what's Quasar?
Quasar is Bittensor Subnet 24 (SN24), a decentralized AI network designed to overcome context length limitations in large language models. It introduces Quasar Attention, a novel mechanism employing a continuous-time formulation and fully matrix-based system, enabling stable processing of sequences up to 5 million tokens without the instability or degradation seen in traditional or linear attention methods like gated delta attention. The subnet incentivizes miners and validators to contribute compute for training and inference of long-context models, demonstrating superior performance on benchmarks such as BABILong and RULER. With open-source weights and implementations available on platforms like Hugging Face and GitHub, Quasar fosters collaborative advancement in decentralized machine intelligence on the Bittensor protocol.
Quasar is a long-context AI subnet on Bittensor for reasoning across large datasets.
Links
x.com/quasarmodelsOperates as a long-context AI subnet on the Bittensor network, designed to handle reasoning tasks across large datasets without losing contextual information.