Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.27256 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908622558068736 |
|---|---|
| author | Tang, Xin Han, Youfang Gou, Fangfei Zhao, Wei Meng, Xin Yu, Yang Zhang, Jinguo Shi, Yuanchun Wang, Yuntao Zhang, Tengxiang |
| author_facet | Tang, Xin Han, Youfang Gou, Fangfei Zhao, Wei Meng, Xin Yu, Yang Zhang, Jinguo Shi, Yuanchun Wang, Yuntao Zhang, Tengxiang |
| contents | Vision-Language Models (VLMs) excel in diverse multimodal tasks. However, user requirements vary across scenarios, which can be categorized into fast response, high-quality output, and low energy consumption. Relying solely on large models deployed in the cloud for all queries often leads to high latency and energy cost, while small models deployed on edge devices are capable of handling simpler tasks with low latency and energy cost. To fully leverage the strengths of both large and small models, we propose ECVL-ROUTER, the first scenario-aware routing framework for VLMs. Our approach introduces a new routing strategy and evaluation metrics that dynamically select the appropriate model for each query based on user requirements, maximizing overall utility. We also construct a multimodal response-quality dataset tailored for router training and validate the approach through extensive experiments. Results show that our approach successfully routes over 80\% of queries to the small model while incurring less than 10\% drop in problem solving probability. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_27256 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | ECVL-ROUTER: Scenario-Aware Routing for Vision-Language Models Tang, Xin Han, Youfang Gou, Fangfei Zhao, Wei Meng, Xin Yu, Yang Zhang, Jinguo Shi, Yuanchun Wang, Yuntao Zhang, Tengxiang Machine Learning Human-Computer Interaction Vision-Language Models (VLMs) excel in diverse multimodal tasks. However, user requirements vary across scenarios, which can be categorized into fast response, high-quality output, and low energy consumption. Relying solely on large models deployed in the cloud for all queries often leads to high latency and energy cost, while small models deployed on edge devices are capable of handling simpler tasks with low latency and energy cost. To fully leverage the strengths of both large and small models, we propose ECVL-ROUTER, the first scenario-aware routing framework for VLMs. Our approach introduces a new routing strategy and evaluation metrics that dynamically select the appropriate model for each query based on user requirements, maximizing overall utility. We also construct a multimodal response-quality dataset tailored for router training and validate the approach through extensive experiments. Results show that our approach successfully routes over 80\% of queries to the small model while incurring less than 10\% drop in problem solving probability. |
| title | ECVL-ROUTER: Scenario-Aware Routing for Vision-Language Models |
| topic | Machine Learning Human-Computer Interaction |
| url | https://arxiv.org/abs/2510.27256 |