# Getting Started Inferix Documentation [Getting Starts](/) * [Overview](/readme/overview) * [$IFX ](/readme/usdifx) * [Resources](/readme/resources) * [Brand Kit](/readme/brand-kit) * [Frequently asked questions (FAQs)](/readme/frequently-asked-questions-faqs) [Inferix Whitepaper](/inferix-whitepaper) * [Introduction](/inferix-whitepaper/introduction) * [High-level description of ANGV](/inferix-whitepaper/high-level-description) * [Implementation of ANGV](/inferix-whitepaper/implementation) * [Decentralized visual computing](/inferix-whitepaper/decentralized-visual-computing) * [Decentralized federated AI](/inferix-whitepaper/decentralized-federated-ai) * [Economic model](/inferix-whitepaper/economic-model) * [Future development](/inferix-whitepaper/future-development) [Worker Node Guide](/worker-node-guide) * [What is Worker Node](/worker-node-guide/what-is-worker-node) * [Worker Node Sales](/worker-node-guide/worker-node-sales) [Verifier Node Guide](/verifier-node-guide) * [What is Verifier Node](/verifier-node-guide/what-is-verifier-node) * [Verifier Node Sales](/verifier-node-guide/verifier-node-sales) [Inferix MVP](/inferix-mvp) * [Tutorial: MVP for designers & GPU owners](/inferix-mvp/tutorial-mvp-for-designers-and-gpu-owners) * [PoR MVP](/inferix-mvp/por-mvp) [Inferix Testnet 2 on Solana & IoTeX \[ENDED\]](/inferix-testnet-2-on-solana-and-iotex-ended) [Inferix Testnet 1 on IoTeX \[ENDED\]](/inferix-testnet-1-on-iotex-ended) [Inferix Explorer](/inferix-explorer) [Team & Achievements](/team-and-achievements) [Community & Events](/community-and-events) [Terms of Service](/terms-of-service) # Overview

## Welcome to Inferix! Inferix is a Decentralized GPU Network for Visual Computing and AI. * For users (3D graphic artists, game developers, enterprises) who need GPU computing power for rendering high-quality graphics, you can use Inferix system to continuously access these precious resources with faster processing time and more efficient spending. * As GPU computing owners, you are enabled to share idle GPUs to InferiX network for rendering 3D graphics, AI inference and earn long-term passive income while simultaneously balancing your main jobs or leisure activities Check out the full dedicated thread to learn more Inferix 👇 {% embed url="" %} ### **The industry of 3D/VR/Render** InferiX's solution meets real-world problems across a range of industries, not only for the AI field but also for high-quality rendering needs

### Can I use Inferix today? **✦ Inferix MVP** is now ready for everyone to try out & make first contributions without registering wallet/logging in, check out 👇 {% content-ref url="/pages/8APOOT3pCAf5649vwkxi" %} [Tutorial: MVP for designers & GPU owners](/inferix-mvp/tutorial-mvp-for-designers-and-gpu-owners) {% endcontent-ref %} **✦** Fill out the Google form now to gain access to the **POR MVP** 👇 {% content-ref url="/pages/KpNdLoy9WvQDcdtanNHP" %} [PoR MVP](/inferix-mvp/por-mvp) {% endcontent-ref %} # $IFX * Token: INFERIX * Ticker: $IFX * Max Supply: $$1,000,000,000$$ $IFX [⚠️](https://emojipedia.org/warning) Attention! Our token has not been launched yet, please be careful with scams.
For a more in-depth look at InferiX Tokenomics, please visit the Economic Model section [👇](https://emojipedia.org/backhand-index-pointing-down) {% content-ref url="/pages/191N73tLGsw6QzECmert" %} [Economic model](/inferix-whitepaper/economic-model) {% endcontent-ref %} # Resources Join our official community to stay updated on the most accurate and timely information! [Official X](https://twitter.com/InferixGPU) • [Warpcast](http://warpcast.com/inferixgpu) • [Website](http://inferix.io/) • [YouTube](https://www.youtube.com/@InferixGPU) • [Medium](https://medium.com/@inferixgpu) • [Discord](https://discord.com/invite/k7rVUYt6Td) • [DePINScan](https://depinscan.io/projects/inferix) • [Explorer](https://dash.inferix.io/workers) # Brand Kit Primary Logo

SVG logo

{% file src="/files/L721RWeiZx3it7vmVppf" %} Inferix Brand Kit (logo, token icon) {% endfile %} # Frequently asked questions (FAQs) ### **✦ Can I use Inferix today?** **Inferix MVP** is now ready for everyone to try out & make first contributions without registering wallet/logging in, check out 👇 {% content-ref url="/pages/8APOOT3pCAf5649vwkxi" %} [Tutorial: MVP for designers & GPU owners](/inferix-mvp/tutorial-mvp-for-designers-and-gpu-owners) {% endcontent-ref %} **✦** Fill out the Google form now to gain access to the **POR MVP** 👇 {% content-ref url="/pages/KpNdLoy9WvQDcdtanNHP" %} [PoR MVP](/inferix-mvp/por-mvp) {% endcontent-ref %} *** ### **✦ When will the Inferix network launch?** **Inferix Network will launch the mainnet after the TGE in November, 2024** *** ### **✦ When will be the Node Sales?** **The verifier node sale will be coming soon. Check all the details here** :point\_down: {% content-ref url="/pages/78qjVvOpb63y4yG0mQcS" %} [Verifier Node Guide](/verifier-node-guide) {% endcontent-ref %} *** ### **✦ What are InferiX min. requirements?** You can read detailed information of minimum requirements for each InferiX Node here 👇 {% content-ref url="/pages/ddqlOaJZ3ROAn4jG8drr" %} [Appendix C: Hardware requirements for nodes](/inferix-whitepaper/appendix-c-hardware-requirements-for-nodes) {% endcontent-ref %} # Inferix Whitepaper This paper introduces Proof-of-Rendering (PoR) and its application in building Inferix's decentralized GPU network. Addressing DePIN Verification, the key challenge facing developers of decentralized physical infrastructure networks (DePIN) recently, Inferix has developed the Active Noise Generation and Verification or Proof of Rendering algorithm. The algorithm is combined with a software layer that includes middleware and client SDK, facilitating connections between 3D creative data systems and the decentralized GPU infrastructure. This creates a unique Decentralized GPU Network for Visual Computing and AI Inference. *This is the web version, original PDF version can be downloaded at:* {% embed url="" %} # Introduction Inferix is a DePIN network of GPUs for visual computing and AI, it is built to bridge the needs of users and hardware owners. Its solution meets real-world problems across a range of industries, not only for the AI field but also for high-quality graphics rendering. Users (e.g. 3D graphics artists, game developers, enterprises) who need GPU computing power for rendering high-quality graphics can use the Inferix system to continuously access these precious resources with faster processing time and more efficient spending. Owners of GPUs can share idle resources to the Inferix network and earn long-term passive income while simultaneously balancing their main jobs or leisure activities. At high-level, Inferix network is naturally a dynamic system where demands of digital content creators and supplies of GPU owners are created continuously over time. Users are concerned with the security and privacy of the system, with the facility of accessing computing resources, as well as with the price that they have to pay for their demands. This section first describes the high-level flow of a decentralized rendering network. Next, we describe one of the main challenges that we have to deal with, that is the authenticity of rendering. [Section 2](/inferix-whitepaper/high-level-description) presents the main idea of the proposed solution then introduces a mathematical model for the Active Noise Generation and Verification algorithm. [Section 3](/inferix-whitepaper/implementation) describes an implementation for the algorithm and its integration into the existing layers of the Inferix network. [Section 4](/inferix-whitepaper/decentralized-visual-computing) presents the main components in the system architecture of the Inferix decentralized GPU network. In [Section 5](/inferix-whitepaper/decentralized-federated-ai), we discuss how to use this network infrastructure for the AI training and inference, then Inferix is actually a GPU network for visual computing and federated AI. In [Section 6](/inferix-whitepaper/economic-model), we present the token economy model of Inferix with a novel algorithm called Burn-Mint-Work for the token issuance problem. Finally, [Section 7](/inferix-whitepaper/future-development) is reserved for ongoing developments in improving the robustness, performance and availability of the network. # Rendering network using crowdsourced GPU The graphics rendering service consists in a network of decentralized machines called *nodes* which are of 3 kinds: *manager*, *worker* and *verifier*. The *managers* are dedicated machines of Inferix while *verifiers* and *workers* are machines joined by GPU owners. The number of *workers* is normally much larger than the number of *managers* and *verifiers*. #### Figure 1

A typical rendering session contains several steps as shown in [Figure 1](#fig_rendering_flow) and explained below: 1. A user creates a rendering job request using the Inferix's plugin for client, this job uploads user's scene data to some *manager*. 2. The rendering task controller of the *manager* receives the rendering job request, then 1. splits it into multiple rendering tasks, each task consists of the scene data and several parameters: range of frames to be rendered, output format, etc. 2. generates corresponding verification keys and sends these keys to the verifying task controller. The rendering tasks will be assigned to *workers* and the verification keys will be sent to the verifying task controller. 3. Receiving a rendering task, a *worker* renders the included scene using the parameters given by the task. When it finishes, it saves the rendered frames to a decentralized storage, then notifies the *manager* by a message containing a unique URL to the result. 4. The verifying task controller of the notified *manager* receives the notification then creates a verification task; this task will be assigned to a *verifier*. 5. Receiving a verification task, a *verifier* 1. checks the authenticity of the corresponding rendered frames, then 2. notifies the *manager* about the verification result. 6. If the rendered frames pass the verification, then the manager notifies the user by a message containing the URL to the rendered frames. Otherwise, these frames are rejected. 7. The user downloads the frames from the storage and manually confirms whether they meet the expectation, if they do not then the user sends a bad result claim to the manager. The *managers* synchronize a database of rendering and verification tasks. That makes the rendering service being both logically and physically decentralized: a graphics scene can be simultaneously rendered by different *workers* and later checked by different *verifiers*, the machines of *workers* and *verifiers* can be also located at different geographical locations. # Rendering verification problem A user submits some graphics work to a *manager*, this work consists of several scenes; each contains information about graphical objects, the camera, light sources and materials. The photorealistic rendering consists of sophisticated computation processes that calculate light properties at surfaces of all visible objects, resulting in 3D rendered images of the scene [\[1\]](/inferix-whitepaper/references#1). One of the most important problems that Inferix has to solve is to verify the *authenticity* of rendered results [\[2\]](/inferix-whitepaper/references#2),[\[3\]](/inferix-whitepaper/references#3), [\[4\]](/inferix-whitepaper/references#4). That means how to ensure that once a user submits a valid scene, then after waiting for an amount of time, the user will receive authentically rendered images. The authenticity can be defined informally as if the result received from the rendering network and the result received when the scene is genuinely rendered by a rendering software are human perceptually indistinguishable. The *workers* who join the rendering network are mostly workstations of GPU owners who want to make profit from their unused computational resources. Respecting the privacy of GPU owners and their resources, besides lightweight open-source software installed to manage the communication with the network, there is completely no control on *workers*. Consequently, there is no constraint to oblige *workers* to render the graphics scene correctly. Indeed, a malicious *worker* may receive a rendering task, but does nothing then uses some forged images as results. Without rendering the scene, the *managers* and users know only superficial features of what the rendered images look like. Obviously, the *managers* and users have no interest in rendering the scene themselves since if they can do that, there is no need to rely on *workers*. Moreover, we cannot deploy any surveillance mechanism on the machines of *workers* due to privacy reasons. Even if we try to do that, this is only a matter of time before a *worker* reverse engineers the mechanism and eventually bypasses it. The situation doesn't seem to favor us: checking the authenticity of something while only having a little knowledge about it, otherwise the attacker has complete information. Naturally, a public-key cryptography approach is using a scheme of *fully homomorphic encryption* (FHE) [\[5\]](/inferix-whitepaper/references#5). The scene is encrypted first by a private key before sending it to *workers*. Given the corresponding public key, the homomorphic encryption software performs the graphics rendering on the encrypted scene without needing to decrypt it. Finally, the encrypted rendered results are returned and decrypted at the user's side using the private key. The advantage of FHE is that the *workers*, even being able to modify the FHE software on their side, cannot interfere with the FHE rendering processes or forge the rendering results without being detected. Unfortunately, this approach is impractical since all state-of-the-art implementations will make the performance of the homomorphic encryption rendering become unacceptable [\[6\]](/inferix-whitepaper/references#6). # High-level description of ANGV To handle this problem, we follow the approach of digital watermarking [\[7\]](/inferix-whitepaper/references#7), [\[8\]](/inferix-whitepaper/references#8) and propose a scheme called *Active Noise Generation and Verification* (ANGV) which is a variant of *proof of ownership*[\[2\]](/inferix-whitepaper/references#2), [\[3\]](/inferix-whitepaper/references#3). Our scheme has several favorable properties: * *Efficiency:* noise generation and verification require much lower computational resources compared with the graphics rendering; the total performance of the system is not affected. * *Fidelity:* the scheme needs to modify the initial scene so the rendered output will be distorted, but the distortion is human perception sub-threshold. * *Robustness:* the embedded noises are robust under rendering enhancements and post-processing operations (e.g. de-noising, anti-aliasing). * *Effectiveness:* there is no need to use special rendering software as in the case of FHE. * *Security:* without knowing the verification key, attackers need the same computational cost with the rendering to bypass the authenticity verification. In current digital watermarking schemes for authentication and ownership verification [\[2\]](/inferix-whitepaper/references#2), [\[3\]](/inferix-whitepaper/references#3), [\[4\]](/inferix-whitepaper/references#4), [\[9\]](/inferix-whitepaper/references#9) invisible watermarks will be embedded into the digital content needed to be protected. The detector (or verifier) tries to extract the watermark from a tested content, then compares the extracted watermark with the original embedded one, if the comparison is passed then the content is authenticated. However, in the context of Inferix's rendering network, the *manager* has access to the image only after the graphics scene has been rendered by *workers*. It is nonsense to embed watermark into the image at this point since the watermarking cannot help to detect any malicious manipulations which may happen before that, i.e. in the rendering process. Our approach is to embed watermarks into the graphics scene submitted by users before sending it to *workers*. The *Active Noise Generation and Verification* scheme consists of two algorithms as described below. # Noise generation In practice, a scene may contain multiple frames, each task of this scene contains some range of frames to be rendered, consequently each worker may render only a subset of these frames. For the simplification purpose, we assume in this section that a scene has only one frame, so the output image is determined uniquely by the scene.

Let $$R$$ denote the rendering process, for each input scene $$G$$, the result of the rendering is an image: $$ I = \mathcal{R} \left(G\right) $$ It is important to note that $$I$$ is actually ***never*** computed, neither by the *manager* in the noise embedding (see also the discussion about [frame sampling](/inferix-whitepaper/implementation/adaptive-noise-spreading)) nor by *workers* in the frame rendering. The equation above represents only equality. Similar with invisible watermark schemes in the literature [\[2\]](/inferix-whitepaper/references#2), [\[3\]](/inferix-whitepaper/references#3), [\[9\]](/inferix-whitepaper/references#9) a noise $$W$$ consists in a random vector of atomic watermarks: $$ W \triangleq \left(w\_1, \dots, w\_n \right) $$ where $$w\_i , \left(1 \leq i \leq n\right)$$ is independently chosen from some normal probability distribution $$\mathcal{N}\left(\mu, \sigma^2\right)$$. Furthermore, $$w\_i$$ has a special structure depending on where it is introduced in the scene $$G$$. The number $$n$$ of atomic watermark signals is chosen around an experimental trade-off between human perception threshold about the image distortion and the false positive ratio of the noise verification. Using a uniformly generated task identification number $$J\_{\mathtt{id}}$$, we calculate a verification key which is a vector of the same size as the noise vector $$W$$: $$ K\_{\mathtt{verif}} \left(S, W, J\_{\mathtt{id}}\right) \triangleq \left( k\_1,\dots,k\_n \right) $$ that will be used later for the noise verification. We have discussed that embedding watermarks into $$I$$ cannot help the authentication, then the noise $$W$$ is not embedded into the image $$I$$ but into the scene $$G$$. Let $$\mathcal{E}$$ denote the embedding function, we now create a watermarked scene: $$ \hat{G} = \mathcal{E} \left(G, W\right) $$ Finally, $$\hat{G}$$ is sent to *workers* for rendering, that results in a rendered image: $$ \hat{I} = \mathcal{R} (\hat{G}) $$ If got accepted, namely $$\hat{I}$$ passes the noise verification which will be presented hereafter, this is the image sent back to the user (recall that $$I$$ in the rendering equation is not computed). The encoding function $$\mathcal{E}$$ and the noise $$W$$ are designed so that the distortion of $$\hat{I}$$ against $$I$$ is imperceptible [\[10\]](/inferix-whitepaper/references#10), [\[11\]](/inferix-whitepaper/references#11) then $$\hat{I}$$ can be authentically used as a result of the graphics rendering. # Noise verification Different from proof of ownership schemes [\[2\]](/inferix-whitepaper/references#2), [\[3\]](/inferix-whitepaper/references#3), the verification of watermark requires a key. Given an image $$J$$ and a verification key $$K\_{\mathtt{verif}}$$, we first try to recover a watermark $$\hat{W}$$ from $$J$$ using a decoding function $$\mathcal{D}$$: $$ \hat{W} = \mathcal{D} \thinspace (J, K\_{\mathtt{verif}}) $$ Next $$\hat{W}$$ is compared against $$W$$, if the deviation is above some threshold $$T$$: $$ \lVert \hat{W} - W \rVert \geq T $$ then $$J$$ will be accepted otherwise rejected.

# Thread model Given rendering tasks each contains basically a watermarked scene $$\hat{G}$$ and some range of frames required to be rendered, the goal of an attacker, namely a malicious *worker* (or in general a group of maliciously colluding *workers* [\[12\]](/inferix-whitepaper/references#12)), is to generate rendered frames that pass the noise verification, with computational costs significantly lower than doing render this range by some conventional rendering software. By Kerckhoff's principle, it is essential that the attackers know the noise generation and verification algorithms, working parameters including trade-offs. But the task identification numbers and the corresponding verification keys are kept secret. Furthermore, we require a strong assumption that the attackers cannot detect the existence of watermarks in scenes. That means attackers can analyze and even modify different watermarked scenes $$\hat{G}$$(s) but they cannot distinguish objects of the noise wrapping vector $$\Omega$$ (discussed in detail in the [noise generation](/inferix-whitepaper/implementation/noise-insertion)) embedded in $$\hat{G}$$(s) from original graphical objects of $$G$$. Otherwise, we assume no constraint on the communication capability of colluding attackers. Not surprisingly, the security of ANGV can be modeled as the problem of sending steganographic messages over a public communication channel with passive adversaries [\[13\]](/inferix-whitepaper/references#13), [\[14\]](/inferix-whitepaper/references#14). Indeed, let us consider some graphics scene, by repetitively receiving rendering tasks for this scene and sending results (both genuinely rendered and intentionally forged), an attacker (or set of colluding attackers) knows a set of accepted and rejected images. The attacker analyzes these tested images to estimate probability distributions $$P\_{\mathcal{S}}$$ and $$P\_{\mathcal{C}}$$ for respectively images that would pass the noise verification and images that would be rendering results of the scene. We use the traditional notations of the steganography literature: $$C$$ for cover-work and $$S$$ for stego-work [\[15\]](/inferix-whitepaper/references#15). The information-theoretic security of ANGV is quantified by the Kullback–Leibler divergence (i.e. relative entropy) $$D\left(P\_{\mathcal{C}} \mathrel{\Vert} P\_{\mathcal{S}}\right)$$ of $$P\_{\mathcal{C}}$$ from $$P\_{\mathcal{S}}$$. Concretely, ANGV is called $$\epsilon$$-secure if $$ \lim\_{n \to \infty} D\left(P\_{\mathcal{C}} \mathrel{\Vert} P\_{\mathcal{S}}\right) \leq \epsilon $$ where $$n$$ is the number of tested images. In particular, $$\epsilon = 0$$ if and only if $$P\_{\mathcal{C}} = P\_{\mathcal{S}}$$, or the attacker cannot distinguish watermarked images from genuinely rendered ones, in this case we have perfect security. ***Remark.*** The distributions $$P\_{\mathcal{S}}$$ and $$P\_{\mathcal{C}}$$ represent partial knowledge of the attacker obtained by analyzing the set of tested images: the larger this set (or the larger $$n$$), the more precise estimation for $$P\_{\mathcal{S}}$$ and $$P\_{\mathcal{C}}$$. # Implementation of ANGV The previous section presents a high-level description of the Active Noise Generation and Verification algorithm. In this section, we discuss in detail the current implementation approaches and proposed trade-off values. # Structure of noise As introduced in the previous section, a noise $$W$$ is a random vector $$\left(w\_i\right)\_{1 \leq i \leq n}$$ where each element is independently chosen from a normal distribution. This atomic watermark is constructed as a rectangular image of periodic patterns as follows: * let fix some values $$M,N$$ for the width and the height of the rectangle, and * let $$\mathcal{X}\_i, \mathcal{Y}\_i$$ be independent and identically distributed normal random variables: $$ \mathcal{X}\_i \sim \mathcal{Y}\_i \sim \mathcal{N}\left(\mu, \sigma^2\right) $$ for some $$\mu$$ and $$\sigma$$, then take $$X\_i, Y\_i$$ be respectively some samples of $$\mathcal{X}\_i, \mathcal{Y}\_i$$. The complex atomic signal $$w\_i$$ is defined by: $$ w\_i(x,y) = A e^{2i\pi\left(\frac{x}{X\_i} + \frac{y}{Y\_i}\right)} \ \left(0 \leq x < M, 0 \leq y < N \right) $$ for some amplitude $$A$$. We observe that $$w\_i\left(x,y\right) = w\_i\left(x + X\_i, y\right) = w\_i\left(x, y + Y\_i\right) , \forall x,y$$ then $$X\_i, Y\_i$$ are actually the horizontal and the vertical periods. ***Remark:*** $$\mathcal{X}\_i$$ and $$\mathcal{Y}\_i$$ are elements of a set $$\left{ \mathcal{X}\_i, \mathcal{Y}\_i \mid 1 \leq i \leq n\right}$$ of independent and identically distributed normal random variables $$\mathcal{N}\left(\mu, \sigma^2\right)$$. The parameters $$\mu$$ and $$\sigma$$ are chosen by analyzing the input scene that is discussed in\~\cref{subsec:noise\_spreading}. ***Proposition 1.*** (Fourier transform of complex atomic signals) $$ F\_i\left(u,v\right) = \frac{A}{M \times N} \frac{\left(1 - e^{2i\pi \frac{M}{X\_i}}\right) \left(1 - e^{2i\pi \frac{N}{Y\_i}}\right)}{\left(1 - e^{2i\pi\left(\frac{1}{X\_i} + \frac{u}{M}\right)}\right) \left(1 - e^{2i\pi\left(\frac{1}{Y\_i} + \frac{v}{N}\right)}\right)} $$ ***Proof:*** Direct calculation (for details, see [Appendix 1](/inferix-whitepaper/appendix)). The given structure of noise has two folds: we empirically find that this form of signal makes the wrapping graphical objects (discussed in the [noise insertion](/inferix-whitepaper/implementation/noise-insertion)) persistent in the rendering of graphics scenes. Furthermore, the distortion raised by any atomic watermark is easily controlled thanks to the simple form of the signal amplitude given in the proposition. The spectrums of atomic signals play a crucial role in the noise verification since they help to distinguish embedded noises from the original image signals. They are also completely determined by the periods $$X\_i, Y\_i$$ given fixed $$M,N$$ since the discrete Fourier transform in the proposition. In turn, these periods statistically rely on the expectation $$\mu$$, we will discuss how to choose this value in the [noise spreading](/inferix-whitepaper/implementation/adaptive-noise-spreading). The length $$n$$ of the noise vector is one of the principal factors which decides the robustness of noise: the higher the value $$n$$, the lower the false positive of noise verification. But this size influences the quality of the rendered image: the lower value $$n$$, the higher fidelity of the rendered images. Consequently, the value $$n$$ is a trade-off between the robustness of the embedded noise and the fidelity of the rendered image, it is empirically chosen to be about 8 to 15. #### Figure 4:

[Figure 4](#random_vector_atomic_watermarks) shows a noise as a vector of $$5$$ atomic watermarks and the Fourier transforms showing the corresponding frequency characteristics. The vital frequencies of energy are clearly shown in the spectrums. For illustration purpose, we take $$X\_i = Y\_i \sim \mathcal{N}\left(25,5\right) \ \left(1 \leq i \leq 5\right)$$, and $$M = N = 512$$. # Noise insertion Given a scene $$G$$, the random vector $$W = \left(w\_i\right)\_{1 \leq i \leq n}$$ is embedded into $$G$$ by 1. wrapping each atomic $$w\_i$$ by a graphical object: let denote it $$\omega\_i$$, then we obtain a vector of objects $$\Omega = \left(\omega\_i\right)\_{1 \leq i \leq n}$$. 2. inserting $$\Omega$$ into $$G$$ so that every $$\omega\_i$$ contributes to the rendered image, namely they distort this image. The distortion is kept to be lower than the human perception of light [\[10\]](/inferix-whitepaper/references#10), [\[11\]](/inferix-whitepaper/references#11). # Geometric constraints By the nature of physically based rendering [\[1\]](/inferix-whitepaper/references#1), an object (or any part of it) in the scene will not be visible if and only if there is no visible light (or in general the light is out of the capability of the sensor) scattered from the surface of the object to the digital camera object. This may be caused by several reasons: the object is not located in the frustum of the camera, is hidden by other objects, or the object is made of some transparent material. Furthermore, the atomic signals $$w\_i \ \left(1 \leq i \leq n\right)$$ should not interfere themselves since this makes the noise verification to be unnecessarily complicated. Consequently, we require the noise embedding to satisfy first the following constraints: * there are no collisions between $$\omega\_i \ \left(1 \leq i \leq n\right)$$, * object vector $$\Omega$$ is completely located in the camera frustum, * no $$\omega\_i$$ is hidden by another object (even partially), including both $$\omega\_j \in \Omega, \ j \neq i$$ and objects of the scene. As [previously mentioned](/inferix-whitepaper/implementation/structure-of-noise), the characteristics of atomic noises in frequency domain are crucial for the robustness of noise verification: we need to restore a certain amount of information about these characteristics from very small distortions made by the objects $$\omega\_i \in \Omega$$ on rendered images. Because of the unavoidable requirement about the fidelity of images, we have to keep these distortions local, concretely these distortions must be well-placed on regions whose locations can be pre-calculated. A practical approach is to constrain the distortion made by $$\omega\_i$$ to be of the same shape as the atomic noise $$w\_i$$. Geometrically, each $$\omega\_i$$ has a *rotation vector* which characterizes the direction of the object in the global coordinate system (i.e. world space [\[1\]](/inferix-whitepaper/references#1)) containing all objects of the scene $$G$$. To keep the rectangular shape of the distortion of $$\omega\_i$$, we require that: * the rotation vector of $$\omega\_i$$ is equal with the rotation vector of the digital camera of $$G$$ for all $$\omega\_i \in \Omega$$. # Distortion region Under constraints about position and direction of noise objects, the imprint of $$\omega\_i$$ on the rendered image is a rectangular region denoted by: $$ k\_i \triangleq \left(x\_i^{\mathtt{ul}}, y\_i^{\mathtt{ul}}, x\_i^{\mathtt{lr}}, y\_i^{\mathtt{lr}} \right) $$ where $$\left(x\_i^{\mathtt{ul}}, y\_i^{\mathtt{ul}}\right)$$ and $$\left(x\_i^{\mathtt{lr}}, y\_i^{\mathtt{lr}}\right)$$ are respectively the upper left and lower right positions in the image coordinate system. It is important to note that $k\_i$ for all $$1 \leq i \leq n$$ can be computed without rendering the scene $$G$$. For the size of distortion regions, similar with the length of the noise random vector, there is a compromise between the robustness of the embedded noise and the fidelity of the rendered frame. The larger the distortion $$k\_i$$, the higher information of $$w\_i$$ can be restored then the higher robustness of the noise verification; but the lower the distortion $$k\_i$$, the higher fidelity of the image. Empirically, we use the bounds $$4 \leq x^{\mathtt{lr}}*{i} - x^{\mathtt{ul}}*{i},\ y^{\mathtt{lr}}*{i} - y^{\mathtt{ul}}*{i} \leq 7$$ for all $$1 \leq k \leq n$$.

