---
title: "CopyLoop: Fast Looped Transformer Decoding via Context Copying and Shallow Screening"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.38822v1"
permalink: "https://feed.astralor.com/items/itm_306c447a67694f4480c0e680"
published: "2026-10-28T20:16:54.000Z"
score: 55
channel: "tech"
tags: ["CopyLoop","Fast Looped Transformer","Decoding","Context Copying","Shallow Screening Looped","Transformers","While","Looped Transformers"]
---

# CopyLoop: Fast Looped Transformer Decoding via Context Copying and Shallow Screening

## At a glance

Looped Transformers increase computational capacity through the repeated application of shared blocks, but their recurrent depth makes conventional full-depth decoding expensive. While existing speculative pipelines acc…

## Key points

- Looped Transformers increase computational capacity through the repeated application of shared blocks, but their recurrent depth makes conventional full-depth decoding expensive.
- While existing speculative pipelines accelerate generation by processing consecutive token positions at varying depths, they are typically limited by sequential decoding overheads.
- We introduce CopyLoop, a training-free decoding algorithm that combines low-cost context-copy proposals with shallow screening and continuous recurrent scheduling.

## Full text

Looped Transformers increase computational capacity through the repeated application of shared blocks, but their recurrent depth makes conventional full-depth decoding expensive. While existing speculative pipelines accelerate generation by processing consecutive token positions at varying depths, they are typically limited by sequential decoding overheads. We introduce CopyLoop, a training-free decoding algorithm that combines low-cost context-copy proposals with shallow screening and continuous recurrent scheduling. Rather than step-by-step generation, CopyLoop proposes multi-token suffixes from repeated context. It then evaluates them early at a shallow recurrent depth to efficiently filter out incorrect branches, processing newer blocks concurrently while earlier blocks are still progressing toward full-depth verification. We evaluate CopyLoop across six model checkpoints from the Ouro, Raven, and Huginn families, covering code generation and multi-step mathematical reasoning. CopyLoop achieves decoding speedups of 2.98x--10.70x over full-depth autoregressive decoding on Base models and 2.74x--3.52x on Thinking models, consistently outperforming or closely matching all competing baselines across evaluation tasks. Furthermore, ablation studies confirm a strong synergy between shallow screening and continuous scheduling, while our analyses show that CopyLoop requires only lightweight model calibration and yields greater speedups on longer generations with reusable contexts.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_306c447a67694f4480c0e680