Macrocosmos Announces IOTA SDK and Liquid Compute at Exploit Summit
IOTA aims to let anyone operate a decentralized data center and train frontier-scale models with up to 10 trillion parameters at up to 3x lower cost than centralized labs.
Macrocosmos
Macrocosmos Tests IOTA With 16B-Parameter Internet Training Run
The Subnet 9 system is training a 16B model across high-end consumer GPUs and distributed compute providers as Macrocosmos stress-tests permissionless AI training.
Macrocosmos Unveils Orion-100B, a 100B-Parameter Distributed AI Training Run
Macrocosmos unveiled Orion-100B, a 100-billion-parameter distributed AI training run on Bittensor that achieved up to 65% of datacenter training efficiency using globally distributed GPUs.