0G Labs partnered with China Mobile to successfully train a 107-billion-parameter AI model using a decentralized approach over standard 1 Gbps network links. Their DiLoCoX framework achieved a 357x improvement in communication efficiency compared to traditional methods by allowing GPU clusters to synchronize less frequently and using innovations like pipeline parallelism and adaptive gradient compression. This breakthrough suggests telecom companies could host distributed AI training without specialized high-bandwidth infrastructure. 0G Labs plans to open-source the technology for independent verification and broader adoption, potentially democratizing large AI model training.
Veea Inc. has announced the commercial release of VeeaONE, a full-stack AI infrastructure platform that integrates edge-to-cloud hardware and software. The platform supports VeeaWare middleware on NVIDIA Jetson and x86/Arm servers, enabling secure, l...
Virtuals Protocol has introduced customizable tokenized indexes on the Robinhood Chain, allowing users to combine multiple assets into a single token. This innovation enables users to create and manage their own asset baskets and earn fees when other...
US spot Bitcoin ETFs saw net inflows of $107.8 million on Wednesday, while Ethereum ETFs attracted $53.8 million, signaling renewed institutional interest in crypto funds. Since their launch in January 2024, Bitcoin ETFs have accumulated over $51 bil...
Kaia has signed a strategic partnership with Xilo Labs to integrate Mochi, an onchain asset management platform, into Kaia's ecosystem. This collaboration allows users to invest stablecoins into diversified portfolios backed by tokenized real-world a...
Bitcoin's price surged to nearly $65,000 following softer-than-expected U.S. inflation data for June, easing fears of Federal Reserve rate hikes. However, on-chain data reveals two main groups of investors—long-term holders who bought near last year'...
