papers/gym-training-platform/sections/09-integration-with-zoo-ecosystem.tex

\section{Integration with Zoo Ecosystem}


Gym is part of the broader Zoo ecosystem for decentralized AI. Gym training jobs that target a model with a sovereign per-LLM chain~\cite{zoo-per-llm-chains} commit gradients, hyperparameters, and ablations directly into chain state via the chain's native instructions, settling on Lux Quasar 3.0 (Lux LP-020) and inheriting GPU scheduling from Quasar\textsc{gpu} (Lux LP-132). For workloads that need confidential compute, attestations anchor on Lux A-Chain (Lux LP-134); for cloud capacity sourcing, operators federate through Lux Cloud (Lux LP-136). The remainder of this section describes the existing component integrations:

\subsection{Experience Ledger}

Training-Free GRPO experiences are stored in the \textbf{Experience Ledger}:
\begin{itemize}
\item Content-addressable storage (IPFS/Arweave)
\item Merkle tree with on-chain root hash
\item Cryptographic verification of all experiences
\item DAO governance for quality curation
\end{itemize}

\subsection{Decentralized Sequential Optimization (DSO)}

DSO extends Training-Free GRPO to multi-agent systems:
\begin{itemize}
\item Agents collaboratively build shared experience libraries
\item Cross-pollination of domain expertise
\item Economic incentives for high-quality contributions
\item Privacy-preserving experience sharing (zk-SNARKs)
\end{itemize}

\subsection{Hanzo Network Integration}

Gym models deploy to Hanzo Network for inference:
\begin{itemize}
\item GPU compute nodes run frozen base models
\item Experience libraries injected as context
\item Proof-of-inference for verification
\item API-compatible with OpenAI format
\end{itemize}

\subsection{Zoo.fund Platform}

Zoo.fund crowdfunds AI research projects using Gym:
\begin{itemize}
\item Researchers propose model training campaigns
\item Community funds via KEEPER token
\item Trained models released open-source
\item Contributors earn inference credits
\end{itemize}