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
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