Multilateral Value Network
Last updated
Last updated
The value flow of ChainOpera network can be divided into 5 aspects based on the machine learning workflow operations.
The ChainOpera platform (hereafter referred to as the platform) charges a tax from all transactions made on Launchpad, with a certain tax rate. This includes creating/staking agents, buying/selling agent tokens, etc.
Agents can provide APIs as a service to the outside world. Calling such APIs would need to pay tokens/points as a fee. The amount of fee is determined by the number of tokens (which reflects the amount of workload for AI computation) of the input & output the API call, as follows:
where Ti means the number of tokens of input, To means the number of tokens of output, and Pt means the price of a singular token.
Such fees from API callers are distributed by the following parties:
LaunchPad would charge a certain percentage of tax from the fee
Agent template developer would receive a percentage of the fee as a reward for the service provided to the agent creator
MCP providers would receive a percentage of the fee as a reward for the service provided to the agent template developer
Agent creator could keep the remaining part as an income
Federated AI platform would charge some percentage of the fee as model serving cost.
In Federated AI platform, the model serving API fee would be distributed by the following parties:
The Federated AI Platform would charge a percentage as a tax
GPU providers would receive some percentage as a reward for devices’ service
Model providers would receive some floating percentage as a reward for the models’ service. A floating percentage here means different model types would be granted differently. For example, an open-sourced model would receive NO shares, while other owned models would receive some certain percentage as a reward.
The platform rewards co-creators (i.e. data contributors/annotators, GPU providers, model developers, etc.) for their contribution to the platform, in the form of points/tokens. Please refer to Co-Creation and The Contribution Model part for detailed information.
Model developers can use data/GPUs on the platform as model training resources. In such cases, the platform would charge a corresponding amount of points/tokens as a fee from this model developer for services provided.
The fee charged by platform is determined by 2 factors:
The type & running time period of GPUs utilized in the training
The type & corresponding amount of data utilized in the training
The platform labels different types of GPUs (i.e. 4090) and data (i.e. text data of 400 bytes) with prices respectively. The overall fee the model developer would need to pay is calculated as follows:
where Pg(k) means the price of Type k GPU, Ng(k) means the numbers of Type k GPU used in the training, Pd(m) means the price of Type m data, Nd(m) means the numbers of Type m data used in the training. Rg & Rd is the calculation ratio for GPU and data fee, respectively.
The platform would also take a concrete percentage of tax from such fees and grant the remaining to associated data contributors / GPU providers as a reward.
