alibaba/TorchEasyRec

An easy-to-use framework for large scale recommendation algorithms.

What it solves

TorchEasyRec is a PyTorch-based framework designed to simplify the development and deployment of production-ready deep learning recommendation systems. It addresses the complexity of implementing state-of-the-art recommendation models, managing large-scale embeddings, and ensuring consistent feature generation between training and serving.

How it works

The framework provides a suite of pre-implemented models for different recommendation stages—including candidate generation (matching), scoring (ranking), multi-task learning, and generative recommendation. It utilizes TorchRec for hybrid data and model parallelism to handle large embeddings through various sharding strategies (row-wise, column-wise, table-wise). It also supports a wide range of data sources including MaxCompute, Parquet, CSV, and Kafka streaming, and integrates with Alibaba Cloud's PAI ecosystem for distributed training and EAS deployment.

Who it’s for

This tool is for machine learning engineers and data scientists specializing in recommendation systems who need to build, scale, and train high-performance models for production environments.

Highlights

  • Extensive Model Library: Includes 20+ battle-tested algorithms such as DSSM, DeepFM, DIN, MMoE, and DLRM-HSTU.
  • C-Scale Scalability: Supports distributed training with mixed precision (FP16/BF16) and zero-collision hashing for dynamic embeddings.
  • Production-Ready Integration: Native integration with Alibaba Cloud PAI-DLC, PAI-DSW, and EAS for seamless deployment and acceleration via TensorRT/AOTInductor.
  • Flexible Feature Engineering: Supports over 10 feature types (e.g., IdFeature, SequenceFeature) and allows for easy implementation of custom features and models.

Related

  • Project
    meta-pytorch/torchrec

    A PyTorch domain library providing sparsity and parallelism primitives for training and deploying large-scale recommender systems with massive embedding tables sharded across multiple GPUs.

  • Project
    datawhalechina/torch-rechub

    Torch-RecHub is a PyTorch framework for building recommender systems with 30+ models, modular design, ONNX export, and support for multiple hardware backends.

  • Project
    PersiaML/PERSIA

    A PyTorch-based parallel training system designed to scale deep learning recommendation models up to 100 trillion parameters on commodity hardware.

  • Project
    pytorch/pytorch

    PyTorch is an open-source machine learning framework that provides GPU-accelerated tensor computation and a flexible tape-based autograd system for building deep neural networks.

  • Project
    alibaba/RecIS

    A unified deep learning framework based on PyTorch for ultra-large-scale sparse and dense computing, specifically optimized for industrial recommendation, advertising, and search systems.