NVIDIA Merlin

An open-source framework designed to accelerate every stage of the recommender system pipeline from data preprocessing to high-performance inference.

NVIDIA Merlin streamlines the development of deep learning recommenders by integrating specialized libraries like NVTabular for ETL and HugeCTR for distributed training. It eliminates the common bottleneck of CPU-bound data processing: moving operations to GPUs to achieve 10x faster training cycles. By supporting industry-standard models and offering the Triton Inference Server for deployment, Merlin enables teams to scale recommendation engines that handle massive datasets (terabyte-scale) with millisecond latency.

https://developer.nvidia.com/nvidia-merlin