1% ============================================================================2% UQO Working Paper No. 33 — TITAN-MoE3% ============================================================================4\documentclass[12pt,letterpaper]{article}5\usepackage[utf8]{inputenc}6\usepackage[T1]{fontenc}7\usepackage[english]{babel}8\usepackage[letterpaper,top=1in,bottom=1in,left=1in,right=1in]{geometry}9\usepackage{mathptmx}10\usepackage{microtype}11\usepackage{setspace}12\onehalfspacing13\usepackage{amsmath,amssymb,amsthm}14\usepackage{booktabs}15\usepackage{tabularx}16\usepackage{multirow}17\usepackage{threeparttable}18\usepackage{siunitx}19\sisetup{input-symbols = {()}, table-format = -1.3, parse-numbers = false}20\usepackage{graphicx}21\usepackage{subcaption}22\usepackage{float}23\usepackage[dvipsnames]{xcolor}24\usepackage[colorlinks=true,linkcolor=NavyBlue,citecolor=NavyBlue,urlcolor=NavyBlue,breaklinks=true,25 pdftitle={TITAN-MoE: Topology-Aware and Predictive Expert Orchestration for Distributed Mixture-of-Experts Inference on Apple Silicon},26 pdfauthor={Simon-Pierre Boucher}]{hyperref}27\usepackage[authoryear,round,semicolon]{natbib}28\bibliographystyle{aer}29\usepackage[font=small,labelfont=bf,labelsep=period,justification=justified,singlelinecheck=false]{caption}30\usepackage{fancyhdr}31\pagestyle{fancy}32\fancyhf{}33\renewcommand{\headrulewidth}{0pt}34\fancyfoot[C]{\thepage}35\usepackage{titlesec}36\titleformat{\section}{\large\bfseries}{\thesection.}{0.5em}{}37\titleformat{\subsection}{\normalsize\bfseries}{\thesubsection.}{0.5em}{}38\titleformat{\subsubsection}{\normalsize\itshape}{\thesubsubsection.}{0.5em}{}39\usepackage[toc,page]{appendix}40\usepackage{enumitem}41\setlist{nosep,leftmargin=*}42\usepackage{algorithm}43\usepackage{algpseudocode}44\usepackage{tikz}45\usetikzlibrary{arrows.meta,positioning,shapes.geometric,fit,calc}46\usepackage{array}47\newcolumntype{L}[1]{>{\raggedright\arraybackslash}p{#1}}48\newcolumntype{R}[1]{>{\raggedleft\arraybackslash}p{#1}}49\graphicspath{{figures/}}5051% --- theorem environments (shared counter, numbered by section) ---52\theoremstyle{plain}53\newtheorem{theorem}{Theorem}[section]54\newtheorem{proposition}[theorem]{Proposition}55\newtheorem{lemma}[theorem]{Lemma}56\newtheorem{corollary}[theorem]{Corollary}57\theoremstyle{definition}58\newtheorem{definition}[theorem]{Definition}59\newtheorem{assumption}[theorem]{Assumption}60\theoremstyle{remark}61\newtheorem{remark}[theorem]{Remark}62\newtheorem{observation}[theorem]{Observation}6364% --- macros ---65\newcommand{\E}{\mathcal{E}}66\newcommand{\Lset}{\mathcal{L}}67\newcommand{\Nset}{\mathcal{N}}68\newcommand{\R}{\mathbb{R}}69\newcommand{\Ex}{\mathbb{E}}70\newcommand{\ind}{\mathbb{1}}71\newcommand{\tps}{tok/s}72\newcommand{\code}[1]{\texttt{#1}}73\newcommand{\titan}{TITAN-MoE}74\newcommand{\gatherqmm}{\code{gather\_qmm}}7576% ============================================================================77% METADATA78% ============================================================================79\newcommand{\WPnumber}{33}80\newcommand{\WPtitle}{TITAN-MoE: Topology-Aware and Predictive Expert Orchestration for Distributed Mixture-of-Experts Inference on Apple Silicon}81\newcommand{\WPsubtitle}{A Measured Systems Study on a Heterogeneous Mac Cluster}82\newcommand{\WPdate}{November 2026}83\newcommand{\WPversion}{0.1}84\input{sections/abstract}85\newcommand{\WPkeywords}{mixture of experts, inference systems, Apple Silicon, expert offloading, predictive prefetching, distributed inference, energy efficiency}86\newcommand{\WPjel}{C63, C88, L86, O33}87\newcommand{\WPauthor}{Simon-Pierre Boucher}88\newcommand{\WPaffiliation}{Département des sciences administratives\\Université du Québec en Outaouais}89\newcommand{\WPemail}{simon-pierre.boucher@uqo.ca}90\newcommand{\WPaddress}{Gatineau -- Pavillon Alexandre-Taché\\283, boulevard Alexandre-Taché\\Gatineau, Québec, Canada J9A 1L8}9192\begin{document}93\input{sections/titlepage}94\input{sections/introduction}95\input{sections/background}96\input{sections/related}97\input{sections/formulation}98\input{sections/architecture}99\input{sections/algorithms}100\input{sections/implementation}101\input{sections/setup}102\input{sections/results}103\input{sections/ablations}104\input{sections/robustness}105\input{sections/limitations}106\input{sections/conclusion}107\newpage108\bibliography{references}109\input{sections/appendix}110\end{document}111