Update to the Lilybit_ Rewards Calculation

Update to the Lilybit_ Rewards Calculation

An updated guide to the new Lilybit_ Reward Calculation for Lilypad Resource Providers

·

15 min read

TLDR Overview:

Primary changes to the original Lilybit_ Rewards calculation

  • Base Lilybit_ points are now assigned by the GPU card type using 3 parameters: VRAM, Cuda core count, and Tensor core count. See mathematics below. RP hashrate will no longer be used, but PoWs are still required to determine the RP uptime multiplier. ** change took effect after Nov/06/2024. Stay up to date in the updates-rp Discord channel.

  • A RP can earn a daily maximum of 20,000 Lilybits per day. **Will be implemented for either the Nov/20/2024 or Nov/21/2024 rewards calculation moving forward. Keep an eye out for an announcement in the updates-rp Discord channel once this is implemented.

  • The slashing mechanism now uses a tiered/progressive system to more fairly slash RPs. Penalties will increase progressively the longer the node remains online. 1-5 days offline will result in a 2.5% slash per day, 5-10 days offline a 5% slash per day, and more than 10 days offline a 10% slash per day.

  • RPs will earn 2 days of a “grace period” after every 30 days of continuous service provided. These 2 days will be applied to 2 subsequent down days recorded by the RP allowing the RP to avoid slashing for these 2 days.

  • Grace Period days do not accumulate to more than 2 days ever. Once used the 30 day count to obtain the 2 days restarts.

  • We are aware that some RPs may have cheated the system, and have multiple data points for this. There will be an audit on cheating before the final allocation of rewards (details will be announced closer to the time). Lilypad reserves the right to alter points allocations based on activity suspected of breaking the spirit of the IncentiveNet Program, not operating in good faith or being deliberately malicious.

Lilybit_ Program Overview

  • GPUs contributing to the Lilypad Network can earn Lilybit_ credits daily.

  • Lilybit_s will be redeemable for Lilypad Mainnet Tokens at TGE. We expect to deliver Mainnet in Q1, 2025, but cannot guarantee this timeline, and delays may occur for technical or operational reason.

  • Approximately 10% of Mainnet Lilypad tokens have been earmarked for the IncentiveNet program (up to 15% depending on demand).

  • The Lilybit_ Leaderboard provides Resource Providers (GPUs) an all in one location to check Lilybit_ rewards earned and node status (the Leaderboard updates once a day with additional credits earned).

  • RPs can use the RP-points dashboard to view Lilybits earned per day.

  • 50-70% of Lilybit_ credits are earmarked to go to compute providers for their essential participation in the Lilypad compute network. The remainder of Lilybit_s will go toward bootstrapping other parts of the network like module creation.

  • The earlier a compute provider participates in the network, the more rewards are available to earn (the algorithm for rewards decreases in phases every 2 weeks by 10% each time, potentially through Mainnet launch).

  • Nodes must be online for a minimum of 4 hours a day continuously (verified with POW efforts) to earn that day’s Lilybit_ rewards.

  • In order to run a Lilypad job, the full resources of the server must be made available to the Lilypad RP. If the server is used for other tasks, jobs may fail. Lilypad cannot provide rewards to RPs that cannot run jobs.

  • Nodes must be kept up to date with the latest Lilypad version to ensure points are earned. Follow along in our Discord to stay in the loop!

  • A 4x multiplier on the base daily rate is available for nodes that are online and taking jobs over the full day.

  • Renting hardware in order to participate in the IncentiveNet phase may disqualify you from claiming tokens due to legal compliance issues.

  • The distribution method for rewards is not currently finalized, but we expect to work with a distribution partner where rewards will need to be claimed within a period of time. Any rewards not claimed will be reallocated back to the lilypad ecosystem development fund.

  • DISCLAIMER: Other legal compliance based on your country’s laws may also disqualify a RP from claiming tokens. As we cannot know the legal situation in all countries when rewards are open for collection - we urge RPs to DYOR. Lilypad may provide an option to convert directly to collateral and continue offering a hardware service in the network.

Lilypad Network Overview

At its base layer, Lilypad offers a coordination platform for compute, bringing together the exponentially rising user demand for compute services & hardware with compute providers who can add new revenue streams and optimized efficiency to their hardware business.

