A Bittensor Subnet Classification Framework
A Bittensor Subnet Classification Framework
Classifiying Bittensor Subnets
This post builds on a piece written by Nick Hotz and I
A subnet in Bittensor serves as a specialized marketplace tailored to a specific task, service, or digital commodity. Managed by an owner, each subnet features a unique incentive system designed to reward participants contributing to the subnet's objectives.
Within a subnet, two key roles are present: miners and validators. Miners execute the designated task or service outlined by the subnet, while validators ensure the integrity of the miners and their outputs.
Subnets can be broadly categorized into two types:
Service-Oriented Subnets – Allow validators to monetize intelligence and other digital commodities produced by miners.
Public Good Subnets – Dedicated to advancing open-source AI and supporting other open initiatives.
Services-Oriented Subnets
Service-oriented subnets enable builders to create monetizable B2C or B2B products or services using the intelligence or digital commodities produced by miners within a subnet. Typically, access to these resources is exclusively granted to validators. Builders looking to leverage the resources produced by miners must either become a validator by acquiring and staking Bittensor’s native asset, TAO, on the specific subnet, or they can access the resources through an API provided by a validator. This exclusivity enables validators to monetize access to a subnet’s resources.
An example of a service-oriented subnet is the Proprietary Trading Network subnet. The subnet incentivizes miners to provide actionable trading signals across various financial markets. Validators within the subnet can either access these signals to make trades themselves, package them to create their own products to sell to other traders or funds, or simply make the signals available through an API for a fee. Critically, validators who understand the miner landscape can create a product that is better than the sum of its parts by aggregating the best miners’ predictions to produce high-conviction trade recommendations.
Other subnets that adopt this subnet business model include:
Vision Subnet – Validators can sell access to a distributed, verifiable model inference network.
FileTAO Subnet – Validators can build cloud storage products and services on the decentralized storage network.
OpenKaito Subnet – Validators can sell access to a decentralized search engine that provides enhanced data retrieval and analysis capabilities to users.
Public Good Subnets
Open-source initiatives have long faced challenges due the lack of financial incentives for contributors, especially in the AI sector where the reliance on voluntary efforts have hindered progress.
While some BigTech companies occasionally open-source their models, it's important to note the difference between this practice and sustained support for community-driven innovation. Relying solely on BigTech for contributing to the open-source AI space limits the community's ability to develop competitive alternatives if this support were to cease.
Bittensor’s public good subnets offer a solution by enabling the open AI community along with other communities focused on collaborative initiatives, such as Folding@home, to monetize their contributions and flourish.
Take the Nous Fine-tuning subnet, for instance. It incentivizes the creation of superior open-source AI models by challenging miners to produce models with the lowest loss using newly generated synthetic data from the leading closed-source model. After training locally, miners upload their models to HuggingFace for validators to evaluate. While Bittensor validators typically have exclusive access to resources produced by miners, the Nous Fine-tuning subnet operates differently, allowing open access as anyone can download the models from HuggingFace.
Consequently, a plethora of open models have been generated and shared, emerging as a public good that propels the open-source AI space forward.
Other subnets following the public good subnet model include:
MyShell TTS Subnet – Incentivizes advances in open-source text-to-speech models.
Pretraining Subnet – Incentivizes advances in open-source pre-trained models.
Omega Labs Subnet – Incentivizes collecting open-source multimodal datasets.
While these subnets may not offer direct revenue-generating opportunities for their validators, their contributions allow Bittensor’s service-oriented subnets to integrate and monetize their resources and services.
For instance, the Vision subnet integrates top models from the Nous Fine-tuning subnet into its distributed inference subnet. This would allow validators on the Vision subnet to offer monetized access to the top open-source models produced by the Nous subnet, akin to Corcel's approach in offering access to leading open AI models through a chatbot interface.
In essence, public good subnets serve as the backbone of Bittensor's service-oriented subnets, comparable to resource extraction operations that provide raw materials to factories for the development of goods for end-users.
Public good subnets are not only repositories for open-source goods and services but also the driving force behind collective efforts toward artificial general and superintelligence.