Torch-RecHubA Lightweight PyTorch Framework for Recommendation Systems

Easy-to-use, extensible, and high-performance recommendation system framework

Ranking Workflows

Cover production-friendly ranking models such as WideDeep, DeepFM, DIN, BST, DIEN, and DCN.

Explore ranking models

Support two-tower retrieval, sequential recall, multi-interest modeling, and recall evaluation.

Explore retrieval models

Train ESMM, MMOE, PLE, AITM, and related multi-task architectures in one workflow.

Explore multi-task models

Explore generative recommendation directions with the same data, trainer, and deployment conventions.

Explore generative models

Export to ONNX, validate runtime behavior, and connect serving-friendly deployment steps.

Open ONNX guide

Plug in Annoy, Faiss, or Milvus for vector indexing and retrieval experiments.

Open vector indexing

Add callbacks, tracking, and visualization to keep experiments reproducible and easier to debug.

Open tracking tools

Reuse one feature schema, data pipeline, and trainer pattern across ranking, matching, and tutorials.

Open data pipeline

RankingMatchingMulti-taskGenerativeONNXVector SearchTrackingVisualization