Rendered watermarked scenes (random vector length 12)

The figure above shows some distortion results of rendering watermarked scenes. From two original scenes, noise vectors of length $$12$$ with different distortion sizes are embedded, then different watermarked scenes are generated. When rendering the scenes containing noises whose distortion sizes are $$7$$ or $$8$$, the distortions are visible under the form of small rectangles dispersed in the rendered images. In contrast, when the sizes are $$4$$ or $$5$$, the distortions are imperceptible. ***Remark:*** While the atomic watermarks are quite large, the distortions made by them on rendered images are constrained relatively small. The figure in the previous section shows atomic watermarks of size $$512 \times 512$$ which are used for watermarking scenes shown in the figure above, their imprints are about $$4 \times 4$$. The sizes of the rendered images are much larger: $$1080 \times 1080$$ and $$1920 \times 1080$$. # Adaptive noise spreading As discussed above, each atomic $$w\_i \in W$$ has its distortion contribution at the spatial region $$k\_i$$ on the rendered image. Since the rendering process contains multiple options to improve the quality of the output (noise reduction, anti-aliasing, etc.), it is severe if the region $$k\_i$$(s) fall into perceptually insignificant regions [\[7\]](/inferix-whitepaper/references#7) of the image because the deliberate distortions raised by $$w\_i$$(s) would be eliminated by the rendering enhancement. Hence, distortions are preferred to be placed in human perceptually significant regions. However, to keep the compromise between the fidelity of the rendered frames and the robustness of noise verification, the strength of the distortion of each $$w\_i$$ must be tuned so that its deviation from locally enclosed regions is within a predetermined bound. We handle this problem using the adaptive noise spreading [\[16\]](/inferix-whitepaper/references#16), [\[17\]](/inferix-whitepaper/references#17), [\[18\]](/inferix-whitepaper/references#18). Research on the human visual perception agrees that the important information of images is located at high energy and low frequency spectral regions [\[10\]](/inferix-whitepaper/references#10). Then before embedding the object vector $$\Omega$$, we render the scene $$G$$ at some low settings to get a sampling instance of the image. Next, we proceed both the spatial and spectral analysis on this instance to get perceptually significant spatial regions, called preferred regions. The atomic noises $$w\_i$$(s) will be placed in these regions. Simultaneously, we tune the expectation value $$\mu$$ used in generating atomic noises so that the energy (statistically given in proposition below) of high frequencies of the noises are sufficiently higher than the threshold used in the [noise verification](/inferix-whitepaper/implementation/noise-verification). ***Proposition 2.*** (Convergence of energies) Let $$\left{ \mathcal{X}\_i, \mathcal{Y}\_i \mid i \in \mathbb{N} \right}$$ be a set of independent and identically distributed normal random variables $$\mathcal{X}\_i \sim \mathcal{Y}\_i \sim \mathcal{N}\left(\mu, \sigma^2\right)$$. Let $$\[X\_i \quad Y\_i]^t$$ be a sample of the random vector $$\[\mathcal{X}\_i \quad \mathcal{Y}\_i]^t$$ and $$w\_i$$ be the signal $$\left(x,y\right) \mapsto Ae^{2i\pi\left(\frac{x}{X\_i} + \frac{y}{Y\_i}\right)}$$ for any $$i \in \mathbb{N}$$. Then the average of discrete Fourier transforms $$\overline{F}*n = \frac{1}{n}\sum\limits*{i=1}^{n} F\_i$$ converges: $$ \overline{F}\_n \left(u,v\right) \xrightarrow\[n \to \infty]{a.s} \frac{A}{M \times N} \frac{\left(1 - e^{2i\pi \frac{M}{\mu}}\right) \left(1 - e^{2i\pi \frac{N}{\mu}}\right)}{\left(1 - e^{2i\pi\left(\frac{1}{\mu} + \frac{u}{M}\right)}\right) \left(1 - e^{2i\pi\left(\frac{1}{\mu} + \frac{v}{N}\right)}\right)} $$ for all $$0 \leq u < M, \ 0 \leq v < N$$. ***Proof.*** Direct application of the continuous mapping theorem. For details, see [Appendix A](/inferix-whitepaper/appendix). It is worth noting that the rendering work of $$G$$ generally contains multiple tasks, each requires to render multiple image frames. But the number of reference instances used for spectral analysis is much smaller, practically less than 1% the total number of rendered frames. While keeping the robustness of ANGV, we can adjust this ratio be even smaller by increasing the size $$n$$ of the noise vector $$W$$. # Verification key generation The constraints and trade-offs discussed in the [noise insertion](/inferix-whitepaper/implementation/noise-insertion) and the [noise spreading](/inferix-whitepaper/implementation/adaptive-noise-spreading) are to ensure the *fidelity* of rendered results and the *robustness* of the verification, but they do not concern the security. Indeed, any attacker knowing the algorithm and parameters including trade-offs, can straightforwardly generate (without rendering the graphics scene) forged images with the same spectral characteristics, finally bypasses the verification. The security is supported using verification keys. Each rendering task has a secret key, in current implementation, this key is also the task identification number $$J\_{\mathtt{id}}$$. When embedding the noise vector $$W$$ into the scene $$G$$, this number is used to compute distortion regions $$k\_i$$ for all $$1 \leq i \leq n$$, the vector $$\left(k\_i\right)\_{1 \leq i \leq n}$$ is called verification key. The computation is modeled as a function: $$ K\_{\mathtt{verif}} \colon \left(S,W,J\_{\mathtt{id}}\right) \mapsto \left(k\_i\right)\_{1 \leq i \leq n} $$ In the operation of the rendering network, the leak of used verification keys is unavoidable. For instance, a *worker* may register itself to become a *verifier* node; when got accepted, it will be assigned verification tasks containing verification keys, then will be able to collect used keys. Even worse, colluding *workers* may exchange collected keys so that each of them will possess a much larger collection [\[12\]](/inferix-whitepaper/references#12). Another possibility is the malicious *workers* may get verification keys from some compromised *verifiers*. Hence $$K\_{\mathtt{verif}}$$ must be designed so that the knowledge about used keys does not leak any information about the next generated keys. The following proposition is necessary for the security of ANGV. ***Proposition 3.*** $$K\_{\mathtt{verif}}$$ is a cryptographic hash function. # Noise verification Given a tested image $$J$$ and a verification key $$K = \left(k\_i\right)\_{1 \leq i \leq n}$$, the goal of noise verification is to recover and check the trails of noises in $$J$$ at all regions $$k\_i$$. For each $$k\_i$$, we pick an atomic enveloping region $$v\_i$$ determined by: $$ v\_i \triangleq \left(x^{\mathtt{ul}}*i - \delta^{x}*{i}, y^{\mathtt{ul}}*{i} - \delta^{y}*i ,x^{\mathtt{lr}}*i + \delta^{x}*{i}, y^{\mathtt{lr}}*{i} + \delta^{y}*{i}\right) $$ where $$\delta^{x}*{i}$$ and $$\delta^{y}*{i}$$ are the width and the height of $$k\_i$$: $$ \delta^{x}\_{i} = x^{\mathtt{lr}}\_i - x^{\mathtt{ul}}*i + 1 \qquad \delta^{y}*{i} = y^{\mathtt{lr}}\_i - y^{\mathtt{ul}}\_i + 1 $$ Since any enveloping region is so small that spectral analysis cannot give reliable results, hence to filter the distortions of noises (i.e. the trails of high energy) we compare gradients of the region and the contained distortion region; one way to do that is using the Laplacian filter. Let $$abla^2$$ denote the Laplacian operator, calculate the mean of each enveloping region $$v\_i$$: $$ \overline{v}*i = \frac{1}{\lvert v\_i \rvert} \sum\limits*{\left(x,y\right) \in v\_i} \left( \nabla^2 v\_i \right) \left(x,y\right) $$ and the mean of corresponding distortion region: $$ \overline{k}*i = \frac{1}{\lvert k\_i \rvert} \sum\limits*{\left(x,y\right) \in k\_i} \left( \nabla^2 v\_i \right) \left(x,y\right) $$ where $$\lvert v\_i \rvert$$ and $$\lvert k\_i \rvert$$ are respectively the area of $$v\_i$$ and of $$k\_i$$. Then compare the deviation (c.f.\~\cref{equ:noise\_recovery,equ:noise\_difference}): $$ e\_i \triangleq \lvert \overline{v}\_i - \overline{k}\_i \rvert $$ with some energy threshold. Using the [noise tuning](/inferix-whitepaper/implementation/adaptive-noise-spreading), we experimentally accept the existence of the atomic watermarked $$w\_i$$ when $$e\_i \geq 5$$. If there is a distortion region where the deviation $$e\_i$$ is lower than the threshold then the image $$J$$ is immediately rejected, otherwise $$J$$ is accepted. ***Remark.*** *From the construction of enveloping regions from distortion regions, the areas can be simply calculated by* $$\lvert k\_i \rvert = \delta^{x}*{i} \times \delta^{x}*{i}$$ *and* $$\lvert v\_i \rvert = 9 \times \lvert k\_i \rvert$$*.* #### Figure 6:

Noise verification using Laplacian filter

The figure on the left shows an enveloping region of size $$9 \times 9$$, its distortion region is of size $$3 \times 3$$ located at the center, numbers at each pixel are the RGB color values. The right one shows the enveloping region after applying the Laplacian convolution. # Threat analysis The threat analysis of ANGV is based on the previously mentioned [security model](/inferix-whitepaper/high-level-description/thread-model). It requires careful and sophisticated settings (it may be worth noting that the rendering problem itself is undecidable in general [\[19\]](/inferix-whitepaper/references#19)) then we refer the details to a technical report. We present only some basic results that can rapidly be proven from current settings. # Attacks on verification keys As discussed in the [verification key generation](/inferix-whitepaper/implementation/verification-key-generation), colluding attackers may know a set of used keys, then use these keys to predict the next keys (this is called *random number generator attack* [\[20\]](/inferix-whitepaper/references#20)). Furthermore, a large enough set of *workers* may also temporarily saturate the rendering task assignment mechanism of the *manager* to control which nodes will be assigned\cite{Lian2007}. These nodes already know the verification keys used for the assigned tasks, then they generate straightforwardly forged images which validate the noise verification. In short, once the verification key generation is predictable, the noise verification will be compromised. This attack is mitigated due the [cryptographic hash property](/inferix-whitepaper/implementation/verification-key-generation) of $$K\_{\mathtt{verif}}$$: knowledge about generated keys does not leak any information about the next keys. # Attacks on noises Another kind of attack is based on analyzing rendered frames to predict the possible positions of distortion regions. Below is a simple result for extreme cases. ***Proposition 4.*** Let $$I = \mathcal{R}\left(S\right)$$ be the rendered frame of some scene $$\mathcal{S}$$, if $$I$$ is a constant signal or white noise then ANGV scheme is perfectly secure. ***Proof.*** We prove for the case of constant signals, the argument for white noises is similar. For simplification, we do not take the constraints about the fidelity of watermarked signals in to account and suppose that $$I$$ is the constant binary signal $$I\left(x,y\right) = 0$$. The distortions of any noise vector of length $$n$$ occurs at distinguished and uniformly random positions $$\left(x\_i,y\_i\right)*{1 \leq i \leq n}$$ on the image, or $$\hat{I}\left(x\_i,y\_i\right) = 1$$ for all $$1 \leq i \leq n$$. Since $$K*{\mathtt{verif}}$$ is a cryptographic hash, $$P\_{\mathcal{C}} = P\_{\mathcal{S}} = \mathcal{U}*{I}^{\otimes n}$$, hence $$D\left(P*{\mathcal{C}} \mathrel{\Vert} P\_{\mathcal{S}}\right) = 0$$. Many researchers observe that the data watermarking can be considered as the communication over noisy channel where the watermarks are signals and the content data is noise [\[8\]](/inferix-whitepaper/references#8), [\[21\]](/inferix-whitepaper/references#21). Under this perspective, the proposition above is actually a special case of the Shannon's noisy channel coding theorem. The spectrums of signals (as depicted in [Figure 8](#fig_trivial_signals)) are Dirac pulses for constant signals and white noises for white noises, then noises can be indistinguishably inserted everywhere. From the attacker's point of view, there is no information to make any significant estimation about the positions of the watermark. #### Figure 8

# Attacks on verifiers In case where *verifiers* are compromised, we employ the consensus mechanism of the Inferix blockchain network: the verification will be executed by several *verifiers* nodes through consensus. In order to minimize computational resource wastage, once two or more nodes reach a consensus, then the rendering result is considered successfully verified. # Performance evaluation We evaluate the execution time of noise insertion and noise verification on several scenes, the lengths of the noise vectors variate from 2 to 40. The tests are executed on a workstation of Intel:registered: Core:tm: i5 2.5 GHz CPU, 32 GB RAM and NVIDIA GeForce RTX3070 GPU. The performance of noise insertion and verification is given in the figures below, detailed data is given in tables in [Appendix D](/inferix-whitepaper/appendix-d-performance-evaluation-data). #### Figures 9-10:

The noise insertion needs to analyze the structure of the input scene to generate and insert noises, that explains the noise insertion experimental results where the execution time, while being proportional with the length of the noise vector, depends importantly on the complexity of input scenes. A loose quantification for this complexity can be observed via the GPU execution times needed to render the scenes, shown in [Table 1](#fig_rendering_time_of_scenes). #### Table 1: | Scene | Rendering time (in seconds) | Rendered frame size | | --------------------- | --------------------------- | ------------------- | | Coca-Cola | 8.09 | 1080x1080 | | Grease Pencil Bike | 4.30 | 2880x1620 | | Blender 3.5 Splash | 11.41 | 1327x1250 | | Bathroom Above Corner | 146.41 | 4000x3000 | | | | | Whereas the noise verification needs only to analyze the distortion regions whose locations are given by the verification key, then the execution time depends mostly on the number of the regions (which is also the length of the noise vector) and slightly on the size of the rendered frame. # Integration We integrate the Active Noise Generation and Verification scheme into the [original rendering flow](/inferix-whitepaper/introduction/rendering-network-using-crowdsourced-gpu#rendering-network-using-crowdsourced-gpu) by placing respectively the noise generation and the noise verification into the rendering task controller of the *manager* and the proof-of-rendering verification of the *verifier*. The completed flow is depicted in [Figure 7](#fig_rendering_flow_with_angv). #### Figure 7

Rendering flow with Active Noise Generation and Verification

In the introductory section of the paper, we have discussed the [problem of rendering verification](/inferix-whitepaper/introduction/rendering-verification-problem), that is to automatically verify whether the submitted scenes of users are genuinely rendered or not. The ANGV is proposed to deal with the challenge, ANGV serves then as a *proof of rendering* (PoR), inspired from the *proof of ownership* schemes [\[2\]](/inferix-whitepaper/references#2), [\[3\]](/inferix-whitepaper/references#3). Other components of the Inferix network simply refers PoR for the underlying ANGV algorithm. # Decentralized visual computing Inferix is a decentralized physical GPU network connected to end users, graphic software, or AI models through a feature-rich software layer that is continuously expanded by Inferix Labs and the community of developers within the Inferix ecosystem. Inferix is built upon the core PoR algorithm, which integrates both on-chain and off-chain verification. This section will outline the system architecture and key design details of the Inferix decentralized GPU system. #### Figure 11:

Inferix decentralized rendering architecture

The system architecture of the Inferix network is described in [Figure 11](#fig_decentralized_rendering_architecture), consisting of three main components: Manager Node, Worker Node, and Client Apps. Data is stored and accessed through a Decentralized Storage System. # Client Apps plugin Inferix provides plugins and APIs that allow traditional graphic design software to easily send 2D/3D graphic data into the Inferix decentralized rendering system and receive photorealistic images or videos in return. At the time of writing this paper, there is only one project on the market offering a GPU-based decentralized rendering solution, but it requires users to use (and pay for) their proprietary rendering engine. Inferix emphasizes the principle of freedom in line with the Web3 spirit, not mandating that end-users use a specific rendering engine or graphics software to leverage its GPU compute network. The Client App Plugin system, developed collaboratively by the Inferix team and developers within the ecosystem, supports most major rendering software and engines such as Blender, SketchUp, Cinema 4D, 3ds Max, and more. # Client API and SDK The core of all real-time rendering engines like Unity, Unreal Engine (UE), Three.js, and Babylon.js, Actif3D [\[22\]](/inferix-whitepaper/references#22) lies in the combination of static space rendering (lightmap baking) and real-time rendering of dynamic elements. Lightmap baking is typically performed by developers during the game build process, but this process often requires several hours and expensive hardware. Traditionally, a large amount of CPU power was used for this task, but recently, most game studios have shifted to using GPUs. However, the costs associated with GPUs remain high, and the long rendering times result in significant waste and expense. Inferix offers a decentralized infrastructure for baking lightmaps at a lower cost. Moreover, by leveraging parallel processing across multiple hardware setups in different locations, Inferix can significantly reduce rendering times for large-scale projects. To support this process, Inferix provides tools for lightmap baking through its rendering system, along with an SDK that enables the integration of these baked lightmaps into various rendering engines. # Manager node Manager Node(s) are computers that handle API load balancing and manage Inferix's services, including Rendering, AI Training/Inference, and Remote PC services. A manager node consists of two components: Service Controller and Job Controller. The rendering cost calculations based on the PoR algorithm will be sent by the Service Controller to the end users, allowing them to decide whether to submit an order. The [Job Controller](https://docs.inferix.io/inferix-whitepaper/decentralized-visual-computing/pages/Di2djAZNb70Mcrxp34yf##fig_rendering_flow) is responsible for dividing rendering jobs into different tasks and assigning them to Worker Nodes for execution. For example, a 1000-frame video rendering job can be divided into 200 tasks, with each task rendering 5 frames. These tasks are pushed into TaskQueue. Available workers that meet the minimum hardware capability requirements, based on the [Inferix Bench index](/inferix-whitepaper/economic-model/inferix-bench-and-ibme/ib-and-ibm), will be randomly assigned to perform the rendering tasks. The rendered results are then aggregated according to the process outlined in the [rendering flow](https://docs.inferix.io/inferix-whitepaper/decentralized-visual-computing/pages/Di2djAZNb70Mcrxp34yf##fig_rendering_flow). # Worker node It is where the actual rendering takes place. Each Worker Node is equipped with a tool called the Render Engine Controller, along with one or several render software or render engines, such as Blender Cycles, V-Ray, Keyshot, D5 Render, Octane, etc. See the [rendering flow](https://docs.inferix.io/inferix-whitepaper/decentralized-visual-computing/pages/Di2djAZNb70Mcrxp34yf##fig_rendering_flow) for more details. The render engines utilized by Worker Nodes typically rely on path-tracing/ray-tracing techniques, which demand substantial compute resources. Some of these engines are open-source and free, such as Cycles, while others, like V-Ray, are paid software. If there are any render engine costs, they will be factored into the rendering price that the Inferix network charges users. # Decentralized storage Data storage and security are critical for Inferix users. The data for a 3D scene typically ranges from a few dozen MBs to several GBs, while AI model data can reach up to several TBs. Managing this data requires specialized methods. # Data categories 3D data stored in the Inferix system includes: * Geometry data of 3D models, characterized by polygons that form the shape of objects. Depending on the render engine, different formats may be used. The higher the number of polygons, the more detailed the rendering result will be; however, the render time will also increase, and more storage space will be required. * Texture data, which is the surface image of the object. Inferix uses data formats with the best compression algorithms for GPUs, such as Basis and KTX, alongside common formats like JPEG, PNG, TIFF, or WebP. * Rendered results in image format * Rendered results in video format * Structural data in JSON format # Multi-level 3D polygon data In the stored data on Inferix, aside from images and videos, the geometry data of 3D models consumes the most storage space. Each time a render is performed, the Worker Node must download this data locally so that the render engines can execute the task. This process can consume significant bandwidth and time. To save bandwidth and reduce download times, 3D models are converted into two levels of detail, known as high-poly and low-poly, and are pre-stored on the Inferix network. The existence of multiple levels of 3D data is referred to as *multi-level 3D polygon data*. # Polygon digester After a 3D data file is stored on Inferix, it may be queried multiple times by Worker Nodes or design software. Depending on the needs of the query, the original data is converted into different levels of polygon detail through a lossy conversion algorithm. This process is handled by the *Polygon Digester* tool within the Inferix storage system. This tool ensures that the appropriate level of detail is provided for each task, optimizing both storage and performance by reducing unnecessary complexity in the 3D models when high detail is not required. # Decentralized storage With the large volume of data involved, using traditional cloud storage models at traditional data centers can be extremely costly. Peer-to-peer (P2P) and decentralized storage networks like IPFS and Filecoin offer a significant reduction in costs while maintaining access speeds comparable to traditional methods. The 3D data storage system of Inferix will predominantly rely on such decentralized networks, leveraging their cost-effectiveness and efficiency for managing extensive data volumes. # Decentralized cache Inferix's 3D data caching system utilizes decentralized CDNs. This approach enhances the distribution and retrieval of data across the network, reducing latency and improving access speeds by caching frequently requested 3D assets closer to the end-users. The decentralized nature of the CDN ensures that caching is distributed across multiple nodes, providing redundancy and resilience while minimizing the load on any single server. This setup aligns with Inferix's broader strategy of leveraging decentralized technologies for efficient data management. # Data security with FHE and TEE Data stored within the Inferix network is categorized into two types: * *Session data* which includes input data along with temporary data generated during the rendering process. This data typically exists for a short duration, ranging from a few minutes to several hours. * *Persistent data* which consists of the output from the rendering process and is stored long-term in the system. Examples of persistent data include images and videos after rendering, or VR scenes created after lightmap baking. Inferix encrypts session data to ensure it remains secure against decryption attacks during data transfer. For persistent data, Inferix offers long-term hosting on its storage system and allows users to share the data publicly or with specific permissions over the Internet. Our surveys show that, on average, over 80% of 3D model data from graphic artists are public data shared on the internet, while nearly 20% are their original creations and need to be protected. Therefore, Inferix will offer data security level options for end-users to choose from. Higher security level options will incur additional costs for computing resources, storage, and bandwidth, which will be added to the service fees that the end-user must pay. There are three components related to data security in the Inferix network: the Manager, Worker and Verifier. We will present a security approach for each of these components below. # Verifier data security enhancement with FHE Verifier is the component that receives the least amount of data among the three main components of the Inferix network. The input data for a Verifier includes a random subset of the rendering job's output along with the algorithm and key to verify it. Rendering jobs that do not require high data security will be executed by standard Verifiers. Otherwise, those that require high data security will be executed by secure Verifiers. Inferix uses Fully Homomorphic Encryption (FHE) [\[5\]](/inferix-whitepaper/references#5), [\[6\]](/inferix-whitepaper/references#6), [\[24\]](/inferix-whitepaper/references#24) technology on secure Verifiers to ensure that end-user data is completely protected from leakage. The hardware requirements for secure Verifiers are higher than those for standard Verifiers, and specifically, these nodes must be equipped with GPUs. #### Figure 12:

[Figure 12](#fig_por_with_fhe) illustrates the operation of the PoR algorithm combined with FHE. All information related to the verifying task including the rendering output data and verification key is encrypted using FHE and sent back to Manager node. The verification result is then sent back to the Manager node for review, while the Verifier remains completely unaware of the content of the verification process or the verification result. # Worker and Manager data security enhancement with FHE The Manager and Worker are the components that receive the most information in the Inferix network. For user data with a high priority on security, Inferix adopts the Trusted Execution Environment (TEE) solution [\[25\]](/inferix-whitepaper/references#25). This solution requires the node to use a CPU that supports TEE, such as Intel SGX or AMD SEV. The additional costs incurred due to the high-end hardware requirements will be factored into the service pricing, allowing users to choose and make decisions accordingly. # Decentralized federated AI In the discussion about [decentralized visual computing](/inferix-whitepaper/decentralized-visual-computing) service, we have introduced the physical GPU network used for graphics rendering. We discuss now how we utilize this GPU network for AI training and inference. These processes are essential items of Inferix's Phase 2 strategy, aligning with the core principles of Web3: openness, decentralization, self-governance, and diversity. Currently, AI advancements are predominantly driven by industry giants like Google and OpenAI, relegating most users to passive roles. This situation runs counter to the principles of Web3 and DePIN. To bridge this gap, we propose an application framework for deploying federated learning models on the Inferix GPU infrastructure in the following sections. This framework is designed not only to reshape the existing landscape but also to elevate the intelligence of the evolving DePIN ecosystem. # Federated learning with TensorOpera Federated learning and its practical benefits have recently started to see widespread application. This article will not delve into the concept of federated learning itself but will focus on applying it to leverage the GPU infrastructure of Inferix. Several foundational projects have developed tools/SDK for federated learning developers. After extensive evaluation, we have chosen the open-source TensorOpera:registered: as the basis for developing the Inferix Federated Learning framework. # Meta LLaMA In its GPU hardware segment, Inferix focuses on devices optimized for graphics rendering, with the RTX3090 and RTX4090 serving as the flagship devices. The TensorOpera:registered: team has released public data on deploying pre-trained models like LLaMA-2 13B or LLaMA-3 7B parameters on the RTX4090. Notably, LLaMA-2 13B inference running on a single RTX4090 using TensorOpera’s ScaleLLM achieves 1.88 times lower latency compared to the same model running on a single A100 GPU using vLLM. For the LLaMA-3 7B, it can run with a token batch size of 256 on a single RTX4090, without additional memory optimization [\[23\]](/inferix-whitepaper/references#23). In their introduction to ScaleLLM, the TensorOpera:registered: team claims that by utilizing this engine with the RTX4090, LLMs can operate with three times less memory, run 1.8 times faster, and be 20 times more cost-effective compared to using A100 GPUs in traditional data centers. Research and experimental benchmarks have shown that we can *train larger LLMs on a larger number of distributed GPUs than in data centers with federated learning*, using Gradient Low-rank Projection (GaLore) [\[23\]](/inferix-whitepaper/references#23), [\[24\]](/inferix-whitepaper/references#24). # Stable Diffusion Stable Diffusion inference can be easily deployed on various Inferix hardware models, ranging from the RTX3070 to the RTX4090 nodes. Additionally, the RTX4090 node is an exceptionally well-suited hardware model for training Stable Diffusion models. # Other AI models Other popular AI models, such as Bark, InstantID, and Whisper, both training and inference, can also run on Inferix GPUs through the TensorOpera AI platform. # Inferix AI From the information presented above, we can conclude that the hardware of the Inferix network is well-suited to serve as an infrastructure for federated AI. Next, we will discuss the design of the Inferix Federated AI system. #### Figure 13:

