INFINITIO GPU WHITE PAPER

A Decentralized Solution for the Global Generative AI Computational Power Market

Infinitio is building a GenAI focused modular L2 to enable a fair access to resources (GPUs, models, developers, data), ensuring efficient allocation across the generative AI ecosystem. This White Paper focuses on GPU allocation.


BACKGROUND

In the field of Generative Artificial Intelligence (GenAI), the GPU computational power market is undergoing significant transformation, yet it confronts a few core challenges:

  • Resource Issue: GenAI products require stable resources, such as computational power and bandwidth, to support their operation. However, the uneven global chip supply and policy barriers make the acquisition of these resources challenging and costly.

  • Data Issue: GenAI models require a substantial amount of user data and local data for pre-training and fine-tuning. When the model scale exceeds a certain threshold, the quality and security of data sources become critical factors.

  • Competitive Issue: Small and medium-sized GenAI startups find it difficult to gain a competitive advantage in this market. The traditional market has monopolized computational power, and protective policies in different countries and regions also lead to access issues for GenAI models.

To address these issues, the Decentralized Physical Infrastructure Network (DePIN) has emerged. DePIN encourages users to share their resources, such as storage space, network traffic, cloud computing power, and energy, to collectively build and maintain network infrastructure. The current market valuation of DePIN is approximately $9 billion, expected to increase to $3.5 trillion in four coming years.

https://www.binance.com/en/square/post/2024-01-05-messari-depin-3-5-2338593099330

DePIN employs token economics to promote the decentralization of global computational power, reducing the risks of computational power centralization and decreasing the cost of computational power for GenAI teams. Its large-scale and diverse user data ecosystem helps GenAI models achieve diversity and security in data acquisition channels, which is crucial for improving GenAI model performance.

Integrating GenAI models, GPUs, and blockchain, DePIN's vision is to harness the power of technology to make the benefits of GenAI accessible to everyone, collectively advancing the development and application of GenAI for the benefit of all humanity. Its goal is to create a fair and efficient global computational power market to meet the demand and supply of various computational powers and promote the development and application of GenAI. By eliminating the entry barriers in the GenAI field, DePIN hopes to extend the benefits of GenAI to every corner of the world.

In summary, we provide a decentralized solution for GenAI teams, helping them address resource, data, and competitive issues, while also offering them a market with immense potential.


INFINITIO.AI OVERVIEW

Vision

Infinitio's vision is to transform the GenAI ecosystem by providing a universal access to fair computational resources, while empowering GPU developers, model developers, and fostering a community dedicated to improve overall operation efficiency and innovation leveraging the blockchain technology.

Mission

Our goal is to create a fair and efficient global computational power market that caters to diverse needs with every demand and supply of computational power, and advances the development and application of artificial intelligence. By eliminating the barriers to entry in the GenAI field, we aim to extend the benefits of GenAI to every corner of the world. We strive to achieve global computational power demand matching, fostering collaboration among developers and large-scale integration of resources.

The Problem

End User

Security and Privacy: GenAI systems may handle sensitive personal data, and ensuring the security of these data and the privacy of users is a significant issue. Moreover, the behavior of GenAI systems may be unpredictable or uncontrollable, which could raise safety concerns.

Convenience & Safety: Users need to enjoy the convenience brought by GenAI while protecting their privacy and safety. However, these two may be difficult to achieve simultaneously, as GenAI systems providing personalized services may need access to users' personal data.

AI Model Developer

Increased development and operational costs: As the scale of GenAI model parameters grows, the cost of developing, running, and migrating these models is also continuously rising, posing a financial challenge to developers.

Steep learning curve: The deployment and maintenance of GenAI models require profound technical knowledge and practical experience. However, there is a relative lack of knowledge on how to develop, maintain, and integrate GenAI systems in practical applications, making the learning curve steep.

High cost of acquiring and using hardware: Professional computing and hardware equipment, such as GPUs, are crucial for training and deploying GenAI models, but their acquisition and usage costs are high, further increasing the burden on developers.

Data privacy and security issues: Ensuring data privacy and security during the training, fine-tuning, and interaction with GenAI models is a significant issue, requiring developers to invest a lot of time and effort to solve.

Uncertainty of GenAI system safety and behaviour: In critical task applications aimed at consumers, the safety of GenAI systems and the unpredictability of their behaviour pose a significant challenge, imposing higher technical requirements and ethical responsibilities on developers.

Trade-off between cost-effectiveness and model performance: Developers need to reduce costs and improve efficiency, but this might sacrifice the model's performance or functionality. To lower costs, developers might need to use cheaper hardware or services, but this could affect the model's operational efficiency and stability.

Computing Resource Provider

Energy consumption issues: Providing large-scale computing services could lead to significant energy consumption. This not only increases operational costs but may also have a negative impact on the environment. Therefore, how to reduce energy consumption and adopt more environmentally friendly energy sources is an issue that service providers need to consider. And computing resource providers who are using green energy need to be specially incentivized or rewarded

Maintenance complexity due to hardware diversity: The diversity of hardware product lines, such as various types and configurations of GPUs, may lead to compatibility issues, thereby increasing the complexity and cost of maintenance.

