---
title: 'Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity'
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
license_note: 'paper is arXiv non-exclusive distribution license (verified on arXiv abs page 2026-11-12); this pointer note is our own words and is CC BY 4.0; we link to the paper rather than reproduce it'
source: https://iinventedlight.com/library/notes/2101-03961-switch-transformers.md
source_license: 'arXiv non-exclusive distribution license'
attribution: 'Pointer by iinventedlight (AI-assisted). Source: William Fedus, Barret Zoph, Noam Shazeer, Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity (arXiv non-exclusive distribution license). This note links to the paper rather than reproducing it.'
author: 'iinventedlight (AI-assisted)'
date: 2026-11-12
status: published
approved_by: "Cole Withington, 2026-11-12 14:44 PT, Librarian chat widget t51s316: \"Yes, publish all 100 second-list arXiv notes\""
checked_on: 2026-11-12
as_of: 2026-11-12
confidence: medium
untested: 'Pointer note only. Paper not reproduced. Abs page and license checked 2026-11-12.'
category: arxiv
topic_area: 'ai-research: efficiency'
tags: arxiv, efficiency, mixture-of-experts, sparsity
arxiv_id: 2101.03961
alias: 2101.03961-switch-transformers
type: pointer
format: pointer
sources:
  - id: S1
    title: 'Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity'
    publisher: arXiv
    authors: 'William Fedus, Barret Zoph, Noam Shazeer'
    url: https://arxiv.org/abs/2101.03961
    version_read: 'v3 (16 Jun 2022)'
    accessed: 2026-11-12
    license: 'arXiv non-exclusive distribution license'
---
# Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity

## Summary
[Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://arxiv.org/abs/2101.03961) (2021) introduced Switch Transformers, a simplified sparse mixture-of-experts design meant to scale transformers toward a trillion parameters at a roughly constant computational cost per example. It made large MoE models much easier to train than earlier, more complex routing schemes.

What it introduced: a sparse expert layer that routes each token to a single expert, trading a huge parameter count for a simpler routing rule and more stable training than some prior MoE systems.

What it led to: later open mixture-of-experts language models, including Mixtral. Live note: [2401-04088-mixtral-of-experts](https://iinventedlight.com/library/notes/2401-04088-mixtral-of-experts/). Sparse expert models remain one of the main alternatives to training a single dense stack at the same parameter count.

Use it when you want the simplified MoE design that preceded later open expert models. Where to read it: https://arxiv.org/abs/2101.03961 (v3 (16 Jun 2022)).

The paper is posted on arXiv under the arXiv non-exclusive distribution license. This note links to it rather than reproducing it. The note itself is CC BY 4.0.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Pointer by iinventedlight (AI-assisted). Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it.

---

Agents: [/agents.md](/agents.md) | [/.well-known/agent-library.json](/.well-known/agent-library.json).
