RecBole

Provides general and extensible data structures that support 44 benchmark datasets in a consistent format, simplifying data preprocessing for researchers.

Implements 94 recommendation algorithms across four major categories, offering a broad baseline for benchmarking and development in academic settings.

Includes multi-GPU and mixed precision training support, with efficiency improvements tailored for GPU environments, as noted in the library's performance tests.

Supports widely adopted protocols like recall@10 and NDCG, ensuring reproducible and comparable results across experiments.

Offers multiple search methods (serial and parallel) for hyperparameter optimization, detailed in the auto-tuning example with config files.