PoC Consensus

PoC (Prove of Contribution) Consensus

AI-worker Internal Synchronization and Task Distribution Mechanism

An AI-worker internally consists of 3 roles, with nodes providing GPU computing power and data services being the most numerous. In this chapter, we collectively refer to nodes providing GPU computing power, data processing, and other resources as Miners. The focus of this chapter is on the synchronization and competition mechanisms of Miners, ultimately providing incentives based on each Miner's contribution. Thus, our mechanism can also be referred to as PoC (Proof of Contribution).

Currently, from the perspective of AI-worker types, there are two categories. One is B2C user-driven, where the AI-worker provides return values based on user needs. The other is research-oriented, where these services are typically not user-driven but research driven, such as cancer research, which requires the calculation of protein models and necessitates a significant amount of coordinated computing power.

Under the Deworker protocol, there are various types of AI-workers. Their demand and service nature determine their strategy for using Miners, which also creates different working models.

From the demand side, it is mainly divided into user-side needs and research-oriented needs. From the Miner management perspective, there are two approaches: one is a distribution system based on Miner priority assessment, characterized by high computing power utilization and the allocation of the number of user demands and incentives based on Node capabilities. The other is the grading of Tasks, where Miners actively choose the type of Task to solve based on computational difficulty and return the results.

Assuming the task of AI-worker1 is LLM distributed inference services, then according to the workspace's API service type, the first mechanism (Miner priority assessment distribution system) is chosen. If the task of AI-worker2 is data crawling and marking work, then the second method, based on task computational difficulty and competition, is more suitable, promoting positive feedback for results and incentives.