A key differentiator and advantage for the Lilypad Network is that unlike other compute marketplaces, which offer hardware in time-based increments requiring users to configure and set them up for the purpose and which generally rely only on a loose reputation system to ensure trust, Lilypad uses a job-based system with verification built into the protocol. (note) This unique mix of on-chain and off-chain features ensures that the compute network is efficient & performant while being verifiably trusted.

Jobs on the Lilypad network are called modules. These modules are containerized scripts (docker images) or arbitrary WASM code with a Lilypad spec attached to them (located on Git Hub), which forms the basis of coordinating match-making between a job and a compute machine. (note) In time, this match-making algorithm will become increasingly sophisticated, with the potential for agents to do the deals for both the nodes and the users.

This module system turns Lilypad into a powerful, performant and competitively priced 3 sided marketplace. Developers have full flexibility over the type of compute they want to run - by creating their own unique modules. They can then hand off the job of computing to the Lilypad network and integrate the intuitive serverless, API-like DX into almost any work flow, and they are even able to attach a fee to the running of the module, bootstrapping a module marketplace system.

It’s this module system that allows Lilypad to become a backend for compute, and for developers to be able to offload the need to host and run large modules to the Lilypad compute platform, as well as giving Lilypad the ability to integrate with a multitude of other agents, protocols and projects to provide an open infrastructure stack which can form the basis of a full-featured decentralised AI ecosystem.

IncentiveNet Overview

Lilypad is working at the cutting edge of what’s possible in research and technology. In order to ensure the Lilypad Network is market ready, performant & secure and to help bootstrap growth, Lilypad is launching the Lilypad Incentivised Testnet (or IncentiveNet)!

The IncentiveNet will run continuously from June 2024 through the launch of Lilypad on Mainnet - expected in late March to April 2025. Lilybit_ rewards credits are provided to participants providing services and utility to the Lilypad Network.

These Lilybit_ reward points will be redeemable, for the Lilypad ERC20 Utility Token at mainnet Launch. Approximately 10% of mainnet LP tokens have been earmarked for the IncentiveNet program (with anywhere up to 15% being provided depending on demand and tokenomics) being allocated to this phase of the Lilypad Network.

Earning Lilybit_ credits is directly relatable to useful work being performed on the network or on behalf of the project goals, with these goals also being holistic in nature to encompass the need for business validation and community education initiatives, and to build a general lifecycle loops for feedback and improvements in the network.

Lilypad IncentiveNet goals:

  • Network Growth Goals (→ supply, jobs and ecosystem)

  • Network Security & Readiness Goals (Testing in Real-world conditions)

  • Community Building, Engagement and Education Goals

  • Market Readiness, Research, Product Testing & Business Validation Goals

The overall distribution of the Lilybit_ reward credits will be heavily skewed in favour of those providing essential services to the network growth. 50-70% of Lilybit_ credits are earmarked to go to compute providers for their essential participation in the Lilypad compute network.

The remainder of Lilybits will go toward bootstrapping other parts of the network like module creation. Lilypad aims to recognise the contribution of all types of community members, including non-technical participants who can also earn rewards for their value add activities in the Lilypad Network. Some of these participatory points will be calculated & applied retroactively & will not be known in advance to the community.

IncentiveNet Timeline

StageMain Incentive FocusDatesRequirementsParallel Effort 1Parallel Effort 2Community
1GPU OnboardingJune-October 2024 (ongoing to mainnet)Installation ReliabilityMetrics DashboardCost testing, & network infra reliabilityOSS docs contributions, DX contributions, community growth (ongoing)
2Module OnboardingNovember 2024 -February 2025Whitepaper, Module Marketplace Reputation system & market matchingNon-determinism with LilypadTokenomicsTutorials & educational threads in multiple languages
3Plugins, Integrations, API/SDK, Downstream apps + DappsJanuary - March 2025Smart Contract Auditing, Game Theory TestingLilypad SaaSMulti-chain deployment, Partnerships & Use CasesIncentivize module and application developers
4Mainnet launch (IncentiveNet may end at Mainnet or continue through Mainnet launch. More on this soon)Late March to April 2025

We're Live & Onboarding GPUs NOW!

Stage 1 Overview

The focus of Stage 1 of the Lilypad Incentivized Testnet is onboarding compute supply - GPU’s - to the network. The incentives for Resource Providers (RPs) will be available to earn throughout the entirety of the IncentiveNet program.

