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

[2503.14075] Growing a Multi-head Twig via Distillation and Reinforcement Learning to Accelerate Large Vision-Language Models Skip to main content Search Submit Donate Log in Search arXiv Press Enter to search · Advanced search Computer Science > Computer Vision and Pattern Recognition arXiv:2503.14075 (cs) [Submitted on 18 Mar 2025 ( v1 ), last revised 10 Apr 2026 (this version, v3)] Title: Growing a Multi-head Twig via Distillation and Reinforcement Learning to Accelerate Large Vision-Language Models Authors: Zhenwei Shao , Mingyang Wang , Weijun Zhang , Zhou Yu , Wenwen Pan , Yan Yang , Tao Wei , Hongyuan Zhang , Jun Yu View a PDF of the paper titled Growing a Multi-head Twig via Distillation and Reinforcement Learning to Accelerate Large Vision-Language Models, by Zhenwei Shao and 8 other authors View PDF HTML (experimental) Abstract: 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. Comments: An extended version of our ICCV paper at this https URL Subjects: Computer Vision and Pattern Recognition (cs.CV) ; Computation and Language (cs.CL) Cite as: arXiv:2503.14075 [cs.CV] (or arXiv:2503.14075v3 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2503.14075 Focus to learn more arXiv-issued DOI via DataCite Submission history From: Zhou Yu [ view email ] [v1] Tue, 18 Mar 2025 09:52:45 UTC (1,713 KB) [v2] Sat, 19 Jul 2025 13:38:51 UTC (3,654 KB) [v3] Fri, 10 Apr 2026 01:08:12 UTC (3,936 KB) Full-text links: Access Paper: View a PDF of the paper titled Growing a Multi-head Twig via Distillation and Reinforcement Learning to Accelerate Large Vision-Language Models, by Zhenwei Shao and 8 other authors View PDF HTML (experimental) TeX Source view license Current browse context: cs.CV < prev | next > new | recent | 2025-03 Change to browse by: cs cs.CL References & Citations NASA ADS Google Scholar Semantic Scholar export BibTeX citation Loading... BibTeX formatted citation × loading... Data provided by: Bookmark Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer ( What is the Explorer? ) Connected Papers Toggle Connected Papers ( What is Connected Papers? ) Litmaps Toggle Litmaps ( What is Litmaps? ) scite.ai Toggle scite Smart Citations ( What are Smart Citations? ) Code, Data, Media Code, Data and Media Associated with this Article alphaXiv Toggle alphaXiv ( What is alphaXiv? ) Links to Code Toggle CatalyzeX Code Finder for Papers ( What is CatalyzeX? ) DagsHub Toggle DagsHub ( What is DagsHub? ) GotitPub Toggle Gotit.pub ( What is GotitPub? ) Huggingface Toggle Hugging Face ( What is Huggingface? ) ScienceCast Toggle ScienceCast ( What is ScienceCast? ) Demos Demos Replicate Toggle Replicate ( What is Replicate? ) Spaces Toggle Hugging Face Spaces ( What is Spaces? ) Spaces Toggle TXYZ.AI ( What is TXYZ.AI? ) Related Papers Recommenders and Search Tools Link to Influence Flower Influence Flower ( What are Influence Flowers? ) Core recommender toggle CORE Recommender ( What is CORE? ) Author Venue Institution Topic About arXivLabs arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs . Which authors of this paper are endorsers? | Disable MathJax ( What is MathJax? ) We gratefully acknowledge support from our major funders , member institutions , , and all contributors. About · Help · Contact · Subscribe · Copyright · Privacy · Accessibility · Operational Status (opens in new tab) Major funding support from

Related documents

• PRISM: Pre-alignment via Black-box On-policy Distillation for Multimodal Reinforcement Learning#149095
• Fast LeWorldModel#1128617
• RL-VLA$^3$: A Flexible and Asynchronous Reinforcement Learning Framework for VLA Training#927227
• Training and Agentic Inference Strategies for LLM-based Manim Animation Generation#798857
• Training and Agentic Inference Strategies for LLM-based Manim Animation Generation#797443