Singa

Singa: Open-Source Distributed Deep Learning Platform

Singa is an open-source distributed deep learning platform that can train large machine learning models over CPUs, GPUs, or clusters. It provides a flexible programming model that supports a wide range of deep learning frameworks and algorithms.

What is Singa?

Singa is an open-source distributed deep learning platform designed specifically for large-scale machine learning workloads. It is created by the DB System Group at Singapore's National University and supported by the National Research Foundation.

Singa aims to efficiently support a variety of machine learning models, algorithms, and frameworks. Its key features and benefits include:

• Flexibility - Singa’s programming interface allows users to implement new deep learning frameworks and mix different frameworks within one model. This offers more choice compared with using a single framework.
• Scalability - Singa can automatically scale up or down across multiple GPUs and multiple machines. This allows faster training for very large models and datasets, without needing to change code.
• Performance - Singa is optimized to utilize the latest hardware technology like GPUs, SIMD instructions, multi-threading, and distributed training to achieve high computational speed and efficiency.
• Usability - Singa integrates closely with Apache Zeppelin and Jupyter notebooks, providing easy and interactive user experiences for research and collaboration.

The Singa framework aims to support many deep learning functionalities including DNN/CNN, RNN, and distributed deep learning out of the box. Popular machine learning libraries like Tensorflow, Caffe, PyTorch, and MXNet can also integrate with Singa.

Singa Features

Official Links