Inferix and TensorOpera integrated architecture

In the [architectural design](#fig_inferix_tensoropera_integrated_architecture), Inferix enables *generative AI artists* and content creators to access AI models trained by *AI Builders* within the Inferix community, as well as models trained by the Inferix Team itself (built-in models). These services can be hosted on the Inferix Manager Node system or on the TensorOpera AI platform. AI Builders have the option to run their models directly on the Inferix infrastructure or through the TensorOpera Bridge. The output result can be hosted in Inferix infra with custom domain option. In addition to handling graphics rendering tasks, Inferix GPU Nodes also serve as Federated Learning Clients by running the Inferix TensorOpera CLI. * *TensorOpera CLI:* the CLI client that is built based on TensorOpera open source with Inferix PoW algorithm integrated. * *Inferix PoW:* general PoW algorithm used to calculate the actual work performed by workers, excluding those involved in rendering tasks. Inferix PoW is based on the Proof-of-Rendering mechanism to calculate the [Inferix Bench](/inferix-whitepaper/economic-model/inferix-bench-and-ibme/ib-and-ibm), incorporating an algorithm to accurately measure the actual working time of a node. # Economic model # GPU compute market for visual computing and federated AI The Visual Computing market was valued at over USD 36.5 billion in 2023 and is projected to grow at a CAGR of more than 23% from 2024 to 2032. A significant trend within this market is the rise of cloud-based rendering services, which offer substantial benefits to designers, content producers, and businesses alike [\[25\]](/inferix-whitepaper/references#25). The global federated learning market was valued at USD 110.82 million in 2021 and is expected to grow at a CAGR 10.7% during the forecast period 2022-2030 [\[26\]](/inferix-whitepaper/references#26). There are approximately 4 million graphic designers working globally, generating an average of about 1.4 billion rendering tasks per year. Traditional 3D graphic design applications such as Blender, SketchUp, 3ds Max, Maya, Cinema 4D, House3D, Actif3D... all require rendering operations to produce images or videos that accurately simulate spaces, characters, animations, and materials as they are designed. Currently, designers using these software tools often need high-end PCs with expensive GPUs. For example, an interior designer often uses a computer worth around 2000$. Each 3D video rendering process usually takes several hours. In terms of software, aside from investing in 3D modeling software, users also need to use a specialized software called a rendering engine, such as VRay (paid) or Blender Cycles (free). There is currently only one decentralized rendering solution on the market, which serves as a competitor to Inferix. However, this solution requires users to utilize a proprietary rendering engine developed by them, which comes with a significant licensing fee. With Inferix, users simply need to install a plugin into their preferred 3D software to access a crowdsourced GPU network, allowing them to submit rendering requests with ease. This approach significantly reduces rendering time compared to using a standard PC and offers substantial cost savings compared to traditional render farm services on the market. # Inferix vision Inferix plans to implement a token economy model to incentivize individuals and organizations worldwide to share their GPU compute resources. This approach is designed to not only lower the cost of GPU resources and enhance efficiency but also create new revenue opportunities and business models for participants in the decentralized network. Inferix envisions creating a fairer, more efficient, and sustainable distributed computing ecosystem through innovative technology and economic models. By challenging the traditional monopoly on GPU power, Inferix aims to promote the equitable distribution and efficient use of GPU resources, fostering the broader adoption and development of visual computing, AI, and blockchain technologies. # $IFX token Inferix tokens ($IFX) are integral to the value exchange within the Inferix ecosystem and the operation of its decentralized GPU network. GPU owners can earn $IFX tokens through the following methods: * *Participation in Inferix visual computing:* Participating in the Visual Computing network is one of the primary ways to earn $IFX token rewards. Participants contribute their computational resources (CPU/GPU, internet bandwidth, storage) to assist with graphic rendering tasks. This includes not only performing rendering work but also contributing to PoR verification. The greater the contribution, the greater the $IFX token reward. * *Contributing to Inferix federated AI tasks:* Participating in the Federated AI network is another key way to earn $IFX tokens. In this process, participants contribute their computational resources to support the training or inference of AI models. Additionally, individuals or groups can earn $IFX token rewards by engaging in activities such as optimizing models and improving training efficiency. * *Running Verifier Node:* The Verifier nodes ensure the integrity and service quality of Inferix network by checking the visual computing or AI workers and their service process. The parameters mainly include liveness, capacity, and quality of service. The checking methods include heartbeat collection, benchmark testing using PoR, link data collection and analysis. * *Engaging in Inferix Governance activities:* Another way to earn $IFX tokens is by participating in the governance of the Inferix network. $IFX token holders can take part in the decision-making process, such as voting on network protocol updates and adjustments to BMW reward parameters. This participation not only enhances the network's democracy and transparency but also allows holders to directly influence the direction of Inferix. Governance participants are rewarded with $IFX tokens based on their level of contribution to the network’s decision-making. This approach is designed to incentivize and empower those who actively participate in Inferix governance, ensuring that the network evolves for the common good of the community. * *Joining Inferix Professional Artist Network:* 3D artists, designers, and architects can join the Inferix Professional Artist Network and utilize Inferix GPU infrastructure services. Based on the volume and frequency of each individual's [IBM](/inferix-whitepaper/economic-model/inferix-bench-and-ibme/ib-and-ibm) consumption, Inferix will implement an $IFX token reward mechanism to encourage end-users to use the service. For GPU owners, please refer to the [node staking and rewards](/inferix-whitepaper/economic-model/node-staking-and-rewards) for more details on how to earn rewards by running an Inferix node. # Burn-Mint-Work token issuance model Increasing the *token velocity* and controlling inflation are critical issues for any utility token. Various issuance models for utility tokens have been proposed, but fundamentally in DePIN projects, there are two commonly used methods: *Work Token* and *Burn-Mint-Equilibrium* (BME). While the Work Token (used by Filecoin project) has the advantage of improving *token velocity*, it requires the provided service to be purely a commodity, with no manual human intervention. On the other hand, BME (used by famous projects like Helium and Factom [\[27\]](/inferix-whitepaper/references#27), [\[28\]](/inferix-whitepaper/references#28)) lacks control over the volume of verified work completed and the ability to adequately penalize substandard service providers [\[29\]](/inferix-whitepaper/references#29). Inferix's BMW (Burn-Mint-Work) is the token issuance mechanism designed to address the creation of tokens based on the amount of work completed and a Node's working capacity. When BMW is combined with the [IBME mechanism](/inferix-whitepaper/economic-model/inferix-bench-and-ibme/ibme), it also helps solve the inflation control issue within the Inferix system, allowing $IFX tokens to be minted flexibly based on the total volume of work completed and the total amount of money users pay for services across the Inferix network. #### Figure 14:

Inferix Burn-Mint-Work token issuance model

Inferix's BMW is an improvement on the BME algorithm, with the important parameter being "work" calculated based on the PoR algorithm. While BME (Burn and Mint Equilibrium) balances only two parameters, burn and mint, BMW balances three parameters: burn, mint, and work. BMW also incorporates penalty mechanisms for substandard providers from the Work Token model. When Inferix completes a graphics rendering or federated AI task $$t$$ for a customer, by the Provider A, the network charges a service fee calculated as follows: $$ \mathcal{F}\left(t\right) \triangleq W\_{t} \times \mathcal{P}\_{A} $$ where $$W\_t$$ is the amount of work completed for $$t$$ measured in IBM (c.f.\~\cref{subsec:governance}), and $$\mathcal{P}\left(A\right)$$ is the unit price for one IBM, which can be adjusted by Provider A. The Matcher algorithm on the Manager Node automatically searches and provides the best options for both the user and the provider. To order services, users must top up a prepaid account with Inferix's internal payment token, called *ifxDollar*, at an exchange rate of $$1 , \text{USD} = 1 , \text{ifxDollar}$$ at the time of the top-up. After completing a rendering job, 80% of the service fee is used to purchase an amount of $IFX tokens from available supply sources (e.g., DEX/CEX exchanges). This amount of $IFX tokens is then burned. Next, 20% of the service fee will be retained and managed by the Inferix Foundation, these funds will be used to reward developers who contribute to the Inferix ecosystem. The issuance of new tokens on Inferix is executed after each *epoch period*, initially set at 72 hours but later adjustable through DAO governance. Suppose after an epoch period $$p$$, the total amount of work completed across the network is $$W\_p$$. In that case, the system will issue a quantity of tokens according to the following formula: $$ \mathcal{T} \triangleq W\_p \times E\_p $$ where $$E\_p$$ is a parameter taken from the Emission Plan, which is planned by Inferix Governance and periodically adjusted by monitoring the [IBME ratio](/inferix-whitepaper/economic-model/governance). The newly issued $IFX tokens are allocated to stakeholders in the network as detailed [later](/inferix-whitepaper/economic-model/token-metrics-and-allocation/token-allocation). This token issuance process must be entirely independent of the token burn process to generate *ifxDollar* mentioned above. When service demand increases, more $IFX tokens are burned, leading to a decrease in the total supply of $IFX, which puts upward pressure on the price of $IFX tokens. This price increase results in fewer tokens needing to be burned to complete the same amount of work, thereby bringing the system back into equilibrium. An increase in the price of $IFX tokens also increases the profitability of providers, attracting more providers joining and increasing supply. When service demand decreases, the opposite scenario occurs, leading to a state of equilibrium. # Inferix bench and IBME # IB and IBM Based on PoR, Inferix can assess the computing power of a Worker Node at a given time and uses a measurement unit called *Inferix Bench* (IB). When IB is multiplied by the node's working time, it results in *Inferix Bench minutes*, abbreviated as IBM. Thus, IBM is the metric used to measure the workload within the Inferix network. To provide a quantitative perspective, 1 IB is defined as the average rendering capacity of a *standard unit node* with 2x RTX4090 GPUs, 32 GB of RAM, 1x Intel Core i9 CPU, SSD storage. This figure is updated daily using DAO mechanism, using benchmarks of a set of sample scenes on 10 nodes. The hardware specification of *standard unit nodes* and the *sample scenes* are also DAO-adjustable. Assuming the average rendering time for one frame of a scene $$G$$ in the sample set is $$T^{0}*{G}$$, then it is not a fixed number, but is instead derived from the combined rendering capacity of the $$10$$ randomly selected *standard unit nodes* at the benchmarking time. The value of $$T^{0}*{G}$$ is influenced by GPU, CPU, storage read/write speeds and network speeds at the time of benchmarking, though the variation is negligible. To determine the IB of any given node $$n$$, Inferix sends render requests for scene $$G$$ (randomly selected) to that node periodically. Assuming the average time it takes that node to render one frame in $$G$$ is $$T\_G$$, the rendering power of $$n$$ is defined by: $$ \text{IB}\left(n\right) \triangleq \frac{T^{0}\_{G}}{T\_G} $$ Thus, the larger the $$T\_G$$, the smaller the $$\text{IB}\_n$$ value. # IBME Short for *Inferix Bench Minutes Efficiency*, is an index used to evaluate the working efficiency of a set of nodes $$\mathcal{N}$$ over a working period $$P$$, determined by the formula of the total IBM paid over the total apparent IBM: $$ \text{IBME}*{P}\left(\mathcal{N}\right) \triangleq \frac{\sum*{n \in \mathcal{N}} \text{IBM}*P \left(n\right)}{\sum*{n \in \mathcal{N}} \text{IBM}\_P^a \left(n\right)} $$ where $$\text{IBM}\_P \left(n\right)$$ is the customer paid IBM for node $$n$$ and $$\text{IBM}\_P^a \left(n\right)$$ is the apparent IBM of node $$n$$ calculated by multiplying the IB of a node by the elapsed time (in minutes) of period $$P$$. We observe that IBME is always a figure below 100% and the higher the IBME ratio, the more efficiently the network operates. #### Figure 15:

Recent reports from DePIN projects have highlighted a significant issue: while the number of nodes participating in the network is very high, many of these nodes are either inactive or active but not receiving any service requests. Inferix solve this problem by controlling the IBME. In addition to planning and adjusting the [Emission Plan](/inferix-whitepaper/economic-model/burn-mint-work-token-issuance-model), IBME also assists Inferix Governance in making decisions related to the use of the Inferix Foundation's funds for promotional activities aimed at increasing supply or attracting service demands. # Price simulation Now, we will build a price simulation for Inferix's GPU rendering service and compare Inferix's price competitiveness with traditional Cloud Rendering services. The details of these calculations are listed in [Appendix B](/inferix-whitepaper/appendix-b-price-simulation-details). The table in [Figure 15](#fig_pricing_simulation) shows the most important information. #### Figure 16:

In this simulation, the calculations are standardized for a single GPU device. The *Provider Price* refers to the minimum service price that a GPU Provider will set to ensure that revenue is always twice the cost per hour of operation. In the best-case scenario, where GPUs receive enough tasks to operate full-time every day, providers can break even in no more than 17 months for the RTX3060, RTX3070, and RTX3080 models, and 24 months for the RTX3090 and RTX4090 models. However, since the RTX3090 and RTX4090 models are better suited for federated AI tasks, their IBME scores may be higher. Moreover, the choice of which GPU model to use for rendering also depends on whether the customer prioritizes shortening the rendering time or prefers a more reasonable price. Therefore, the decision on which device to invest in depends on the specific conditions of each individual in particular market conditions. A crucial point to note here is that with the proposed *Provider Price*, Inferix’s service price is the equivalent of only from 12% to 38% of the average price in the traditional cloud rendering market at the time of writing this paper [\[30\]](/inferix-whitepaper/references#30), [\[31\]](/inferix-whitepaper/references#31), [\[32\]](/inferix-whitepaper/references#32). This means that providers can adjust the price higher to shorten the payback period. ***Price of 1 IBM***: Based on the previously discussed calculation method for [IBM](/inferix-whitepaper/economic-model/inferix-bench-and-ibme/ib-and-ibm), the price of 1 IBM is approximately $0.016 according to this pricing simulation. # Token metrics and allocation * Token: INFERIX * Ticker: $IFX * Max Supply: $$1,000,000,000$$ $IFX The total supply of $IFX is determined based on the assumption that the Inferix network will serve approximately 4 million graphic artists and around 10 million end-users utilizing generative AI services. With 1 billion $IFX tokens in circulation, the *token velocity* of Inferix is expected to reach a minimum of 10.0, according to the Equation of Exchange $$M \times V = P \times T$$. # Token allocation The token allocation of $IFX is depicted in [Figure 16](#token-allocation-chart), that includes: #### Figure 17:

* ***Investors:*** Inferix raises capital from investors, including Seed, Strategic, KOLs, and Public Investors, using 22.5% of $IFX total supply. * ***Liquidity and listing:*** 0.6% for DEX and CEX. * ***Community grants fund:*** 14.4% for: * initial airdrop for OG members, * initial grants for GPU providers, * education grants for using Inferix in schools. * ***Ecosystem fund:*** 50% for: * staking reward, * PoR reward using BMW, * Inferix platform expansion projects, * Inferix hackathon and bug fixes, * Inferix DAO building, * [Guaranteed node buyback](/inferix-whitepaper/economic-model/node-sale-and-penalty-pool) program. * ***Inferix foundation:*** 12.5% for: * Inferix Labs operation activities, * Inferix global expansion. # Token vesting The token vesting is shown in [Table 2](#tab_token_vesting). #### Table 2: Token vesting | | | | --------------------- | -------------------------------------------------------------------------------------------------------------------------- | | Seed | 6 months cliff, then 12 months linear vesting. | | Strategic | 6 months cliff, then 12 months linear vesting. | | KOLs | 10% at TGE, 3 month cliff, then 15 months linear vesting. | | Public sale | 10% at TGE, 3 months cliff, then 4 months linear vesting. | | Liquidity and listing | 100% unlocked at TGE, circulated when listed | | Community Grants Fund | Unlock 3 million $IFX at TGE for community airdrop. Later schedule is under DAO governance. | | Ecosystem Fund | Locked at TGE, gradually unlocked under the BMW model for GPU providers and Ecosystem partners. Release all within 4 years | | Inferix Foundation | 12 months lock, then 24 months linear vesting. | # Governance $IFX token holders participate in the network’s governance decisions, including protocol updates, reward policy adjustments, and so on. This decentralized governance structure ensures the democratic and transparent nature of the network. Governance decisions are made through a weighted voting mechanism where $IFX holdings determine the voting weights. This encourages long-term holdings and active participation in network governance. The Inferix Network backend infrastructure is built on three core components: Worker Node, Manager Node and Verifier Node. With the expectation of serving 4 million designers, Inferix will have around 1 million Worker nodes, with a significant portion coming from the PCs of the designers themselves. In a scenario where 4 million render jobs are processed daily, the Inferix network would require approximately 2500 Manager Nodes to efficiently coordinate and handle the PoR algorithm. In this scenario, Inferix would need approximately 25000 Standard Verifier Nodes to ensure the entire network operates seamlessly with low latency. A standard verifier node is a PC that meets standard verifier [hardware requirements](/inferix-whitepaper/appendix-c-hardware-requirements-for-nodes). # Node staking and rewards Inferix allocates up to 50% of its total token supply as rewards for its node system, sourced from the [Ecosystem Fund](/inferix-whitepaper/economic-model/token-metrics-and-allocation/token-vesting). Additionally, Inferix will distribute 80% of its service revenue to the nodes through [BMW model](/inferix-whitepaper/economic-model/burn-mint-work-token-issuance-model). To earn these rewards, node owners must stake a certain amount of $IFX tokens through purchasing a node license and actively run their nodes to participate in the Inferix network. We present next the reward mechanism for the nodes in the Inferix network. # Worker Workers are the most critical component of the Inferix network. The reward pool for Worker Nodes constitutes 75% of the Ecosystem Fund and is distributed through the BMW model. In other words, three-quarters of the 80% of the network's service revenue is allocated to Workers. To receive BMW rewards, each Worker must hold an amount of $IFX tokens at any given time. This token amount acts as a penalty fund in case the Worker fails to meet Inferix's standards. The minimum stake amount of node $$n$$ can be calculated as follows: $$ S\left(n\right) \triangleq 2 \times \mathcal{E} \times \text{IB} \left(n\right) \times P\left(n\right) $$ where $$\text{IB}\left(n\right)$$ is the benchmark value of the node (c.f.\~\cref{equ:inferix\_bench}), $$P\left(n\right)$$ is the price per IBM defined by the node provider himself and $$\mathcal{E}$$ is a constant defined by number of minutes of the Epoch period, that is $$72 \times 60 = 4320$$. For example, if a provider owns a node with 32 GB RAM, Intel:registered:Core:tm:i9 CPU, and an RTX4090 card, its computing power will be approximately 0.5 IB. To participate in the Inferix network at a service price of $2 per hour, it is needed to hold a quantity of $IFX tokens equivalent to $$0.5 \times 2 \times 4320 \div 60$$ = 72 USD. *Inferix Edge:* computers are specially designed to participate in the Inferix worker network. These machines are streamlined by removing non-essential components and focusing on enhancing the CPU, GPU, RAM, and motherboard to optimize performance for Visual Computing and federated AI tasks. # Verifier Verifier is also an important component in the Inferix network and is allocated 7.5% of the total $IFX token supply for the reward pool. Inferix mobilizes a total of approximately 25000 Verifier nodes through node sale programs. The first node sale program will be announced before TGE. In order to run a Verifier node, we do not need to invest in powerful hardware like Worker nodes, it is possible to run the verification on a mobile device (see [Appendix C](/inferix-whitepaper/appendix-c-hardware-requirements-for-nodes) for more details). # Manager The Inferix network requires 2500 Manager nodes, which will be sourced through Manager Node License sales. Initially, when the mainnet goes live, the *Inferix Foundation* will deploy no more than 100 in-house nodes of this type to ensure network availability. The plan for Manager node sales will be announced before the TGE. The reward pool for Manager Nodes is 1.25% of the total supply of $IFX tokens. # Penalty pool *Penalty Pool* is a token fund collected from penalties imposed on Worker nodes that violate Inferix standards while performing assigned visual computing or AI inference tasks. The fund is used as a reward for Verifiers who actively contribute to ensuring the efficient operation of the network. # Node sale and guaranteed node buyback # Node sales Node sales are crucial programs for building Inferix's compute network. Participants in these programs are required to stake a certain amount of $IFX tokens by purchasing a node license. The number of $IFX tokens needed varies depending on the type of node and the timing of the license purchase. Node sale programs will be organized into tiers based on the order of the sale. As a general rule, the earlier the license is purchased (the lower the tier), the cheaper the price. Licenses purchased before the TGE are referred to as *whitelisted* and participants who acquire whitelisted licenses will receive more benefits and discounts compared to those who purchase after the TGE. # Guaranteed Node Buyback In order to ensure the stability of node supply, increase transparency, and safeguard the rights of network contributors, Inferix introduces the very special *Guaranteed Node Buyback* program. * ***Eligibility:*** Node license holders with a participation rate of over $66.7%$ in executing Inferix network tasks, and who have never been blacklisted, will be eligible to participate in this program. * ***Timeline:*** Starting six months after the TGE, node license holders will have a 7-day window to participate in the program. * ***Execution options:*** * either *buyback 100%* with $IFX, value equivalent to the price of nodes denominated in USDT at the time of node purchase, there are 30-days linear vesting from time of buyback. * or *immediate buyback 80%* in ETH, value equivalent to the price of nodes snapshot in ETH at time of purchase. To ensure transparency and security, Inferix will integrate the *guaranteed node buyback* program into smart contracts and collaborating with reputable, top-tier third-party auditors. A portion of the repurchased node licenses will be reassigned to current active node operators to further boost engagement. The remaining licenses will be sent to the Inferix treasury, with possible applications including distributing node operation rewards to $IFX stakers, reselling nodes, or burning them. **Guaranteed Node Buyback Fund**: To facilitate the *guaranteed node buyback*, **10,000,000 $IFX** will be allocated from the *Ecosystem Fund*, as the *Guaranteed Node Buyback Fund*. Node license holders will retain all previously distributed airdrops, even after participating in the program. # Future development # PoR and NFT minting for graphics creative assets During the process of creating their work, graphic artists frequently engage in rendering activities. The results and traces generated during rendering on Inferix can be used as proof of artistic creation. Naturally, PoR combined with native blockchains can be utilized to mint NFTs that fully capture the artist's creative process. In the future, the Inferix ecosystem could potentially host projects related to this field. # ZKP and PoR communication Zero-knowledge proofs (ZKP) have recently garnered significant attention from developers in the Web3 space. At present, applying ZKP to heavy edge computing systems, such as GPU-based rendering or AI, faces challenges due to the high computational resources required, leading to reduced processing performance. However, the Inferix development team believes that ZKP will become increasingly valuable for our use cases in the future. Therefore, we are continuing to explore experimental approaches that combine ZKP with PoR. We will publish these experiments in future papers. # Inferix RemotePC In the future, Inferix plans to develop Inferix Remote PC, a service that allows end-users to directly access Workers via remote desktop. This will enable users to have greater control over installing the necessary software on the worker machines, potentially reducing service costs. For example, if a user has a license for a render engine that Inferix’s workers do not have, they can still utilize it with Inferix's computational hardware through Inferix Remote PC. # Rendering professional network In its plan to utilize the Community Grants Fund, Inferix has programs aimed at developing a community of graphic artists and students from universities related to visual computing. These initiatives will result in the creation of a Rendering Professional Network capable of connecting end-users with professional graphic creators. This will foster sustainable growth for Inferix. # References **\[1]J. F. Hughes, A. van Dam, M. McGuire, D. F. Sklar, J. D. Foley, S. K. Feiner, and K. Akeley, Computer Graphics: Principles and Practice, 3rd ed. Addison-Wesley, 2014** **\[2] M. Wu and B. Liu, "Watermarking for image authentication," in Proceedings of International Conference on Image Processing, 1998.** **\[3] M. M. Yeung and F. Mintzer, "An invisible watermarking technique for image verification," in Proceedings of International Conference on Image Processing, 1997.** **\[4] R. B. Wolfgang and E. J. Delp, "Fragile watermarking using the vw2d watermark," in Security and Watermarking of Multimedia Contents. SPIE, 1999.** **\[5] C. Gentry, "Fully homomorphic encryption using ideal lattices," Proceedings of the 41st Annual ACM Symposium on Theory of Computing, 2009.** #### \[6] C. Gentry, "Computing arbitrary functions of encrypted data," Communications of the ACM, vol. 53, no. 3, 2010. **\[7] C. Marcolla, V. Sucasas, M. Manzano, R. Bassoli, F. H. P. Fitzek, and N. Aaraj, "Survey on fully homomorphic encryption, theory, and applications," Proceedings of the IEEE, 2022.** **\[8] I. J. Cox, J. Kilian, T. Leighton, and T. Shamoon, "Secure spread spectrum watermarking for multimedia," IEEE Transactions on Image Processing, 1997.** **\[9] I. J. Cox, M. L. Miller, and A. L. McKellips, "Watermarking as communications with side information," Proceedings of the IEEE, 1999.** **\[10] S. A. Craver, N. D. Memon, B.-L. Yeo, and M. M. Yeung, "Can invisible watermarks resolve rightful ownerships?" in Storage and Retrieval for Image and Video Databases, 1997.** **\[11] N. Jayant, J. Johnston, and R. Safranek, "Signal compression based on models of human perception," Proceedings of the IEEE, 1993.** **\[12] H. R. Wu, A. R. Reibman, W. Lin, F. Pereira, and S. S. Hemami,"Perceptual visual signal compression and transmission," Proceedings of the IEEE, 2013.** **\[13] Q. Lian, Z. Zhang, M. Yang, B. Y. Zhao, Y. Dai, and X. Li, "An empirical study of collusion behavior in the maze p2p file-sharing system," in 27th International Conference on Distributed Computing Systems, 2007.** **\[14] I. J. Cox, M. L. Miller, , and A. L. McKellips, "Watermarking as communications with side information," Proceedings of the IEEE, 1999.** **\[15] C. Cachin, "An information-theoretic model for steganography," in International Workshop on Information Hiding, 1998.** **\[16] F. A. P. Petitcolas, R. J. Anderson, and M. G. Kuhn, "Information hiding - a survey," Proceedings of the IEEE, vol. 87, no. 7, 1999.** **\[17] M. D. Swanson, B. Zhu, and A. H. Tewfik, "Transparent robust image watermarking," in Proceedings of 3rd IEEE International Conference on Image Processing, 1996.** **\[18] C. I. Podilchuk and W. Zeng, "Image-adaptive watermarking using visual models," IEEE Journal on Selected Areas in Communications, vol. 16, no. 4, 1998.** **\[19] S. Voloshynovskiy, A. Herrigel, N. Baumgaertner, and T. Pun, "A stochastic approach to content adaptive digital image watermarking," in International Workshop on Information Hiding, 2000.** **\[20] J. H. Reif, D. J. Tygar, and A. Yoshida, "Computability and complexity of ray tracing," Discrete and Computational Geometry, vol. 11, no. 3, 1994.** **\[21] J. Stern, "Secret linear congruential generators are not cryptographically secure," in 28th Annual Symposium on Foundations of Computer Science, 1987.** **\[22] J. Chou, S. S. Pradhan, and K. Ramchandran, "On the duality between distributed source coding and data hiding," in Conference Record of the Thirty-Third Asilomar Conference on Signals, Systems, and Computers, vol. 2, 1999.** **\[23] Actif3D Team. (2023) The no-code vr/ar platform for creator economy. \[Online]. Available: ** **\[24] I. Chillotti, M. Joye, and P. Paillier, "Programmable bootstrapping enables efficient homomorphic inference of deep neural networks," in Cyber Security Cryptography and Machine Learning, 2021.** **\[25] OMTP. (2009, May) Advanced trusted environment: Omtp tr1. \[Online]. Available: ** **\[26] TensorOpera. (2023, Nov) Scalellm: Unlocking llama2-13b llm inference on consumer gpu rtx 4090, powered by fedml nexus ai. \[Online]. Available: llm-inference-on-consumer-gpu-rtx-4090-powered-by-fedml-nexus-ai/** **\[27] ——. (2022) Scalellm: Serverless and memory-efficient model serving engine for large language models. \[Online]. Available: ** **\[28] Global Market Insight. (2024, May) Visual computing market size by component (hardware, software), by display platform (interactive whiteboards, interactive kiosk, interactive table, interactive video wall, monitors), by industry verticals, regional outlook and forecast, 2024 - 2032. \[Online]. Available: ** **\[29] Polaris Market Research. (2022, Sep) Federated learning market share, size, trends, industry analysis report, by application (industrial internet of things, drug discovery, risk management, augmented and virtual reality, data privacy management, others); by industry vertical; by region; segment forecast, 2022 - 2030. \[Online]. Available: ** **\[30] A. Haleem, A. Allen, A. Thompson, M. Nijdam, and R. Garg, "Helium: A decentralized wireless network," Helium Systems, Tech. Rep., 2018.** **\[31] P. Snow, B. Deery, J. Lu, D. Johnston, and P. Kirby, "Factom: Business processes secured by immutable audit trails on the blockchain," Factom Protocol, Tech. Rep., 2018.** **\[32] K. Samani. (2018, Feb) New models for utility tokens. \[Online]. Available: ** **\[33] Super Renders Farm. (2024) Render farm pricing. \[Online]. Available: ** **\[34] RebusFarm. (2024) Render farm prices and discounts. \[Online]. Available: ** **\[35] Fox Renderfarm. (2024) Render farm pricing. \[Online]. Available: ** # Appendix A: Proofs ### Fourier transform of complex atomic signals Substitute the atomic signal: $$ \begin{equation\*} w\_i\left(x,y\right) = A e^{2i\pi\left(\frac{x}{X\_i} + \frac{y}{Y\_i}\right)} \ \left(0 \leq x < M, 0 \leq y < N \right) \end{equation\*} $$ into the discrete Fourier transform: $$ \begin{equation\*} F(u,v) = \frac{1}{M \times N} \sum\limits\_{x = 0}^{M - 1} \sum\limits\_{y = 0}^{N - 1} w\_{i}(x,y) e^{2i\pi \left(\frac{x u}{M} + \frac{y v}{N}\right)} \end{equation\*} $$ We have $$ \begin{equation\*} F\left(u,v\right) = \frac{A}{M \times N} \sum\limits\_{x = 0}^{M - 1} \sum\limits\_{y = 0}^{N - 1} e^{2i\pi\left(\frac{x}{X\_i} + \frac{y}{Y\_i}\right)} e^{2i\pi \left(\frac{x u}{M} + \frac{y v}{N}\right)} \end{equation\*} $$ Moreover $$ \begin{align\*} &\sum\limits\_{x = 0}^{M - 1} \sum\limits\_{y = 0}^{N - 1} e^{2i\pi\left(\frac{x}{X\_i} + \frac{y}{Y\_i}\right)} e^{2i\pi \left(\frac{x u}{M} + \frac{y v}{N}\right)} \ =&\sum\limits\_{x = 0}^{M - 1} \sum\limits\_{y = 0}^{N - 1} e^{2i\pi x \left(\frac{1}{X\_i} + \frac{u}{M}\right)} e^{2i\pi y \left(\frac{1}{Y\_i} + \frac{v}{N}\right)} \ =&\left(\sum\limits\_{x = 0}^{M - 1} e^{2i\pi x \left(\frac{1}{X\_i} + \frac{u}{M}\right)}\right) \left(\sum\limits\_{x = 0}^{N - 1} e^{2i\pi x \left(\frac{1}{Y\_i} + \frac{v}{N}\right)}\right) \end{align\*} $$ Using geometric sum formulae: $$ \begin{align\*} =& \frac{1 - e^{2i\pi M \left(\frac{1}{X\_i} + \frac{u}{M}\right)}}{1 - e^{2i\pi \left(\frac{1}{X\_i} + \frac{u}{M}\right)}} \frac{1 - e^{2i\pi N \left(\frac{1}{Y\_i} + \frac{v}{N}\right)}}{1 - e^{2i\pi \left(\frac{1}{Y\_i} + \frac{v}{N}\right)}} \ =& \frac{1 - e^{2i\pi \frac{M}{X\_i} + 2i\pi u}}{1 - e^{2i\pi \left(\frac{1}{X\_i} + \frac{u}{M}\right)}} \frac{1 - e^{2i\pi\frac{N}{Y\_i} + 2i\pi v}}{1 - e^{2i\pi \left(\frac{1}{Y\_i} + \frac{v}{N}\right)}} \ =& \frac{1 - e^{2i\pi \frac{M}{X\_i}}}{1 - e^{2i\pi \left(\frac{1}{X\_i} + \frac{u}{M}\right)}} \frac{1 - e^{2i\pi\frac{N}{Y\_i}}}{1 - e^{2i\pi \left(\frac{1}{Y\_i} + \frac{v}{N}\right)}} \end{align\*} $$ Hence: $$ \begin{equation\*} F\left(u,v\right) = \frac{A}{M \times N} \frac{1 - e^{2i\pi \frac{M}{X\_i}}}{1 - e^{2i\pi \left(\frac{1}{X\_i} + \frac{u}{M}\right)}} \frac{1 - e^{2i\pi\frac{N}{Y\_i}}}{1 - e^{2i\pi \left(\frac{1}{Y\_i} + \frac{v}{N}\right)}} \end{equation\*} $$ ### Convergence of energies The random variables $$\mathcal{X}\_i, \mathcal{Y}\_i \left(i \in \mathbb{N}\right)$$ are independent then two variables: $$ \overline{\mathcal{X}}\_n = \frac{\mathcal{X}\_1 + \dots \mathcal{X}\_n}{n} \quad \text{and} \quad \overline{\mathcal{Y}}\_n = \frac{\mathcal{Y}\_1 + \dots \mathcal{Y}\_n}{n} $$ are independent for all $$n \in \mathbb{N}$$. Since $$\mathcal{X}\_i \left(i \in \mathbb{N}\right)$$ are independent and identically distributed and $$\mathcal{X}\_i \sim \mathcal{N}\left(\mu, \sigma^2\right)$$ $$ \overline{\mathcal{X}}\_n \xrightarrow\[n \to \infty]{a.s} \mu $$ by strong law of large numbers; similarly $$\overline{\mathcal{Y}}\_n \xrightarrow\[n \to \infty]{a.s} \mu$$. Hence $$ {\[\overline{\mathcal{X}}\_n \quad \overline{\mathcal{Y}}\_n]^t} \xrightarrow\[n \to \infty]{a.s} {\[\overline{\mathcal{X}} \quad \overline{\mathcal{Y}}]^t} $$ for some independent and identically distributed variables $$\overline{\mathcal{X}} \sim \overline{\mathcal{Y}} \sim \mu$$. Let us fix some $$0 \leq u < M$$ and $$0 \leq v < N$$, then it is obvious that the function $$ F\_{\left(u,v\right)} \triangleq \left(x,y\right) \mapsto \frac{A}{M \times N} \frac{\left(1 - e^{2i\pi \frac{M}{x}}\right) \left(1 - e^{2i\pi \frac{N}{y}}\right)}{\left(1 - e^{2i\pi\left(\frac{1}{x} + \frac{u}{M}\right)}\right) \left(1 - e^{2i\pi\left(\frac{1}{y} + \frac{v}{N}\right)}\right)} $$ is continuous. We define the variable: $$ \mathcal{F}*{\left(u,v\right)} \colon \Omega \to \mathbb{R}^2 \ \omega \mapsto F*{\left(u,v\right)} \left(\mathcal{X} \left(\omega\right),\mathcal{Y}\left(\omega\right)\right) $$ for some random variables $$\mathcal{X}$$ and $$\mathcal{Y}$$, then $$\mathcal{F}*{\left(u,v\right)}\left(\omega\right)$$ is nothing but the Fourier transform of the atomic signal whose the horizontal and the vertical period are $$\mathcal{X}\left(\omega\right)$$ and $$\mathcal{Y}\left(\omega\right)$$ respectively. Let us consider the sequence $$\left(\mathcal{F}*{\left(u,v\right)}^n\right)\_{n \in \mathbb{N}}$$ where: $$ \mathcal{F}*{\left(u,v\right)}^n \triangleq \omega \mapsto F*{\left(u,v\right)} \left(\overline{\mathcal{X}}\_n \left(\omega\right),\overline{\mathcal{Y}}\_n\left(\omega\right)\right) $$ By the continuous mapping theorem: $$ \mathcal{F}*{\left(u,v\right)}^n \xrightarrow\[n \to \infty]{a.s} \omega \mapsto F*{\left(u,v\right)} \left(\overline{\mathcal{X}} \left(\omega\right),\overline{\mathcal{Y}}\left(\omega\right)\right) $$ Since we have proved that $$\overline{\mathcal{X}} \sim \overline{\mathcal{Y}} \sim \mu$$ then the limit of the sequence is the constant variable $$F\_{\left(u,v\right)} \left(\mu, \mu\right)$$. Also, since $$F\_{\left(u,v\right)}$$ is continuous: $$ \lvert F\_{\left(u,v\right)}\left(\overline{\mathcal{X}}\_n, \overline{\mathcal{Y}}*n\right) - \overline{F}*{\left(u,v\right)}\left(\mathcal{X}\_n, \mathcal{Y}\_n\right) \rvert \xrightarrow\[n \to \infty]{} 0 $$ Hence $$\overline{F}\_{\left(u,v\right)}\left(\mathcal{X}\_n, \mathcal{Y}*n\right) \xrightarrow\[n \to \infty]{a.s} F*{\left(u,v\right)} \left(\mu, \mu\right)$$. ***Remark.*** The proof does not require that $$X\_i$$ and $$Y\_i$$ have the same distribution. Indeed, if $$X\_i \sim \mathcal{N}\left(\mu\_{x},\cdot\right)$$ and $$Y\_i \sim \mathcal{N}\left(\mu\_{y},\cdot\right)$$ then $$\overline{F}\_{\left(u,v\right)}\left(\mathcal{X}\_n, \mathcal{Y}*n\right) \xrightarrow\[n \to \infty]{a.s} F*{\left(u,v\right)} \left(\mu\_x, \mu\_y\right)$$. # Appendix B: Price simulation details [Table 17](#fig_price_simulation_details) contains price simulation detailed data. We are calculating the Provider's input costs, which include electricity costs and hardware depreciation. The calculations are standardized for a single GPU device. #### Table 17

