DeepYardDeepYard
Q

Quasar

18B parameter MoE foundation model optimized for long-context agent applications

Open SourceFree

About

Quasar is an 18B parameter Mixture-of-Experts foundation model with 2B active parameters, specifically designed for long-context agent applications. Built by Silx AI with a hybrid transformer architecture, it leverages decentralized training through Bittensor's distributed network. The model excels at processing extended contexts while maintaining computational efficiency through its MoE approach, making it suitable for complex agent reasoning tasks.

Details

Language
Patterns

Tags

open-sourceautonomousframeworkpythonmulti-agent