Pressure to provide efficient services: Service providers need to provide efficient and stable services to meet user needs, but this may require a large amount of energy and human resources, increasing their workload and costs.

Need to support a variety of hardware and software: To meet the needs of different users, service providers may need to support a variety of different hardware and software, which could also increase their workload and costs.

Operations and Technical Community

Audit and security issues: Model auditing, data security, and code auditing are all important issues, but these tasks may require a lot of time and expertise, which places high demands on the human resources and technical levels of service providers.

Hardware and software compatibility issues: Due to the diversity of hardware and software, such as heterogeneous migration of GPU SDKs, there may be compatibility issues, increasing the complexity of development and maintenance.

Pressure to provide high-quality services: In order to meet customer needs, third-party service providers need to provide high-quality services, but this may require a lot of resources and expertise, increasing their workload and costs.

Need to support a variety of hardware and software: To meet the needs of different customers, service providers may need to support a variety of different hardware and software, which could also increase their workload and costs.

What is needed

Is there a one-stop service like booking.com in the field of DePIN, integrating GenAI developers, GPU, technical communities and blockchain?

In creating a global computing power market, we can identify the following main participants:

AI Developers: They need efficient computing resources to develop and run complex GenAI models. Their main problems include increased development, operating and migrating costs, steep learning curves, high costs of hardware acquisition and use, data privacy and security issues, uncertainty of GenAI system safety and behavior, as well as the trade-off between cost-effectiveness and model performance.

Computing Resource Providers: They provide the necessary hardware and computing resources for GenAI developers to use. Their main problems include energy consumption issues, maintenance complexity due to hardware diversity, pressure to provide efficient services, and the need to support various hardware and software.

Operations and Technical Third-Party Service Providers: They provide various services, such as model auditing, data security, code auditing, etc., to help GenAI developers and computing resource providers solve various problems. Their main problems include audit and security issues, hardware and software compatibility issues, pressure to provide high-quality services, and the need to support various hardware and software.

Our goal is to create a “decentralized computing power matching network,” which will be similar to booking.com in the travel market, however we provide professional value-added service compared with booking.com, with a greater emphasis on meeting diverse user needs. It will take advantage of blockchain's strengths, combined with smart matching, trust and security mechanisms, and community-driven methods to create a fair, efficient, and secure global computing power market.

Features

This project aims to construct a decentralized, secure, and efficient one-stop global computing power marketplace that satisfies the demands and supplies of various computing capabilities, promoting the development and application of artificial intelligence.

Automated Computing Power Matching Market: This feature requires the establishment of an automated, global computing power matching market, enabling automatic matching between AI developers and computing resource providers. This necessitates strong algorithmic support and an effective market operation mechanism.

Decentralized Task Queue: This feature involves creating a decentralized task queue, allowing AI developers to submit their computing tasks to this queue, and computing resource providers can retrieve tasks from the queue for computation. This feature demands efficient, stable task management and scheduling mechanisms, as well as robust fault tolerance and recovery capabilities.

Distributed Resource Matching Queue: This feature involves establishing a distributed resource matching queue to enable effective matching and allocation of computing resources on a global scale. This feature requires robust resource management and scheduling capabilities, as well as efficient resource matching algorithms.

Token Reward Mechanism: This feature involves creating a token reward mechanism to incentivize computing resource providers, AI developers, and value-added service providers to participate in this market. This feature requires a fair, transparent reward distribution mechanism and a robust token economy model.

Security and Privacy Protection: This feature requires the system to effectively protect user data security and privacy, including AI developer's model data, computing task data, and computing resource provider's hardware information, etc.

Solution to Energy Consumption: This feature requires the system to effectively manage and schedule computing resources to reduce energy consumption, and it also needs to provide additional incentives to computing resource providers using green energy.

Hardware and Software Compatibility: This feature requires the system to balance various different hardware and software to meet the needs of different users, including different types and configurations of GPUs, performance requirements at different stages, and different AI models and algorithms.

Audit and Security Issues Resolution: This feature requires the system to conduct effective model audits, data security, and code audits to ensure the system's safe and stable operation.

Community-driven:

  • GPU Acceleration Services: Community members can provide GPU optimization and acceleration services with various types and configurations to meet the different computational needs of AI developers. To encourage community members to provide higher quality services, an evaluation and incentive mechanism may need to be established, enabling providers to receive rewards according to the quality of their services.

  • AI Model Deployment: Community members can share their experiences and techniques in AI model deployment, helping others to better deploy and optimize their AI models. This may require the establishment of an open knowledge-sharing platform, allowing community members to easily share and access information.

  • Optimization and Migration of AI Models: Community members can share their experiences and techniques in optimizing and migrating AI models, helping others to better optimize their AI models to suit different hardware environments and computational needs. Community members can share their experiences and techniques in optimizing models and deployments under different computational power costs, helping others to better optimize their AI models to suit different computational power costs. This may require the establishment of a cost optimization sharing mechanism, allowing community members to easily share and access cost optimization solutions.


SOLUTION

We are a distributed computing power matching marketplace, focusing on providing computing power services that can meet a variety of user needs. Therefore, we are not concerned with the cost structure of a single computing power network.

