UltraViT: Latency-Optimized On-device Vision Encoder for Large Vision-Language Models

  • 类型:arxiv
  • 标识:2607.23373
  • 链接:https://arxiv.org/abs/2607.23373
  • 主分类:multimodal
  • 形态:application
  • 被引:0
  • 被引来源:Semantic Scholar + OpenAlex
  • S2被引:0
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:Extensive experiments demonstrate that the on-device latency-informed design combined with the tailored training strategy establishes a new state-of-the-art for efficient LVLM encoding, significantly outperforming existing encoder-centric baselines while operating on-device at nearly 1.7xthe speed.
  • OpenAlex ID:W7171564273
  • OpenAlex DOI:10.48550/arxiv.2607.23373
  • DOI:10.48550/arxiv.2607.23373
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://doi.org/10.48550/arxiv.2607.23373
  • OpenAlex更新:2026-08-25
  • 待LLM分类:否
  • 标题中文:UltraViT:面向大视觉-语言模型的端侧延迟优化视觉编码器
  • TLDR中文:大量实验表明,结合端上延迟感知设计与定制化训练策略,建立了高效 LVLM 编码的新 SOTA,在端上以近 1.7 倍速度运行的同时显著优于现有以编码器为中心的基线。
  • 来源文件
  • /inbox/tom/_candidates/2026-07-29-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]