Image generated with Stable Diffusion v0.9 run on Lilypad: lilypad run sdxl-pipeline:v0.9-base-lilypad3 -i Prompt="techno cyber jungle with a community receiving news " -i Steps=200
🌎 Overview
The Lilypad Network is a serverless, distributed compute network offering internet-scale data processing for AI and other next-gen use cases. We aim to enable the open source AI revolution to flourish using crypto-economic mechanisms to provide competitive compute services.
Lilypad believes that compute is a fundamental building block in a truly open web and AI stack. The Network aims to provide access to a universal user base - empowering developers to efficiently execute internet-scale computation tasks directly from a smart contract (on any EVM-compatible chain), CLI or traditional SaaS flow, with use cases across multiple verticals including AI, ML, gaming, DeFI & DeSci.
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 Program (or IncentiveNet)!
The IncentiveNet will run continuously from June 2024 through to the official launch of Lilypad on mainnet - anticipated in Q4 2024, and will offer rewards to those participants providing services, utility and value to the Lilypad Network in the form of Lilybit_ credits.
The Lilypad Incentivised Testnet Program will focus on specific areas of network growth in tranches of reward activities leading up to mainnet release, and will run parallel reward tracks alongside the main stage network growth goal.
Stage 1 will focus on onboarding GPU Supply to the Lilypad Network, with Lilybit_ rewards being awarded for time on the network (up to a 4x multiplier), power of compute brought to the network (based on hashrate from POW submissions) and consistency of being online. The earlier a provider joins the network, the more Lilybit_ points that will be available to them also, with this focus area also making up the bulk of Lilybit_ rewards that will be awarded in the IncentiveNet program.
Sign up to the Lilypad Newsletter or turn on your Twitter - sorry X, alerts to hear when this goes live first!
⚒️ Engineering Update
In more exciting news for Lilypad, we are thrilled to announce that the Lilypad Network is deploying to the Arbitrum Sepolia Testnet!
We are thrilled to work with this Ethereum Layer 2 network, including a growing community of devs and users to offer serverless compute jobs from both smart contracts and CLI. Lilypad processes jobs and uses Arbitrum for verification and validation of the off-chain compute.
Lilypad is fully EVM-compatible and will be deployed to multiple chains, however, Abritrum’s low Time to Finality (TTF) and high transaction throughput (TPS) provide the performance and scale required to operate the Lilypad economy. The consistently low fees also make Arbitrum an optimal choice for Lilypad compute users.
Lilypad Core
To support the launch of the Incentivised Testnet Program and the ongoing development of Lilypad, core Lilypad components have been moved over to more robust infrastructure using a variety of continuous integration and delivery pipelines! This was a huge push from our team to ensure network reliability as Resource Providers on-board and a wide variety of workloads are run on the network. These new CI/CD workflows help our team to reduce turnaround time for network upgrades and will be essential as the IncentiveNet grows into Mainnet by Q4 2024. In addition, this data sets the baseline for working on larger changes, like adding end to end tests for jobs on the network. This helps development move faster and with confidence knowing we aren’t introducing regressions and bugs.
Internal logging systems have been built to maintain tight feedback loops with node providers, jobs run on the network, and modules added to Lilypad. Our team uses the data to understand what jobs have been run on the network through the full life cycle of a job. In turn, our team can identify network bottlenecks and troubleshoot ad hoc issues.
Lilypad Apps
To leverage the new Lilypad metrics and logging system, our front-end team has been hard at work building out a variety of metrics dashboards. The first metrics dashboard give us a network level view of all Lilypad nodes at any given point of time. A Resource Provider leaderboard with data on the top RPs will launch alongside the Incentivised Testnet. In addition, the Lilypad website has been overhauled and will launch soon!
Stay connected on the Lilypad Discord and follow along on X/Twitter for the latest on the community leaderboard and new website!!
