SYLIN

Together compiler

by Together AI · Framework · Text & code

Agents search for smaller GPU-operation groups to fuse instead of hand-tuned megakernels; 28% higher throughput on Llama 3.2 1B on B200

StatusResearch