---
format: "aidr-story-markdown/v1"
id: "445dd1b17f284a5e09bf932249b8ae3f09fe28cc35db7b349d6105109166b72b"
canonical_url: "https://aidr.today/445dd1b1?lang=en"
title: "Loopback Transformer: Towards a Unified View of Residual Looped Transformers"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-16T00:00:00.000Z"
category: "Research"
topics: ["llm"]
source_urls: ["https://huggingface.co/papers/2611.07226"]
summary: "The Loopback Transformer reduces parameter count by repeatedly applying shared blocks. On ClimbMix, Loopback reaches lower validation loss than standard Transformer baselines with 40-75% of their Transformer parameters at reference scales from 191M to 3B. It also has higher mean accuracy over thirteen downstream tasks at 1.4B and 3B."
---

# Loopback Transformer: Towards a Unified View of Residual Looped Transformers

> [Open the canonical story](<https://aidr.today/445dd1b1?lang=en>)

**Published:** 2026-11-16T00:00:00.000Z
**Category:** Research
**Topics:** llm

## Summary

The Loopback Transformer reduces parameter count by repeatedly applying shared blocks\. On ClimbMix, Loopback reaches lower validation loss than standard Transformer baselines with 40\-75% of their Transformer parameters at reference scales from 191M to 3B\. It also has higher mean accuracy over thirteen downstream tasks at 1\.4B and 3B\.

## Sources

- [Story source](<https://huggingface.co/papers/2611.07226>)