Decentralization and Meeting Diverse Needs: We leverage blockchain technology to construct a decentralized computing service marketplace. This marketplace allows entities possessing computational resources to directly transact with users in need, effectively eliminating intermediaries and reducing transaction costs. Furthermore, we utilize Distributed Computing (DePIN) to further reduce the cost associated with GPU usage. With this comprehensive solution, we aim to reduce the overall cost within the GenAI industry while catering to the diverse needs of users.

Intelligent Demand Scheduling and Resource Matching: We've developed an intelligent matching and demand prediction system using advanced algorithms. This system matches the most suitable computational resources based on AI developers' needs, such as computational power, price, hardware, and software requirements. This efficient matching mechanism is the heart of our network. The demand matching system utilizes oracles on the blockchain and AMM (Automated Market Maker) algorithms to balance market supply and demand, thereby enhancing resource utilization and meeting diverse user needs.

Dynamic Pricing and Contract Management: We've established a dynamic pricing system that adjusts prices in real-time based on market supply and demand. Additionally, we've developed a contract management system to oversee contracts between users and computational power service providers. These mechanisms ensure fair and transparent transactions and can resolve potential disputes that may arise during transactions.

Top-tier Data Rights and Privacy Protection: We employ cutting-edge data security technologies and strategies to ensure the highest level of protection for model data ownership and usage rights. Moreover, we ensure the privacy of data used or inputted by AI surpasses current standards. Technologies such as zero-knowledge proofs and homomorphic encryption could be considered to achieve this goal. [TODO 1]

Cross-Industry AI Implementation Guidance: We aim to establish an open-source community that provides professional guidance and support, assisting various industries in implementing and applying AI. Community members can share their experiences and knowledge, collaboratively solve problems, and drive the rapid development and widespread application of AI.

Ongoing Improvements by a Large Developer Community: By fostering a large developer community, we allow community members to continuously enhance the platform and introduce new features. We can design a reward mechanism to incentivize community members to actively participate in the platform's development and improvement.

Provision of Customized Services: We offer a flexible platform where third-party technology service providers can deliver highly customized services to users. Users can find the most suitable services on this platform based on their specific needs. This not only caters to the unique needs of companies but also enriches the diversity and flexibility of our platform. In this model, our role is to provide and maintain the platform, with specific services being delivered by third parties. We ensure the security and transparency of transactions via blockchain technology.

Developers

Nvidia's CUDA stack was developed in 2006, allowing developers to optimize Nvidia GPUs to accelerate their workloads and simplify GPU programming. There are 4 million CUDA users, with over 50,000 developers actively using CUDA for development, boasting a robust ecosystem of programming languages, libraries, tools, applications, and frameworks.

Reward Incentive: The biggest motivation for developers is the income incentive that can be realized by contributing their technology.

Market Share Expansion: Participating in the construction of the technical community can help GPU developers expand their market share. They can reach more potential customers through this platform, including other AI developers, research institutions, enterprises, etc.

Technological Innovation: By participating in such projects, GPU developers can learn about the latest technology trends and demands in practice, which will help them to carry out better product development and innovation.

Community Participation and Brand Enhancement: In this decentralized computing power market, GPU developers can not only provide computing resources but also participate in the construction of the community, providing technical support, sharing experiences, etc. This can not only enhance their brand influence but also help them establish better customer relationships.

Sustainability: Developers can better utilize their resources, reduce waste, and achieve sustainable development.

Modules

  1. Automated Computing Power Matching Market

  2. Decentralized Task Queue

  3. Distributed Resource Matching Queue

  4. Token Reward Mechanism

1. Automated Computing Power Matching Market

We will build a decentralized trading platform that allows providers and demanders of computing power to trade directly. In addition, we will introduce smart contracts to automatically execute transactions, ensuring the fairness and transparency of the transactions. Furthermore, we will provide oracles on the blockchain, introducing the mechanism of prediction markets, allowing users to predict and invest in future computing power demand and prices, which can further improve the liquidity and efficiency of the market. Considering the mismatch in the supply and demand of the GPU market, we will also provide computing power contracts in the form of futures and spot, priced in our tokens.

Decentralized Trading Platform: We aim to establish a decentralized trading platform, enabling providers and users of computational power services to transact directly. This approach eliminates the need for intermediaries, reduces transaction costs, and caters to a wide variety of needs.

Smart Contracts and Oracles: We plan to implement smart contracts to automate transaction execution, ensuring the fairness and transparency of each transaction. Moreover, we will introduce Oracles on the blockchain, integrating the mechanism of prediction markets. This will empower users to forecast and invest in future computational power demands and prices, thereby enhancing the liquidity and efficiency of the market.

Dynamic Pricing and Contract Management: We propose to develop a dynamic pricing system that adjusts prices in response to market supply and demand. Concurrently, we will devise a contract management system to oversee contracts between users and computational power service providers. These mechanisms aim to ensure transactional fairness, transparency, and the resolution of potential disputes.

Computational Power Futures and Spot Contracts: Given the imbalance of supply and demand in the GPU market, we intend to offer computational power contracts in the form of futures and spot contracts, using our tokens as the standard pricing measure.