* *Electricity Cost* is calculated based on the power consumption of the device multiplied by the assumed unit price of $0.20/kW h. Hardware depreciation costs are calculated based on a general assumed depreciation rate of 25% per year for both the GPU and other PC components. * *Provider Price:* The suggested unit price for Providers is calculated as double the Provider's input costs per GPU per hour. * The input cost for Inferix is calculated by adding the storage costs to the Provider price. This storage cost is based on the assumption that each 3D scene requires 1 GB of short-term storage for 96 hours and 100 MB of long-term storage, with a rate of $0.05/GB per month. * *Inferix Service Price* is calculated by adding a 20% commission for Inferix Foundation to the input cost. # Appendix C: Hardware requirements for nodes * Worker node: The minimum requirements for a single license for manager node are as follows: * NVIDIA Tesla T4 or higher GPU * 2 $$\texttt{x}$$64 Intel:registered:Core:tm: 2.1 GHz CPU * 16 GB RAM * 100 GB disk space * 100 Mbit/s internet connection * Manager node: The minimum requirements for a single license for manager node are as follows: * 6 $$\texttt{x}$$64 Intel:registered:Core:tm: 2.1 GHz CPU * 64 GB RAM * 1000 GB disk space * 100 Mbit/s internet connection * Secure manager node: The minimum requirements for a single license for secure manager node are as follows: * 6 $$\texttt{x}$$64 Intel:registered:SGX Core:tm: 2.1 GHz CPU * NVIDIA GeForce RTX3090 GPU * 64 GB RAM * 1000 GB disk space * 100 Mbit/s internet connection * Standard verifier node: The minimum requirements for a single license for standard verifier node are as follows: * 1 $$\texttt{x}$$64 Intel:registered: Core:tm: 2.1 GHz CPU * 8 GB RAM * 10 GB disk space * 10 Mbit/s internet connection * Mobile verifier node: The minimum requirements for a single license for mobile verifier node are as follows: * Octa-core Cortex:registered:-A55 2.2 GHz CPU or equivalent * 4 GB RAM * 10 GB disk space * 10 Mbit/s internet connection ***Remark.*** *Mobile verifier node can be used only for PoR verification tasks. This type of node cannot run other types of verification.* * Secure verifier node: The minimum requirements for a single license for standard verifier node are as follows: * 1 $$\texttt{x}$$64 Intel:registered:SGX Core:tm: 2.1 GHz CPU * NVIDIA GeForce RTX3090 GPU * 8 GB RAM * 10 GB disk space * 30 Mbit/s internet connection * Standard unit node: The hardware requirements for a standard unit node must be ***exactly*** as follows: * 1 Intel:registered:Core:tm: i9 CPU * 32 GB RAM * 20 GB SSD disk space * 100 Mbit/s internet connection # Appendix D: Performance evaluation data The detailed data for the performance evaluation is given tables below. In any table, each row shows the execution time (in second) of inserting a vector noise into a Blender scene whose name is given in the table name, and the time (also in second) of verifying the distortions regions on a frame rendered from the scene. The tests are proceeded on a workstation of Intel:registered:Core:tm:i5 2.5 GHz CPU, 32 GB RAM. The noise insertion and noise verification do not need GPU. #### Table 3: | | | | | ------------- | -------------- | ----------------- | | Vector length | Insertion time | Verification time | | 2 | 0.747 | 0.343 | | 5 | 0.785 | 0.845 | | 7 | 0.822 | 1.059 | | 10 | 0.871 | 1.525 | | 12 | 0.923 | 1.855 | | 15 | 0.982 | 2.352 | | 17 | 1.061 | 2.612 | | 19 | 1.062 | 2.929 | | 20 | 1.092 | 3.081 | | 21 | 1.094 | 3.217 | | 23 | 1.163 | 3.508 | | 25 | 1.204 | 3.786 | | 27 | 1.241 | 4.182 | | 29 | 1.330 | 4.475 | | 30 | 1.335 | 4.673 | | 32 | 1.394 | 5.059 | | 35 | 1.472 | 5.585 | | 37 | 1.490 | 5.982 | | 40 | 1.612 | 6.409 | Coca-Cola #### Table 4: | | | | | ------------- | -------------- | ----------------- | | Vector length | Insertion time | Verification time | | 2 | 0.920 | 0.477 | | 5 | 1.300 | 1.077 | | 7 | 1.524 | 1.412 | | 10 | 2.085 | 1.976 | | 12 | 2.239 | 2.286 | | 15 | 2.550 | 2.722 | | 17 | 2.865 | 3.068 | | 19 | 3.153 | 3.453 | | 20 | 3.269 | 3.527 | | 21 | 3.439 | 3.747 | | 23 | 3.592 | 4.123 | | 25 | 3.975 | 4.546 | | 27 | 4.279 | 4.913 | | 29 | 4.594 | 5.313 | | 30 | 4.666 | 5.557 | | 32 | 5.087 | 5.812 | | 35 | 5.434 | 6.265 | | 37 | 5.729 | 6.654 | | 40 | 6.229 | 7.116 | Grease Pencil Bike scheme #### Table 5: | | | | | ------------- | -------------- | ----------------- | | Vector length | Insertion time | Verification time | | 2 | 3.959 | 0.546 | | 5 | 7.125 | 0.922 | | 7 | 7.468 | 1.251 | | 10 | 8.938 | 1.642 | | 12 | 9.934 | 1.871 | | 15 | 12.773 | 2.443 | | 17 | 13.721 | 2.751 | | 19 | 14.028 | 3.083 | | 20 | 15.351 | 3.228 | | 21 | 15.766 | 3.347 | | 23 | 17.384 | 3.717 | | 25 | 18.587 | 3.978 | | 27 | 21.182 | 4.281 | | 29 | 21.549 | 4.523 | | 30 | 22.224 | 4.706 | | 32 | 23.549 | 5.037 | | 35 | 27.202 | 5.444 | | 37 | 29.063 | 5.764 | | 40 | 29.277 | 6.087 | Blender 3.5 Splash scheme #### Table 6: | | | | | ------------- | -------------- | ----------------- | | Vector length | Insertion time | Verification time | | 2 | 2.313 | 0.657 | | 5 | 5.012 | 1.152 | | 7 | 10.228 | 1.498 | | 10 | 15.495 | 1.997 | | 12 | 17.667 | 2.415 | | 15 | 19.627 | 2.964 | | 17 | 23.986 | 3.311 | | 19 | 26.091 | 3.738 | | 20 | 27.808 | 3.816 | | 21 | 28.326 | 4.080 | | 23 | 31.014 | 4.479 | | 25 | 32.523 | 4.770 | | 27 | 34.605 | 5.151 | | 29 | 36.062 | 5.462 | | 30 | 36.456 | 5.710 | | 32 | 38.921 | 6.031 | | 35 | 42.441 | 6.496 | | 37 | 44.138 | 6.919 | | 40 | 49.456 | 7.458 | Bathroom Above Corner scheme # Worker Node Guide All about Inferix Worker Node [**What is Worker Node**](/worker-node-guide/what-is-worker-node) [How do the Worker Node work](/worker-node-guide/what-is-worker-node/how-do-the-worker-node-work) [Worker Node Rewards](/worker-node-guide/what-is-worker-node/worker-node-rewards) [How to run Worker Node](/worker-node-guide/what-is-worker-node/how-to-run-worker-node) [What is the Worker Node License (NFT)](/worker-node-guide/what-is-worker-node/what-is-the-worker-node-license-nft) **Worker Node Sales** [Guide to Purchase Worker Nodes](/worker-node-guide/worker-node-sales/guide-to-purchase-worker-nodes) [Worker Node Sale Timeline](/worker-node-guide/worker-node-sales/worker-node-sale-timeline) [Node Supply, Price, Tiers and Purchase Caps](/worker-node-guide/worker-node-sales/node-purchase-caps) [Guaranteed Node Buyback](/worker-node-guide/worker-node-sales/guaranteed-node-buyback) [How to get Node Whitelisted?](/worker-node-guide/worker-node-sales/how-to-get-whitelisted) [Smart Contract Addresses](/worker-node-guide/worker-node-sales/smart-contract-addresses) [User Discounts & Referral Program](/worker-node-guide/worker-node-sales/referral-program) [Worker Node Purchase FAQ](/worker-node-guide/worker-node-sales/worker-node-purchase-faq) [ABKK Collaboration FAQ](/worker-node-guide/worker-node-sales/abkk-collaboration-faq) # What is Worker Node Worker Node is one of the fundamental components of the Inferix network backend infrastructure: * **Worker Node:** The most essential component in the network, responsible for the bulk of the network’s rendering and processing tasks. Inferix will mobilize about 100,000 Worker Nodes in total. Workers receive rewards from the network’s service revenue (75% of the Ecosystem Fund). Each Worker must hold a certain amount of $IFX tokens as a penalty fund to ensure they meet Inferix standards.\ \ All worker nodes must be equipped with GPUs .The hardware requirements for them is defined [HERE](/inferix-whitepaper/appendix-c-hardware-requirements-for-nodes).
* **Verifier Node:** These nodes ensure the reliability and accuracy of the network. Inferix will have about 25,000 Verifier Nodes, and they are crucial for maintaining the integrity of the system. Unlike Workers, Verifier nodes don’t require heavy hardware and can be run on mobile devices. They receive 7.5% of the $IFX token supply as rewards.
* **Manager Node:** Manager nodes coordinate the overall network and are critical for handling the Proof of Render (PoR) algorithm. Inferix plans to have 2,500 Manager Nodes to manage millions of render jobs daily. The reward pool for Manager Nodes is 1.25% of the $IFX token supply. ***Read more:*** [How do the Worker Node work](/worker-node-guide/what-is-worker-node/how-do-the-worker-node-work) [Worker Node Rewards](/worker-node-guide/what-is-worker-node/worker-node-rewards) [How to run Worker Node](/worker-node-guide/what-is-worker-node/how-to-run-worker-node) [What is the Worker Node License (NFT)](/worker-node-guide/what-is-worker-node/what-is-the-worker-node-license-nft) # How do the Worker Node work The graphics rendering service consists in a network of decentralized machines called *nodes* which are of 3 kinds: *manager*, *verifier* and ***worker***. The *managers* are dedicated machines of Inferix while *verifiers* and ***workers*** are machines joined by GPU owners. A typical AI inference or graphic rendering session contains several steps here we explain main Worker Nodes in short-terms :point\_down: ([read the complete documentation about Worker Nodes HERE](/inferix-whitepaper/decentralized-visual-computing/worker-node)) The task controller of the *manager* receives the AI/rendering job request, the job will be split to several tasks and assigned to *workers.* After the tasks finished, the *Worker* sent the results back to manager and they will be verified by Worker Nodes. After the results are successfully verified, an amount of rewarded IFX tokens will be sent from Inferix Blockchain system to *worker.* *** *Penalty Pool* is a token fund collected from penalties imposed on Worker Nodes that violate Inferix standards while performing assigned visual computing or AI inference tasks. The Penalty Pool is used as a reward for Verifiers who actively contribute to ensuring the efficient operation of the network.
# Worker Node Rewards For their role in Inferix Network, Worker Node operators will receive $IFX tokens. \ 37.5% of the total $IFX token supply is allocated for Worker's reward pool. This is the biggest reward pool of Inferix token economy system. # How to run Worker Node ### 1. **Meet with the minimum requirements:** > The minimum requirements for a single license for Worker Node are as follows: > > * NVIDIA Tesla T4 or higher GPU > * 2 x64 CPU cores 2.1 GHz > * 16 GB RAM > * 100 GB disk space > * 100 Mbit/s internet connection ### 2. How to run the Inferix Worker Follow these steps: * Download and install Inferix worker software: * If you own a computer with GPU equipped, join Inferix [Mining Staking](https://provider.inferix.io/) platform[ ](https://sharedmine.inferix.io/)as a GPU provider and start receiving rewards ### 3. Don't have a computer meets the requirements? Don't worry, you still be able to get rewards by joining our [SharedMine ](https://sharedmine.inferix.io/)platform. # What is the Worker Node License (NFT) The Worker Node License (as an ERC721 NFT) allows you to be part of Inferix ecosystem, earning rewards by running a Worker Node Client. You can choose to run your Worker Node Client on your on devices matching the min. requirements. Check the min. req. and how to run the worker node [HERE](/worker-node-guide/what-is-worker-node/how-to-run-worker-node). You can [purchase](/worker-node-guide/worker-node-sales/guide-to-purchase-worker-nodes) your NFT either from the [Node Sales](/worker-node-guide/worker-node-sales) or through secondary market in the future. You also can refer the official [Smart Contracts Addresses](/worker-node-guide/worker-node-sales/smart-contract-addresses) of Worker Node sales. Worker Node License NFTs are non-transferable from the purchase wallet for the first year. Read more information on the Worker Node Purchase FAQ :point\_down: {% content-ref url="/pages/Qmq0QRbBoHAXZpm8n0rH" %} [Worker Node Purchase FAQ](/worker-node-guide/worker-node-sales/worker-node-purchase-faq) {% endcontent-ref %} # Worker Node Sales You will be able to acquire Worker node licenses through Node Sales, crucial programs for building Inferix's compute network. There will be **2 type of Node Sales** for you to participate: * **Whitelist Sale (if eligible).** * **Public Sale.** Node Sale programs will be organized into tiers based on the order of the sale. As a general rule, the earlier the license is purchased (the lower the tier), the cheaper the price. Licenses purchased before the TGE are referred to as Whitelisted licenses. Participants who acquire Whitelisted licenses will receive more benefits and discounts compared to those who purchase after the TGE. (see info about whitelist [HERE](/worker-node-guide/worker-node-sales))\ \ **Inferix is offering 100,000 nodes across 30 tiers**. Each tier has a limited number of licenses, and prices increase as the tiers progress. Here’s a brief outline of the pricing: * Tier 1: $300 per node * Tier 2: $350 per node * Tier 3: $400 per node * Tier 4: $450 per node * Prices continue to rise until Tier 30, where the price reaches $1,750 per node **The nodes will be available for purchase directly on our Node Sale Page** ([**https://worker.inferix.io/**](https://worker.inferix.io/)). Please ensure you have enough **USD₮0** *(USDT token on Arbitrum One network)* in your wallet to complete the transaction.\ \ Once purchased, **your node will be represented as an ERC-721 NFT, which will be airdropped directly to your wallet.** This will happen 72 hours after the node is purchased. *** During the **public sales,** there are limits on how many node licenses you can buy: * **Tier 1:** Max of 10 nodes per wallet * **Tier 2:** Max of 20 nodes per wallet * **Tier 3:** Max of 25 nodes per wallet * **Tiers 4 - 8:** Max of 30 nodes per wallet **Tiers 9 - 30:** There is **no limit** on the number of nodes you can purchase in these tiers. Each tier has a limited number of licenses, so early purchases will be more cost-effective. ***Read more:*** [Guide to Purchase Worker Nodes](/worker-node-guide/worker-node-sales/guide-to-purchase-worker-nodes) [Worker Node Sale Timeline](/worker-node-guide/worker-node-sales/worker-node-sale-timeline) [Node Supply, Price, Tiers and Purchase Caps](/worker-node-guide/worker-node-sales/node-purchase-caps) [Guaranteed Node Buyback](/worker-node-guide/worker-node-sales/guaranteed-node-buyback) [How to get Node Whitelisted?](/worker-node-guide/worker-node-sales/how-to-get-whitelisted) [Smart Contract Addresses](/worker-node-guide/worker-node-sales/smart-contract-addresses) [User Discounts & Referral Program](/worker-node-guide/worker-node-sales/referral-program) [Worker Node Purchase FAQ](/worker-node-guide/worker-node-sales/worker-node-purchase-faq) [ABKK Collaboration FAQ](/worker-node-guide/worker-node-sales/abkk-collaboration-faq) # Guide to Purchase Worker Nodes ### A video walk-through on how to purchase a node: {% embed url="" %} **Step 1 - Get USDT tokens to buy the Node and for gas fee.**\ *Nodes can be purchased using* [USD₮0](https://arbiscan.io/token/0xfd086bc7cd5c481dcc9c85ebe478a1c0b69fcbb9), the bridged [USDT ](https://etherscan.io/token/0xdac17f958d2ee523a2206206994597c13d831ec7)on [***Arbitrum One***](https://docs.arbitrum.io/build-decentralized-apps/public-chains#arbitrum-one) ***Network**.* {% hint style="info" %} Don't know how to get USDT on Abitrum One Network? * Go to [Arbitrum Bridge](https://bridge.arbitrum.io/), connect your Metamask wallet, in "**Select Network**" dialog set "**Testnet mode**" to **OFF**, choose your "From" and "To" network and convert your USDT token assets to USD₮0 * If you don't have USDT token in your wallet, go to a centralized exchange, such as Binance, buy USDT and then withdraw it to your wallet address using the Arbitrum One network {% endhint %} **Step 2 - Go to the Node Sale Page (**[**https://worker.inferix.io**](https://worker.inferix.io/)**) and connect your wallet.** \ Please take into consideration r*ewards from nodes will be airdropped to the same wallet used to purchase it, and this cannot be changed.* **Step 3 - Select available tier and click “Purchase”**\ There are a total of 30 tiers of nodes available for purchase. For detailed information about each node read ["Worker Node Sales"](/worker-node-guide/worker-node-sales) and ["Worker Node Purchase FAQ"](/worker-node-guide/worker-node-sales/worker-node-purchase-faq) **Step 4 - Approve the purchase of your Inferix Node License**\ Click on “Approve” on the sale page. You will be prompted to sign a contract in your wallet. Once it has been signed, accept. {% hint style="info" %} Make sure you have enough ETH *on Arbitrum One network* in your wallet to cover gas fee! A node purchasing order typically incurs gas costs equivalent to about $0.10–$1.00 worth of ETH. Not sure how to get ETH on Arbitrum One network? * Go to [Arbitrum Bridge](https://bridge.arbitrum.io/), connect your Metamask wallet, in the "**Select Network**" dialog set "**Testnet mode**" to **OFF;** choose your "From" and "To" networks then convert your ETH assets to Arbitrum One ETH * If you don't have ETH token in your wallet, go to a centralized exchange, such as Binance, buy ETH and then withdraw it to your wallet address using the Arbitrum One network {% endhint %} ### **Congratulations you have now successfully purchased your Inferix Node License!** \ **What’s next? Important info!** :point\_down:\ \ Within 72 hours after purchasing Inferix nodes, investors will receive ERC-721 NFTs to the wallet addresses used during the Inferix node sale. The NFT symbolizes ownership of the purchased Inferix Worker node. Keep in mind that in order to run nodes, you will need a PC that match the min. requirements for EACH node license you buy. ([Min. hardware req. ](/inferix-whitepaper/appendix-c-hardware-requirements-for-nodes)[HERE](/inferix-whitepaper/appendix)) There are limits on the number of nodes you can buy **during whitelist sales** and **public sales.** (see info [HERE](/worker-node-guide/worker-node-sales)) The NFTs will be airdropped to buyer wallets directly, there's no claiming process. The NFTs will be non-transferable for the first year after the node sale. **a** After this lock-up period, you’ll be able to transfer or sell your node on the secondary market. You can also join our [Guaranteed Node Buyback](/worker-node-guide/worker-node-sales/guaranteed-node-buyback) program after purchasing nodes. # Worker Node Sale Timeline