Aims of Stage Main Aims

• Onboard long-term Compute Providers (GPUs) to the network

• Encourage feedback on tooling, logging, metrics & other network requirements • Build relationships with compute providers and the community

Other Aims

• Build, develop & test network reliability, market matching, costing mechanisms, verification & performance

• Encourage Opensource & documentation contributions

• Encourage educational content creation

• Grow & engage a supportive and knowledgeable community

Lilybit_ Reward Point Allocation - Compute Providers

Early Resource Providers (RPs) participating in the Lilypad IncentiveNet Program will receive Lilybit_ points for providing compute power to the network, which will be redeemable for Lilypad tokens in future.

The amount of rewards available to Resource Providers is based on:
• The GPU card info returned from Nvidia-smi. Currently, only NVIDIA GPUs are initially eligible.

• Time online & accepting jobs on the network daily (PoW and daily rewards multiplier)

• How early a node joins IncentiveNet to provide compute

• Continuity of time on the network throughout the IncentiveNet rewards period leading up to mainnet and the Lilypad Token Generation Event (TGE).

💡
Note: Lilypad reserves the right to allocate bonus points - for highly participatory parties, or to deduct points from actors found to be cheating the overall program spirit or operating in a way that is not constructive to the growth of the Lilypad Network and it’s overall goals.

Lilybit_ Reward Allocation Mechanism’s

Base Rewards

Resource Providers contributing (Nvidia) GPU power to the Lilypad Network can earn a base number of daily Lilybit_ credits when they are online and accepting jobs for a minimum of 4 continuous hours in any one day. The Base Rewards available to a node (or RP) will be determined by

• The Time a node is online each day (in continuous 4 hour segments)

• The Phase of the IncentiveNet Program (where a phase is a 2 weekly time period)

• The Compute Power a node provides to the network (measured in MHash/seconds)

• The Continuity of Participation a node has in the network over time

Fortnightly Reward Phases: Phase Function 𝛽=0.9

The base number of rewards available each day will depend on the phase that the IncentiveNet is at, with a phase being equal to a time period of 2 weeks. In order to provide an incentive for early adopters (the early frog gets the best flies!), the earlier the phase, the more rewards will be available, with a step down method used to reduce the number of rewards available each phase (every 2 weeks).

Each phase the number of Lilybit_ rewards able to be earned for the same compute task will be reduced by 10%. This means that the earlier a RP joins, the more points are available to be earned, with the step down delta being equal to 0.9, meaning the number of points for a given action in each phase is only worth 90% that of the phase before it. This function is shown in Figure 1 and can be described by the mathematical equation below:

📐
Pph[i+1] = 𝛽 Pph[i]

where

P = Lilybit_ reward points

ph = phase, • 𝛽 = 0.9

i = phase number

Points Multiplier per Phase Graph

Figure 1: Points Multiplier per phase

Points Multiplier per Phase Table

Below are the first 5 phases alongside their corresponding dates & the multipliers:

Phase NumberPhase multiplier (= 0.9^phase number)Dates
00.9^0 = 1June/24 - July/14
10.9^1 = 0.9July/15 - July/28
20.9^2 = 0.81July/29 - Aug/11
30.9^3 = 0.729Aug/12 - Aug/26
40.9^4 = 0.6561Aug/27 - Sept/08
50.9^5 = 0.596049Sept/09 - Sept/24
… to Mainnetlate March to early April 2025
💡
Special 3 week calibration & release phase → all other phases are 2 week periods

GPU Card info for rewards

In order to fairly provide rewards for GPUs tuned for running AI workloads, we are rewarding based the card information itself. Lilybits are awarded based on the GPU info and daily rewards multiplier. The pertinent card aspects used for points allotment are:

  • VRAM (measured in MB)

  • Cuda core-count

  • Tensor core-count.

This equation is given by the following:

$$R_{smi}[k] = P_{ph}[i]P_{smi}[k]$$

where:

  • Psmi is the points-value assigned to a given card. P_smi is calculated by

$$P_{smi}[k] = 16e^{(4\times10^{-5}(x[k]-14,072))}$$

where x is the sum of a GPU’s VRAM, Cuda core count, and Tensor core count. In other words:

$$x = N_{VRAM} + N_{CUDA} + N_{TENSOR}$$

NOTE: Two GPUs of the same make and model could have small variations in VRAM, Tensor cores, and Cuda cores, thus the exact number of Lilybit rewards may vary slightly from GPU to GPU.

