Automatic Expert Discovery in LLM Upcycling via Sparse Interpolated Mixture-of-Experts

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Shengzhuang, Wei, Ying, Schwarz, Jonathan Richard
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915343736242176
author Chen, Shengzhuang
Wei, Ying
Schwarz, Jonathan Richard
author_facet Chen, Shengzhuang
Wei, Ying
Schwarz, Jonathan Richard
contents We present Sparse Interpolated Mixture-of-Experts (SIMoE) instruction-tuning, an end-to-end algorithm designed to fine-tune a dense pre-trained Large Language Model (LLM) into a MoE-style model that possesses capabilities in multiple specialized domains. During instruction-tuning, SIMoE automatically identifies multiple specialized experts under a specified sparsity constraint, with each expert representing a structurally sparse subset of the seed LLM's parameters that correspond to domain-specific knowledge within the data. SIMoE simultaneously learns an input-dependent expert merging strategy via a router network, leveraging rich cross-expert knowledge for superior downstream generalization that surpasses existing baselines. Empirically, SIMoE consistently achieves state-of-the-art performance on common instruction-tuning benchmarks while maintaining an optimal performance-compute trade-off compared to all baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2506_12597
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Automatic Expert Discovery in LLM Upcycling via Sparse Interpolated Mixture-of-Experts
Chen, Shengzhuang
Wei, Ying
Schwarz, Jonathan Richard
Machine Learning
We present Sparse Interpolated Mixture-of-Experts (SIMoE) instruction-tuning, an end-to-end algorithm designed to fine-tune a dense pre-trained Large Language Model (LLM) into a MoE-style model that possesses capabilities in multiple specialized domains. During instruction-tuning, SIMoE automatically identifies multiple specialized experts under a specified sparsity constraint, with each expert representing a structurally sparse subset of the seed LLM's parameters that correspond to domain-specific knowledge within the data. SIMoE simultaneously learns an input-dependent expert merging strategy via a router network, leveraging rich cross-expert knowledge for superior downstream generalization that surpasses existing baselines. Empirically, SIMoE consistently achieves state-of-the-art performance on common instruction-tuning benchmarks while maintaining an optimal performance-compute trade-off compared to all baselines.
title Automatic Expert Discovery in LLM Upcycling via Sparse Interpolated Mixture-of-Experts
topic Machine Learning
url https://arxiv.org/abs/2506.12597