Demand Prediction and Matching: We will design a demand prediction system that forecasts future computational power needs based on users' historical demand and market trends. Simultaneously, we will create a demand matching system to align users' needs with the resources of computational power service providers. These mechanisms will help balance market supply and demand while enhancing resource utilization.

Data Rights and Privacy Protection: We commit to utilizing the most advanced data security technologies and strategies to ensure the highest level of protection for model data ownership and usage rights. Additionally, we aim to protect the privacy of data used or inputted into AI beyond current standards.

Community Model Optimization and Task Scheduling: We plan to harness the power of the community and combine it with dynamic task scheduling mechanisms to provide comprehensive one-stop services. This includes not only matching the supply and demand of computational power but also covers aspects such as model optimization and migration to meet users' evolving needs. This approach will boost the platform's flexibility and user experience while stimulating community engagement and innovation.

Function Specifications

User Interface: Provides a user-friendly interface that allows users to conveniently browse market information, submit demand orders, query transaction records, real-time market dynamics, computational power price trends, user reputation history, and other information.

User Authentication: Implements user registration and login functions, ensuring that only authenticated users can participate in transactions. A digital wallet is used as the authentication method to ensure the security and authenticity of user identities.

Order Management: Implements functions such as user order submission, order query, and order cancellation. The order management module needs to handle user transaction requests and ensure the validity and consistency of orders.

Trading Matching Engine: Responsible for matching orders from suppliers and demanders to find suitable trading counterparts. The matching engine needs to consider factors such as order price, quantity, and priority, and provide efficient matching algorithms to ensure the fairness and efficiency of transactions. In the blockchain environment, transaction matching is carried out in a decentralized manner. Each node can participate in transaction matching, improving the fault tolerance and availability of the system.

Smart Contract Execution: Automatically executes preset rules, such as transaction matching, delivery, and settlement. The code of the smart contract is public, ensuring the openness and transparency of transactions. Smart contracts can automatically execute delivery and settlement operations after the two parties reach an agreement, ensuring the fairness and transparency of transactions. Smart contracts also need to implement other functions such as order status updates, payments, and dispute resolution.

Oracle Services: Introduces Oracle to provide inputs of external data and events. Oracle can be used for predictions of computational power demand and price, acquisition of market information, etc., providing users with decision support and investment opportunities. Oracle can provide real-time market information, such as the price of computational power, supply and demand conditions.

Dynamic Pricing System: Adjusts the price of computational power in real time according to market supply and demand and other factors. The dynamic pricing system can automatically adjust prices according to market changes to provide a fair trading environment and a reasonable pricing mechanism. Dynamic pricing is implemented through smart contracts.

Contract Management System: Manages contracts between users and computational power service providers. The contract management system can handle operations such as contract creation, updating, and termination, ensuring the validity and compliance of contracts. The status of each contract is recorded on the blockchain, ensuring its immutability and traceability.

Data Security and Privacy Protection: Adopts advanced data security technologies and privacy protection strategies to ensure the confidentiality and integrity of user data. At the same time, it needs to ensure sufficient protection of user privacy and handle user data in compliance with regulations.

Security and Risk Management: Implements security measures to prevent fraud and malicious behavior. At the same time, it establishes a risk management mechanism to monitor trading activities, detect and respond to risk events in a timely manner. Various security measures are implemented through smart contracts, such as limiting abnormal transactions and setting transaction limits. At the same time, a risk management mechanism can be established, such as monitoring market risks through Oracle and issuing early warnings.

Log Recording and Auditing: Records system operations and transaction records for auditing and tracking purposes. Log recording and auditing functions can help ensure the traceability and compliance of transactions. Through AI + smart contracts, automatic auditing can be achieved, improving the efficiency of auditing.

2. Decentralized Task Queue

The main goal of this queue is to become a general-purpose decentralized task scheduling system, allowing users to post their tasks to the system, and then the system will automatically allocate computing resources to execute these tasks. This system can adopt a consensus mechanism similar to blockchain to ensure the fair distribution and execution of tasks. Users/miners/service providers can all earn rewards by completing tasks.

Decentralized Task Scheduling: The system needs to be able to accept tasks from users and automatically allocate computing resources to execute these tasks. This process needs to be decentralized, meaning no single entity or institution controls the distribution and execution of tasks.

Task Distribution and Execution: The system needs to adopt a consensus mechanism similar to blockchain to ensure fair distribution and execution of tasks. This means that the execution rights and results of each task should be public, transparent, and verifiable.

Flexible Task Types: The system needs to be able to accept various types of tasks, including but not limited to data processing, model training, algorithm execution, etc. This means that the system needs to have sufficient flexibility and scalability to accommodate various task requirements.

Resource Utilization: The system needs to be able to efficiently identify and utilize computing resources, including CPU, GPU, memory, storage, network bandwidth, etc. This means that the system needs to be intelligent enough to dynamically adjust resource allocation based on task requirements and resource availability.

Function Specifications

Task Submission and Reception:

This process can be carried out via smart contracts, ensuring fairness and transparency during the task submission and reception process. The smart contract records task information (such as task ID, submitter, required resources, etc.) on the blockchain, but the specific content and execution results of the task are stored in an off-chain environment to improve efficiency and protect privacy.