Compute Power Multiplier - measured with GPU card type

GPU Proof of Work calculations (PoW) run on RPs, but will only be used for the daily uptime multiplier. This PoW will run each hour on the hour and once randomly during each hour to ensure nodes are participating in good faith. PoW info is used to verify a node is online as well as the daily rewards multiplier and slashing.

The Proof of Work mechanism is shipped with the Lilypad Node installation, and nodes - identified by their unique wallet address, will listen for the PoW smart contract event to be triggered. Once the event is broadcast, a node has one minute to calculate and submit their GPU PoW.

Time Online Multiplier

Nodes can earn more rewards for consistent uptime! The baseline requirement for a node to earn Lilybit_ rewards in any one day is to be online and accepting jobs for a 4 hour continuous period.

In order to incentivize people to onboard compute and keep that compute active, Nodes can earn up to 4X the daily rewards. The time multiplier calculation is based on an exponential curve rate of the number of 4 hour continuous time windows a node provides resources. The equation for this is given by:

$$R_{day}[j]=1.3195^{(N_{POW}−1)} \sum_{k=1}^K=(R_{smi}[k])$$

where

K is the number of POWs in a day

Npow is any block of 4 consecutive POWs passed.

1.3195^(Npow-1) is the daily time multiplier that makes it such that any RP who stays online all day and passes all POW tests will receive a 4X multiplier to their daily points -> since 1.3195^(Npow-1) = 1.3195^(6-1) = 4.

Note that any amount beyond Npow= 1, will earn some multiplier on an exponential slope.

Continuous Node Service & Point Slashing

Lilypad is seeking to reward node providers who see Lilypad as a long term project with serious potential to disrupt, innovate and grow.

To achieve the goal of incentivizing real resource providers to the Lilypad Network, a slashing mechanism has been implemented which will deduct points from nodes that become unresponsive or stop providing resources to the Lilypad Network.

This mechanism disincentives both those actors seeking to farm points on multiple wallets and those actors seeking only to earn valuable early points without making a long term commitment to provide compute power and help build out the Lilypad decentralized compute network. If a node stops providing compute for the minimum 4 hour window on any one day, their total points accumulated in earlier deployments will be slashed. The slashing mechanism now uses a tiered/progressive system to more fairly slash RPs. Penalties will increase progressively the longer the node remains online.

  • 1-5 days offline: 2.5% slash per day

  • 5-10 days offline: 5% slash per day

  • More than 10 days offline: 10% slash per day

A grace period for RP downtime is now included in the slashing mechanism (Oct-01-2024). RPs will earn 2 days of a “grace period” after every 30 days of continuous service provided. These 2 days will be applied to 2 subsequent down days recorded by the RP allowing the RP to avoid slashing for these 2 days. Grace Period days do not accumulate to more than 2 days ever. Once used the 30 day count to obtain the 2 days restarts.

For a user offline for a consecutive 24 hours, the slashing calculation is given by

🧮
Rday[j] = (1-S)Rday[j-1]

where

S is the percent of points slashed If you are offline for 24 hours, your points will drop to (1-S) of what they were on the day before, and each consecutive 24 hour period offline results in another S slashing.

NB: This slashing calculation only applies to rewards earned for providing compute to the network and not other participation tasks in the network that are subject to their own mechanisms. So for example if you earned 10000 rewards for contributing compute and 100 for an OSS contribution (for example), only the 10000 Lilybit_’s would be subject to the slashing.

Caveats

Some of the mechanisms that allocate rewards to resource providers may change in future iterations. As much as is possible, these changes will be designed to NOT impact a node in a negative way, however there may be some instances, especially considering that IncentiveNet is validating Lilypad’s underlying protocol logic and code, where this is not the case. Lilypad will do it’s best to ensure that all voices are heard however and strive to keep the allocation of rewards fair.

Lilypad also reserves the right to alter points allocations based on activity suspected of breaking the spirit of the IncentiveNet Program, not operating in good faith or being deliberately malicious💡NB: There will also be incentives for cheating the game theory later in incentiveNet- so report the logic bug and Lilypad will be more than happy to apply retroactive points for this service once the game theory testing mechanism is live (we’ll also do our best to mathematically model the impact of not seeing this bug and reward to this effect)!

Did you find this article valuable?

Support Lilypad Network by becoming a sponsor. Any amount is appreciated!