Pith. sign in

REVIEW 5 cited by

Let the Expert Stick to His Last: Expert-Specialized Fine-Tuning for Sparse Architectural Large Language Models

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2407.01906 v2 pith:QIUKJONV submitted 2024-07-02 cs.CL cs.AIcs.LG

Let the Expert Stick to His Last: Expert-Specialized Fine-Tuning for Sparse Architectural Large Language Models

classification cs.CL cs.AIcs.LG
keywords expertsfine-tuningllmspefttasksexpert-specializedmodelsactivated
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Parameter-efficient fine-tuning (PEFT) is crucial for customizing Large Language Models (LLMs) with constrained resources. Although there have been various PEFT methods for dense-architecture LLMs, PEFT for sparse-architecture LLMs is still underexplored. In this work, we study the PEFT method for LLMs with the Mixture-of-Experts (MoE) architecture and the contents of this work are mainly threefold: (1) We investigate the dispersion degree of the activated experts in customized tasks, and found that the routing distribution for a specific task tends to be highly concentrated, while the distribution of activated experts varies significantly across different tasks. (2) We propose Expert-Specialized Fine-Tuning, or ESFT, which tunes the experts most relevant to downstream tasks while freezing the other experts and modules; experimental results demonstrate that our method not only improves the tuning efficiency, but also matches or even surpasses the performance of full-parameter fine-tuning. (3) We further analyze the impact of the MoE architecture on expert-specialized fine-tuning. We find that MoE models with finer-grained experts are more advantageous in selecting the combination of experts that are most relevant to downstream tasks, thereby enhancing both the training efficiency and effectiveness. Our code is available at https://github.com/deepseek-ai/ESFT.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Preserving Long-Tailed Expert Information in Mixture-of-Experts Tuning

    cs.LG 2026-04 unverdicted novelty 7.0

    A new SFT framework for MoE models combines bias-driven sparsification with gated condenser experts to retain long-tailed expert information, outperforming DenseMixer and ESFT by over 2.5% on math reasoning and common...

  2. MoE$^2$-LoRA: When MoE Models Meet MoE-style Low-Rank Adaptation

    cs.CL 2026-07 conditional novelty 6.0

    Routing LoRA adapters with the frozen base router's logits plus a shared cross-layer adapter pool gives the best PEFT accuracy and retention on three MoE backbones.

  3. Behavioral and Representational Evidence of Binomial Ordering Preferences in Large Language Models

    cs.CL 2026-06 unverdicted novelty 6.0

    LLMs recover dominant binomial orders from corpora but align less closely with exact preference distributions, with preference strength partially encoded in middle-to-late layers and manipulable via steering.

  4. Unveiling Language Routing Isolation in Multilingual MoE Models for Interpretable Subnetwork Adaptation

    cs.CL 2026-04 conditional novelty 6.0

    MoE models isolate high- vs low-resource languages in expert routing; training only the isolated target subnetwork (RISE) lifts low-resource F1 by up to ~11 points with little cross-lingual loss.

  5. Application-Driven Pedagogical Knowledge Optimization of Open-Source LLMs via Reinforcement Learning and Supervised Fine-Tuning

    cs.CL 2026-04 unverdicted novelty 4.0

    EduQwen 32B models optimized via RL then SFT set new SOTA on the Cross-Domain Pedagogical Knowledge Benchmark and surpass Gemini-3 Pro.