NVIDIA-Merlin/HugeCTR

HugeCTR is a high efficiency GPU framework designed for Click-Through-Rate (CTR) estimating training

What it solves

HugeCTR addresses the challenge of training and deploying large-scale deep learning recommendation models, specifically those with very large embedding tables that are often too massive for standard frameworks to handle efficiently.

How it works

It is a GPU-accelerated framework that optimizes the workflow for recommendation systems. It utilizes model-parallel training and multi-node distribution to handle massive datasets and model sizes. The framework provides a high-level Python interface for model definition and includes a Sparse Operation Kit for GPU-accelerated sparse training and inference operations.

Who it’s for

It is designed for data scientists and machine learning practitioners who need to build, train, and deploy high-performance recommender systems at scale.

Highlights

  • GPU Acceleration: Optimized specifically for high performance in recommendation benchmarks like MLPerf.
  • Scalability: Supports multi-node training and model-parallelism to manage large embeddings.
  • Developer Friendly: Offers a high-level Python API, comprehensive documentation, and a HugeCTR to ONNX converter.
  • Training Efficiency: Supports mixed precision training to speed up the learning process.

Related

  • Project
    NVIDIA/recsys-examples

    A collection of optimized recommender models and GPU-accelerated components for large-scale ranking, retrieval, and generative recommendation systems.

  • Project
    NVIDIA-Merlin/Merlin

    NVIDIA Merlin is an open-source library suite that accelerates the end-to-end creation of large-scale recommender systems using GPU acceleration for feature engineering, training, and inference.

  • Project
    InfiniTensor/InfiniTensor

    A high-performance inference engine for GPUs and AI accelerators that enables efficient model deployment and academic validation.

  • Project
    meta-recsys/generative-recommenders

    A framework for generative recommendation systems that reformulates recommendation as a generative modeling problem to overcome compute bottlenecks and scale to trillion-parameter models.

  • Project
    NVIDIA-NeMo/Curator

    A GPU-accelerated data curation framework for building repeatable pipelines to load, filter, deduplicate, and transform massive text, image, video, and audio datasets for AI training.