Release v0.0.2

Main feature

1. Support PaddlePaddle v1.5.1
2. GPU inference engine, with a benchmark report on image classification model using SE_ResNeXt50_32x4d (see doc/GPU_BENCHMARKING.md)
3. Sparse parameter service and API with a demo. In this release we support sparse parameter read/write only within Paddle Serving process locally; in next release we will support distributed sparse parameter service.

First Release

PaddlePaddle inference serving framework