\ May 29, 2025 Whitelist wallet address submission (For all node tiers by all partners) May 29, 2025 Whitelist winners announcement May 30, 2025 10:00 AM (UTC) Inferix Node Sale (Whitelist & Public) # Guaranteed Node Buyback In order to ensure the stability of node supply, increase transparency, and safeguard the rights of network contributors, Inferix introduces the very special ***Guaranteed Node Buyback*** program: ***Eligibility:** Node license holders with a participation rate of over $66.7% in executing Inferix network tasks, and who have never been blacklisted, will be eligible to participate in this program.*\ [*Free node airdrop*](https://docs.inferix.io/worker-node-guide/worker-node-sales/pages/NdB5EkLfGQNVCF513jKl#id-3.-free-node-airdrops) *winners will not be eligible to participate this program* ***Timeline:** Starting six months after the TGE, node license holders will have a 7-day window to participate in the program*. **Execution options:** * either *buyback 100%* with $IFX, value equivalent to the price of nodes denominated in USDT at the time of node purchase, there are 30-days linear vesting from time of buyback. * or *immediate buyback 80% of* the price of nodes denominated in USDT at time of purchase. To ensure transparency and security, Inferix will integrate the *guaranteed node buyback* program into smart contracts and collaborating with reputable, top-tier third-party auditors. A portion of the repurchased node licenses will be reassigned to current active node operators to further boost engagement. The remaining licenses will be sent to the Inferix treasury, with possible applications including distributing node operation rewards to $IFX stakers, reselling nodes, or burning them. *** **Guaranteed Node Buyback Fund**: To facilitate the *guaranteed node buyback*, **50,000,000 $IFX** will be allocated from the *Ecosystem Fund*, as the *Guaranteed Node Buyback Fund*. Node license holders will retain all previously distributed airdrops, even after participating in the program. \
# How to get Node Whitelisted? #### 1. **Node Whitelist Program** Obtain a Node Whitelist spot for our Worker Node Sales by participating in: * **Whitelist drop for Animoca Brands Japan Node Holders**\ *Click* [*HERE* ](https://docs.google.com/forms/d/e/1FAIpQLSdmw6dGY5RKqkXjCUdiB4EqnBlGlWZv4OIhWN0kiGsuREVcww/viewform?usp=header)*to join ABKK Whitelist Program* * **Other events with our strategic partners** * [**Our referral program for Influencers & Community Leaders (KOLs)**](/worker-node-guide/worker-node-sales/referral-program)\ *(If you are interested, contact our team by sending email to )* #### 2. Free Node Airdrops Some community campaigns by Inferix will airdrop node licenses completely free to individuals who achieve top performance in completing specific campaign tasks. Inferix will use tokens from the [Community Grants Fund](/inferix-whitepaper/economic-model/token-metrics-and-allocation/token-allocation) to reimburse the equivalent amount of USDT for the licenses distributed for free. All decisions regarding these free node airdrops are made through DAO governance. However, these free airdrop nodes will not be eligible to participate in [Guaranteed Node Buyback](/worker-node-guide/worker-node-sales/guaranteed-node-buyback) program. #### 3. The Winners This is the list of whitelisted winners of Worker Node Sale, with corresponding Tier number:

#	Wallet Address	Tier
1	0x9bd9f9e4b051ab0346efc3d740193096fdcb7d49	1
2	0x2B97eb170a57fa2B5ea499b9f0176Ef587c6F54d	1
3	0xe97aB3E085906Febb2a82d58b05F7359bf7cbF3a	1
4	0xF8053C193A98dc542dd6e19900Eaae86DAE60160	1
5	0xba71D40B8A82F57cCE8AC4adb9cA72eB25807Fd9	1
6	0xB2d1Fc01f2c9a45549dddAb5A13CAc750F7f73FA	1
7	0x9dE4481fbD28a2A770474f18e37e6825eb1C5696	1
8	0x97ac402a98c33ca0194bc51377159c635ce2f672	1
9	0x79b3fCD92fa1275D6dc30Bd4Bd7be4e3A29568C6	1
10	0x053b8Ebe5C8aeE5Ca41fdb86DE4849D5Be6E3c77	1
11	0x9ff786Fda13110BABb55751E54446899388519f9	1
12	0x77Dd6dF601E62f9dB48ce77790e17F773B5cB50B	1
13	0xa134686727C1E4828c6761a18EbD2315c0FfF147	1
14	0x3d4294100A8bb0242Cd735e8b1F38dEC5C1408Ff	1
15	0x97142Fa7E481f256C94055598CA679340Ae7Ea00	1
16	0xF6f731f1638Cf3e7b355Addb5e5874Ed27125344	1
17	0xF6f731f1638Cf3e7b355Addb5e5874Ed27125344	1
18	0xd4fea4F45BA37c800CD31628A26D8B2Db04186a6	1
19	0x55BE9cb776a3D4722b9da3b833c247a82CE6d151	1
20	0x9B18e079abDB7691Fad7C1e1584fb915b1577938	1
21	0x36Cd4e5766e09c4e86D04a8f8812021E84644ED6	1
22	0x71ccf365d0f38869a77e66679958e2e74c88c77d	1
23	0x7cdb87d626b3fdf25ffcbbec8d3ae74498fc4415	1
24	0xb8e06ec74c3960702499Fe09D18cC79309Ed919F	1
25	0x628f13c0d9009c023aa6912DeeBff33E2d165014	1
26	0xEade324087F1860DF8E651159ECCe0B665E087Ae	1
27	0xbf4649c2E0Ee2Da4286DE89cED7eC0Bee379ddeF	1
28	0xB812fE6100e7217F3805Ce65d7eBEb7921af2125	1
29	0x9e4e5f46CB54c780676567AD0c53719e71e1Ac8c	1
30	0x46B16E517fE6A511a7942A69B6A70026213b5F28	1
31	0x13B6D002DBa1D81Ff9fc96A75Df4D8ee275Fd96E	1
32	0x4B96952b8A35Aafd7c6a4e40e85f59785F772A96	1
33	0x77C7aea8ABBdDd502946Ae7A387ee8BE91ea7Be4	1
34	0x44F4f80c3be6E940862768a42f75EA7275f1c7A8	1
35	0xeA30eF18E95f7Ff519A4eDc3a28E859C060525B1	1
36	0x33fe4A6B3E79615067E75bDa042F8820D7666d82	1
37	0x8E18053925DCd475f62CD613CDc73903f67FBF8B	1
38	0x302f35c3958a757439123df66EC183D715A67993	1
39	0x01631C94d189Eb4F95Bf691109E7526973B6EeAC	1
40	0x01631C94d189Eb4F95Bf691109E7526973B6EeAC	1
41	0xf5777334409900d5Ad6aB26628Bc0fA95eAD611D	1
42	0x8850a37930accc83bc048f9bfdf1d3e90479cb8a	1
43	0x031523B4082791c92BF01d58E2EEA3234DE80ec7	1
44	0x695773becca5391aaaeea1378647b8330efefe20	1
45	0x9BEa06db1bCAD999Ce9f86839C91e17C6D817311	1
46	0x6dEf40C44173250EA13a65210cD009865f4435BD	1
47	0xAb9Ff3b61c1fbC40D1ac6610b3b1093805165fb9	1
48	0xff9556aa94500534C2703CBf3f99748FFD155b9d	1
49	0x6edf4B57c948e094ce97701114cCF6dc7f987237	1
50	0x2fa5fcc93ac9b373ef0065708330f7cac493fea7	1
51	0x3d3722c8cAEDA1e445144aCdA23eaF3b4F43E9e4	1
52	0xf60Ea5e0a621d1DE96AA967b6E79909Aa461849E	1
53	0x1981e15dD25072d5159413b8A0508364f59d7Ff9	1
54	0xD078838FDA1BcDAC629b3eA9F9E0763cc1eEAc44	1
55	0x6e29DFD2B1c03439e3D9e7bCD261d35f8ed2418e	1
56	0x7f819De86a4d4AeD6fFF42aEDb7a573E3cf8F694	1
57	0x0D054059a4Ff96cFab5901A872D3C31D73Ee26b4	1
58	0x949bA6738da3aa8F98e6137387b658493729170a	1
59	0xd7F3CC7616b0C168E303debdFd0Ab1Fae10e971B	1
60	0x8773b0970afE7E31574896F615109C3f4326a673	1
61	0x703Ff8197d45Af8f1ec723a01eC208B5c1F511de	1
62	0x7f1804dDCAf785c247c0359c852A85347cB0C2b3	1
63	0xB6444e585a7BD87cB532eCbf19887a699c929394	1
64	0x2a18805a7299525cF66B3251D908a2804385466a	1
65	0x40C9E701a275C6962a5d81Eb935471369d82c072	1
66	0xA3F1B620dbc281c8A4A80C47a36e19b44D20E149	1
67	0x1D313B53f5573abf0DD81B9a30FD91062DE0867d	1
68	0x7A22B8F5bc5aD93d09Ff36789019460e76eF21ED	1
69	0x6052d1B30C7E028a861B31FDdA17E44528526B16	1
70	0x3457C198a56590c0720208912293337F41a7eA65	1
71	0x428686d0D926922d95797Fc854Fa604D51Ea67FD	1
72	0xd5A7c61A40195Ae9c2E396126A31Be5B14dc6c67	1
73	0x14B63B8D5D35eC901784438d6911Df1207AAa647	1
74	0x32A0EFDcF7c5b3371f5C76C834bEa6f5dC64837A	1
75	0x8827F917712f017e2D71164E489f50Be29583d7E	1
76	0x4E3f07a0526EbedC30e5Cedc900F8923dc1A5Ba9	1
77	0x7332F7a30918DbDF693058435Fc246EfFea73576	1
78	0x445Ee1F00e8f9bb19323711F8372F6d0BBae7038	1
79	0x23B51A3125Ba4Bad01261f7C8126f0D471207ACE	1
80	0xE7B54BA100f02EC9193CB2236e789ada6b002059	1
81	0xD18dfF76324874a67b1376dc3FC918d6BC2A4195	1
82	0x8B6c97593E1ebfd0A2907c650F0F4c66B62aC014	1
83	0x33a757E2b100a7891ed0a4BE982DbfAC7aaf7327	1
84	0xfC54829dd57d6306024Eb9d762e574918A9aA6AA	1
85	0xD5c483e0FD4bF46b7ad975B8c67cdD8c5DdD5930	1
86	0xd5DB2aB1d8427Cd0031cbb8eE4F91Bf2eF1a3DDD	1
87	0x92A3e61D8B28c7dBbF32A4249fB4b2CDEf55c7Dd	1
88	0x5B590db729a1771117144080AEF006c540f40fdA	1
89	0x0B0F816c1FCc2aeec0b41D10bC1bA33EdFB67957	1
90	0x04e6C5245731629A875F41Eb23ab4bfeadE1a2Fb	1
91	0x9C0dCC940F1a3996b57EdD74b587785E6fA237aC	1
92	0xaFF5971916f9c59ba68b90E50CeCa00B64D869fE	1
93	0x4D97D50159b963CE7DE405934e1584459f8775cf	1
94	0x609a2f5c649C97b79d66d787d476090c494caDcD	1
95	0x473CA8150553772d628E7E8e880c7Cf4C88F7e86	1
96	0x88F4FCfa7364Bbb95D011C206948a84561b90350	1
97	0xBDCd45e973A5F0eF9d0Fc0B7D12Ec693c5543736	1
98	0x950417cF22359940b50286ff48F18A64B1FF1F0D	1
99	0x877081a5872F6de5cf99e4b954aCc4e8E516647C	1
100	0xCc2f894B85509cb2F39cFC54C1e8b075424bfb37	1
101	0xAd761f9ed016c5f207ED8D5A5B8Eaae1F2504aD6	1
102	0x0DfA24DD60FBa55a190681968DAafF5Df08A3926	1
103	0x695B215d55C0C8C23895e2D8566941570CdF5813	1
104	0x55131C108b4b24F24d3674d322c1f654E6fe71af	1
105	0x068F06574e2B7Cc1bcD7cd89de4af360B2A19BFA	1
106	0x2aDeb33780dFa830868E931D221bFB57747bF101	1
107	0x2db4620DC59E31c7ddD6100AE2cC74CAe1234B10	1
108	0xBB6b23D9Cc9b9B344261cBEbd2FE5bF05Fe72575	1
109	0x6c07651CDA3F191e8342eAa1557B93C720FD7A4A	1
110	0x6068583430dF9a61e9717d65e4D31E9627DA27e3	1
111	0x71567aCa11C5ab9Aeb090Ef005A2340C5Da34b41	1
112	0x4DBfDD888Bc54fAAB3CC81ea256f732B031e3F9a	1
113	0xB387B72aE0797F5161546c9FEE78901B23135777	1
114	0x8292525C12ce1FcbB1f1c53F317cD0E2FB901671	1
115	0x09733F557c74F08BaeD81F875F7730e0B85B3c6E	1
116	0x345D0aB7eab38cBc0267Ff98Fa187c898f599C49	1
117	0x2227dc78c0782DDD68D170afc44086079f959e16	1
118	0x43e36aa7bC79aA8c6f56f3c009bCfd01e5326B38	1
119	0x34f3133A2207262f2a15BA6b9a0687cEd5714188	1
120	0x68AE0bbddAcd3d2d930a0bbc4C8FDF9DEfe452Ca	1
121	0xe8705Ff18188987a0A675B8754A6f35Ed8f7060f	1
122	0x718C6345b65Fa7AA4cBfA856d9Aa52E551591723	1
123	0x38077D7Cfa87268BdD367023CcCB28AB27B47d7d	1
124	0xC97a25f93A6cBdE9C30174ba621456844bC474f8	1
125	0x5e37BAb8BDaE325F0d1950ae415433d32CcdD04f	1
126	0x22F3fBa6C22697ED8b10eC324CaDde69E43fDCe7	1
127	0xDeE4C200CC2D95FF4bdF375688F89011581C69f5	1
128	0x77d0317bdea6E9a4841B52C182E05731dF09583B	1
129	0x6FD6A577C09DA0a0Ec9e8ec28F54c96e91528Ee6	1
130	0x2c89D1c7eB6A78AE2Fd3B54A350D30A96Ec06B15	1
131	0xcdA98433a19F2a761A73CA40904f69A55E7FdFA6	1
132	0x156017f631e73099Cdf2d38371763bC161dda86b	1
133	0x8a7A14FC747069F2955bd40b1BB08e86669aAFFe	1
134	0xcfeE217FCf249bD060FDF28c92f3E2db8C671038	1
135	0x9600aeAfc989E0B583b1e8C102D2dd7f00ce9F3E	1
136	0x93E7f9aE64Ea39F6B852708427282457C9e29b52	1
137	0x1807B2d69c5C1C3d8BC65707B6BfB4757424Bf32	1
138	0x8e1e62054Ff0d1130c7455C221CcEe416bF8E6FA	1
139	0x3bAcefD9a4A061E91D5d875E4593e49F0bf06dC6	1
140	0x6C0fb749483eC1f2256768566A7C50b474c77CB2	1
141	0xACDA474B2dE8ba8e27d0a45D800bB9520F7E2667	1
142	0xF15042a7E6BBa240126969Eb66DF16f1437E768B	1
143	0xe7D726269c447E864D822Bae927EB85607E74c09	1
144	0x0D829d4Ac52751d977bb15D0bDd9666dD444E561	1
145	0x55226Fd7b947a6cf6a64456D237256D73a6009e3	1
146	0xf652dF96e19aA6dF607E89A7049b7f094269da33	1
147	0x7a1896738c551311956bFe203B995f8165D0a479	1
148	0x69bbd15C9cE0b382A4C74e6E328D9627a3dB2Ce8	1
149	0x2CE154768Aae2095Cc1F039C644Fd0DCC10CfdF4	1
150	0xcE84ae05B916Ad976FA1E3693F047A1bD2A8A01a	1
151	0xe28c74B86afe8488120B05FdB186128de3EC35E3	1
152	0x0b22a639D358D4bE8d7182c166ac047D939eE585	1
153	0xb93dbfB09F63186F684B0eAd7Bf1d2E568519C6A	1
154	0x076240D68C411BFa4Cfbe9cfaA5D126c7E9C6D51	1
155	0x70E34C256227b5093dDA9aEda1Ba8aa846C2722e	1
156	0x92b52366D3342ed3E07B2E183A77D9b473a543e9	1
157	0xc1721A88a4494fD39fD26e8a8f5F54901ada8c5d	1
158	0x3aB3cE3a63Cbf19EE4b0AC9e3a99F7a31666209a	1
159	0x8401153894c8Dd7098b859231C938266BD8d3141	1
160	0x3839f334dAE75CA150484d916d18b338E16d45CF	1
161	0x0F279C1057705F44A6E4da94371ff415925203e7	1
162	0x7E1acBB38eb4771a768AF5C598E81fdFff37a877	1
163	0xbb93f740eA98997569601883Af6C2Ad89bAa4596	1
164	0xfD4324613A8133fEe7228861650300775ee92eD5	1
165	0x128e73A342e6d2f6e360370D2FeFD6E26921fe3a	1
166	0xc52dD127406a5eBB8b845F963376b3997C517A86	1
167	0xac6E22727957C019041B5Ab1774f5dC261E6fb3c	1
168	0xE8f3e78808e38c4535C95Eaef5Aad8C7Bd192D2D	1
169	0x4E7fFf4D7eaAa5026d8A416B1E0570B918aC4DC5	1
170	0x9Ab8D3baE4Ee512F9FAd640F10b991b1951aBFfD	1
171	0x0863809E128f67944a934027e39F2780B28c33D7	1
172	0x7Dea38569F2430e52aa3A0E1bB67533722E2296C	1
173	0x3111c62a40FECC7c8BAAbB383c923DaF73e4Ff87	1
174	0xbb06f7B59d33bdD162589F9528b0FE8CD63C4707	1
175	0x2C3eD0211d5EA74Ce3dA545b7AF217e4284eA030	1
176	0x562f66c8Dc5AB8968d46686A1F28bB27438b52e0	1
177	0xF52b20129381e545dCcf95C8950FE12b0Fe3a533	1
178	0x358cFE338F7ca44F427fe869C706b95425095AdA	1
179	0x44CaFEA2fb65209220383377D1cd52De635e4938	1
180	0x8eDd9297EEa7D8E6054f7BD2cb022b94a7E0B710	1
181	0x5C90a8fbF68Ca2A347bF91B2EC77dE8ed30ea4A4	1
182	0xF562d25896408083F7eb712d3A905932b8fED00B	1
183	0x6340906cA14133F5f3C8A83526AA6de45ED9B338	1
184	0x583251B299b932eb1710CEc2b41153F01adE3841	1
185	0x621D08bBe2Ab511254a40A714fbDDCF35828C0e6	1
186	0x1643E33eE3F75a3E489fe3F54316Bf3F4fD7B76a	1
187	0x90Fb854EeA61Efc684d258578B7ACf27A59Be8C0	1
188	0x0827a1026B2E216EB8bB19C5fafA1f87326d3EE1	1
189	0x0d8Abcf5029B3E3797c23BB0DeE3320072A29669	1
190	0x48718268d87b9102B47c547F5C3001c355EbE96A	1
191	0xf5213b1a50F27F574E7aB3F67F16c8d7Cf03cF1E	1
192	0xD217751266Bcbf5B8DDE9aE77Bb77926AB4f42e0	1
193	0x0c018255dc32C2E0bfBeFA3D77420fd2Bea6CBb8	1
194	0xF5925eE74BB34641359b138B99d9b34A079dEca9	1
195	0xc8cAEb793c16731DaEcEC3dD58a565c79B8291C0	1
196	0x946B4BdF922c824B84B2eA1589965164b5d6DE42	1
197	0x538AE82A0c04a1BFE8055b16DEbBeB67bAA263F4	1
198	0x334160B79E812D5dF64C6Ae79FFC32E4E5E64412	1
199	0xE116961BF7387ce5864d516Bf31E199e84917Bb4	1
200	0xF2daBEAAdf8E906f513F96Cf6A63e457a448a290	1
201	0x8A83b9EbCC8E7335a8E593B8A8b2953563778e0E	2
202	0x54c1c3B9A80fFC96Ff4296A7078F9E1721D8e83b	2
203	0x3db7C21D72E50602A13284a33b0ab9AfbBd0F6b7	2
204	0x24c4501F356d29288A4bf3ebcF55e69DE02CEF11	2
205	0x0211E961E48f165150e3c4467f1778708793D519	2
206	0xf22A01B4B32676574DEDb9E7DE37f36F3c3ee7A8	2
207	0x479cabf43ED78a6A5b5b49654adA65C8B3017420	2
208	0x013B8ED355388286242CA8d6a394CE17bB4BaaEa	2
209	0xB3001378E22A946d9Ed8e3aA598262a35295F6cb	2
210	0xD158FaAD0f13dFE7e4B89B1eBCf0b85E65940Cd9	2
211	0x5d6E7863d0026503a2910C7cDC758Ef5a8624903	2
212	0x14Eb6c7c0198fA3216d7D7494c2BF4a9cf454b5c	2
213	0xc0FAB02cf8D61562f9b56AC90E1E5DC6245B7c1e	2
214	0x2523020a7D1F8E0001dBB49016cD34196590bC1D	2
215	0x28Ed326E625610f808F344F2621cD9d30129Eb06	2
216	0xC81eA5632F1465550aAcf67d170cd76F292FA6f0	2
217	0x8dBB06955428B48C39758838200183d456Bbc651	2
218	0xbe7F80137AcC92f2A5fd36F490644B55F41f54E8	2
219	0x09c0d7B4CC30Eb367ecE1E8C66B91aDa5a88AA1A	2
220	0x63a269aF3797F2eC9e4906aF6F7BD28448AD2659	2
221	0xcF592Ffe468b30cDa0F520d045A3CAdcB047eEB5	2
222	0xd4E9156250137Cf9171068e06f897A425e6Be6fD	2
223	0x32522F2979DD3A1f9a955623b90aB24dda63F943	2
224	0x318244fF583f0BF00c0e9AE288E8359ff0F952a2	2
225	0x15E4b0F2E4bb7e2656cFE1924bE043a53Aa927A3	2
226	0xff7b614a5E2E80a219F1c675538Dd0544DAaA28d	2
227	0x4cd265Fb1A48bBAabf68Ab74F1157f2D7b738aDb	2
228	0x0871B89Bf567e8FDA5846A0Bf181c22A8C9EE02A	2
229	0x917A025B3b218C1E097225eDdE8EBe5a12cC473E	2
230	0x9a4C67d09635Ea065C041c60a9b019e948F782b3	2
231	0xc97A727CDc4B2A051d884da9F37Cd048a105d25e	2
232	0xD52b4b358221ecd97bED6f060C5AFD06Eb7A6A60	2
233	0x2dCE68C96284eCd1A35e6c53091c683a479b0720	2
234	0x68adD88Ac27826649Ee74129b3F75D7Ae8f974d1	2
235	0x4dc505cB0590ac9b2D622dab1cc625A4Be0a880F	2
236	0xe8bD74FcF5f171251326F00E34021F790D0D6ED8	2
237	0xcb22E3eBA89C35b316E275f5b2D8b4a6500ADA83	2
238	0xaA49adCDB63974d56Aa14B849fCAb0732D1C7394	2
239	0xBC420a487CCeFF5D64A08524580cE960f2c23Dce	2
240	0x2169f9a08499ad4D0bF5db8438be01f87aE9a9F0	2
241	0xA68a0aCb0B5b35F550EA1813dC6612e5302913be	2
242	0xcaeCBa97Ec6fA39faf89d350081fE6c851509afE	2
243	0xC34659238C0DFE65C7c1AAFfdB5d3576A1c1E612	2
244	0xeE4cfbbf856C590D526DDcbD7B25C07522B94c99	2
245	0x63DD2b88e6df4282ac0Db575b2070E9D97855Bb7	2
246	0x52D17F60f17665a5907E4e52100563Ff6bB4e641	2
247	0x1AcD1139a8Fcfe8390d7E5a74262E3FDFDF4D310	2
248	0x549823d7E9e2A4d17a6D03e565dDeA3a3C0EEaa6	2
249	0x8C58F67267795F6aDb39bf97191CE7229aab7047	2
250	0xe6bF81279213E99ee576055d549c92A0CEC1f19E	2
251	0x369C9C6a92E2C1535A18f543Dc6c57c364244903	2
252	0x26944c12f1f932f362D6Af2F8804E782A5DDA440	2
253	0x8B5154CfCb4357Ddb7d9EE97468ae12e2e69EbB7	2
254	0x2ec9343ecBdc71429A951a80A1D0aAeb42935993	2
255	0x6f3E2066CBE3D1869B2c72Fc49E6A16f05E74d96	2
256	0x1Da6f3F8a292B6327b9A5b5Bd891DF21Aa184C72	2
257	0x125FfE86e544c3Ab4Addf2D3850694324ab95CbD	2
258	0x9B2c523bA2da2c6D66a412d0418cAF97755C4301	2
259	0x8B4c0CC33fa32B033d59F0c08E7dA07c52a3D173	2
260	0xf43164DC84E69314B1F3e313979800bc4AeEa06e	2
261	0x118bC1D8f6442638E9eed146F06BAB9fA9Aea251	2
262	0xE368aBFc27322af4720E7500eB5D2AB0C26caE87	2
263	0x90db4C532095d89805D581f462B0bd622B4aA9A8	2
264	0xE151b116364568f4689898e37f1c9657A02F20d0	2
265	0xbCfD9d9425878c8dE1C08480513648CeD4Bb06A7	2
266	0xE37fe1ce25E6C2618aD92dd4f16139e7e51ac576	2
267	0xEd814dE95bB6dD186Ce40354A060Be7349c51d6F	2
268	0xb64D154ebC6b1b1bB36E8d47cB182bb9013c00FB	2
269	0x6F85cD4Af6B4C936F8007aD454c5473768ec6b6A	2
270	0x2b79Fd0B50F598aae6b402609B33aaabEB72D093	2
271	0x5bCeF42dE1F6aaabE59A44F32d0f0Ab7Fef54d34	2
272	0x1580650D5B249Cf04dCBBD6D163c145506f9DF92	2
273	0x362c7ea1E8Fe38ADB04E1038A165D6CEbf45971B	2
274	0x08188a1f4E91c47202453D0944b128b561875090	2
275	0x79CD1d056348f91BCC4c22714cc1e8DA2D88C4cd	2
276	0x8faBc9ca3745462aC1e46cAE032059BE8FA2603E	2
277	0x8f8cdc8BE3a71F4903dC74Bf0D2D30645D21F5a8	2
278	0x0d7ea5E2f85297D99511d5786e40e6b2d5eEafE9	2
279	0x51fa8a0bbbc66ed168399342FE02ff876B6dc421	2
280	0xBfbef6E5dEad58f8D868C150F9ed855D37dF2fBa	2
281	0x005b57b79AAEC5e7C71c82EA199959b2A5de0506	2
282	0x7D973529D708784aE15320bDE518fc6E2141a0d5	2
283	0x272148f8F7d7532B2fD3dC67209553AeA0b00eB5	2
284	0xFC560c1a03c8F5403D6192d940675c663b129411	2
285	0x41883F63938A6E82B550878F57d77044754b06AF	2
286	0x78c0F3f90ca937cc55E3A64F9490A271b903c9D6	2
287	0x09f0Cf813aa011BfAA733448B64d6999d384CFCe	2
288	0x89B42bF72FaC7E17b0B58Ab8D3adA1E7A4390530	2
289	0x63dBc96536036C8a3bf5a6863729268CDe9E600f	2
290	0xd6FFB7Cb6069571E3C6A27Ea97eB8e41F0F9fEf8	2
291	0x7764690337D875EB06403434771B65B00B2D0337	2
292	0x3c5be2E6F4d5C92147B973144826b279d8A70A7a	2
293	0x32dB26735294b52b722EeF3129ad536F030B3536	2
294	0xdf946e2114c99a15AB6408B32b6686DaB6268428	2
295	0x047fFB4c07E231A0d838bD701643Ca71A7142413	2
296	0x047fFB4c07E231A0d838bD701643Ca71A7142413	2
297	0x6edf4B57c948e094ce97701114cCF6dc7f987237	2
298	0x33f8efebed89f35cb9481197c17294c4573bdb1b	2
299	0xba71d40b8a82f57cce8ac4adb9ca72eb25807fd9	2
300	0x3613B63164caE1a604a7671f8EA6dea2b4CbABac	2
301	0xf962A2c0C4A8a4C57C1f9fe53c508C87f6d9df1c	1
302	0x9b2aa18Edf8e2bE2BA70FfF2E8feAE0C2c135CA0	1
303	0x946C21B6379E06006a87c8f1475A623cbe41242a	1
304	0x29D100dd25Db80C3Deefb6647b7Eb030F35841bc	1
305	0x766BEd71A20bA3E50ea2cFFb46BA21185C54e566	1
306	0xdfc419881F833598b45c554816D03c6842259834	1
307	0xEc1DDE951A199e9b15A18366147e9130D520f16F	1
308	0x67781f23B5dee77A7A12D9020a10a56aD8378BAa	1
309	0x8F506b781967B4c95A22537c99ff2d3042133b68	1
310	0xBe17b494212685952D4cC359f9aCda981E2CA3dB	1
311	0x7da5709D28F3dfC191d6792Aee85193e5bA48D78	1
312	0xe5de1DBdB2158f00eA0CAb4FCE92F75632D956f4	1
313	0x4F7963c2a04B98e33B99c18BfDA65bB2530a617d	1
314	0x8e910bE74A3271d11BAa9B050CF100d8A6b223a7	1
315	0x1506745C43BD5468ba40f23A56f3189429Cc1D27	1
316	0xf274478259C29C03F83381177d8FF06D0E7f50C8	1
317	0xB2d2BD2968Cf1cb32fb675609F94289178563f17	1
318	0x12052a4298265Da4e62DD10218E1260f516dF72e	1
319	0x6052933e954040785F77edaBd5DB2C2368Ba4c96	1
320	0xf86c5405E7C1694F4CCFC9AC9BE8A9CE896470fc	1
321	0x4C5f4243dB617a443814C5e4883260c768C7e02E	1
322	0x8666e9404FcB2ab0C24723be17bFfCDf21278358	1
323	0x8569dBffeDB2bf1Fff9CCfeCD0E59444833dBB07	1
324	0x6a6A4a7d4e705f5bE616d68E1AA8FE075cb654df	1
325	0x3Da6FC194340bC7c3D571f628a28956491688418	1
326	0x9ef8eb1E89a94297325681497A0Ad344a8a38d62	1
327	0x724Ecac7F38e58ad62E2695F72f97C4F920A39bA	1
328	0x9Aa308598b42b8ab5ce400dA966eE0a155368E08	1
329	0x2ca3D912a65F5942584AE3dfbf668a35Fcb0124F	1
330	0xFADf9Edf52F157157a265A2804F988dE2422F697	1
331	0x6818A24C0bf86BaeE3ac55F64e63A4067a3840c7	1
332	0xb0207F301bA488344508661e612d8a1863AED866	1
333	0x1B45b6b5b470Af6eC3194A979Be866830918af09	1
334	0x3a51DF7C5b007877fF2d982b7D3e373ecDDbd973	1
335	0xb7798b779Bf4dD44EdfbEb5F484d27EfB7Ae377a	1
336	0x5033cc8Ee4c83D70A174F3D7Cf0e816024436547	1
337	0x36c8370B331e5107926e0a6ebefEC863C447959A	1
338	0xB7776184D468A98B0D9302FC199EaF64E71382fe	1
339	0x669EC4Ae6194e839Dd323Dd27D12F65475008d34	1
340	0x1F3030a9843EB458Bb592a10a353BFDc51f61B0C	1
341	0xA4615Ce4c387ED34bC39F4Ce49f8676CbD947A17	1
342	0xeD617DCE567990056a6081BB1F560e5f9d020226	1
343	0x4Ea1dFbe9110b366549691f307B1a56b15c677cE	1
344	0xaB3AD82941b4262825a4084f3910585c12D3E804	1
345	0x4Cfbbff96eA96EaeeD536FD880874c73B0baF218	1
346	0x0765878abf0AA225BF71192C87D6096DeB127107	1
347	0xc207D8d7da80a6CE6D30E22c2B3f1870599FF179	1
348	0x22a7fCC6a5b39B07D59AdD99Db09Cd78a2358A8c	1
349	0x2d478187437157dA34E2e5bcdf48Cdef4b154340	1
350	0x8e53bec8a9fdE50F499248B5fA88345f5C858E8C	1
351	0xcdf341bA41028E3856A083c688c1369180832a8F	2
352	0x9528F8637d46aEC29E46C8bc51011d5b637Cab08	2
353	0xD2E33895D9Fd7942729d376B73C9220b8eE54f8c	2
354	0xB791fBFdE0902cD02a122bDD16F306E38E75D1a7	2
355	0xbB971E2ef2a98381c7FDB9c13F4AE99a5592dDBC	2
356	0xB7dd5741FB78911fBEb7D5f272A6E66686e95484	2
357	0xce20CAa5BEA1A679EB26EB2D3a0C01ad649C99ca	2
358	0x6FAAf1FECC14DE26683cD1BFe1CEA95AfD67af8d	2
359	0xee69e096ce84D4F14CF21c29d9800b07439a232a	2
360	0xf5CDb68dc02AAE25f53e1F8A7021eaAdaA773586	2
361	0xF1157590AC6d5d0Ebe6922A4f03b514371CDF8A9	2
362	0x4D3749BF7827E182e1553Dd7Cdf261d3bC2a0a26	2
363	0xd3bF7F2A88DfCdF925349BdeBBFFE0cA5805D35F	2
364	0x5A720762DAA6772fbFCba6C19a6c5aAf32Ce65C1	2
365	0xeb3af2d15a09711eD91Df15A7Ab7cf6C7fFCA34F	2
366	0x8294FD6b0846561De37beE32A65427Ff66D72081	2
367	0x87e783120E3906e54FDbfd90D00c05F502C5e058	2
368	0xbEBEC60735A0588f21735376cd281DBd08FB03B0	2
369	0xe5F4fAf6068421ddBb14aEE41cE4c7B4037B4820	2
370	0x21BbA1bD8877743F0221C70D6240d0eaF9E38321	2
371	0xdBB7a78Be8d7096027CEb0E1529Cf50A2a491cC5	2
372	0xc23e9a67b300770617433f7dc33c6a74869e42a1	2
373	0xA72Cfa475e93281ba2618FB6f8bE2A73700C8764	2
374	0x5D7B40165c2121165F37a51DcF987FAEA485F617	2
375	0x9B042a1c9971EEc2c9Fa72bB0f4660fe1fA7Be17	2
376	0xCbcD7320184560cb7405F09d84FeD191698aF79a	2
377	0x03c0Cb2D955F58229510bb690fAC9bF5E1a02219	2
378	0x5aE5034A8168712625855cE2E70AdFa50eEf4309	2
379	0xB16922CBEE50c6516c9F75b55536F39008f33bca	2
380	0x0966C80bf0A8182206EC80589B5d30a5223695a4	2
381	0x05757cA970e98C441A6144085135F9FCb7E921e5	2
382	0x1968c7fFcC36dDdC383d8e9B7307b7ecaC81b1Fd	2
383	0x5c3D2F4d5Aa9D85071bC48018760E8afD6beE35f	2
384	0xc4978926fC8457b094eD603d6c1be93964562BEd	2
385	0x267718F207E8116203527d276D263Bd9689Fb5df	2
386	0xbb854aC9569a46945D9c88E7716B9f9008F42645	2
387	0xB951F6B65554591F6084CD5b77b6C7f7AdE32925	2
388	0x297F01AD1217D82FC04A1B0075bC2384FeA2eA07	2
389	0x4184724Cac95C1f748a7aF48c7829de5860Fe380	2
390	0xe054BF25ea92E9f037276c68070fC809b6B4D343	2
391	0x869A976c871752e7d56098d3B537e4cF128D8418	2
392	0x0C481Be8549095140C63f7a3d49DC57c8A9c1BEC	2
393	0x22A16337ea66E5c56121aB55183b44A930d10Ca3	2
394	0x33B71873518DE10670116C25B315C2C0e2734273	2
395	0x767E55cf115E580764fAe2300E410C24EE220dc7	2
396	0x9763e53f77d45055252cce111fc5e2000ce710b9	2
397	0x2429A6a80ecd3840f609deF474065Eb141F11b57	2
398	0xc7AAFE4D85aE73A944765b39b832946F310f89b8	2
399	0x2bab9437165Aa132571115B44144184C83207a55	2
400	0x16A57C9406304e6595f6de8bE2A926c725BC62d0	2
401	0xE9AC851f7AE78e87E612c80dD97F800539C201e4	2
402	0x9E92054E23D8f050dB5839E480105Cc84397D9Ae	2
403	0x0c61A5843A7cc89b3760199220A559433c20738e	2
404	0x15Ca4ED8346596D7BD93114dDC1C3017424c914e	2
405	0x300f296CF36C25aB5a3D450e6ea46A8B19E757a7	2
406	0x75e41101ea6f08019CFDd8182E01d52986Fa6704	2
407	0xC08562f2Db01cC01Ff2A8bAFED1740Fe2FE82938	2
408	0xD0Cb41a92bbE7d75500660d037206b47a310556E	2
409	0xD24887bd0473068306cd198184ac46899cce9b0D	2
410	0x733a179ad13BE6A797AdB482C94C71025ce7E1A5	2
411	0x60D71E9E0A5Fd482C6931bff4C64e6A6D3Dc949D	2
412	0x27052D064F0193714B60049578635948A5AF323e	2
413	0x0262c71Cc717Ec617190424D99a5F73DF54fDB8b	2
414	0xc38B9D138Ca005bA3D457FcC84773e513af53ef9	2
415	0x84136a878a4210Ebb9851A0Cd18162F38eEe0062	2
416	0xdc52ca73d03D041f27ec98B778261117d323ed5D	2
417	0x384A27B52097A25B71D82248f875650d9035bB43	2
418	0x624362C640d1b7C6E491e0038a5aDA67bDc0BC3D	2
419	0x526640872a60AF5Ad166100831B40c422b503915	2
420	0xc8ac8C87d6740E382CD214041035e829968ADD7E	2
421	0x5B7D021869b262312DBD3ED36222F7240B14cEe2	2
422	0x922147938dfD97be7Dd913F547036b9eD06eAdB6	2
423	0xbDCb47FBeB6E4d718c8186fcd562d408dF8779F9	2
424	0xD078838FDA1BcDAC629b3eA9F9E0763cc1eEAc44	2
425	0x21E5b22b8946f65BD01506A3437A83A882024785	2
426	0xD9F575419eB4BBc0e7E5E6f29087572795DAD81f	2
427	0x8c3B21CdAa97A73aD89115c63D6b86D1cf0B22f8	2
428	0x4104677C860a743b046F90da60274e28F0D231B7	2
429	0xD1763dd6AD2F89c07A1f5CA2853593EEf337b6D6	2
430	0x4a18e8B282Ae8A4DeDE5Dfb1aFfbfBB78F4BF9b9	2
431	0xe647AB2F780d5797874Faca4af39c3aa4939D716	2
432	0xc3286F270849f135f5CC035C7423b2D89C35b306	2
433	0x0D629060728FF89829CBC4cb75226cE08EC37DB7	2
434	0x0700708234bFaB229c5EDDB036bEe75480f2719B	2
435	0x134D0804879e61679dBa2f18d6eD57Bea1908Dd5	2
436	0xa315b14CE102861aD07787Cd11a625BF91b2B626	2
437	0x39CDc57A43A0fF39674e6036afec5E516A40054A	2
438	0x494edBF45a271107F0fBdd05225Bd4C71E71b351	2
439	0x0B123805D91c0F638DD76B13ce5B5fDbEF89Faf7	2
440	0x5AA6e90b717a3c4fB8C1B4A1F18ACAEF94875415	2
441	0xd9b05CFc700A3665a8aBccd0725c3436094C2b86	2
442	0xB04Dc52C46c0245659FD8225FB48935d392Ea2b6	2
443	0x45035Ec63713ff0D128be77cE256C38a292625c2	2
444	0x6841cD230e092343e1EdE8f456d9bC0a609dcFF7	2
445	0xC4b6A1c79e902f438e44C22149C7A4Cdc794a6ea	2
446	0x0baB6e752Dc30b0E46D6162092f1959402787134	2
447	0xcA7D9FF021bC616999335cb8F71a33cf90E072DB	2
448	0x869A60fC797538EB497098756C7de91793a48ECb	2
449	0x5BCC3030A593bD5bC5a81E0e30561aB15cF2BB61	2
450	0x8A20176bE5c8113c0c0CEb4F671E9FcA2b5cef91	2
451	0xF01df43B02be63226BF6aA5173685eE78236C540	2
452	0x09D9909F6EE5d769BE84e69f2A278AB80F722784	2
453	0xdEd4962D36d832D5a834F930D3D05070d836e1A1	2
454	0x621D0c7048997c8703bb52F4A219d638C80b5F6b	2
455	0xB30eC84b95EEAbC17ac32969332412bc6265bE38	2
456	0x16a49482FAdF302e5E64511d1E2dCFf92591A85C	2
457	0xd4d7046047dea6dc552ecF8001bbE5f107CEC4cf	2
458	0xcA0fCA1eC1C89F1168eE87a248B3aedc8E42a5D8	2
459	0xdF7F1e9A184300c1243fC7cd4B49bAF750908Ec8	2
460	0x79EF9EADF7c763dE9Caf50c7798d3cd8f2714198	2
461	0xB27181270f1f025c4f3041E391e0Fd606C838253	2
462	0xf04828B90FA19d2c98E92192b66ccE948003d91b	2
463	0x6e29DFD2B1c03439e3D9e7bCD261d35f8ed2418e	2
464	0xb644Fbedad87a5D1D9dEE77cC03a9e9Ed85Aa346	2
465	0x410B63b8d7dcb4be16327868c0dCD7a8bc194c36	2
466	0xBAafbFa8C71DbD35288EDBfbed955cA20ca006E9	2
467	0xEf551a865F6bcBE5D3b993971dD829A9eaAec85e	2
468	0xeE1814f6deef2222B9924fF04f7B5a6a47cC5b36	2
469	0x3AC88030D49B3aA19F2b85780B8c476083c53CcB	2
470	0xbAbe7ad5b560261A1f52845aFC57fcbE390482c4	2
471	0xA7d26E59B811B33dD3182D33eA9ABaC2929b733a	2
472	0x09b8baEec7F9EE07738e1345e7a4Ef2fDfC9B368	2
473	0x6347cFdA65e11d68bCaFb2eB2B12484B4eB29FC4	2
474	0x4513812AdD99BaA69010ee271BC49fa08A7E1F64	2
475	0xE85C968Bc0dB9F8138e925e333c744966896Cec4	2
476	0xf66405C0b62C27eE3A792Ff02F4EbE3936758246	2
477	0x7f819De86a4d4AeD6fFF42aEDb7a573E3cf8F694	2
478	0x882C7d77c4feeFfcc4F649594Faa865e85BC35a5	2
479	0x9479b0c99F3f67e9Dd58F31995DccDE6f413997f	2
480	0xFaBad7BFf5438AAD05Db85Cf8d9363C704E7F267	2
481	0x0D048fAf8C96a38aE725E163E82F84DF393193d8	2
482	0xaf100Ae4948c90bc06aB7F5A0194306d23E12076	2
483	0x0f4D97A3627c50013d5B195BB7EcDb4297451c45	2
484	0xa5bA5eA85eaF0Efa2c7660558EA3CC2cA1a93c02	2
485	0xA73D77E9Da4cBD0dAd3aFcca8B66Fb9a521147Ca	2
486	0xbaBfA9e7B7C66D4649815d47c536621D80a715B9	2
487	0xbd27A0Caa1CB86Be8EBd79Cc4475681f5F34141F	2
488	0x6299Ab6197aab216Edec253AD6b9b0e1Db111EBb	2
489	0xC35f334A9ac7FD7Da85E21a5b6Abfe15e34D768f	2
490	0x002951BA2c3573cE24ad0374C4a5aBaCbF584ce7	2
491	0x9eEA7602EfF463982aA9E5702202b0987848CE8e	2
492	0x0958c9d7c60f83D73a90fA33223e163a0e63648D	2
493	0x8BEF453C5F4a6ac04a6A83d55235C46279ac8782	2
494	0x9A3bFEDBB2F0C2aE3F7f02c223e1eD9cd53df1BA	2
495	0x50e5B1D0FC07c50dA3f2D64470535055c31F2287	2
496	0xDd8D1C4FAEa5f120d0A2E6FC5Eb5180e4534c145	2
497	0x0D054059a4Ff96cFab5901A872D3C31D73Ee26b4	2
498	0x471A2C3C6dEa73e8F7f077c68916880a6896ED9F	2
499	0x2d456Add8cB4e0E655D7F7B77f04eb57f6ace6fa	2
500	0x43a870b7464ed3fDF0f653F2A09787fb6dA04277	2
501	0xe6F5745e7158443D4ca1DADb035920A7Db61c7EB	3
