Brief
Hypertensor is based on the following papers, whitepapers, projects, and research.
Paper
Paper
Paper
Paper
Paper
Paper
Paper
Paper
Introduction
Hypertensor is a decentralized machine learning network. It is a peer-to-peer network for machine learning models to ensure the future of Artificial Intelligence is available to everyone across the globe without the need of relying on and the control of centralized data centers like OpenAI.
Summary
Hypertensor gives users the ability to contribute computational power towards machine learning models in a decentralized network. In return for contributing computational power, incentives are rewarded. Incentives are based on a peer-to-peer ranking system that is based on throughput, model hosting, availability, ping, stake balance, and other parameters. Hypertensor is built incorporating Hivemind, a library for decentralized deep learning across the Internet. Hivemind is built using libp2p, a well established p2p library that many blockchains use, such as Ethereum and many others. We put forward the concept of incentivizing the contribution of computational power towards machine learning models by integrating blockchain technology that will validate throughput, secure the models, and act as a payments & transactions infrastructure.
Overview
The network is similar to Bittorent, a peer-to-peer network across the internet for downloading files. We use the Kademlia algorithm which is a Distributed Hash Table that gives the ability to connect many computers into one network. Per model, each peer contributes a piece of the model and combines with other peers in the network doing the same. Users can access this network to train, fine-tine, and sample models as they would with their favorite platforms like Pytorch and Huggingface.
Subnets
Hypertensor subnets are decentralized peer-to-peer networks that contribute to the overall network and AI economy, enabling the serving of AI models across thousands of servers and fostering the decentralization of AI. This system operates similarly to a blockchain, but instead of computing blocks of transactions, it computes AI models. Clients and end-users can utilize these models just like any other, including LLAMA2, ChatGPT, image diffusers, and more. This decentralized approach enhances scalability, resilience, and innovation in AI applications, ensuring a robust and diverse ecosystem for AI development and deployment.
Hypertensor employs an optimistic approach to inference validation. Similar to how Ethereum blockchain validators verify transactions and other validators, subnet node validators in Hypertensor subnets perform inference validation on other nodes to ensure they are honestly hosting the models. If a node is found to be dishonest, a consensus must be reached to remove that node from the subnet, resulting in various penalties, including slashing.
Blockchain
Hypertensor will utilize Substrate to build our blockchain. It will act as a validation and incentives network. At at a high-level, the blockchain will be responsible for validating the peer-to-peer network of those hosting models, and be responsible for releasing incentives to peers based on the consensus of measuring computational data from the peer-to-peer network.
The blockchains consensus mechanism will be using a nominated proof-of-stake (NPoS) mechanism. Node operators and validators must stake a minimum required balance and will receive rewards per block to be split between validators and peers hosting models.
Conclusion
While decentralized machine learning is a technological breakthrough, the way it shapes the human race is going to be dependent on how the technology is cultivated. As we have seen in finance, centralization has brought upon a high level of corruption that without blockchain technology would be improbable to defeat.
Acting now -- for the betterment of the human race to ensure the use of AI doesn't establish majoritively through centralized data centers that can control the training, or what consumers can train, is imperative. This is the moment we will look back on in history and be overjoyed that companies like Hypertensor and its competitors began the journey of decentralizing AI.