Apache SINGA

Apache SINGA, an Apache Software Foundation top-level project, is a distributed deep-learning library for training machine learning models across multiple GPUs, with the notable added capability of deploying trained models directly inside relational databases for querying. It handles automatic gradient calculation via computational graphs, memory and parameter optimization, ONNX interoperability for exchanging models with other frameworks, and time-profiling for individual operators — with installation available via Conda, Pip, Docker, or building from source. Current release 5.1.0, with adoption reported at organizations including Alibaba and Citigroup, points to genuine production use beyond academic experimentation. As distributed deep-learning infrastructure, it requires real ML-engineering expertise to deploy effectively; it's free and open source under the Apache License with no cost to use. Good fit for a team doing distributed deep-learning training who specifically wants to deploy trained models inside a relational database for direct querying, a less common capability among mainstream frameworks.