LoopCache paper proposes faster long-context self-speculative decoding for looped transformers
A new arXiv paper, LoopCache, introduces a method to accelerate long-context self-speculative decoding for looped transformers, which reuse a shared block across recurrent steps to gain effective depth with compact models.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-19 05:00 · arXiv cs.CL
LoopCache: Accelerating Long-Context Self-Speculative Decoding for Looped Transformers