Task Submission: Users submit tasks via an interface or API, which includes details of the task, type and number of required resources, expected completion time, etc. After the task is submitted, the task information is stored on the blockchain via a smart contract, ensuring the immutability and transparency of the task information.

Task Pool: The submitted tasks first enter a task pool. This task pool is on-chain, and all computing nodes can view the tasks in the task pool. The task pool can be sorted according to factors such as the time of task submission, the difficulty of the task, etc.

Task Selection: Computing nodes select tasks from the task pool based on their capabilities and resources. To ensure fairness, computing nodes can select tasks through a blockchain-based consensus algorithm, such as Proof of Work (PoW) or Proof of Stake (PoS). This process can be implemented via a smart contract to ensure fairness and transparency.

Task Execution: Computing nodes execute tasks off-chain. During the execution process, computing nodes need to generate a computation proof to prove that they have indeed performed the computation. This computation proof can be generated using verifiable computation technology.

Result Submission and Verification: Computing nodes submit the computation results and computation proof to the blockchain. Other nodes on the blockchain or dedicated verification nodes will verify the correctness of the computation proof. If the verification passes, the computation result is accepted; otherwise, it is rejected.

Task Scheduling and Allocation:

Off-chain Task Scheduling: To improve system performance, we can carry out the task scheduling process off-chain. We can design a decentralized task scheduling algorithm that decides how to allocate tasks based on various factors (such as the computing power of nodes, network bandwidth, storage space, etc.). This algorithm can run on each node, and each node can decide which tasks to accept based on this algorithm.

On-chain Task Recording: Although task scheduling is done off-chain, we still need to record the task allocation results on-chain. This can be implemented through a smart contract. When a node accepts a task, it needs to record this information in the smart contract. This way, other nodes can verify whether this node indeed has the right to execute this task.

Computation Proof: To verify that a node has indeed executed a task, we can use Verifiable Computation technology. When a node completes a task, it needs to generate a computation proof that proves it has indeed executed this task. This proof can be verified on-chain to ensure that the node has not cheated.

Support for Multiple Types of Tasks: The system should have the ability to handle various types of tasks, including data processing, model training, algorithm execution, etc. The specific design and implementation may be adjusted according to the specific needs and environment of the project. When designing a distributed system that supports multiple types of tasks, we need to consider the following key components:

Task Type Definition and Abstraction: First, we need to define and abstract various different types of tasks, such as data processing, model training, algorithm execution, and model optimization, etc. Each type of task should have a clear interface definition, including information on the task's input, output, execution environment, resource requirements, etc.

Task Adapters: For each type of task, we need to design a task adapter that transforms a specific type of task into a general task interface. This way, we can use the same system and algorithm to handle various types of tasks.

Resource Matching and Scheduling System: This system should be able to precisely match appropriate resources according to the characteristics and requirements of the task, and dynamically schedule tasks and resources. For example, for model training tasks that require a large amount of computational resources, the system should prioritize assigning them to nodes with powerful computing capabilities.

Support for Parallel and Distributed Processing: For large-scale data processing and model training tasks, the system needs to support parallel and distributed processing. This may require the design of complex task splitting and merging algorithms, as well as data synchronization and consistency guarantee mechanisms.

Automatic Optimization Mechanism: For model optimization tasks, the system needs to support automatic parameter adjustment and optimization. This may involve integrating advanced optimization algorithms and frameworks, such as Bayesian optimization, reinforcement learning, etc.

Fault Tolerance and Recovery System: The system needs to have strong fault tolerance and recovery capabilities to deal with various issues during task execution. For example, the system should be able to automatically detect and handle task failures, reschedule tasks, or recover from errors.

User-friendly Interface: The system needs to provide a user-friendly interface and API, enabling users to easily submit tasks, view task status, obtain task results, adjust task parameters, etc.

Plugin System: Considering the expandability of the system, we can design a plugin system. Users or developers can add new task types to the system by simply implementing the task adapter interface, transforming a specific type of task into a general task interface.

Containerization and Virtualization: To handle tasks that require a specific execution environment, we can use containerization and virtualization technologies. This way, we can provide each task with an isolated, controllable, and replicable execution environment.

Reward Distribution: The computational nodes that complete tasks will receive corresponding rewards. The logic for the rewards can be found in the BUSINESS MODEL & TOKENOMICS section.

3. Distributed Resource Matching Queue

This solution can be referred to as the "Distributed Resource Matching Queue". It is designed as a decentralized resource management system that allows providers to post their resources into the system, and then the system automatically matches and allocates these resources to complete tasks. It uses scalable blockchain smart contracts to ensure the fair distribution and use of resources. In addition, resource scheduling will refer to an evaluation mechanism, with the scale of supply and demand services as the core, to assess the quality of resources and services. The design philosophy of this solution is to liberate resource management from centralized control, making it an open, fair, and transparent market, thereby promoting the fair distribution and efficient use of global computational resources.

Decentralized Resource Management: The system allows any entity with computational resources to publish their resources to the system, eliminating the need for intermediaries.

