Route Experts by Sequence, not by Token

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wen, Tiansheng, Wang, Yifei, Feng, Aosong, Ma, Long, Liu, Xinyang, Wang, Yifan, Guo, Lixuan, Chen, Bo, Jegelka, Stefanie, You, Chenyu
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915894091841536
author Wen, Tiansheng
Wang, Yifei
Feng, Aosong
Ma, Long
Liu, Xinyang
Wang, Yifan
Guo, Lixuan
Chen, Bo
Jegelka, Stefanie
You, Chenyu
author_facet Wen, Tiansheng
Wang, Yifei
Feng, Aosong
Ma, Long
Liu, Xinyang
Wang, Yifan
Guo, Lixuan
Chen, Bo
Jegelka, Stefanie
You, Chenyu
contents Mixture-of-Experts (MoE) architectures scale large language models (LLMs) by activating only a subset of experts per token, but the standard TopK routing assigns the same fixed number of experts to all tokens, ignoring their varying complexity. Prior adaptive routing methods introduce additional modules and hyperparameters, often requiring costly retraining from scratch. We propose Sequence-level TopK (SeqTopK), a minimal modification that shifts the expert budget from the token level to the sequence level. By selecting the top $T \cdot K$ experts across all $T$ tokens, SeqTopK enables end-to-end learned dynamic allocation -- assigning more experts to difficult tokens and fewer to easy ones -- while preserving the same overall budget. SeqTopK requires only a few lines of code, adds less than 1% overhead, and remains fully compatible with pretrained MoE models. Experiments across math, coding, law, and writing show consistent improvements over TopK and prior parameter-free adaptive methods, with gains that become substantially larger under higher sparsity (up to 16.9%). These results highlight SeqTopK as a simple, efficient, and scalable routing strategy, particularly well-suited for the extreme sparsity regimes of next-generation LLMs. Code is available at https://github.com/Y-Research-SBU/SeqTopK.
format Preprint
id arxiv_https___arxiv_org_abs_2511_06494
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Route Experts by Sequence, not by Token
Wen, Tiansheng
Wang, Yifei
Feng, Aosong
Ma, Long
Liu, Xinyang
Wang, Yifan
Guo, Lixuan
Chen, Bo
Jegelka, Stefanie
You, Chenyu
Machine Learning
Artificial Intelligence
Information Theory
Mixture-of-Experts (MoE) architectures scale large language models (LLMs) by activating only a subset of experts per token, but the standard TopK routing assigns the same fixed number of experts to all tokens, ignoring their varying complexity. Prior adaptive routing methods introduce additional modules and hyperparameters, often requiring costly retraining from scratch. We propose Sequence-level TopK (SeqTopK), a minimal modification that shifts the expert budget from the token level to the sequence level. By selecting the top $T \cdot K$ experts across all $T$ tokens, SeqTopK enables end-to-end learned dynamic allocation -- assigning more experts to difficult tokens and fewer to easy ones -- while preserving the same overall budget. SeqTopK requires only a few lines of code, adds less than 1% overhead, and remains fully compatible with pretrained MoE models. Experiments across math, coding, law, and writing show consistent improvements over TopK and prior parameter-free adaptive methods, with gains that become substantially larger under higher sparsity (up to 16.9%). These results highlight SeqTopK as a simple, efficient, and scalable routing strategy, particularly well-suited for the extreme sparsity regimes of next-generation LLMs. Code is available at https://github.com/Y-Research-SBU/SeqTopK.
title Route Experts by Sequence, not by Token
topic Machine Learning
Artificial Intelligence
Information Theory
url https://arxiv.org/abs/2511.06494