Quasar

Quasar optimizes flash-linear attention kernels for long-context LLM inference (10M+ tokens). Miners develop and submit optimized kernels that are verified through logit-level inference verification, making ultra-long context processing fast and efficient.