Automatic Resource Matching and Allocation: The system can retrieve tasks from the Decentralized Task Queue, automatically match and allocate resources to meet user needs. This can significantly improve resource utilization and reduce waste. The system can dynamically schedule resources based on current demand and resource availability. This can better balance supply and demand, improving the system's response speed and efficiency.

Upgradable Smart Contract Templates: The system uses upgradable blockchain smart contracts to manage the allocation and use of resources. The contracts use our token for pricing. This ensures fair and transparent transactions, preventing any form of fraud or abuse.

Resource and Service Quality Assessment: The system will refer to an evaluation mechanism centered on the scale of supply and demand services to assess the quality of resources and services. This can help the platform system automatically find the most suitable resource and service allocation strategy, and it can also incentivize providers to offer higher quality resources and services.

Function Specifications

Dynamic resource management is a crucial aspect of the decentralized computing power matching market. It not only requires consideration of how to effectively allocate and adjust resources, but also how to implement this process on the blockchain.

Resource Registration: Computation nodes need to register their resources on the blockchain, which might be implemented through smart contracts. Registration information could include resource types, quantities, performance, etc.

Computational Resources: These are the basic resources, including computational units like CPUs, GPUs, TPUs, etc. Managing these resources requires a dynamic scheduling algorithm that can adjust based on task requirements and resource availability.

Storage Resources: These include storage devices like RAM, hard drives, etc. Managing these resources also requires a dynamic scheduling algorithm, which can adjust based on task requirements and resource availability.

Network Resources: These include network parameters like bandwidth and latency. Managing these resources requires a dynamic scheduling algorithm that can adjust based on task requirements and resource availability.

Other Resources: Other resources that can be abstracted as file descriptors of virtual devices.

Resource Verification: The system verifies the authenticity and accuracy of the resource information provided by the nodes by running a series of benchmark tests and integrity checks. These tests and checks can be run by the nodes themselves, and the results submitted to the system. To prevent fraud, the system can use a decentralized consensus algorithm, Proof of Stake (PoS), to confirm the correctness of the results.

Resource Allocation: The system needs to dynamically allocate resources based on task requirements and resource availability. The system uses a dynamic pricing algorithm to determine the token generation rate to incentivize resource providers to provide more and better resources.

Resource Adjustment: For the resource adjustment module, we consider the following aspects:

Dynamic Adjustment: Resource providers should be able to dynamically adjust the resources they have registered. For example, if a node adds new hardware (such as a more powerful CPU, more RAM, etc.), they should be able to update their registration information on the blockchain.

Automatic Adjustment: In some cases, the system may need to automatically adjust resource allocation. For example, if the demand for a task suddenly increases, the system may need to automatically allocate more resources to that task.

Demand-driven Adjustment: Resource allocation and adjustment should be primarily driven by market demand. If the demand for a certain resource increases, then the nodes providing that resource should be able to receive more rewards.

Penalty and Reward Mechanisms: To encourage nodes to provide accurate and timely resource information, the system can set up a penalty and reward mechanism. For example, if a node provides incorrect resource information, they may lose some of their staked tokens; conversely, if a node provides accurate resource information and responds to system adjustment requests promptly, they may receive additional rewards.

Transparency and Traceability: All resource adjustment operations should be recorded on the blockchain to ensure their transparency and traceability.

Resource and Service Quality Assessment: The resource and service quality assessment module is a key component in the distributed resource matching queue, with the main goal being to provide a fair, transparent, and reliable mechanism to assess the quality of resources and services in the system.

Resource Quality Assessment: Resource quality assessment is mainly achieved by detecting and measuring the performance, reliability, and stability of computation nodes.

Service Quality Assessment: Service quality assessment is mainly achieved by measuring the efficiency, correctness, and timeliness of task execution.

Feedback Mechanism: The feedback mechanism provides a way for users and resource providers to evaluate the quality of resources and services.

Reward and Penalty Mechanism: The reward and penalty mechanism rewards and penalizes resource providers based on the results of the resource and service quality assessment. For example, if a resource provider provides high-quality resources, the system can automatically give more rewards; if a resource provider provides low-quality services, the system can automatically give penalties.

4. Token Reward Mechanism

The token reward system is divided into two parts: the resource-adjusted computing power utility token (PToken) and the Governance token (GToken).

The resource-adjusted computing power utility token, PToken, allows users to earn tokens by providing computing power, completing tasks, providing resources, etc. These tokens can be used to purchase services and earn profits on the platform. The generation of tokens is linearly related to the scale of the deployment and inference of the overall artificial intelligence model. The essence of the computing power token is a type of voucher backed by the credit of the total available forecasted computing power . The generation speed of resource-adjusted computing power utility token is a forward fulfillment commitment for miners to provide forecasted computing power.

The governance token, GToken, allows token holders to participate in the platform's decision-making, which can further enhance the fairness and transparency of the platform.

TOKENOMICS

Gtoken and Ptoken Utilities and Value Proposition

Infinitio Decentralized Autonomous Organization (DAO): The DAO community will play an important role in adjusting the parameters of the PToken generation rate model, auditing tasks, and verifying carbon footprint proofs. To ensure its fairness and effectiveness, we will explain in the GToken module how to set up some voting rules and mechanisms to prevent malicious voting.

