modify hip readme (#436)

README_HIP.md

@@ -1,30 +1,32 @@
-## Lightseq for HIP Quick Start
-Now lighteq hip only support training.
+## Lightseq HIP Quick Start
+LightSeq supports CUDA backend and HIP backend. Now LightSeq HIP only support training.
 ### Build
 
-1. 设置环境变量
+1. ENV Setting
 ```shell
 export C_INCLUDE_PATH=${ROCM_PATH}/rocblas/include:${ROCM_PATH}/rocrand/include/:${ROCM_PATH}/hiprand/include:${ROCM_PATH}/hip/include/hip:$ROCM_PATH/hip/include/hip/hsa_detail:$ROCM_PATH/hipcub/include/hipcub:$C_INCLUDE_PATH
 export CPLUS_INCLUDE_PATH=$C_INCLUDE_PATH
 export LD_LIBRART_PATH=${ROCM_PATH}/rocblas/lib/:${ROCM_PATH}/rocrand/lib:$LD_LIBRART_PATH
 export LIBRARY_PATH=${ROCM_PATH}/rocblas/lib/:$LIBRARY_PATH
 ```
-2. 编译lightseq
+2. Compile
 ```
 verbose=1 ENABLE_FP32=1 ENABLE_DEBUG=1 CXX=hipcc CC=hipcc python3 setup.py install bdist_wheel
 ```
-编译后在dist下生成whl文件,可以方便的迁移到其他平台环境使用。
-### Fast training from Fairseq
+After compiling, whl file is generated under dist, which can be conveniently migrated to other platforms.
+
+### Fast training From Fairseq with LightSeq HIP
 
 You can experience lightning fast training by running following commands,
 Firstly install these requirements.
 
 ```shell
-pip install lightseq fairseq sacremoses
+pip install fairseq sacremoses
 ```
-python层同cuda环境运行方式完全一致，更多示例可以参考[README.md](README.md).
+Users need no modification with python training. It works exactly the same as CUDA environment under the HIP backend. See [README.md](README.md) for more examples.
+
+You can set HIP_VISIBLE_DEVICES to specify which gpu card for training. Example of training transformer with LightSeq HIP is as follows:
 
-如需指定加速卡运行，可使用HIP_VISIBLE_DEVICES指定。使用lightseq训练transformer示例：
 ```
 export HIP_VISIBLE_DEVICES=0
 lightseq-train /public/DL_DATA/wmt14_en_de_joined_dict  \