A unified PyTorch recommendation library with 94 benchmark algorithms.
A comprehensive PyTorch-based framework for recommendation algorithms and benchmarking.
RecBole is a Python and PyTorch-based library designed for reproducing and developing recommendation algorithms. It suits data scientists and researchers building general, sequential, context-aware, or knowledge-based recommendation systems.
It includes 94 recommendation algorithms categorized into general, sequential, context-aware, and knowledge-based models. The library offers a unified data structure, 44 formatted benchmark datasets, and GPU-accelerated execution supporting multi-GPU and mixed precision training.
It suits researchers and machine learning engineers developing or benchmarking recommendation algorithms. It does not suit non-technical users or those looking for an out-of-the-box production recommendation API without Python coding.
Covers general, sequential, context-aware, and knowledge-based recommendation algorithms.
Provides pre-processed benchmark datasets ready for immediate model training.
Optimized for GPU environments with multi-GPU and mixed precision training capabilities.
Includes built-in search methods in both serial and parallel for optimal configurations.
Free under its MIT licence, with the only cost being running it yourself on your hardware.
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
RecBole requires Python version 3.7 or later and PyTorch version 1.7.0 or later.
The library provides formatted copies of 44 benchmark recommendation datasets.
See Zekai first in Google