Motif-Upcycling: Structure-Preserving Adaptation of Transformer Models

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Author: Lirov, Kharki
Format: Recurso digital
Language:English
Published: Zenodo 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866901630973116416
author Lirov, Kharki
author_facet Lirov, Kharki
contents <p>We introduce Motif-Upcycling, a structure-preserving framework for adapting pretrained Transformer models. The key idea is that common feed-forward modules, including SwiGLU FFNs, can be exactly factorized along their intermediate channel axis into motif-aligned components. With neutral routing, the factorized module computes the same function as the original pretrained block at initialization. We further introduce Scale-Aware Residual Control (SARC), an identity-preserving control motif that modulates the magnitude of trainable residual interventions relative to the residual stream. We also propose Emergence as Coupled Budget Thresholds (ECBT), a conditional model showing that apparent capability cliffs can arise from multiplicatively coupled motif effectiveness curves under uniform budget allocation.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19840136
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Motif-Upcycling: Structure-Preserving Adaptation of Transformer Models
Lirov, Kharki
computer science
Machine learning
natural language process
deep lear
neural network arch
<p>We introduce Motif-Upcycling, a structure-preserving framework for adapting pretrained Transformer models. The key idea is that common feed-forward modules, including SwiGLU FFNs, can be exactly factorized along their intermediate channel axis into motif-aligned components. With neutral routing, the factorized module computes the same function as the original pretrained block at initialization. We further introduce Scale-Aware Residual Control (SARC), an identity-preserving control motif that modulates the magnitude of trainable residual interventions relative to the residual stream. We also propose Emergence as Coupled Budget Thresholds (ECBT), a conditional model showing that apparent capability cliffs can arise from multiplicatively coupled motif effectiveness curves under uniform budget allocation.</p>
title Motif-Upcycling: Structure-Preserving Adaptation of Transformer Models
topic computer science
Machine learning
natural language process
deep lear
neural network arch
url https://doi.org/10.5281/zenodo.19840136