LoRA Diffusion: Zero-Shot LoRA Synthesis for Diffusion Model Personalization

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Smith, Ethan, Seid, Rami, Hojel, Alberto, Mishra, Paramita, Wu, Jianbo
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866916505659113472
author Smith, Ethan
Seid, Rami
Hojel, Alberto
Mishra, Paramita
Wu, Jianbo
author_facet Smith, Ethan
Seid, Rami
Hojel, Alberto
Mishra, Paramita
Wu, Jianbo
contents Low-Rank Adaptation (LoRA) and other parameter-efficient fine-tuning (PEFT) methods provide low-memory, storage-efficient solutions for personalizing text-to-image models. However, these methods offer little to no improvement in wall-clock training time or the number of steps needed for convergence compared to full model fine-tuning. While PEFT methods assume that shifts in generated distributions (from base to fine-tuned models) can be effectively modeled through weight changes in a low-rank subspace, they fail to leverage knowledge of common use cases, which typically focus on capturing specific styles or identities. Observing that desired outputs often comprise only a small subset of the possible domain covered by LoRA training, we propose reducing the search space by incorporating a prior over regions of interest. We demonstrate that training a hypernetwork model to generate LoRA weights can achieve competitive quality for specific domains while enabling near-instantaneous conditioning on user input, in contrast to traditional training methods that require thousands of steps.
format Preprint
id arxiv_https___arxiv_org_abs_2412_02352
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LoRA Diffusion: Zero-Shot LoRA Synthesis for Diffusion Model Personalization
Smith, Ethan
Seid, Rami
Hojel, Alberto
Mishra, Paramita
Wu, Jianbo
Machine Learning
Low-Rank Adaptation (LoRA) and other parameter-efficient fine-tuning (PEFT) methods provide low-memory, storage-efficient solutions for personalizing text-to-image models. However, these methods offer little to no improvement in wall-clock training time or the number of steps needed for convergence compared to full model fine-tuning. While PEFT methods assume that shifts in generated distributions (from base to fine-tuned models) can be effectively modeled through weight changes in a low-rank subspace, they fail to leverage knowledge of common use cases, which typically focus on capturing specific styles or identities. Observing that desired outputs often comprise only a small subset of the possible domain covered by LoRA training, we propose reducing the search space by incorporating a prior over regions of interest. We demonstrate that training a hypernetwork model to generate LoRA weights can achieve competitive quality for specific domains while enabling near-instantaneous conditioning on user input, in contrast to traditional training methods that require thousands of steps.
title LoRA Diffusion: Zero-Shot LoRA Synthesis for Diffusion Model Personalization
topic Machine Learning
url https://arxiv.org/abs/2412.02352