FastJEV: Understanding Redundancy for Compact JEV Inference

Key points

• In this paper, we study these sources of redundancy and present FastJEV for compact candidate evaluation.1
• We first introduce shared context anchoring to reuse recurrent initial states and omit unused final recurrent caches.1
• Through controlled experiments, we show how candidate overlap and branching structure affect the execution cost of history reuse.1
• These findings motivate designing sharing granularity and execution schedules together for efficient JEV inference.1

Sources (1)

• [1]FastJEV: Understanding Redundancy for Compact JEV Inference
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 07:11 AM

  JEV models make multimodal decisions by directly scoring candidates.

  In this paper, we study these sources of redundancy and present FastJEV for compact candidate evaluation.

Extractive summary: sentences quoted from the sources.