🖥️ Tokenomics & IncentiveNet
In just the last few weeks, our tokenomics team has built out mechanisms to support the launch of the Incentivised Testnet Program as well as outlined the roadmap for Mainnet Tokenomics. In the short run, our goals are to incentivise GPUs to join the network and grow the developer community of module creators and application devs. In the long run, we aim to provide a fair compute economy that rewards useful work at a global scale!
Incentivised Testnet Program
A points metric system has been developed to reward Resource Providers for contributing compute to the Testnet. In the first incentivised phases, RPs who join the network early will earn extra points. Keep an eye out for the launch of a Lilypad community Resource Provider leaderboard coming soon! On-board GPUs to earn points, and feel free to reach out to our team in Discord.
The network must measure the power of a Resource Provider (GPU) and provide a means to prove the uptime of a given RP. To solve this, our team is building out a Proof of Work mechanism to verify RP participation. The proof will be run by each RP to prove participation in the network. This allows our team to assure users of the Lilypad Network that the compute paid for is the compute received. The PoW mechanism will also help protect Lilypad against negative outcomes such as Sybil Attacks.
A crucial component of the Lilypad economy is the module marketplace. Later phases of the Incentivised Testnet Program will focus on growing this community of module creators and users. One of the many parts of this phase will be the launch of a voting mechanism for onboarding modules! More on this soon.
Non-Determinism and Lilypad
The team is currently modifying our game-theoretic mechanisms to deal with non-determinism in compute jobs! Initially, the Lilypad network was designed only to run Deterministic compute jobs Simply put, when a model is run on Lilypad an input to the model was required to yield the same output each time. The output could not be variable with the same input, which was a limiting factor as many AI models are Non-Deterministic. This shift in thinking came about due to feedback from developers in the community. In the long run, this shift will be a huge upgrade to the network's capabilities!
🔬 Verification, Training, and Privacy Research
Research has been conducted in verification, training, and privacy for large language models (LLMs). The intention is to cross-reference feedback from customers and developers with modern mechanisms to provide security and privacy when running LLMs on Lilypad.
In the verification domain, we explored probabilistic and non-probabilistic approaches, including distance-based inference verification, model fingerprinting, and homomorphic encryption.
In the training domain, we explored probabilistic and non-probabilistic approaches, including training losses distance comparison and, as in the inference domain, model fingerprinting and homomorphic encryption.
Furthermore, we surveyed and reviewed papers on privacy-preserving machine learning techniques, decentralised and federated learning. Finally, we analysed the choices of other (decentralised) AI/data providers in the field, including Bagel, Hyperbolic, and SnowFlake.
⚡ Developer Events & DX Updates
Weekly AMA Sessions!
To help answer questions and gain insights from our team, Lilypad is hosting AMA sessions every Wednesday! Every other week, the time of the AMA alternates to ensure we cater to RPs and devs across the globe! Find an AMA time that works for you on the Lilypad Calendar.
Javascript CLI Wrapper
Lilypad is committed to providing a world-class developer experience! With this in mind, we launched a JS CLI wrapper api used to send simple POST requests that execute jobs on the Lilypad Network. Easily build a front end application using Lilypad and the JS CLI Wrapper as your AI/ML backend!
Execute a Lilypad job from a Smart Contract
A new Hardhat template for calling Lilypad Jobs from Smart Contracts is here! Create a Dapp that runs AI inference from an immutable EVM contract using decentralised compute from Lilypad.
🚀 Partnerships
Exciting Partnerships Expanding the Lilypad Ecosystem
We’re thrilled to announce (soon) several new partnerships, formalised through recent Memorandums of Understanding (MOUs). These collaborations mark a significant step towards unlocking innovative use cases for our decentralised compute network.
What's on the Horizon?
While these project collaborations are still in their early stages, we're actively exploring groundbreaking integrations that combine Lilypad's distributed computing power with our partners' unique capabilities. Some of the most promising areas include:
- On-Chain Databasing & Storage:
Leveraging subnet and blockchain based tech for enhanced security, transparency, and trust.
- Cross-Network Standardization:
Improving interoperability and making it easier for different blockchains to work together seamlessly.
