HiRED: Attention-Guided Token Dropping for Efficient Inference of High-Resolution Vision-Language Models

Documenti analoghi

• QPART: Adaptive Model Quantization and Dynamic Workload Balancing for Accuracy-aware Edge Inference
  di: Li, Xiangchen, et al.
  Pubblicazione: (2025)
• iMAD: Intelligent Multi-Agent Debate for Efficient and Accurate LLM Inference
  di: Fan, Wei, et al.
  Pubblicazione: (2025)
• Polymorph: Energy-Efficient Multi-Label Classification for Video Streams on Embedded Devices
  di: Ghafouri, Saeid, et al.
  Pubblicazione: (2025)
• SLED: A Speculative LLM Decoding Framework for Efficient Edge Serving
  di: Li, Xiangchen, et al.
  Pubblicazione: (2025)
• PAINT: Paying Attention to INformed Tokens to Mitigate Hallucination in Large Vision-Language Model
  di: Arif, Kazi Hasan Ibn, et al.
  Pubblicazione: (2025)