mirror of
https://github.com/hanzoai/papers.git
synced 2026-10-29 07:34:31 +00:00
zeekay
0d1d121451
Fix shared/ include paths to ../shared/ for 55 subdirectory papers so
that hanzocover.sty and lstlang.tex resolve when building from each
paper's own directory (the build pattern used by latexmk per paper).
Build fixes:
- hanzo-operative/sections/03-screen-understanding-pipeline.tex: move
\COMMENT from after \FOR{...} onto the \STATE line so the algorithmic
env doesn't mis-parse.
- hanzo-network-architecture (abstract + sections 04, 12, 13): escape
unicode mu (U+03BC) as $\mu$ for pdflatex.
- ntt-transform/sections/08-benchmark-results.tex: same mu escape.
All 55 Tier-2/3 papers now build clean with latexmk -pdf.
68 lines
3.1 KiB
TeX
68 lines
3.1 KiB
TeX
% Hanzo Platform: PaaS for AI-Native Application Deployment
|
|
\documentclass[11pt,twocolumn]{article}
|
|
\usepackage{../shared/hanzocover}
|
|
\usepackage{amsmath, amssymb, amsthm}
|
|
\usepackage{mathtools}
|
|
\usepackage{bm}
|
|
\usepackage{graphicx}
|
|
\usepackage{booktabs}
|
|
\usepackage{hyperref}
|
|
\usepackage{enumitem}
|
|
\usepackage{algorithm}
|
|
\usepackage{algpseudocode}
|
|
\usepackage{xcolor}
|
|
\usepackage{float}
|
|
\usepackage{subcaption}
|
|
\usepackage{multirow}
|
|
\usepackage{listings}
|
|
\input{../shared/lstlang}
|
|
\usepackage[numbers]{natbib}
|
|
\hypersetup{colorlinks=true,linkcolor=black,citecolor=blue,urlcolor=blue}
|
|
|
|
\newtheorem{theorem}{Theorem}
|
|
\newtheorem{lemma}[theorem]{Lemma}
|
|
\newtheorem{definition}{Definition}
|
|
\newtheorem{proposition}[theorem]{Proposition}
|
|
|
|
\lstset{
|
|
basicstyle=\ttfamily\small,
|
|
breaklines=true,
|
|
frame=single,
|
|
numbers=left,
|
|
numberstyle=\tiny,
|
|
xleftmargin=2em,
|
|
}
|
|
|
|
\title{Hanzo Platform: PaaS for AI-Native Application Deployment}
|
|
\author{
|
|
Marcus Chen \quad
|
|
David Wei \quad
|
|
Zach Kelling \\
|
|
\textit{Hanzo AI Research} \\
|
|
\texttt{[email protected]}
|
|
}
|
|
\date{December 2021}
|
|
|
|
\begin{document}
|
|
|
|
\hanzocoverpage
|
|
|
|
\begin{abstract}
|
|
We present \textbf{Hanzo Platform}, a Platform-as-a-Service (PaaS) designed specifically for deploying AI-native applications---systems that incorporate LLM inference, vector databases, model serving, and GPU compute as first-class primitives. Traditional PaaS offerings (Heroku, Vercel, Railway) treat AI workloads as an afterthought, requiring manual configuration of GPU instances, model serving infrastructure, and inference APIs. Hanzo Platform introduces three innovations: (i) an \emph{AI-aware build system} that automatically detects model dependencies, optimizes container images for inference workloads, and configures GPU scheduling, reducing deployment time for AI applications from hours to minutes; (ii) a \emph{unified resource model} that treats LLM API access, vector database provisioning, GPU allocation, and traditional compute/storage as a single resource plane with declarative configuration; and (iii) a \emph{cost-optimized autoscaler} that dynamically scales inference endpoints based on token throughput, request latency, and cost budgets, achieving 38\% lower inference cost compared to static provisioning. We evaluate Hanzo Platform against five competing PaaS offerings on deployment time, operational cost, scaling responsiveness, and developer experience across 15 representative AI application architectures. Results demonstrate 4.7x faster deployment, 38\% lower cost, and 2.1x better scaling responsiveness. Production deployment data from 14 months of operation with 2,800+ applications and 47M inference requests validates the approach.
|
|
\end{abstract}
|
|
|
|
\input{sections/01-introduction}
|
|
\input{sections/02-architecture}
|
|
\input{sections/03-ai-aware-build-system}
|
|
\input{sections/04-unified-resource-model}
|
|
\input{sections/05-cost-optimized-autoscaler}
|
|
\input{sections/06-developer-experience}
|
|
\input{sections/07-security-and-compliance}
|
|
\input{sections/08-evaluation}
|
|
\input{sections/09-production-deployment}
|
|
\input{sections/10-related-work}
|
|
\input{sections/11-discussion}
|
|
\input{sections/99-conclusion}
|
|
|
|
\end{document}
|