The parameter server is a distributed system scaling to industry size machine learning problems. It provides asynchronous and zero-copy key-value pair communications between worker machines and server machines. It also supports flexiable data consistency model, data filters, and flexiable server machine programming.

NOTE: We are refactoring this repository in the lite branch, there is a simpler communication interface, and application codes will be moved to other DMLC repositories soon.

• Document
• Wiki
• How to build
• Examples
  • Linear method, Linear method with Cloud
  • Deep neural network, see CXXNET and Minverva