New Method Targets Speculative Decoding Slowdowns in Vision-Language Models
A paper introduces VisHead-Spec, a head-routed visual attention approach for speculative decoding in vision-language models, aiming to preserve speedups under high concurrency.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-18 05:00 · arXiv cs.LG
VisHead-Spec: Head-Routed Visual Attention for Multimodal Speculative Decoding