XAI-Link

Inspiration

At XAI-Link, we're driven by the potential of marrying Web3 with AI. Our project delves into the groundbreaking realm of decentralized training for Large Language Models (LLMs), revolutionizing how these models are developed and enhanced.

What XAI-Link Does

XAI-Link pioneers in enabling decentralized training of LLMs. Our unique approach incorporates a 'proof of model improvement' mechanism, where the model's pricing correlates directly with its accuracy, fostering a more dynamic and effective development environment.

How We Built It

• AI/MLOps Engine: Developed in Python, forming the backbone of our project.
• LLM Training: Utilized the methodologies from Andrej Karpathy's tutorial for effective LLM training.
• Frontend Development: Employed HTMX for a responsive and interactive user interface.
• Backend Infrastructure: Built with FastAPI/Python, ensuring robust and scalable server-side functionalities.
• Smart Contract Integration: Implemented Solidity smart contracts for dynamic model pricing based on accuracy.

Challenges

• Orchestrating complex infrastructure components.
• Multifaceted challenges in training the LLM.
• Synchronizing multi-threading in the MLOps engine.
• Seamless integration of frontend and backend systems.
• Incorporating Chainlink with the Anacostia API for broader functionality.

Accomplishments

• Successfully trained an LLM from scratch with two distinct datasets.
• Implemented decentralized storage for all training data on IPFS.
• Developed a mechanism to pull data from our MLOps engine to interact with smart contracts.

Learnings

• The complexities and intricacies of building, training, and integrating LLMs with blockchain technology.
• The immense potential these technologies hold in democratizing AI and creating new, decentralized AI Model Economies.

Future Directions for XAI-Link

• Enhancing our MLOps engine, Anacostia, with additional integrations.
• Aiming to democratize AI using decentralized technology, fostering new business models and economies in the AI domain.

Technical Documentation and Resources

• Chainlink Functions: Supported Networks | Sepolia Functions
• Funding Contracts: Fund Your Contract
• Consumer Address: 0x2B8dD33b0248e62254821664dCc6667cF48073F4
• Subscription ID: 1845

Future Work

• Implementing Chainlink automation for dynamic data retrieval from IPFS.
• Enabling users to mint their pipelines as NFTs, with both manual and suggested pricing.
• Introducing a leaderboard showcasing top contributors based on the popularity of model versions.