Introducing #FEDSTR, A Decentralized Marketplace for Federated Learning and LLM Training on the NOSTR Protocol
We propose an approach that builds upon the existing protocol structure with end goal a decentralized marketplace for federated learning and LLM training. The concept is similar to that of data vending machines; the service providers receive parts of a dataset and they return a trained AI model on the input data. An instance of such implementation, involves job chaining and multiple communication rounds between the customer and service providers. The optimization algorithms include the classical federated learning and the Distributed Low Communication algorithm (DiLoCo) for training LLMs (in a synchronous fashion). We propose validation rules for every job output (or job feedback) to ensure accurate computation before finalizing payments. An application of #FEDSTR could be @dataMachine 🤖, when the training of the AI model is executed by vending machines.
Feedback and comments are greatly appreciated. A preprint of our work is available: https://tinyurl.com/fedstr
@PABLOF7z @Sovereign Engineering @Gigi @DETERMINISTIC OPTIMISM 🌞 @MW
#introductions
