Technology Architecture

How does this work?

HEXS consists of a factory smart contract as an entrance to an Agent creation. Smart Contract (Factory) deploys agent smart contracts and related instances for each model being fine-tuned by developers. These agents' smart contracts set up a Decentralized Autonomous Organization (DAO) contract with them to hold the future owners of this model. The owner joins the agent's DAO upon initialization, and new users of the fine-tuned model automatically become members.

Agent Contract holds an AI Agent Ledger that manages records of training shares and other metadata throughout the fine-tuning process. Pre-trained models sourced from open-source hubs like Hugging Face serve as the foundation and are fine-tuned using leased computation resources, such as GPUs from the Akash Blockchain.

Subsequently, the Agent playground is utilized to run and test the fine-tuned model's performance across various parameters and features, comparing it with other models on the marketplace. Developers then publish their research and deploy their fine-tuned models on the platform's Marketplace for monetization, enabling end-users to access and utilize them via API keys.

When a fine-tuned model is deployed, the AI Agent Ledger updates training shares and other values within the DAO accordingly with the distributed training formula. After deployment, end-users can purchase or access free APIs (per the model settings) to integrate these fine-tuned models into their applications, leveraging their specialized capabilities. This process facilitates developers in sharing and monetizing their work while providing valuable AI resources to end-users.

HEXS consists of the following,

Hexs Smart Contract(Factory):

The Hexs Smart Contracts(Factory) deploys agent smart contracts and other instances for every model which the developer is fine-tuning. The pre-trained models after getting finetuned are known as agents.

Agent Smart Contract: The agent smart contract initializes an AI agent Ledger and a DAO. The owner becomes the member of that agent's DAO at its initialization and each new user of that finetuned model keeps on becoming its member.

AI Agent Ledger: AI Agent Ledger is responsible for recording training shares and other metadata such as model trends, ratings, etc. These recorded data are later shared with the DAO.

Open Source Hubs: Pre-trained models are leveraged from open-source hubs like Hugging Face. These pre-trained models will be finetuned using leased computation resources like GPUs from the Akash Blockchain.

Akash Blockchain: Akash Blockchain offers computation power and resources which is an inexpensive and feasible supercloud network to train AI models.

Agent Playground: Agent Playground tests the finetuned model's working on various parameters to compare its benchmark scores and features with other models on the marketplace.

Marketplace: Marketplace allows developers to publish their research and deploy their work like the finetuned model on the platform so that they can be monetized and the end-users can use them. When a fine-tuned model is deployed in the marketplace, the AI Agent Ledger updates the training shares and other values in the DAO after deployment end-users can purchase APIs of these finetuned models to build applications using them.

User flow,

The process begins with a developer downloading a model (fine-tuned or pre-trained) from the marketplace. Then the developer proceeds to fine-tune the downloaded model to a preferred niche using leased computational resources from Akash Blockchain.

Later, the developer deploys the model on the marketplace so that end-users can use it for a fee which would generate revenue or sell the model to interested buyers altogether.

Distributed training formula,

The formula for calculating Distributed Training on Hexs incorporates user engagement, ratings, and model popularity, averaging these factors to determine the overall score.

Who'll be using HEXS,

LLM Developers:

LLM developers refine pre-trained Language Model Models (LLMs) or agents within the fine-tuned models space on the Hexs platform.

Application Developers:

Application developers utilize inexpensive APIs provided by the Hexs platform to develop their applications.

Users:

Users generate inexpensive inferencing from the models deployed on the Hexs marketplace using credits.