Blockchain

Introduction

Hypertensor is built using the Substrate framework. Substrate enables developers to build future-proof blockchains optimized for any use case by giving developers the ability to build application logic through the FRAME environment by building modules called pallets.


Repository

GitHub

https://github.com/hypertensor-blockchain/hypertensor


FRAME

Hypertensor (Substrate) FRAME (Framework for Runtime Aggregation of Modularized Entities) pallets are modular modules that are the fundamental building blocks of the Substrate blockchain. They are used to build runtime logic for blockchains and can provide anything from business-level operations, and application logic, to library-level functionality.

Pallets

More pallets are established but not included in the v1.0 logic. Documentation will be released as their integrations are utilized.


General Terminology

Blockchain Validator

A validator for the blockchain

Model Validator

A validator hosting AI models. Peer, Model Peer, and Model Validator are used interchangeably in this documentation.


Devnet Versioning Roadmap

v1.0

Blockchain

The first version of testnet primarily focuses on peer consensus, including peer staking, block rewards, and peer emissions.

Goals

  • Form consensus amongst peers hosting AI models and incentivize them.

Features

  • Each model validator (peer) is required to stake a minimum balance towards the model they are hosting.

  • Each model validator is required to send in data on each peer on each epoch to form consensus.

Possible Feature Introductions On Intermin Versionings Up To v2.0 or for v2.0

  • Introduce a MinModelPeers per model instead of being universal.

  • Introduce a "dishonest" model peer function for peers to vote dishonesty.

  • Update storage value titles to more relevant titles.

  • Update peer consensus logic to include peer states and states logic.

  • Increase the decimal point precision.

  • Add claiming IP addresses per account_id as IP => AccountId and whitelisting other accounts to use to allow multiple model validators per IP address.

  • Allow updating IP address.

  • Create a minimum stake per peer based on the total peers instead of one flat rate.

  • Remove some of the computation-heavy logic in the peer consensus mechanism in favor of increasing the total amount of available subnet space. This decision will be based on the feedback from the community.

  • Adding logic to require consecutive out-of-consensus blocks before the removal of a peer.

  • Update benchmarking code and weights to use dynamic benchmarking for extrinsic functions.

  • Apply a version of the EDA (Excess Distribution Algorithm) to peers per model based on stake balance.

  • Introduce a piecewise curve functionality for the stake rewards vault based on the optimal count of model peer slots.

Notes

  • Model hosting will focus on LLM's through open-sourced HuggingFace models. In the future, we will be implementing other AI categories.


v2.0

Blockchain

The second version of testnet will expand model validation and add model democracy, plus make any necessary changes to the peer consensus mechanism.

Goals

  • Update peer consensus mechanism.

    • Optimize.

    • Add slashing logic.

    • Add Peer states and states logic.

    • Add peer dishonesty voting.

  • Update storage value titles to more relevant titles.

  • Test democratization for modal initialization and removal, and validation of artificial intelligence through both on-chain logic and human intervention.

    • Voting through model validators on a time and balance weighted mechanism.

    • Vetting machine learning models through randomly chosen peers multiple times per epoch to ensure the results returned are what is expected, models are online, and model validators are online.

  • Test model entry fee based on a per live peer model.

Features

  • Each model will go through a democratic voting process to be initialized.


v3.0

The third and possibly final version of testnet will focus on including a POS mechanism. This is a separate system from peer staking and will coexist together.

Goals

  • Add staking for general accounts separate from model validator (peer) staking.

  • Add upgradeable blockchain mechanics.

Features

  • Users will be able to stake towards validators in return for yield.

Notes

  • The preliminary testnets all use a POA mechanism, an Authority-based round-robin scheduling (Aura), and doesn't require staking. We plan to migrate into a NPoS mechanism using Blind assignment of blockchain extension (BABE) slot-based scheduling.

  • Upgrading to a NPoS mechanism will require an inflation mechanism based on the supply staked.


Future

Hypertensor is an ongoing effort from contributors. The future may include introducing smart contracts, a robust governance system, liquid staking, a layer 2 blockchain, or other features.


Following

There may be a following version or versions after the third version of testnet but this has not been finalized.


Eco-Friendly

Hypertensor is an eco-friendly blockchain that will integrate a proof of stake mechanism for the mainnet and a proof of authority mechanism for the preliminary testnets. Over the years following mainnet, we plan to adopt the use of sustainable energy over the majority of or all of our nodes and advocate the use of sustainable energy for all node operators.

We plan to implement a carbon accounting system to estimate and track carbon emissions and quantify and analyze the impact. Understanding our emissions will allow us to make the most informed and impactful decisions to make changes to the network to reduce our environmental impact.

Last updated