Mesh LLM is an open-source project endorsed by Jack Dorsey that pools idle Nvidia GPUs into a decentralized peer-to-peer network for running large language models. The system uses a « Skippy » pipeline to partition massive models like a 235 billion-parameter Mixture of Experts across multiple machines. Performance reaches roughly 16 tokens per second on local area networks and 10 to 25 tokens per second over wide area networks. A related project called Meshtrain rewards GPU contributors with MeshCoin through a P2P local ledger, integrated into Block’s Buzz app. The zero-central-server architecture ensures data privacy as queries never touch external data centers.
Source: Read the original article
