Part 5/11:
Another noteworthy development is Petals—a decentralized platform enabling the operation of 100-billion-parameter language models across distributed GPUs.
How Petals Works
Modeled on peer-to-peer computing concepts like SETI@home and Folding@Home, Petals fragments large models into smaller layers hosted across multiple user machines. Participants contribute their GPU processing power to run segments of the model, collaboratively enabling inference without centralized infrastructure.