502	0x4918263e2cDE2F1093aD138E4D290639bc1Afaad	3
503	0x949bA6738da3aa8F98e6137387b658493729170a	3
504	0x1D0e18a64A5CC946A5fa616ab403C046Ce6A7cA9	3
505	0xFA396E9B4Da0F4C48dFbAF46cC293dadB6810379	3
506	0xd7F3CC7616b0C168E303debdFd0Ab1Fae10e971B	3
507	0xAaAC86F45659be57Ae5f7156176011B0dc587743	3
508	0x8773b0970afE7E31574896F615109C3f4326a673	3
509	0x1dFC0E5d8340691b07c344DeDf7420C9a4317585	3
510	0x5020D3A51611988F5E57df8B2aBC3611cdE4b59E	3
511	0x273537A65B2Ff6eE3386f118036BFd2a0BA4799a	3
512	0x14c9c0cc1B9183e6d2bf92bF6FA2364b47201DF5	3
513	0x341759716244035E5e707af66497bf324B0Bdb53	3
514	0x94B3586E86dd72507F934D6A704Dd13a4C900E2D	3
515	0x3b53b2eB3D5f9632626E2043C81f74a34cc92Edf	3
516	0x08D7E53D338bA72BA5f3Bd2Db6E1D96CD6047c06	3
517	0x7Bee4dDd3E41dc59E7b60B2960492f89590f95C9	3
518	0x558Aedf7e4F9EFF79538a47D2d904A3E7945F62B	3
519	0x104bBA12c3264C49D5326A23D1f284d81833a080	3
520	0xE6e536C9cf50c645cc759397Fe801E3fD197369C	3
521	0x2C3453170F2F834317cb91145D7fa0CA68F792a3	3
522	0x76524E66A664822ad30528B08a7Dca46C226ff76	3
523	0x6DdF19b75B2072b757493Ab4084F2c8180FC82Ae	3
524	0xBacFAB854A50171199124e1773df68B4d4DB3325	3
525	0x6801Ea4a07759539e83bDfEe58342a0901b8F710	3
526	0xf6ad2e11D2D935684bB7B1C8cA9c61cD21a17bDe	3
527	0x1fE80d3e80E66F63914dD6ebDfB7cbf53c28668A	3
528	0x22BD850287AFe8630941aa50a6cAC81a71c98f39	3
529	0xAD2Cf9a0296705560d02e93a1B0808db5894C9B4	3
530	0xbfe657F7e18253F4A47122480981E0bF1F20a1CD	3
531	0x2037d5df0728C58013131A9347c53FFa4C47573b	3
532	0x453bFCaE275018990D8690A2Ab1c0A4E526f977F	3
533	0xA755F59d6cEDdF9ea7557689F9f1fa1Ef246B9e3	3
534	0x82b68d929CE396B6f07993fd6dc2030A3dE63DEE	3
535	0x4BFe86c9B4fDBdB87859082F5Fa5d1903ab0E375	3
536	0xE69d5A1807663D59BaB698056F49eB2Ee100c6EA	3
537	0x79CB1ab1A43fbF624Eb6c2f6F92332597c379bc2	3
538	0x36f07dA7AbbaA8d3408f1617A7a892545A580c8A	3
539	0xc84036C5A122b5486DcEe7ac6db4e41110f61E2e	3
540	0x783aC13b8E10825268F297955BACcd0d87B2F551	3
541	0xd86edd3a7BfCab905D09F71b23194795fb7B9e48	3
542	0x2cCCEAb2726c57fed4F02B2829a38cdd84fa34D0	3
543	0xc604A989bc886A4e1f2d5767B01300fcE6dc73AD	3
544	0x25AcA82e65703ec85cA2d4CB9a0db9BDFEc52392	3
545	0x703Ff8197d45Af8f1ec723a01eC208B5c1F511de	3
546	0x8723B7fd03dCe5869f076f78fE44551C4147F029	3
547	0xCdf7734EA6Df47E4c01229b7063B47C571914e50	3
548	0x2EA29D8C8658F79bD2979616A65a70850e03cf2c	3
549	0x0E9d25BB4DF74e6ce738e516f01AFde3BCE1a549	3
550	0x0b95e2B1B697A6Ada5316d10d673Fb008a68ca3f	3
551	0xf245f433139945E78e9FA75B9d687A54f2e801b6	3
552	0x0f0F46D1848a951d89758370c768f9C5b84F7241	3
553	0xE379F706b6b00a60E3BFA40eFF5B20A138c3bb2c	3
554	0xbD3c46aA3364aFff4B02BF6E0331346E885375FA	3
555	0x35B3b7Cf3101dC2f8eb67A5AEf29010729BfD969	3
556	0xcbdB030859a9dBa02400949869A9aFb2d098bAA4	3
557	0x40567fbbA4Ecb64080cd3D8a8b6E144A543ed890	3
558	0xf887e3FDCC2e4a3B5904A09B183ae5F68AF97102	3
559	0x49A80A18b9da8C31c059fB6a688a70aF38e21c66	3
560	0x028f10304C5db7108D8E4ED3d6fCdC074a1193D2	3
561	0xe001b0e676e9f36EfBb3c80df48a94A6D2483720	3
562	0x1b1fDD7B818d35B06b3dA3E460244d5523659A3B	3
563	0x760eA0631077a3FcB28B2B82d565cC9f019b0820	3
564	0x7f1804dDCAf785c247c0359c852A85347cB0C2b3	3
565	0x93861A4Dcf1E25AffD5d81Bb0Eb9e57012d47338	3
566	0x098C668029A9d7902968eaF9E76e658354a876cA	3
567	0x0e0655f1319c641d9C0a4eE19C3e5ab75A97C69b	3
568	0x8062ffE7F77DAed0be1FdDEe091ce4A5bACB4461	3
569	0xb338e2f66F931D304841AE5F7E93ca8D6aa87FAf	3
570	0xD4A5A33a8bFE5789bCFd3b08C94D993B95B3Babb	3
571	0xc7975B84C455a9640187d165cC5fe3A22A8c0fFb	3
572	0x060CA554b671f38BA5c1a3b7D6981a36176DB49a	3
573	0x03cBa1288Ff812ee9aD329A6A70D8345a1086d27	3
574	0x92743B70Ac2DEb0694D92da399476b7b4a9257aF	3
575	0x4e85B8fc1C4e01b833255e0829b9d81c12A1478B	3
576	0xddA884dD8862F7A1b04714c8735ec8dec5AF28D9	3
577	0x3Aa56aeD2e1ac5584fa72D2f870b79dba15CceaD	3
578	0xe91aE11Bc203E63539A4681B1cBDBd5956E68412	3
579	0xd3CB391792bBab9FfA5c47854Fd8185f0f2d5c85	3
580	0x935744d3c4164dC96D3e5620a6508fa0a709e10F	3
581	0xc46956b304F122D014f7ae052A615d3891047Cd6	3
582	0x7199d6eD85b8f24326699F6890B5d6005C5fc78B	3
583	0x2Df0C3FCc5d6eDaAcF7aeBD3d5bF5654557584F1	3
584	0xE0D4cADA981943f5D8Af499381D03d90DcfFACDb	3
585	0xF76Dd3564e89628e95a98dBD8d768d4920198dfD	3
586	0xDE4ceba1e280EC01F115C6782EE2E395e3D53cEd	3
587	0x5c8EB884a4b9977AdDfb0876eE6f34a08510ebDc	3
588	0x111f36C7Bfc7ce2Ba5c11488D6c60f2A9A6FE202	3
589	0x76bcBE742C2f43aD8333171025E5B376e393Be00	3
590	0x8D1a8bf08ff453d455bdBAdadc1437E77cBf83Bb	3
591	0xa99FeB1a79043Aa606e71c02985EF7eaA81f9533	3
592	0xa035871DF14d490f1Be072873Ffb3E3B1Dc7e2b4	3
593	0x7c69576f721450c3bF35e6C5dA78f4d2ad3704B3	3
594	0x00694CC523bDf0970Bb6c721EAA862e011BE4fE4	3
595	0xE3ba9e36C2A42660045d7E001666652aA9b6D544	3
596	0x8bce15d3aDDdc060f4E55b8AEd72a2140468197b	3
597	0xB6444e585a7BD87cB532eCbf19887a699c929394	3
598	0xBF85aF4A8028af3c0F87Ef3839bBe29e99FE787E	3
599	0x2a18805a7299525cF66B3251D908a2804385466a	3
600	0x26f93D83740FC4030F5Dd5bF23640C095302117F	3
601	0x14e7599c6Fb545F2924FAc913520C6976a766F62	4
602	0x50060574e18ECB1D32a458a96b49c8a0E0415340	4
603	0x484a478A410EF1F243b9e5cE14f52750cbfF3178	4
604	0xe3F42ab3A64D70a2bF0914a74fF4Ff4182E7cdb5	4
605	0xFe5C0cd626C112a0027176549f9228cF77CBC616	4
606	0x4Ca25063efDc8ab86dd8ED0e595D847aEa78B1dF	4
607	0x4c67503f189582Ef647f9351848Ad85E672A0877	4
608	0xAe4aBD3D7DC2bE798D720820a3295e0BF3500672	4
609	0x395656d42ecF3debf00dEEAb4eF5641dda90Ec3B	4
610	0x914cb62c7Fd768caC0A544d4B39C0fC9959D8537	4
611	0x4Df0eA13e6263B1f2Ee89C19f0435BA7149BF66D	4
612	0xD091d50C230C77CF298d91BC5B60AE83153fD968	4
613	0x68A8F0645745cB5e852c4D35eF7B86367e538eCF	4
614	0xA8B34999587ddaf8868f9e64bf363A0E8192eEB9	4
615	0x5E67292b8441787eeF4ddBbF0229FB5Add19C665	4
616	0xBe18D55a8cDFf86D2ca212C53c12fF6Da837364f	4
617	0xA90A867Fb62eCa6A4cE5DF3Ee4042E6984F405e7	4
618	0x3A8A6d33a912761ca24E367a622c9F9D1c5Fd52f	4
619	0x989D30d1032Bc1F45d92E707E80FcB80A69a67E2	4
620	0x3313bb1f4D4f070BF00ffD4733F85E7C7E50D8D9	4
621	0x87778eaD1E3a435fbCcB697B298b0EeCf9236549	4
622	0xD15F7B47233B4E2e82D0117Bb744435518A6e634	4
623	0xd2E5fF83BC7d8D4BA37Cd25e6B1fAF2aB9e4CD9c	4
624	0xeA41aFFeF8E36622A58c4813D83b59FDBd8b6E49	4
625	0x1f1d9383b0C38f0482d62F0029bACB3b701bB311	4
626	0x4A19fBD35E30bE8B33ddd0ce7C36E3772baFE713	4
627	0xC9D1F773a8dd8a25fE232C5b3092E7d58Fdd9084	4
628	0xCaC38456cD733AaD505048bBDB62D50047a0993A	4
629	0x6d5737383F06A78DBD8825656C4C0581C924DfFf	4
630	0xc92e77F0D996820c52216c637661d00623D86b13	4
631	0xF4387Bb6bCc4b17d2D01F15Ef4aa3EA72f4f8dCA	4
632	0xa2C3fa5F636a0C6C787e24688E8D253c30138Bcd	4
633	0xf448D28241a45E8509B8B2ea926A2dbA095664FD	4
634	0xFFb8dD9e1565284EA30b646a1f0d5BEF74Af4354	4
635	0xd775985acBb923A9977a8C57251f946046DA8A0F	4
636	0xc8B0f10A275a06A6913A9f7351707E3AdB70f2fc	4
637	0x40C9E701a275C6962a5d81Eb935471369d82c072	4
638	0x76FF422e9CfcE9F12291607797b8b8F6f67f13C2	4
639	0x91Dce339Aa91bEc237fa1870D561209DeF84a2d7	4
640	0x0Ca14076918A5EA758731049FEfb1732fA34AC25	4
641	0x0E1337f7357736cb217b96dd47898fDDd03a73DB	4
642	0x38F4293D563B21cB15760E8fCEC2f852210A6210	4
643	0xA3F1B620dbc281c8A4A80C47a36e19b44D20E149	4
644	0x7C1c20c0165c0B7E6d5A651C2eDC0eD1895547eE	4
645	0x8916b324448b741E0aE0f474030f59C0875BBe27	4
646	0x038a2b6F62A448D24F9B2B33363BB865346dbD8b	4
647	0x8E985F54C57E75237AdCebf88223f7CBceE2D0f2	4
648	0x324f4a474A77D5B1e64dC897f0191E453d3CC2c1	4
649	0xC3E7b2F4917871B887616ce380A30285655dD5cF	4
650	0xAEA3c8Fb21bDA5D7f2E9E12b88853a67d01fEda8	4
651	0xbA6F981a0623190d9E7a8554ED2f4E27FDe6DD28	4
652	0x0C510eF8b186008fB111b27305767c305Dff2fB1	4
653	0x7e43E9A881285af86690363464b651Fcd7A262A1	4
654	0xD35e8380041672DA5E27E33D63870f0047b08e58	4
655	0x41cc7585Bbe4D9E42c64AD699f7fCB265FC8EA42	4
656	0x136878B77e084d8aaa35A9FEe50E7C16806Fe012	4
657	0xe45DcC88d5da2DFF1EE1751AF024bA6df0B75131	4
658	0x1D313B53f5573abf0DD81B9a30FD91062DE0867d	4
659	0xAE623Ea44955b3b05D3b417d7D81c6F9e864dD92	4
660	0x083c173380105194F33261291826f7b61cf860B1	4
661	0xeE59C734d40fFB847Cc6B6973d50F9be10eE4a94	4
662	0x7d14b69D258C47EDdB19Eb68D5fb3371A40Fd1FA	4
663	0x7A22B8F5bc5aD93d09Ff36789019460e76eF21ED	4
664	0xf686d43CDf7Fe4b73EF623BCf31361B53250b46F	4
665	0x23003F66813BFa87BEA6A4d4B6fEd156352E054c	4
666	0x250Ce10dBfa7ca0Ed3d33953f814D69b218958F2	4
667	0xfF11526d86D5b9B64D7b4AF6F3f48CAb5fED6e9E	4
668	0x0D33846aD14aAc9FBd7abBb1C137bAE457A9CE18	4
669	0xfd73Feaf2a594E122c93038C9080D56590CA0d94	4
670	0xD963EB8f24c421EE9e658f7c1B3c6E7eAc4B42A1	4
671	0xcc97d4975098E46dB608C41Cf77D2D944a7a9Df3	4
672	0xE7445A8F061EE771e1AABcBd3958e21b2528338f	4
673	0x60E42f4CDB7dC094dAc46c6C700d7a021029A9b2	4
674	0xaA165ff0cD3D29d2b0B7A8fcDcE878DFD6CB1FFE	4
675	0x210A836B207cC0eaa0e83417075CA3b7d63FB268	4
676	0x5F7fa1637e430C33E20b37D0F599e975C595543c	4
677	0x95780f7120Ea644A55fda6853A9e07C51fcAF344	4
678	0x2584c88DB5B5E22A1e8CD6018FEF40A3a7a0D83F	4
679	0xFe42Ca2435F8C988DA3E03ffa4e4f3f950F10890	4
680	0x14Ee109379D30346480C4D97fec713D47E2FBf3A	4
681	0x7F85d132E9cc74454741fC6B04C1fc5a44dCDdcb	4
682	0xdffAee2C5B10F84e5c29C87510baf86c75745122	4
683	0xcD53D940F85ae7CFDA8a8380144d684e403191Df	4
684	0xA2617A042D5a5b2939d1f84298a7DaE7b5547564	4
685	0x267E46b4F980C8B5752B4E7B424D2b21DcFfCCfe	4
686	0x34C9B68d89B325F0F8eb8dC09c1aa9b0D2e4fD34	4
687	0x79fEBe49327ff78F23cf225Ea4c6b8b74ad3C26b	4
688	0xb08EB93d9d71925B820fB49eec467e69047603c5	4
689	0x97d6183415cAFbfF08e391ADec84De3a6436fe16	4
690	0x77a6201a3bB4f21DEd46522Bb984EEf6fC3b4232	4
691	0xeD54Df24f5FDc55653DB6f4FB4A6a0363Fe5Df4E	4
692	0xDbaEDad85B3EB718bee4347d119855d1A0be0549	4
693	0x138Cfc825956EA493694d3e05f3f48b38De778dA	4
694	0xCE1A5C7211184E843139dBb0AB4F8104F3e9513F	4
695	0x5dCA2e462Bc9A3b22c5A14EC461901E3c03294cA	4
696	0xa2843a94b8b7B8A0cF45F22597f3C212CA0a84B0	4
697	0x266bc4e87C2c77302a8b94416B933bfB209827aD	4
698	0x7Ea99063e5Ab67A8Be6a2C726023BF2EAB77b0D3	4
699	0xa6fEd282364efB35542Fea7aE27Cd4D6fd9C2e51	4
700	0x270FEb767B37EDD2c5E463da47c62Dc24C5dB209	4
701	0xE1851d7d00b9887F986c26Af4BcFf2c957130cb1	5
702	0xe75418C6FdE2711CF6F3c7b65B17618EB7Bf2f72	5
703	0xCbF1931a13800726EdbF02E98DDcB0261d3D031b	5
704	0x6052d1B30C7E028a861B31FDdA17E44528526B16	5
705	0xD98d4661e17235C5ddA210559B1Eed7e4698C709	5
706	0x522Dad16C98639972aa1eb6b6D78826E889Aa48B	5
707	0xf559e98dd6da8Cd8EE1452412C89AED3186d8F64	5
708	0x2AD1B8a48679822f0BfAa2d36a75DFC930B52CB6	5
709	0xDB02a8d55dB9Ab7bCBaBFBfece6De520D9b97682	5
710	0x0A430f82A15EB754741BDE6e00e9210caAB06BbA	5
711	0xadb14041fC69DcE0B01C994DDb9eD22881fBA53E	5
712	0xfc1aFf88D5a07023e1196332b849d71217340A8e	5
713	0xA2B64E2B1115B6C8366116Cef9d09f5B04A93b03	5
714	0x2600587DC3b18fC61DA8cD33054983792D227669	5
715	0xB8aec73dB1Cc470C4B8704fdAFAB10AbBAE569F1	5
716	0x17d62688D3B7682BFd8309cfE5297848523d473A	5
717	0x2bC245880ABaD127E489242e69dCd3C4E0b02482	5
718	0x7634108A66E6C095630DA9eE77e31A52074Ab268	5
719	0x64293658520D72FDF12bdE5e55ae7ABd9fC36cCC	5
720	0x78f35BC010762C4D1554749afB83B3cC8baA8229	5
721	0xA126d0EC88Dc874A5ef649e40fA8153FD00301B8	5
722	0x44880B226C0a0967F7CEA42B08e986954a335FD9	5
723	0x97b2AD8c3a5e107686BD8924827B9b5eeAf94dF0	5
724	0x3457C198a56590c0720208912293337F41a7eA65	5
725	0x617FE9559cbd2de10870cfD24F23d072cF05FdA8	5
726	0x09D0b24B621dAa75A66CA89E51F9Dc8798727271	5
727	0xA70b60145F8f70e408C771985CEDAa11Ba6132C0	5
728	0x86C11d18Dc9642C7E9fcF1E69d40F5a505d374DD	5
729	0x3D3a416057Cff24a76490b8750974d220A441588	5
730	0x530BF6D56E34a2b1A15bb6983041cAf4fa789888	5
731	0x962D6ba4b5d82fef18E386cE9e7E0D2f4bB2E7C6	5
732	0xc6729f5Fd74c76aBA923A8dd8205012612fd657F	5
733	0x7451d13d5bc25f8c0e77a9Ad48B6c41ae47ca8dF	5
734	0xEc7f20Cb3f2b0d4966E1F7eA262e5fb0A49Ad24c	5
735	0x470696b491a9F5c58e8978bc5F5A56a52C73CC66	5
736	0x428686d0D926922d95797Fc854Fa604D51Ea67FD	5
737	0x33dE7A01f22Edf8aBef2fc234fb82564a87f43CE	5
738	0xde32C78C54159Ce4150752F3820584bE5335C29E	5
739	0x0613636d1FDd724091f02F7F33ca24CD1E829364	5
740	0x9cCcF958991f89761a672aa5A3Cbc72889Ef0274	5
741	0xD78f0EBe8ff98cf38f30104216291cb85f1e4E1f	5
742	0x988717F96da94227E8361c2512484507625803dc	5
743	0x0E8d02175427Aa403cb9042A82c2FacDBaAc2aC6	5
744	0xE715547C88dd6A9fB8bCab8E9a8fCfd2E24D46b4	5
745	0x7123C3fb14105a8afCd2a0f7Bfe7D9b671DCd9bf	5
746	0x4f38e678f292D9126c83476dc77B426e9ae6b26C	5
747	0x6cAf94589e85251E644B8cc62bb01D6AB5c200B9	5
748	0x7385b44F874278c693556B41dd17B58A0717c83f	5
749	0x115Fce763e4DDFD0E064b1dE842A56f60378622c	5
750	0x651F8442Fc15D5A65F451C22F5555eFcB8937da3	5
751	0xD4F6d643a4FCDb4F1Caa61Bb4991cD8eAFf5c0b8	5
752	0x62626CB7C84Acc9Ee2562d94A6BB408dCfB3B2d7	5
753	0x4cFCa07cA38ab86349fC9Dc9eDcA13CDc201f0c0	5
754	0xd0c0f29e32Ca42c863A78E21D882C7B1892F5590	5
755	0xF80174cb2bdF0c1812AB1D29A56407C17317eB98	5
756	0x8fdc974370D70CF30866381a60d094e535ABc00F	5
757	0x9ae69385eacb679388f3FB3e407397d6F2b8a7D8	5
758	0xAD7a82CaBD20D6f23bCB8Dc04f39bA0883Ae34e9	5
759	0xb7b349adBBCf911c36d93D8fd3727248F818cC10	5
760	0x7E7f4282EbDF6D0a2Ff505c6b5e7376199873D4c	5
761	0xeefb738F9768D6562d488d80c728C834B52DDd5c	5
762	0xE5E742Bc2f3AbD73bf80bbBFe71fB759DB5ebBba	5
763	0xd5A7c61A40195Ae9c2E396126A31Be5B14dc6c67	5
764	0x321f855D8597d71730e8C9BB165A653B48AbEDa7	5
765	0x6EEbA3d0820c3f7bb041C9914B303e201Dda77c7	5
766	0xaed2E087f2E4787dA305d030AD91075c4614f75f	5
767	0x7DC914C7b7e37C18d1523BFCe8E4b5ae4DBe31ba	5
768	0xA436d95D655344bbE51fd356f868d1AF600803aF	5
769	0x551eEBA23E7ad3B1c510d0c1E462bF4D10b23F69	5
770	0x45683A1B202ef5cf3B9eC46Fa4047CdF2c980a1D	5
771	0x591c1616C71F93b566409573864216FF38F1828D	5
772	0xA5E8Dd3fB57530d0bC7D173CF8384872e30239C9	5
773	0x2C24e74ABb66b598185cEaDD7854488880AfC882	5
774	0xaFb6634914F52743Dd8fA2a19744FFDF9f66Edd5	5
775	0xC010EFaAE4e011B08d69eC69e349F16d469b14D6	5
776	0xF08E833847e67F4EED7C65d3abF74e1F71F7a9cd	5
777	0xF807EF4D0cDFE89ddD10807074B8D4fC9C339f17	5
778	0xa7Fc22543B430a15c9C41b9606E938Ca6dc05196	5
779	0x54982659DC7ED7f48C53f7C86D25f07FB120186A	5
780	0xBD3CFA8Bb6dF0f49bf8342596008a5846A639134	5
781	0x3cF1E99EA8603a40210708B9a5FEb68C087CEB95	5
782	0x81E89247284Ad768fdF52d426D7b0A4C8FFbafcb	5
783	0x058337bF43Bb3d5112F35A6dF5B816049e4994Fa	5
784	0x0241E667468D8996b2Fa0BA1E3a8A31C88acBBEB	5
785	0x7BE41f1CA07616E75094c2468f4A561716F23c54	5
786	0xDB62cb222f472CD4fAce2dcD8E2a4A67efe0432c	5
787	0x09a40ad21f995832Ee57114A84Ffac1Eb4D01F0e	5
788	0x1aC71B606f37Dc1E3CbA9F87F0Fa4844E163C80b	5
789	0x57783f732A2565a512626A87cc1156f5d50EDE42	5
790	0x6300b44B2B14acF2867fDFb26c98B8F7B3BC2ff5	5
791	0x41c99b5daeF2114Ff06DF5473A2eE9Cb9c639bAf	5
792	0x7A68b6E3ed6735B6E00ab1630BBD156610CdBCF4	5
793	0xCA878e3Eae0179daCd43660f225cB5FDe7f97F12	5
794	0x42f1F4e7ecc354C9D8C685dE939bb5F718345cb4	5
795	0x4aE568147131E9C9BE194a338F680405e7081492	5
796	0xc47A8786f1F7f9b44FFF91cE83Ef6A202Dc62118	5
797	0x7E532D7133496B4a4b44c0cE24C92187403D0450	5
798	0x14B63B8D5D35eC901784438d6911Df1207AAa647	5
799	0x70Ae84C3372860887B4689E5D96a6B159b11F4eA	5
800	0x4AB9307307260c1063fDEc456253B40fD7dB1220	5
801	0xD822DCB3b901Cdb7B5699e914e84B9E6bE8D953C	6
802	0xFdA522Bc62C5c07a64208fb9d04Af5B0019d5bFA	6
803	0xb81D292a0B915fB46c477a102B1E7e0dF4F340c6	6
804	0x607ac8c87472d2163043ea3a170b7Ff4e1e3484e	6
805	0x137b7AfF9a25789A69399303521a3a119dC26B42	6
806	0x06af2468d19b21F41D020b2efcf99E8B3f6043A3	6
807	0xD297675f901795BBbdF76828D5DcdF302399FECd	6
808	0x605Fa5edB31a5572550C6185518Da510eddd14e4	6
809	0x0C4Acc1bD77f1E906876A78838782663dCb0dd7A	6
810	0xf6AC124451B675c119736134A1d9d51F5098F6e9	6
811	0xd527dB8d3C2e9FB3EC4Da5E1A9ba094F31ebC00E	6
812	0x862EACBBa9AE1b80D96176AFEf6f80A6A9136742	6
813	0x572CACD3f1987CBF76a96568784F05d41b929C31	6
814	0x7a23A699Aa2380740A274f60534d66136312E44B	6
815	0x549F49BFAb465a76963D2860Cce0D0B0FE9E527C	6
816	0x93e9Cb024a4Ec9af284A7996d454C1B179E4a7CF	6
817	0x880F4e2cEC0aB1431261b4E08afa95D7891BC1B6	6
818	0x7A00b69e3415f0cD2AFD9a48b507F8FbC68e423a	6
819	0x40ea8f16F00eB3ad6f33BBf8224C43D10a968892	6
820	0x91fD6e86e3f01B9dcEE51BCbf3153a982D40Fa2a	6
821	0xcBD97893cbCDCF9DA3Ca34E6139179e1228AE451	6
822	0x3E5f44b54F15B535B8063b6Cc4A7D96F0A54c3AA	6
823	0xd597293F6b46Eb8Ae8102e9C2B6a3CAdEB12E146	6
824	0xEca1Cb6f6379A5208Fd5dCd35455FdE7C9e94c61	6
825	0xA29fCdeFEbFFaB4f630C8e53Fe8c5B2a68e6fb0F	6
826	0x114EF1B178C84E21e0090EA738d8A269695Fa5Fc	6
827	0x9CD5aecF6Cd410918f5D6aF373cb433A973f085B	6
828	0x6adeae0Bdd30FBC19F7e542bA561019E955cf159	6
829	0x91DBCc0682d01E9cc2365B3aaa9B6A9C17f42230	6
830	0x9B3C6c6E26E0236423896Fc03d353d985a834135	6
831	0x50C0a530258415408c3eDDd0f53F01F5b230D5D0	6
832	0x6416235EDf8B831f1d2D9F4030101EecFd8A3277	6
833	0x138748Bd7e56EB926d7503a48Cc8fD99d613Ac72	6
834	0x2C48F7293Cbd165EeDBf06b6E6696668EB203280	6
835	0xF4F8aa9114C63717E1E82bEFf7F7B00c14B0033c	6
836	0x32A0EFDcF7c5b3371f5C76C834bEa6f5dC64837A	6
837	0x60d1ccd0FBB3dA87E4AA8c947274047004074220	6
838	0x2F0caCcEA5ca9330B269cc99239E91955Dfe270d	6
839	0x2Aaa095d99274C37ea624da596FB9685E0f60b7E	6
840	0x8E44874954A4CBf4D1dc9c7cc3D920D3c10689B0	6
841	0x1d5b178f02Cb8DA8A1cCEdC14a71C6D4f67fBFA5	6
842	0x816480667f8Fd83F6cca873d34eB694b634028B5	6
843	0x28e8EF91edF42ef40Db63481f0114d87B631Af40	6
844	0x0cC3803a8EFeB95C42809134B72bBa98ec91b6a3	6
845	0x634B9956CF8207eA8d3F4b940c7B8482E98Ea2b5	6
846	0xD1eD858b83f94cbA8AfCB5bDe52D5d2D1990060A	6
847	0xd4c7BA27736114caeE814c1b76B35CB317C7574E	6
848	0xF2758A26738e393108DB2b408E4b4AC150f1BFC0	6
849	0xC04eA01EFf13185E64295fAf9746B7efbC833218	6
850	0x2F3ad5f1DbB02fbE74DD6E5114a52519DF0Db456	6
851	0xE734326Fb7dc82D27ABE6a3fB1E8194288734030	6
852	0x66444258b7a637b34c6B6e737Bb90c5133738918	6
853	0xD11D17601C9e1095531F4d47B3AC0bE694F8F652	6
854	0xdEfA3a665A8627BaED0fcf2238e645f1fa475187	6
855	0x26e31088a94c0e8b87fCB9CA051Cc5eE7D3f76da	6
856	0x00B3FDb024Ee70B9FeB9CDF24DDb932205e10419	6
857	0x3673bd73b178a7aD00718ab3a3928a7EC3513b23	6
858	0x0b9301f5a7D89481696FD862e5C3aE29867AD1a8	6
859	0xFddfD07C2422283447883a604A8686fCF8Ac7D9c	6
860	0xA79fBd8d7F9590bCcd9DeF9438280Fb814EF7a4f	6
861	0x41610435bBc8C01f0EA092e6e9D08047cb9f10c3	6
862	0x8801a59eb8FFF315992aeAACd62ca2e352A51939	6
863	0xC2f1a60EF8d261A392Ef6021f01cf6e61650fCF4	6
864	0xCacEaa6E4fE45c66fF36eE6401965637a8882B1E	6
865	0x45B9EF5cBC3E05A5740a81ec81045087d116fda2	6
866	0x86DeC831440700d2c0F689211554b7b1138Aef6C	6
867	0xB4AA64F4bAF74BDBB094DD9c45EfA221fd70b0c8	6
868	0x3a5D13107587220DF796096Cb76889826e1b55a2	6
869	0x8f7d0600629aAFd14F8Ca4fEDaCF983395F5E492	6
870	0x7A4AD7E218f435e0e95F0070F60bF97C7D01878d	6
871	0x29b399eb5937Cb606658B23CA6d186dedf958a00	6
872	0xe53Fb124A9aFd1dAe22B4B4becE3eD40ea9faE34	6
873	0xB4cce59fC0e1e5119582659601e4EaE0AE89C4a0	6
874	0x1451E471A7291721eAA2653530D787624C9DA730	6
875	0xCe84834d0b0084888b258fc6F88Ab2C80666E638	6
876	0x4403A515ba77bf77b673d55418Cd1fB144C41669	6
877	0x8827F917712f017e2D71164E489f50Be29583d7E	6
878	0x84d0A5A461B006013bd296cB3FcC6408446f7F91	6
879	0xce9ce49130D21f11165c329122d16f853B07b6B6	6
880	0xCaFA5f02Cc07FB2e47aE87821E8a9CaA372FDED2	6
881	0x60011813B630771a55edBFc406782b102244339D	6
882	0x147c6624D5D537563C1683894E9eB3F9fa097fA7	6
883	0x640C6b41AE30FAA7242160e589fE6915EC2665d7	6
884	0xd7ED442abF0bCBc8B1810Fc322B81FfB4FD4d4eA	6
885	0x9bE83f4d058d2C7DD047C579d0a458F497b47127	6
886	0xB15c432d36D3976156d4a8ECBe9b136D7eCC2507	6
887	0xB31284F180C1B80B8B33F7448d05C074475B325D	6
888	0x95D014238036D29686cC74e39d80580E708b9c24	6
889	0x70B15Ef7D67967809052bB7908bc7C64CB8cdD80	6
890	0x0D6997dDCa28d8747dfDE5fc368E305688aD7695	6
891	0x4E3f07a0526EbedC30e5Cedc900F8923dc1A5Ba9	6
892	0x927dc69bb9Ac369A78CD6cA141FC6aE002085C03	6
893	0xe94D38961d1C04153ff17775e2e285B61C93AdBB	6
894	0xe5206c4cdE2D46fDFE247C02D758dB28a7abf95A	6
895	0x0f8F56D35eB2a7aB04eC3E38A642b8DB1e911EdA	6
896	0xF73f8759FE310F272b7580FcEBdac53b5A604835	6
897	0x32624fb46C87B64b12F440C8A4A8F2E68C542473	6
898	0xF34c396a358C0cD1dB70F4A722194E35d8Be8d85	6
899	0x7332F7a30918DbDF693058435Fc246EfFea73576	6
900	0x1a77e72f0Ca846e07E018E9B8c4b14C48Fb90211	6
901	0x610A704B547e7330e48F25af30Eaf014852277b7	7
902	0x445Ee1F00e8f9bb19323711F8372F6d0BBae7038	7
903	0xB7f14c739A90E557B189A3ec053c7A86516E99cA	7
904	0x6068Cc36b49bd1fb19Ca81DB56821cEd6d036A00	7
905	0x1d0d4CE7495d965ef58c656e7dC17A058886EF35	7
906	0x6489d3E2dE66f5c6e4e5fB95e9c7913397101663	7
907	0xc072518Aaab53330Cb339F4B93713125116ED283	7
908	0xB2dACEC3149D3F47267542d40a569C8A3A6f5dD9	7
909	0x8688f29ACd3747BEE758dcc9Aa254649Fe3C242b	7
910	0x36bA52D46E88360F5c4940ae9a3757892437cC50	7
911	0x1b4b9654667c8E0DCdA3Ee88567d11FF3a6a714E	7
912	0x8257C0f3C549000A405ebeb00f47d85fAeD16541	7
913	0x1EaEc107Ee0Bd4ef49f98080d8F1C5975E763cDE	7
914	0x31B8D96775A9A592Bc66F38CA02C845c968C6026	7
915	0x1Be121284c1464BF403Bd3D7B295C0Fc15AA2258	7
916	0xc4be07d1CDE1C6cac09Fed4Df57b9DBD1835dD39	7
917	0x16a761837d72cBFadE8b5068de318Ca4f6a36704	7
918	0xC52450dD359F8468aE9df1ccAEE1673CC3F00408	7
919	0xc2d706C5d8e5C60Fe59caf67027834dA48f2165f	7
920	0x0c5354CeE5bC9dd395a59308785062bBda7cf218	7
921	0x6e68369A4ada90823244304C32b82D6E0659a040	7
922	0xAF56778B83A181546Cbef855A36D654E5c3E3DF1	7
923	0x17Ed0DC946371A4c18706f1f670E7e5F8d044081	7
924	0xc2c0196C32168C6DE5afC347A8822159183854e5	7
925	0x390C915127562f992D6409C11EF8e32b4CeB60f4	7
926	0xDd35A53717b925535803c4490598C4f860C0179D	7
927	0x67Afb99557BbbD487f59274ABf6ec037c6036F77	7
928	0x25eAD8d1d16c07F121efF3cECE56a2711e187e11	7
929	0xc6C3Ab8C469a090C5a259899E516e468711a5d36	7
930	0xDE3fBE14B5A10a1E3E97c74CD1F4E6C6013E203c	7
931	0xF6f99Da292C522D79f76b4590Aa3a78d215B1E03	7
932	0x24763F8Ba4853300532429285C9c75930d7B7CcE	7
933	0x054149bb0B8E38F84219DE289582DDA996810dD2	7
934	0x6649A39Bf344e34cE04f23F9De31A4b7B7f38E4E	7
935	0xE142Da34881B1E98689f58986AF9C449146063B1	7
936	0x5c18CC1616d18a2DF6C5A3Ff5066c187Bc5AE860	7
937	0x644fa2C1a50A61986CF2F889bccDAf0BB2cE5373	7
938	0x23B51A3125Ba4Bad01261f7C8126f0D471207ACE	7
939	0xa870003356D07Baa9aebB94de7F5DB1315AE0433	7
940	0xB2c4dEb2af7e323CBa0188E583C1fd24BF50f4de	7
941	0x18758A3d6007f52d993d66cD4BCacF7d04214E25	7
942	0x26A5B952915C01eC353a650F5FEa71cF7eE3F695	7
943	0x203B391ab777f344Ba17deBDaE1D8fEeebC963Bb	7
944	0x5B626aD6ADe72316436Aa2eA59421176bF0cA00a	7
945	0x86479a8D72d5180665f25A2adf44351E58c0bE3A	7
946	0x8C1D6244533f8bA9E82Ce1b99ccbEd0f48086008	7
947	0xD16Be2070b17C9aDecB89d182beD7eA6a8Eb9fa6	7
948	0xd872e4E5E6efCf0d720c36a7B89B1BD965b4dac7	7
949	0xBf77282FA5C014B7cEf018F4bA975a416580CbE8	7
950	0x12c73b8B3cA68ECcad166b0a7BB55e97C564b278	7
951	0x27ab758eB0a9b4368B4619268B0E96a6FE369078	8
952	0x0af4DC57650a9e7c44F6d1c34dc1Cec3295a0ac6	8
953	0xb527523BF745a49AbC6163C6343af2bcDf7a3519	8
954	0xc09b19cdA1d71BDDc3239BE46f16790A93460de1	8
955	0x1Af515D99d40c8f4EcFd88D658bADfef8CDC56dE	8
956	0x2493B75e4F705922C1CF193986654b94a8172988	8
957	0x9f3c0A084BEe7F7bb0268526e106e40075A2C786	8
958	0x04e2B9b4C7E88E0094ECe69A926adFd939C47ac8	8
959	0x83df518f98B3961a5B6107c8BeB13A86b4927485	8
960	0x6B02EA7D309eCbd1aBBEc32096b3A0DB79edE880	8
961	0xf0181BEC49c6D92A35d6A1BBB8B44f1a4b308A8B	8
962	0x358Ff0D8438E64aD45546D9Ec109dbAaF2fFF250	8
963	0x81b95bD586762239C9e65a339d82FBC900eB385B	8
964	0x3ec43755BDE3CebE7a422397F50D9d1A5E1D1160	8
965	0x84d3448d2A0810F036Ce378f9c74D9fA3F2765D7	8
966	0x538dF652B134b0BF803fc1D17a66194ea60B8c55	8
967	0xA14100000CDaFa273cf27A265e72deF6874443c7	8
968	0xFaADE9B1d392F44338484c8128483efe03B61b29	8
969	0x9e27C20A75d2532a2d5DD8d0592ACF449835eB18	8
970	0x97fe2469CA74d2d6e381421A262837501DDb6AF9	8
971	0x1df15B6F846D2A5F92EFd452582dF187A3ba379a	8
972	0xd15BCCDAAb85d6a3Df725131F54d6D9f1De5a5B1	8
973	0xf5A05A7808C1C64eB0e6355164C4C68b1708188a	8
974	0xAD8D480C3621EEca6a78a95E0Ec1Ae2478D908c7	8
975	0xF6724d0C4D09f3590bFf358A32481f3E9FeEe534	8
976	0xC906936F2c0a2444532C8b79Dda53B124c9a12c2	8
977	0x4bD456BCd82162fEe257aAe37FFE34cBECA45503	8
978	0x76d21D57beba5617418Cd0FBf8500373c8b87C70	8
979	0x0c9200D96c4E909143a813E3891c46Cc152eD8dB	8
980	0xFBDC47661Bc97CE7B2E4331c9d3A64030a509c6f	8
981	0x43E6C2F3bC31dD9d66802c0589be635d4e535e93	8
982	0xbbf24B5020B5ae46360df208a346985110e66593	8
983	0x04df7Eb6C7b67D96Bf6AD7b4d8f2440FA80d266b	8
984	0x20dA35dC49282d010e659864b41467aF492aFcB4	8
985	0x282526d6fFEE0264Af7469bb2B978Ee392B1C814	8
986	0xa03C634333474fc0dEc5F8f87820Cf4F411F0d9e	8
987	0x7b87076d7b000B75f0C1474688ebFde6E9c2eE4d	8
988	0xdF0d333378FD95FF823B460492C1CA14f6C17499	8
989	0xFACeF439bCD8429F0b478D84662fbd21E4b4C307	8
990	0xD90f4b37FdF3375BC715E61637B2d2a028095e40	8
991	0x080d2c81B4f59e2fA67FE07Bf3dc6608CDD8DE42	8
992	0xc2dc9aE2Be4cC75AF2e12b322f5a922a38499556	8
993	0xe994dfC65E5fE35fd36f07657f62304130E60231	8
994	0x85c3Afa0d2fd118067E599F9eE580045bd75E01C	8
995	0x4967ee68B14AF27803B22C5245243de2dC55dC05	8
996	0xE7B54BA100f02EC9193CB2236e789ada6b002059	8
997	0x847eC417cDF74aDD32cd6256066cDf760E5c3012	8
998	0x5D9D556D9e65ab71b6040D2818aCe2e2Bc076B49	8
999	0x66E56b630e695A4955388973FAB073Da5A42D977	8
1000	0x6DE59ac89a2Ee9146c8ae8d8b2fF6336C1149932	8

