mirror of
https://github.com/hanzoai/papers.git
synced 2026-10-22 17:20:17 +00:00
zeekay
f9db4f5737
Bylines, paper authorship, correspondence addresses and the founder blurb under each post named an individual. They name the organisation that published the work instead: the brand on posts and pages, the research byline on papers, and research@ in place of a personal address. Both are what these repositories already used elsewhere — nothing new was coined. Co-authors are untouched. Where a paper named someone else alongside, they stay named; only the one byline changed. Bibliography entries, running heads, PDF metadata and affiliations move with it, so a citation still resolves to whoever the cited paper names. Sample and fixture data uses the placeholder name these tests already use elsewhere, so no assertion changed meaning.
64 lines
2.5 KiB
TeX
64 lines
2.5 KiB
TeX
\documentclass[11pt,twocolumn]{article}
|
|
\usepackage{../shared/hanzocover}
|
|
\usepackage[utf8]{inputenc}
|
|
\usepackage{amsmath,amssymb,amsthm}
|
|
\usepackage{graphicx}
|
|
\usepackage{booktabs}
|
|
\usepackage{algorithm}
|
|
\usepackage{algorithmic}
|
|
\usepackage{hyperref}
|
|
\usepackage{xcolor}
|
|
\usepackage{listings}
|
|
\input{../shared/lstlang}
|
|
\usepackage{tikz}
|
|
\usetikzlibrary{shapes,arrows,positioning,fit,backgrounds}
|
|
|
|
\definecolor{codegreen}{rgb}{0,0.6,0}
|
|
\definecolor{codegray}{rgb}{0.5,0.5,0.5}
|
|
\definecolor{codepurple}{rgb}{0.58,0,0.82}
|
|
\definecolor{backcolour}{rgb}{0.95,0.95,0.92}
|
|
|
|
\lstdefinestyle{mystyle}{
|
|
backgroundcolor=\color{backcolour},
|
|
commentstyle=\color{codegreen},
|
|
keywordstyle=\color{codepurple},
|
|
numberstyle=\tiny\color{codegray},
|
|
stringstyle=\color{codegreen},
|
|
basicstyle=\ttfamily\footnotesize,
|
|
breakatwhitespace=false,
|
|
breaklines=true,
|
|
captionpos=b,
|
|
keepspaces=true,
|
|
numbers=left,
|
|
numbersep=5pt,
|
|
showspaces=false,
|
|
showstringspaces=false,
|
|
showtabs=false,
|
|
tabsize=2
|
|
}
|
|
\lstset{style=mystyle}
|
|
|
|
\title{Hanzo ML: A Production ML Framework for Model Serving, A/B Testing, and Feature Stores}
|
|
\author{Antje Worring, Hanzo AI Research\thanks{[email protected]} \\ \textit{Hanzo Industries \quad Lux Industries \quad Zoo Labs Foundation} \\ \texttt{[email protected]}}
|
|
\date{July 2022}
|
|
|
|
\begin{document}
|
|
|
|
\hanzocoverpage
|
|
|
|
\begin{abstract}
|
|
We present Hanzo ML, a Rust-based machine learning framework optimized for serverless deployment and fully homomorphic encryption (FHE) compatibility. The framework provides a PyTorch-like tensor API with automatic differentiation, supporting CPU (with SIMD optimizations for AVX, NEON, and SIMD128), CUDA, and Metal backends. Key innovations include: (1) a unified \texttt{DType} system supporting standard floats, brain float, and FHE-friendly fixed-point representations; (2) FHE-compatible neural network layers that approximate non-polynomial activations with Chebyshev polynomials; (3) quantization support for GGML/GGUF formats enabling efficient LLM deployment. Benchmarks demonstrate 3.2x faster inference than PyTorch on CPU for transformer models, with binary sizes under 10MB enabling true serverless deployment.
|
|
\end{abstract}
|
|
|
|
\input{sections/01-introduction}
|
|
\input{sections/02-background}
|
|
\input{sections/03-architecture}
|
|
\input{sections/04-fhe-compatible-layers}
|
|
\input{sections/05-quantization-support}
|
|
\input{sections/06-evaluation}
|
|
\input{sections/07-transformers-library}
|
|
\input{sections/08-discussion}
|
|
\input{sections/99-conclusion}
|
|
|
|
\end{document}
|