NVIDIA Merlin™ is an open-source framework designed to build high-performing recommender systems at scale. It empowers data scientists, machine learning engineers, and researchers to streamline the entire recommender pipeline, from preprocessing and feature engineering to training, inference, and deployment. Merlin components and capabilities are optimized to support the retrieval, filtering, scoring, and ordering of hundreds of terabytes of data, all accessible through easy-to-use APIs.

Features

• Merlin Models: Provides standard models for recommender systems, including high-quality implementations from machine learning (ML) to advanced deep learning (DL) models on CPUs and GPUs.
• Merlin NVTabular: A feature engineering and preprocessing library designed to manipulate terabytes of recommender system datasets, significantly reducing data preparation time.
• Merlin HugeCTR: A deep neural network framework for recommender systems on GPUs, offering distributed model-parallel training and inference with hierarchical memory for maximum performance and scalability.
• Merlin Transformers4Rec: Streamlines the building of pipelines for session-based recommendations, making it easier to explore and apply popular transformers architectures.
• Merlin Distributed Training: Supports distributed training across multiple GPUs, leveraging model parallelism to scale training.

Benefits

• Better Predictions: Achieve better predictions and increased click-through rates with optimized models and tools.
• Faster Deployment: Accelerate deployment to production with streamlined workflows and easy-to-use APIs.
• Interoperable Solution: Designed to be interoperable within existing recommender workflows, supporting data science, ML, and DL on CPUs or GPUs.
• Scalability: Optimized to handle hundreds of terabytes of data, ensuring scalability for large-scale recommender systems.
• Comprehensive Support: Includes libraries, methods, and tools that address common challenges in building and deploying recommenders.