Merge pull request #294 from Tencent/jiaruifang/accurate_timer

accurate timer

Merge pull request #277 from Tencent/develop

chunk size searching

Merge pull request #271 from Tencent/develop

upgrade to v0.4.4

Update README.md

hotfix a bug caused by copy_ instream

polish readme.

HyrbidPS schedules chunk schema in both static mode and eager mode. P…

…erformance is 50% for fp16 training

HyrbidPS schedules chunk schema in a eager mode. Performance is not s…

…o good because huge overhead on data moving between CPU and GPU