OmniOpt: Taxonomy, Geometry, and Benchmarking of Modern Optimizers

2026-07-04

AI Summary

This work addresses the lack of systematic guidance in optimizer selection for large-scale model training, where existing approaches are fragmented and struggle to accommodate diverse computational, memory, and task requirements. The authors propose a five-stage meta-pipeline that unifies the update mechanisms of optimizers and introduce a norm-constrained Linear Minimization Oracle (LMO) to achieve geometric unification. Building on this foundation, they establish a novel two-dimensional taxonomy that jointly considers optimizer families and optimization objectives, along with a comprehensive benchmarking framework spanning multiple domains, model scales, and tasks. Systematic evaluation reveals critical trade-offs among optimizers in terms of performance, robustness, and efficiency, providing a practical coordinate system and empirical basis for both optimizer selection and future algorithm design.

Abstract

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.

Problem

• optimizer selection
• large-scale model training
• benchmarking
• taxonomy
• optimization methods

Innovation

• optimizer taxonomy
• norm-constrained LMO
• meta-pipeline
• cross-domain benchmarking
• training objective trade-offs