GToken: Governance Token

Purpose: Gtoken serves as the DAO governance token within Infinitio. Its primary function is to empower holders with voting rights, enabling them to submit proposals and vote on various initiatives. This democratic approach ensures that the platform evolves in alignment with the community's interests and priorities.

At the heart of Infinitio's decentralized governance, the Infinitio DAO empowers active community members by rewarding them with governance tokens (GToken). These tokens are not just a symbol of loyalty but a tool of empowerment, granting holders the right to vote on pivotal project decisions. Voting processes are streamlined through smart contracts, ensuring that the outcomes are automatically and transparently executed.

Governance Token Staking

The Infinitio DAO enhances community engagement by allowing users to stake their GTokens. This staking mechanism not only confers voting rights but also actively involves token holders in decision-making processes, reinforcing their influence within the community.

Staking Rewards for Governance Tokens

Staking GTokens is incentivized with rewards, promoting sustained community engagement and participation in the governance of Infinitio.

Foundation-Issued Task Rewards

To drive community contribution, the foundation allocates GTokens from its treasury for specific tasks. These tasks, essential for deploying large-scale AI models—such as migration optimization, data preprocessing, security auditing, and copyright management—support the infrastructural needs of the Infinitio Depin network. Developers who create AI tools utilized in these deployments earn GTokens, fostering a cycle of innovation and reward within the ecosystem.

Utility of Infinitio Governance Tokens

Beyond governance, GTokens also function as a versatile payment method within the Infinitio network.

Payment Method for Commercial AI Models: For commercially deployed applications and models on the Infinitio platform, GTokens serve as a viable payment option, facilitating transactions and enhancing liquidity.

Public Incubation Capital: This initiative enables token holders to sponsor and incubate commercial AI projects on the Infinitio network. Sponsorship through GTokens supports projects in two key scenarios:

Projects can use GTokens to manage their ecosystems before issuing their own tokens, boosting demand for GTokens.

For projects planning to issue their own tokens but lacking initial capital, sponsorship through GTokens helps them leverage the Infinitio platform’s resources to kick-start their operations.

Governance Token Distribution

Infinitio governance tokens enable community members to earn through task completion. Holding these tokens entitles members to a proportional share of voting rights, with all significant decisions being made within the DAO’s robust smart contract-enabled system. This decentralized framework not only allows for the trading and staking of tokens for additional rewards but also supports ecosystem project operators in utilizing GTokens for decentralized decision-making related to their projects.

In the Infinitio ecosystem, Gtoken and Ptoken play pivotal roles in facilitating governance and computation payment, respectively. Here's an overview of their functions and importance:

Ptoken: Computation Payment Token

Purpose: The resource-adjusted computing power utility token, PToken, allows computing power providers to earn tokens by providing computing power, completing tasks, providing resources, etc. These tokens can be used to purchase services and earn rewards on the platform.

Utility: The generation of Ptoken is linearly related to the scale of the deployment and inference of the overall artificial intelligence model. The essence of the computing power token function is similar to "voucher,” backed by the credit of the total available forecasted computing power. The generation speed of resource-adjusted computing power utility token is a forward fulfillment commitment for miners to provide forecasted computing power. For computational resources, Ptoken enables efficient transactions within the marketplace. This model ensures a balanced approach to ecosystem growth, combining democratic governance with practical utility.

The generation rate of PToken is determined by multiple factors, including the amount of resources registered by miners, the quality and quantity of tasks completed, and environmental factors:

Miner Resource Registration: Miners need to provide detailed specifications and performance test results of their hardware at the time of registration. To ensure the authenticity of this information, we propose a hybrid online and offline verification method, including verification using smart contracts and regular online performance tests. If a miner's actual performance does not match their registration information, we will implement a penalty mechanism. Miners need to pledge a certain amount of PTokens as a deposit, and if their information is found to be false, they will lose this deposit.

PToken Incentives and Verification: The generation rate of PToken will be dynamically adjusted according to the amount of resources registered by the miners. To prevent miners from maliciously increasing their resources, miners need to first pay a deposit, pledging GToken governance tokens. Moreover, miners can only receive PTokens after they start providing services and complete a certain number of initial tasks. These tasks need to be verified by validators. Validators, designed with reference to POS, are distributed task verification nodes that run specific system tasks. They become validators by pledging GToken governance tokens.

Task Creation and Scheduling: Tasks will come from real user demand, and users need to pay a certain number of PTokens as the cost of the task. To prevent malicious users from creating a large number of meaningless tasks, we will dynamically adjust the gas fee of the task based on its difficulty, importance, or other factors. The gas fee, used to pay for task creation, task scheduling, and task verification costs, is the computational power token PToken.

Futures Resource Commitment: Miners need to commit to their futures resources. We introduce a reputation system and a deposit system to ensure that miners fulfill their commitments. The reputation system will increase the reputation value and rewards of miners who fulfill their commitments, while the deposit system requires miners to pledge a certain number of PTokens as a deposit. If they do not fulfill their commitments, they will lose these deposits.

