等我一期双色球
@xhy941249849
Two decentralized AI projects I've been following recently: @fortytwonetwork: AI inference platform from @monad_xyz ecosystem: AI inference platform deployed on high-performance L1 Monad, distributing AI inference tasks to consumer devices (retail investors-civilian devices) around the world to build a dynamically expanding network Highlights: The core lies in efficient distributed inference capabilities + high performance of Monad chain, completing multiple nodes within milliseconds, and practicing real-time and low-latency characteristics to the extreme. Flexible task allocation and communication mechanisms also make large-scale node collaboration more stable
0 reply
0 recast
1 reaction
等我一期双色球
@xhy941249849
@PluralisHQ: Focuses on decentralized AI training and proposes the concept of "Protocol Models": a communication-efficient model parallel method designed specifically for decentralized training. Latest news: A major breakthrough was announced on June 3: For the first time, it was proven that large language models can be distributedly trained on consumer devices connected to the Internet without losing speed or performance Official original text: https://x.com/PluralisHQ/status/1929916173530849439 Generally speaking, AI models are fed by GPU computing power, which requires high-speed network connections in addition to GPU costs. It is difficult for retail investors to have tickets to participate in it, but Pluralis splits large language models and uses low-rank compression (compression rate exceeds 100 times) to achieve distributed training on consumer devices without losing speed and performance, solving the high cost and unsustainability of centralized training.
0 reply
0 recast
0 reaction