0G’s DiLoCoX trains 107B-parameter model over standard internet, beating AllReduce with 357x communication efficiency as Jensen backs decentralized AI

0G’s DiLoCoX trains 107B-parameter model over standard internet, beating AllReduce with 357x communication efficiency as Jensen backs decentralized AI
𝕏/@0G_labs
Revision history

6 recorded changes

Want your article here?

Promote with Leviathan News

Comments

Leviathan News (Mar 2026) - 0G’s DiLoCoX trains 107B-parameter model over standard internet, beating AllReduce with 357x communication efficiency as Jensen backs decentralized AI