2024)
- [AppAgent: Multimodal Agents as Smartphone Users](https://mlsanity.com/paper/2312.13771/) (2023)
- [LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Models](https://mlsanity.com/paper/2403.15388/) (2024)
- [SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference](https://mlsanity.com/paper/2410.04417/) (2024)
- [VisionZip: Longer is Better but Not Necessary in Vision Language Models](https://mlsanity.com/paper/2412.04467/) (2024)
- [LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding](https://mlsanity.com/paper/2404.16710/) (2024)
- [PyramidDrop: Accelerating Your Large Vision-Language Models via Pyramid Visual Redundancy Reduction](https://mlsanity.com/paper/2410.17247/) (2024)
- [Mobile-Agent: Autonomous Multi-Modal Mobile Device Agent with Visual Perception](https://mlsanity.com/paper/2401.16158/) (2024)
- [A Survey on Efficient Inference for Large Language Models](https://mlsanity.com/paper/2404.14294/) (2024)
- [mPLUG-DocOwl: Modularized Multimodal Large Language Model for Document Understanding](https://mlsanity.com/paper/2307.02499/) (2023)
- [Beyond Text-Visual Attention: Exploiting Visual Cues for Effective Token Pruning in VLMs](https://mlsanity.com/paper/2412.01818/) (2024)
- [AuroraCap: Efficient, Performant Video Detailed Captioning and a New Benchmark](https://mlsanity.com/paper/2410.03051/) (2024)
- [MAmmoTH-VL: Eliciting Multimodal Reasoning with Instruction Tuning at Scale](https://mlsanity.com/paper/2412.05237/) (2024)
- [Fit and Prune: Fast and Training-free Visual Token Pruning for Multi-modal Large Language Models](https://mlsanity.com/paper/2409.10197/) (2024)
- [ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Models](https://mlsanity.com/paper/2412.00447/) (2024)
- [FastVLM: Efficient Vision Encoding for Vision Language Models](https://mlsanity.com/paper/2412.13303/) (2024)
- [Multi-Stage Vision Token Dropping: Towards Efficient Multimodal Large Language Model](https://mlsanity.com/paper/2411.10803/) (2024)
- [Cross-modal Information Flow in Multimodal Large Language Models](https://mlsanity.com/paper/2411.18620/) (2024)
- [Efficient Large Multi-modal Models via Visual Context Compression](https://mlsanity.com/paper/2406.20092/) (2024)
- [Filter, Correlate, Compress: Training-Free Token Reduction for MLLM Acceleration](https://mlsanity.com/paper/2411.17686/) (2024)
- [A Stitch in Time Saves Nine: Small VLM is a Precise Guidance for Accelerating Large VLMs](https://mlsanity.com/paper/2412.03324/) (2024)

## Built on by (most cited first, 15 of 15)

- [Video-MMLU: A Massive Multi-Discipline Lecture Understanding Benchmark](https://mlsanity.com/paper/2504.14693/) (2025)
- [VideoARM: Agentic Reasoning over Hierarchical Memory for Long-Form Video Understanding](https://mlsanity.com/paper/2512.12360/) (2025)
- [SpecVLM: Enhancing Speculative Decoding of Video LLMs via Verifier-Guided Token Pruning](https://mlsanity.com/paper/2508.16201/) (2025)
- [FlashVID: Efficient Video Large Language Models via Training-free Tree-based Spatiotemporal Token Merging](https://mlsanity.com/paper/2602.08024/) (2026)
- [HiDrop: Hierarchical Vision Token Reduction in MLLMs via Late Injection, Concave Pyramid Pruning, and Early Exit](https://mlsanity.com/paper/2602.23699/) (2026)
- [N"uwa: Mending the Spatial Integrity Torn by VLM Token Pruning](https://mlsanity.com/paper/2602.02951/) (2026)
- [IVC-Prune: Revealing the Implicit Visual Coordinates in LVLMs for Vision Token Pruning](https://mlsanity.com/paper/2602.03060/) (2026)
- [Catching the Details: Self-Distilled RoI Predictors for Fine-Grained MLLM Perception](https://mlsanity.com/paper/2509.16944/) (2025)
- [Beyond Surrogate Gradients: Fully Differentiable Token Pruning for Vision-Language Models](https://mlsanity.com/paper/2605.28051/) (2026)
- [SpecVLM: Fast Speculative Decoding in Vision-Language Models](https://mlsanity.com/paper/2509.11815/) (2025)
- [Enhancing Target-unspecific Tasks through a Features Matrix](https://mlsanity.com/paper/2505.03414/) (2025)
- [SpecFLASH: A Latent-Guided Semi-autoregressive Speculative Decoding Framework for Efficient Multimodal Generation](https://mlsanity.com/paper/2505.12728/) (2025)
- [Progressive Video Condensation with MLLM Agent for Long-form Video Understanding](https://mlsanity.com/paper/2604.02891/) (2026)
- [DiffPrune: differentiable information throttling for token pruning in vision-language models](https://mlsanity.com/paper/2608.01985/) (2026)
- [ConsensusDrop: Fusing Visual and Cross-Modal Saliency for Efficient Vision Language Models](https://mlsanity.com/paper/2602.00946/) (2026)

## Abstract (arXiv)

Large vision-language models (VLMs) have demonstrated remarkable capabilities in open-world multimodal understanding, yet their high computational overheads pose great challenges for practical deployment. Some recent works have proposed methods to accelerate VLMs by pruning redundant visual tokens guided by the attention maps of VLM's early layers. Despite the success of these token pruning methods, they still suffer from two major shortcomings: (i) considerable accuracy drop due to insensitive attention signals in early layers, and (ii) limited speedup when generating long responses (e.g., 30 tokens). To address the limitations above, we present TwigVLM -- a simple and general architecture by growing a lightweight module, named twig, upon an early layer of the base VLM. Compared with most existing VLM acceleration methods purely based on visual token pruning, our TwigVLM not only achieves better accuracy retention by employing a twig-guided token pruning (TTP) strategy, but also yields higher generation speed by utilizing a self-speculative decoding (SSD) strategy. Taking LLaVA-1.5-7B as the base VLM, experimental results show that TwigVLM preserves 96% of the original performance after pruning 88.9% of the visual tokens and achieves 154% speedup in generating long responses, delivering significantly better performance in terms of both accuracy and speed over the state-of-the-art VLM acceleration methods. Moreover, we extend TwigVLM to an improved TwigVLM++ variant by introducing a novel multi-head twig architecture with a specialized pruning head. TwigVLM++ improves pruning quality via a two-stage training paradigm combining a distillation learning stage and a pruning-oriented reinforcement learning stage, and further accelerates inference via a tree-based SSD strategy.