# Smart Contract Addresses #### Smart contracts on Arbitrum One Network

#	ContractName	Address
1	InferixNodeLicense	0x7874b66Bd15E9Ad89d92c03B570d702E96319D55
2	InferixNodeSaleConfiguration	0xBF53b5B55661e6394a5828d835297879676cf500
3	InferixNodeSale_10_false	0x33bF840FE2F4842FA857F8f7cd05fd6C3227A257
4	InferixNodeSale_11_false	0xe5c7187347cC8C7111f3Ee6b7fD3F12fbBC50B12
5	InferixNodeSale_21_false	0x925b793Ab856F8f0d6941B58B46136E443D9d55a
6	InferixNodeSale_23_false	0x258ff7f0BCAcf9FC302B3a2866Bafc21a4E95D79
7	InferixNodeSale_24_false	0x2F59FF055D6e2ffbFF61928B9E85A6ca1fc67A50
8	InferixNodeSale_25_false	0x26C44b3564c3e9F6bf0A5214c6862A5a9C357A34
9	InferixNodeSale_5_false	0xefEDc1b808462945a3215B90C450b27A161DB958
10	InferixNodeSale_5_true	0xd3CA106d0A093e89Db07eB05c3E518b6ef2dea8D
11	InferixNodeSale_6_false	0x83BFeA1F8340EB8f7B55a452D66d68C49a2Ef377
12	InferixNodeSale_6_true	0xf19d3E6fB4288721Bf2BaeD033C511A5dA23eF02
13	InferixNodeSale_12_false	0xFF071db08e57691a7C3C7cBd1D5BCaEb59837851
14	InferixNodeSale_13_false	0x28d997CF749185c9eE0d20C1A963B32031Ef1154
15	InferixNodeSale_14_false	0x6b92599698d73df4DAA341379EA2F201A4fa8530
16	InferixNodeSale_15_false	0xccDe2180d19AFAE692684Ae192E613c341B26bB4
17	InferixNodeSale_16_false	0xf694660a7CA066549fff0EDb1ce57AAF910252B2
18	InferixNodeSale_17_false	0xDE15721278432a8083db49dd517E48a2919B7906
19	InferixNodeSale_18_false	0x4E58315c81Ac6DA3ee4c538C5554e5029E7Ba972
20	InferixNodeSale_19_false	0x6621E17CA8F83BDF1c2307c23e6a75b9c456B3a9
21	InferixNodeSale_1_false	0x3C749fd61A8DDc10592118C815749213022ed5df
22	InferixNodeSale_1_true	0xe00C4B4AB94688FB26590ee0BbBc9bFfdAB2974A
23	InferixNodeSale_20_false	0x621B1860aa0F5501eB473BF547719b335c5b1777
24	InferixNodeSale_22_false	0x2bCC3dE9797e238D34FA440BA8a32B1a2488e4Eb
25	InferixNodeSale_26_false	0x0de1Da26C1a04462378daa31b5732Ff17c99c2d2
26	InferixNodeSale_27_false	0x5C58f256360d032198ae74723D2270195FC465b9
27	InferixNodeSale_28_false	0xFaA0ec2149875FeF3e192d54EfffFe206bb4Fb7a
28	InferixNodeSale_29_false	0x5DAB72671202EE7A7C6e8A94E6359F3034c0EB60
29	InferixNodeSale_2_false	0x65AE9B1B603ACc3e50a8987982757e5B51d15f60
30	InferixNodeSale_2_true	0xbE06a74dE20Eb7720cd239a28D670387D50310cc
31	InferixNodeSale_30_false	0xea54F64e279c6cE6784Ae82fA59Ee3B38a13AfD9
32	InferixNodeSale_3_false	0x6B5f5E5a875116B959fB50d221B38d050973a3Eb
33	InferixNodeSale_3_true	0x0A4Cc09a0c2774C9b2785Bf6Ec2003Ad38a1D142
34	InferixNodeSale_4_false	0xd604544758B1181d0616CD8Fb228822e3a14Cc38
35	InferixNodeSale_4_true	0x6A3F2aAc59616cE0F8bCb0A84C180eccE2D0A610
36	InferixNodeSale_7_false	0xC77DA032aF809533990E576dc13b3367d16127b8
37	InferixNodeSale_7_true	0xc44D7783452Eb21b599EBB0857d87Ce7Bcc667bA
38	InferixNodeSale_8_false	0x8F51F53307A3a1e3227EA09D829a3BaD7C590063
39	InferixNodeSale_8_true	0x72757F9442Ee3c71E46c41F4133CCf40EBB5812E
40	InferixNodeSale_9_false	0xBB0d61d3B80D48b05ad4daf4AD781c052Fd4B998

Will be announced before Sales Open day # User Discounts & Referral Program ### **How do node buyers create their own referral code?** It is their wallet address. Anyone who purchases a node can refer someone else by passing them their wallet address to use as a code ### I am using a referral code for the Node Sale. What discounts do I get? There are 2 types of referral codes available for use during the node sale:

Referral Code Type	KOLs/Partner referral codes	Individual User referral codes
Rebate Amount to Buyer	10%	5%
Commission to Referrer	Varies	5%
How to get it?	You can get a code from Inferix’s KOLs/partners either by clicking on their official referral code links or by copying their links into the referral code field during the node sale checkout process	Every buyer can have their own referral code. The wallet address you use during the node sale is your personal referral code. Users can share their wallet addresses as referral codes with other potential buyers
How does it work?	Clicking on an official Inferix’s KOLs/partners referral link will automatically take you to the Inferix node sale page, where you will need to confirm that you are applying the code. Alternatively, you can copy the code and paste it during the buying process	When someone makes a purchase with your wallet address as their referral code, you will get a 5% commission from their purchase, while they will get a 5% rebate. Likewise, if you use the wallet address of someone who already made a node license purchase, they will get a 5% commission and you will receive a 5% rebate
When do users receive Rebate/Commission	The Inferix team will process and airdrop the amount back to users two weeks after the sale. Users will receive the rebate/commission amounts in USDT	The Inferix team will process and airdrop the amount back to users two weeks after the sale. Users will receive the rebate/commission amounts in USDT

### What benefit do users get for successful referrals? Users with successful referrals (users used your code to purchase a node) will receive a 5% rebate in USDT ### Will referred users who input their friend’s wallet address as a code automatically get cashback? No. A manual check will be done to verify if the referrer has actually purchased a node. Once confirmed, only then will the rebate be airdropped to the referee ### How will I receive my cashback? The rebate you will receive is based on your total node purchase amount in USDT. After the team has verified you are holding a Verifier Node during the cashback period, you will be airdropped your USDT amount # Worker Node Purchase FAQ 1**. How many nodes are available for purchase?**\ Inferix is offering 100,000 nodes across 30 tiers. Each tier has a limited number of licenses, and prices increase as the tiers progress. *** **2. What are the steps to purchase a node?**\ To purchase a node, there's two types of sales: * Whitelist Sale (if eligible): If you are whitelisted, you'll have node slots secured to buy. The Whitelist Sale has 8 tiers of pricing, with increasing prices. * Public Sale: The remaining number of nodes will be available in the Public Sale on a first-come, first-served basis. Public sale prices will increase across the 30 tiers. The purchasing of Whitelist Sale and Public Sale will be started at the same time. A detailed guide, and a video walkthrough to purchase a node is available here :point\_down: {% content-ref url="/pages/8Yeyi6poySzU5Dy8yqz0" %} [Guide to Purchase Worker Nodes](/worker-node-guide/worker-node-sales/guide-to-purchase-worker-nodes) {% endcontent-ref %} *** **3. Who is eligible for the Whitelist Sale?**\ Whitelist participants include: * Whitelisted (WL) users * OG (Original Group) members * Key Opinion Leaders (KOLs) * Node Sale Partners * Participants of events and campaigns *** **4. What is the purchase token for the sale?**\ Nodes can be purchased using [USD₮0](https://arbiscan.io/token/0xfd086bc7cd5c481dcc9c85ebe478a1c0b69fcbb9), it is the bridged [USDT ](https://etherscan.io/token/0xdac17f958d2ee523a2206206994597c13d831ec7)on Arbitrum One network during both the Whitelist and Public sales. {% hint style="info" %} **Don't know how to get USDT on Arbitrum?** * Goto [Arbitrum Bridge](https://bridge.arbitrum.io/), connect your Metamask wallet, set "Testnet mode" to OFF and convert your USDT token assets to USD₮0 * If you don't have USDT token in your wallet, goto a centralized exchange, such as Binance, buy USDT and then send it from your Binance account to your wallet address using the Arbitrum One network {% endhint %} *** **5. Is there a purchase limit for nodes?**\ Tiers 1 - 8: There are specific purchase caps per wallet for these tiers: * Tier 1: Max of 10 nodes per wallet * Tier 2: Max of 20 nodes per wallet * Tier 3: Max of 25 nodes per wallet * Tiers 4 - 8: Max of 30 nodes per wallet Tiers 9 - 30: There is no limit on the number of nodes you can purchase in these tiers. *** **6. How does the pricing work?**\ The price of each node increases with each tier. Here’s a brief outline of the pricing: * Tier 1: $300 per node * Tier 2: $350 per node * Tier 3: $400 per node * Tier 4: $450 per node * Prices continue to rise until Tier 30, where the price reaches $1,750 per node. Each tier has a limited number of licenses, so early purchases will be more cost-effective. *** **7. Where can I purchase nodes?**\ The nodes will be available for purchase directly on our [Node Sale Page](https://verifier.inferix.io). All node purchases are conducted using USDT tokens on the [Arbitrum One](https://docs.arbitrum.io/build-decentralized-apps/public-chains#arbitrum-one) network. Please ensure you have enough USDT in your wallet to complete the transaction. *** **8. How will I receive my node?**\ Once purchased, your node will be represented as an ERC-721 NFT, which will be airdropped directly to your wallet. This will happen 4 weeks after the sale is complete. *** **9. Can I transfer my node after purchase?**\ Nodes will be locked for a period of a year after the sale. After this lock-up period, you’ll be able to transfer or sell your node on the secondary market. *** **10. What rewards do I earn as a node holder?**\ As a node holder, you will earn $IFX rewards via our Staking Mining or Shared Mine programs. *** **11. What are the rules for releasing rewards for Workers？**\ The rule of node rewards is defined in section VI. Burn-Mint-Work token issuance model in White paper. Inferix releases rewards after each Epoch period of 72 hours. The rewarding amount for each period is planned in the Emission Plan and the Ecosystem Fund will be released all within 4 years. *** **12. How does the referral program work?**\ Inferix offers a referral program for both KOLs and users: * **KOLs/Partners:** You’ll receive a unique referral code to share with your audience. For every sale made through your referral code, you'll earn a commission (the rate may vary depending on the KOL/partner agreement) along with a 10% rebate on each node purchase. * **Users:** Your wallet address used for purchasing the node will serve as your personal referral code. You will receive a 5% commission and a 5% rebate on any node purchases made using your code. Check the detailed information here :point\_down: {% content-ref url="/pages/6DYIgJhMux1FlPji9p2u" %} [User Discounts & Referral Program](/worker-node-guide/worker-node-sales/referral-program) {% endcontent-ref %} *** **13. When will the Inferix network launch?**\ Inferix Network will launch the mainnet after the TGE in Q3, 2025 # ABKK Collaboration FAQ **1. How will the distribution of the FREE nodes of ABKK giveaway will work**\ The Free Node winners of Animoca Brands Japan (ABKK) will receive node licenses in the form of ERC-721 NFTs to their wallet address after submitting their email address and being verified by Inferix on the Node Airdrops page. *** **2. Users who won the FREE nodes have to purchase another node during the sale to get their reward?**\ Free node winners will receive $tIFX airdropped tokens to run mining with a Tesla T4 or RTX 4060 GPU. If he want to run with higher GPU hardware to receive more rewards, he need to buy more $tIFX to do staking mining *** **3. FREE Node winners will be eligible for buy back program?**\ Free node winners will not be eligible to participate in the [Guaranteed Node Buyback](/worker-node-guide/worker-node-sales/guaranteed-node-buyback) program. *** **4. How many FREE nodes were given out in total on ABKK's Collaboration giveaways?**\ A total of 50. *** **5. How many nodes can a wallet whitelisted by ABKK purchase?**\ A wallet can purchase up to the maximum number of nodes defined per tier in the [*Node Supply, Price, Tiers, and Purchase Caps*](/worker-node-guide/worker-node-sales/node-purchase-caps) document. However, ABKK may allocate fewer nodes per wallet than the specified Cap-Per-User in each tier, due to the limited number of whitelist slots available for ABKK. *** **6. Who are ABKK whitelisted winners?**\ The whitelisted winners from ABKK community are announced by ABKK on its public media channels, and their wallet addresses are included in to the list of all whitelisted winners of Inferix Worker Node Sales. See the full list of winners [HERE](/worker-node-guide/worker-node-sales/how-to-get-whitelisted). *** **7. Who are ABKK free node winners? When will the free nodes be distributed?**\ The free nodes for ABKK winners will be airdropped on June 15th. See the complete winners list here:

Wallet Address
0xba7606DD75DA1Dd64ECF2D113593d85a9eae6389
0xd1c32B425BC5c6091AD5f3B5603Be1Af769D44d7
0x4C8E82E289BcB30Bf70354A2C82a78d957DFA415
0x97BA8847EB83138891ad70722Bc771b64a96FcBe
0x2BC45D459B11cCC88BE0cBdA1a59B3Df14761BA5
0x040f62aEcc0F2162d6EA1C59eD02a66d4952E767
0x25fA0FB82AF5c8DC8b0ac010871Ace9311C17423
0x5Bc3911BEAc22F1903fa8eA1D9126eB014E73cc8
0xdfA7b841F7971ddeaB8857eFA723585b25b185B1
0x8be4227d72c01fb0DFe45e5340f7e09Ac56DB340
0x9C688C3b1fb3d4F041ca4d1C58cC861c33CbA81B
0x0Cf6207aBF1b1EeA1239526d12a574329834bBcb
0x46daFacfE602b06652D3a71aDBEeCC5Fb45675e1
0x5aAdcE00717F7ceCf15Ee52605d91B5a4b4fF529
0x757836AE97B58833F8E96A1240E6CbAbe1955707
0x53D271eA69a2bE4B7f5CD8e8A557CC0DFFC155F5
0x6a02e3134ede4f975D8844422bF364c846Ba1dcf
0x8A8F511d82A79BcD3DE2b62F5a4cd134b75bE519
0x9521cC0C403A33C2E1077D38DE4297604eDADA26
0xef99c381cec2edb1480e5b6407c5780d897d3876
0x8f8693d84be2c354d3a5520b8599d6b66e2927e0
0x6d121153cf21df9e8a51b83240e27fa5c3b037cb
0xbd680338b781ccbdaafdf25e902f54b4a4974600
0x924a47e7b2a5b593487abe79cc65309039b18dce
0xb4177d40ab21dcc27198f64cff13d0c16196b58c
0xC9305676AbA446F76FAD31BFFa14a7CdFB6348Ce
0x9B18e079abDB7691Fad7C1e1584fb915b1577938
0x2395C09b5dD159A0A76be765FB59b19a336a19C3
0xDf1a05478ae466Cf8074e6d93A9d802a87A5a508
0x2F88e64E4586B497f9Bd0f572c905A6C4ba5534D
0x6D9ab96D61ca088980eDd9Cc346417daE90FA913

# Verifier Node Guide All about Inferix Verifier Node
# What is Verifier Node Verifier Node is one of the fundamental components of the InferiX network backend infrastructure: * **Worker Node:** The most essential component in the network, responsible for the bulk of the network’s rendering and processing tasks. Workers receive rewards from the network’s service revenue (75% of the Ecosystem Fund). Each Worker must hold a certain amount of $IFX tokens as a penalty fund to ensure they meet Inferix standards. * **Verifier Node:** These nodes ensure the reliability and accuracy of the network. Inferix will have about 25,000 Verifier Nodes, and they are crucial for maintaining the integrity of the system. Unlike Workers, Verifier nodes don’t require heavy hardware and can be run on mobile devices. They receive 7.5% of the $IFX token supply as rewards.\ \ **Verifier** is the component that receives the least amount of data among the three main components of the Inferix network. The input data for a **Verifier** includes a random subset of the rendering job's output along with the algorithm and key to verify it. Rendering jobs that do not require high data security will be executed by **standard Verifiers.** Otherwise, those that require high data security will be executed by **secure Verifiers.**\ \ The hardware requirements for **secure Verifiers** are higher than those for **standard Verifiers**, and specifically, these nodes must be equipped with GPUs.\ (read min. requirements[ HERE](/inferix-whitepaper/appendix-c-hardware-requirements-for-nodes))
* **Manager Node:** Manager nodes coordinate the overall network and are critical for handling the Proof of Render (PoR) algorithm. Inferix plans to have 2,500 Manager Nodes to manage millions of render jobs daily. The reward pool for Manager Nodes is 1.25% of the $IFX token supply. # How do the Verifier Node work The graphics rendering service consists in a network of decentralized machines called *nodes* which are of 3 kinds: *manager*, *worker* and ***verifier***. The *managers* are dedicated machines of Inferix while ***verifiers*** and *workers* are machines joined by GPU owners. A typical rendering session contains several steps here we explain main Verifier Nodes in short-terms :point\_down: ([read the complete Rendering Flow HERE](/inferix-whitepaper/introduction/rendering-network-using-crowdsourced-gpu)) The rendering task controller of the *manager* receives the rendering job request, the rendering tasks are assigned to *workers* and the verification keys will be sent to the verifying task controller. The verifying task controller receives the notification and then it creates a verification task; **this task will be assigned to a *****verifier***. After receiving a verification task, a *verifier:* 1. checks the authenticity of the corresponding rendered frames. 2. notifies the *manager* about the verification result. Then, if the rendered frames passed the verification, the manager notifies the user by a message containing the URL to the rendered frames. The user manually confirms whether the results meet the expectation, if they do not then the user sends a bad result claim to the manager. *** *Penalty Pool* is a token fund collected from penalties imposed on Worker Nodes that violate Inferix standards while performing assigned visual computing or AI inference tasks. The Penalty Pool is used as a reward for Verifiers who actively contribute to ensuring the efficient operation of the network.
# Verifier Node Rewards For their role in InferiX Network, Verifier Node Operators will receive $IFX tokens. \ 7.5% of the total $IFX token supply is allocated for Verifier's reward pool. # How to run Verifier Node ### 1. **Meet with the minimum requirements:** > **Standard Verifier Node** > > the minimum requirements for a single license for Standard Verifier Node are as follows: > > * 1 $$\texttt{x}$$64 CPU Core 2.1 GHz > * 8 GB RAM > * 10 GB disk space > * 10 Mbit/s internet connection > **Mobile Verifier Node** > > The minimum requirements for a single license for Mobile Verifier Node are as follows: > > * Octa-core (2.2 GHz Cortex-A55 or equivalent) > * 4 GB RAM > * 10 GB disk space > * 10 Mbit/s internet connection
> > ***Remark.*** *Mobile verifier node can be used only for PoR verification tasks. This type of node cannot run other types of verification.* > **Secure Verifier Node**\ > The minimum requirements for a single license for standard verifier node are as follows: > > * 1 xx64 Intel®️SGX Core™️ 2.1 GHz CPU > * NVIDIA GeForce RTX3090 GPU > * 8 GB RAM > * 10 GB disk space > * 30 Mbit/s internet connection ### Detailed guide on how to run the Inferix Verifier Client will be available soon. # What is the Verifier Node License (NFT) The Verifier Node (as an ERC721 NFT) allows you to be part of InferiX ecosystrm, earning rewards by running a Verifier Node Client. You can choose to run your Verifier Node Client on your on devices matching the min. requirements. Check the min. req. and how to run the verifier node [HERE](/verifier-node-guide/what-is-verifier-node/how-to-run-verifier-node). You can [purchase](/verifier-node-guide/verifier-node-sales/guide-to-purchase-verifier-nodes) your NFT either from the [Node Sales](/verifier-node-guide/verifier-node-sales) or through secondary market in the future. The Official [Smart Contracts Addresses](/verifier-node-guide/verifier-node-sales/smart-contract-addresses) will be released soon, so stay tuned for further updates. Verifier Node NFTs are non-transferable from the purchase wallet for the first year. Read more information on the Verifier Node Purchase FAQ :point\_down: {% content-ref url="/pages/6YCklZa1VNw9MnXcHm9J" %} [Verifier Node Purchase FAQ](/verifier-node-guide/verifier-node-sales/verifier-node-purchase-faq) {% endcontent-ref %} --- [Next Page](/llms-full.txt/1)