Environmental Factor: We will consider the carbon footprint of miners as a factor in the generation rate of PToken. Miners need to provide proof of their carbon footprint. To prevent miners from forging proof, the DAO community will conduct audits and verifications.

How to address the issue of excessive centralization of computing power?

The most significant incentive weight parameter is the demand adjuster.

Dynamic reward adjustment: When it is detected that a miner or a few miners have excessively high computing power, the reward token generation rate is nonlinearly reduced. This way, miners have no incentive to concentrate their computing power as it does not lead to more rewards.

Encouraging decentralization of computing power: Miners can be encouraged through reward mechanisms to distribute their computing power across multiple nodes. When a miner's computing power is distributed across multiple nodes, they can receive additional rewards.

Community governance: Through community governance, community members can participate in decision-making. By voting, community members can decide whether measures should be taken to adjust the token generation rate.

Introducing randomness: Introduce some form of randomness into blockchain mining, such as considering random factors in selecting miners for block validation, not just their computing power. This can help reduce the advantage of miners with high computing power.

Token rate generation adjustment algorithm factors:

T(t) = B * (Q(t) + C(t) + H(t) + S(t) + D(t) + M(t)) * N(t) * R(t) * E(t)
  • T(t) is the token generation rate for miners at time t.

  • B is the base token generation rate, a constant that represents the token generation rate for miners under the most basic conditions.

  • Q(t) is the service quality factor, ranging from [0,1], representing the quality of service (including network speed, GPU computing power, CPU computing power, promised service uptime, etc.) provided by the miner at time t.

  • C(t) is the community contribution factor, ranging from [0,1], representing the miner's contribution to the community (such as developing new tools, providing feedback, etc.) at time t.

  • H(t) is the historical performance factor, ranging from [0,1], representing the miner's historical performance (such as the stability of service quality, reputation, etc.) at time t.

  • S(t) is the service stability factor, ranging from [0,1], representing the stability of service (such as fault rate, availability, etc.) provided by the miner at time t. Service quality includes factors like network speed, GPU computing power, CPU computing power, promised service uptime, etc. These factors are nonlinear and require consideration of the demand regulator's parameter constraints. The focus is on solving the problem of miners with more computing power generating tokens faster. Hence, a nonlinear incentive function is adopted.

  • D(t) is the service diversity factor, ranging from [0,1], representing the diversity of service types provided by the miner at time t. Miners who provide a greater variety of services should receive higher ratings. We need to design an indicator that can measure service diversity. For example, miners who offer both 24G 4090 cards and 80G A800 cards should receive more token rewards.

  • M(t) is the market demand factor, ranging from [0,1], representing the market demand for the type of service provided by the miner at time t. It represents the total amount of API visits by all network users and total token consumption over a period of time at time t.

  • N(t) is the number of miners factor, ranging from (0,1], representing the total number of miners on the entire network at time t. As the number of miners increases, the value of N(t) decreases to reduce the token generation rate for each miner.

  • R(t) is the demand regulator factor, ranging from (0,1], representing the trend and rate of demand changes over a period of time at time t. When demand increases, the value of R(t) will increase to increase the token generation rate.

  • E(t) is the environmental factor, ranging from [0,1], representing the miner's carbon footprint proof at time t. Miners who hold carbon footprint proof will receive incentives for the generation rate.

Design rationale:

Service Stability Factor (S): This newly introduced factor is to better assess the stability of the services provided by miners. The stability of the service is very important to users, so this factor can encourage miners to provide more stable services.

Service Diversity Factor (D) and Market Demand Factor (M): These two factors are to better reflect the degree of match between the types of services provided by miners and market demand. Through these two factors, miners can be encouraged to provide more diverse services and adjust the type of services according to market demand.

Number of Miners Factor (N) and Demand Regulator Factor (R): These two factors are to balance the supply and demand relationship of the entire network. Through these two factors, the token generation rate for each miner can be adjusted according to the actual situation of the number of miners and demand on the entire network.

Environmental Factor (E): This factor is to encourage miners to take environmental measures and reduce their carbon footprint. Through this factor, miners who hold carbon footprint proof can be rewarded, increasing their token generation rate. This factor is adjusted by DAO-style certification.

Governance Allocation and Vesting

GToken Total Initial Issuance: 100 million

Inflation: Yes , Interest used to pay for staking governance tokens.

The following table shows the allocation of the initially issued Ptokens, as well as purposes of usage.

INFINITIO ECONOMY RULES

The economic values of the Infinitio platform is embedded by Gtokens plus Ptokens and how the Infinitio DAO Treasury operates.

As a tooling platform, Infinitio Finance project has certain preset rules of governing its economy in order to bootstrap its adoption, while maintaining transparency and integrity even at the very early stage of its launch.

As the ecosystem becomes more and more mature, the Infinitio DAO, which represents the entire Infinitio token holder community (including but not limited : Ptoken related technical community , the end-user community formed by deploying and releasing application projects on the Infinito platform for end users to use.), will be maintaining the platform in a decentralized fashion, and can change these rules as they deem necessary, through the very transparent and accountable procedures supported by the platform itself.

Last updated