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

Meta has developed Triton Low-Level Extensions (TLX), an open-source extension to Triton that introduces explicit warp-group and cluster-level orchestration for GPU machine learning workloads. TLX bridges the gap between high-level productivity models and low-level hardware control, enabling engineers to adopt new GPU features on NVIDIA and AMD hardware. It currently powers production workloads across Meta, delivering performance improvements in recommendation system training and inference, as well as large-scale LLM pretraining. Open article →