How We Built TLX: A Hardware-Native GPU Compiler for Production ML at Scale

Meta detailed Triton Low-Level Extensions (TLX), an extension to the Triton GPU compiler that adds explicit warp-group and cluster-level orchestration so engineers can use new hardware capabilities while keeping Triton's programming model. Meta reported that TLX has demonstrated performance improvements in RecSys training, RecSys inference, and large-scale LLM pretraining. The work targets production ML at scale on modern GPU architectures.