Papers
arxiv:2607.04033

OmniOpt: Taxonomy, Geometry, and Benchmarking of Modern Optimizers

Published on Jul 4
ยท Submitted by
Cheng Tan
on Jul 7
#1 Paper of the day
Authors:
,
,
,
,
,
,
,
,
,
,
,

Abstract

OmniOpt presents a unified framework for optimizer selection in large-scale model training by combining meta-pipeline transformations, norm-constrained linear minimization oracles, and a cross-domain benchmark to systematically analyze optimizer families and their trade-offs across different training objectives and model scales.

Optimizer selection for large-scale model training has become a system-level design decision constrained jointly by compute, memory, tuning budget, and task diversity, yet the landscape of over one hundred methods remains fragmented. We therefore present OmniOpt, a unified survey and benchmark cookbook of optimizers for the research community. OmniOpt rests on four coupled components. First, we treat every optimizer update as a structured transformation through a five-stage meta-pipeline, and show that most methods engage only one or two of these stages. Second, we use norm-constrained linear minimization oracles (LMOs) to unify different optimizers. Third, these two views ground a dual-dimension taxonomy, one dimension assigning each method to a mechanism family and the other recording the measurable training objectives it aims to improve. Fourth, and at the core of this paper, we instantiate the full taxonomy in a unified cross-domain benchmark spanning representative optimizers, model scales, and training regimes from language model pretraining to image classification, systematically analyzing each method family across multiple effect objectives and laying out their trade-offs. OmniOpt thus supplies the research community with an operational coordinate system for selecting optimizers under explicit mechanism and objective assumptions, and charts a direction for the future development of the optimizer community.

Community

Optimizer selection for large scale model training is constrained by compute, memory, tuning budget, and task diversity, yet the landscape of over one hundred methods remains fragmented. We introduce OmniOpt, a unified survey and benchmark cookbook for the research community. OmniOpt has four coupled components. First, we model every optimizer update as a five stage meta pipeline and show that most methods engage only one or two stages. Second, we use norm constrained linear minimization oracles (LMOs) to unify different optimizers. Third, these two views ground a dual dimension taxonomy: one axis assigns each method to a mechanism family, and the other records the measurable training objectives it targets. Fourth, we instantiate this taxonomy in a comprehensive cross domain benchmark covering representative optimizers, model scales, and training regimes from language model pretraining to image classification. We systematically analyze each method family across multiple effect objectives and detail their trade offs. OmniOpt thus provides the community with an operational coordinate system for selecting optimizers under explicit mechanism and objective assumptions, and it charts a clear direction for future optimizer development.

Sign up or log in to comment

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2607.04033 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2607.04033 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2607.04033 in a Space README.md to link it from this page.

Collections including this paper 0

No Collection including this paper

Add this paper to a collection to link it from this page.