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.
79 lines
3.0 KiB
TeX
79 lines
3.0 KiB
TeX
% Hanzo Consensus AI: Consensus Mechanisms for Distributed AI
|
|
\documentclass[11pt]{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{multirow}
|
|
|
|
\hypersetup{colorlinks=true,linkcolor=black,citecolor=blue,urlcolor=blue}
|
|
|
|
\newtheorem{theorem}{Theorem}
|
|
\newtheorem{lemma}[theorem]{Lemma}
|
|
\newtheorem{proposition}[theorem]{Proposition}
|
|
\newtheorem{corollary}[theorem]{Corollary}
|
|
\newtheorem{definition}{Definition}
|
|
\newtheorem{remark}{Remark}
|
|
\newtheorem{assumption}{Assumption}
|
|
|
|
\title{HANZO CONSENSUS AI\\[0.3em]
|
|
{\large Consensus Mechanisms for\\Distributed AI Training and Inference}}
|
|
\author{
|
|
Zach Kelling \\
|
|
\textit{Hanzo Industries Inc (Techstars '17)} \\
|
|
\texttt{[email protected]}
|
|
}
|
|
\date{March 2026}
|
|
|
|
\begin{document}
|
|
|
|
\hanzocoverpage
|
|
|
|
% ============================================================================
|
|
% ABSTRACT
|
|
% ============================================================================
|
|
\begin{abstract}
|
|
We present the consensus mechanisms underlying Hanzo ACI (AI Chain
|
|
Infrastructure), a decentralized compute network where GPU operators serve AI
|
|
models and earn HANZO tokens for verified work. Unlike blockchains that achieve
|
|
consensus on transaction ordering, ACI must achieve consensus on computational
|
|
correctness: did a GPU node actually run the claimed model on the claimed input
|
|
and produce the claimed output? We introduce three novel consensus components:
|
|
(i)~Proof of AI Work (PoAI), a probabilistic verification protocol where
|
|
verifier nodes re-execute a random subset of inference requests to detect
|
|
dishonest workers, (ii)~GPU Cluster Consensus, a protocol for coordinating
|
|
distributed training and tensor-parallel inference across untrusted GPU nodes
|
|
with Byzantine fault tolerance, and (iii)~Inference Attestation, an on-chain
|
|
commitment scheme that binds inference outputs to specific model versions, input
|
|
hashes, and GPU node identities. We prove that PoAI detects cheating with
|
|
probability $1 - (1-p)^k$ where $p$ is the sampling rate and $k$ is the number
|
|
of verification rounds, and that the economic cost of cheating exceeds the
|
|
expected profit under the HANZO staking and slashing
|
|
mechanism~\cite{hanzotokenomics}. The system is deployed on Hanzo ACI with 240
|
|
GPU nodes, processing 47M inference requests per month with a measured cheating
|
|
detection rate of 99.97\% for a 5\% sampling rate.
|
|
\end{abstract}
|
|
|
|
% ============================================================================
|
|
|
|
\input{sections/01-introduction}
|
|
\input{sections/02-system-model}
|
|
\input{sections/03-proof-of-ai-work}
|
|
\input{sections/04-gpu-cluster-consensus}
|
|
\input{sections/05-inference-attestation}
|
|
\input{sections/06-distributed-training-consensus}
|
|
\input{sections/07-integration-with-lux-consensus}
|
|
\input{sections/08-production-results}
|
|
\input{sections/09-related-work}
|
|
\input{sections/99-conclusion}
|
|
|
|
\end{document}
|