- Novel Web3 Applications (dApps):
Creating new tools and experiences that wouldn't be possible without a decentralized compute infrastructure.
Building a more private and user-centric Web3 by leveraging decentralised tech to protect individual data and enhance online autonomy.
Project Spotlight
The Rare Compute Project and Lilypad are collaborating to unlock the power of GPUs into the Lilypad Network, significantly enhancing our computational capabilities. This initiative allows for the provisioning of GPU power via containers directly to users, opening up opportunities for advanced scientific research and AI applications. By enabling GPU rental on an hourly or credit basis, the project supports a wide range of use cases, including decentralised scientific computing and rare disease research. This collaboration aims to tell impactful stories and demonstrate the critical role of decentralised computing in scientific advancements.
Join us on Discord to witness the real-time development of this exciting project!
Stay Tuned! We'll keep you updated as these projects progress and have partnership announcements soon. Expect to see exciting new use cases, deeper integrations, and even more ways to leverage the power of decentralized compute with Lilypad.
HackFS Project Highlights
It was a pleasure participating in the HackFS hackathon! Lilypad saw 15 project submissions, with teams running AI agents, AI model marketplaces, and AI inference with encrypted results on the network. Our team greatly appreciates all feedback received from the hackers and has already acted on the input!
Winners are yet to be announced, but we are excited to see Lilypad used with networks like Filecoin, Lit protocol, Fleek, and more!
Our team has been impressed with the innovation from the HackFS submissions and has had a tough time judging. Some notable mentions go out to the following projects:
CypherCraft aims to simplify the process of training models on private datasets. These datasets can have custom access control conditions which enable specific type of users or model providers to use this private datasets. CypherCraft automates the process of deploying ml model scripts by auto-containering, training the models on the Lilypad compute network, and saving the models weights on IPFS nodes.
Nextown made a marketplace to run AI models. First users subscribe to the Nextown channel to run pre-existing AI models or to be able to add new models. Then they select AI models from the UI, which are executed on the LilyPad decentralised compute network. The results are then encrypted using the user’s public key to ensure privacy via Lit Protocol and stored on the lighthouse. This setup ensures that only the user can decrypt and access the results using their private key. The site is deployed on Fleek CDN and can be accessible via its ENS name which nextown.eth, or directly with is DNS name which is nextown.xyz. In addition to these, the platform also hosts the ml models as an REST endpoints that can be used for inference and verify model correctness through zkML.
Stay tuned for the winner announcement!!!
Lilypad Chinese Community Lead Taosheng, presented the Lilypad network at Chengu University in Sichan China! Chengu University is the only program in China that offers a blockchain major. Taosheng provided an overview of how the Lilypad network aggregates GPUs and provides infrastructure to run the module market.
TimeDAO wrote an incredible deep dive on the Lilypad Network. Learn more about the pillars of the network and our roadmap in this blog! Our team is super grateful for large amounts of participation inbound from the Resource Provider and developer community in Asia!
Consensus 2024
The vibes at Consensus 2024 were electric! DePin and DeAI events brought over 15,000 devs, teams, and investors to Austin Texas for a week of collaboration and learning. DeAi Day brought the AI community together for talks about open source AI and the need for open compute networks. With the advent of large, centralised AI players, the need for open, decentralised AI only continues to grow!
🔮 What's Next?
Our team is gearing up for the launch of the Lilypad Incentivised Testnet Program. The IncentiveNet will roll out in phases over the next months. The initial focus is related to on-boarding Resource Providers with a variety of GPUs and then incentivising developers to add modules and build applications.
Our research team is heads down developing the roadmap for advanced Lilypad functionality (such as the ability to run non-deterministic models) as well as planning the roadmap to Lilypad mainnet! Tokenomics research is in full swing and we aim to release further research papers in early Q3 2024.
💬 Have more questions or looking for guidance? Join our community AMA hosted every Wednesday! Chat to us on Discord and follow along on X/Twitter for the latest news and updates!
