Metaspore Online Serving is an online application framework developed by DMetaSoul that supports online algorithm business services and low-code mode. Contains three sub-modules serving, experiment-pipeline, and feature-extract.
-
serving: Model computing service, providing Sparse DNN model, twin tower model, Spark ML model, SKLearn model, XGBoost model prediction service and Milvus vector recall service.
-
experiment-pipiline: Experiment Workflow SDK, a real-time dynamic update, low-code mode algorithm experiment flow framework. Supports common online functions such as dynamic update experiments, multi-layer orthogonality, and experimental cut flow.
-
feature-extract: A framework for automatic feature generation, a framework for automatically generating JPA code for features based on Yaml files. Supports multiple DB sources and provides feature interfaces conveniently.
Complete example: Movielens Demo example link
| Documentation | |
|---|---|
| serving | Doc |
| experiment-pipeline | Doc |
| feature-extract | Doc |