DeepYardDeepYard
Q

Quasar

10B parameter foundation model with 2M token context using Gated Linear Attention

Open SourceFree

About

Open-source foundation model framework built on Qwen3.5-9B that achieves 2 million token context windows through Gated Linear Attention (GLA) architecture. Unlike transformer-based models with quadratic complexity, Quasar uses linear attention for efficient processing of extremely long contexts. Ideal for researchers and developers working with large documents, codebases, or multi-turn conversations requiring extensive context retention.

Details

Language
Patterns

Tags

open-sourceframeworkragpythonself-hosted