Created by: ellinyang

1）paddle版本：Paddle v2 2）GPU ：Tesla P4 3）系统环境：Linux version 3.10.0_3-1-0-3 (gcc version 4.8.2 20140120 (Red Hat 4.8.2-16) (GCC) ) 训练信息：多卡 ；图像大小300，初始学习率0.00025

单机跑SSD时没有问题，用分布式训练时跑几个batch便会出现以下问题：

................................Thread [140458398054144] Forwarding detection_output,
*** Aborted at 1558425098 (unix time) try "date -d @1558425098" if you are using GNU date ***
PC: @                0x0 (unknown)
*** SIGFPE (@0x7fbf03bc1a18) received by PID 52 (TID 0x7fbf04ebf700) from PID 62659096; stack trace: ***
    @     0x7fbf04897390 (unknown)
    @     0x7fbf03bc1a18 __expf_finite
    @     0x7fbf03bc942f expf
    @     0x7fbeb51d4899 paddle::decodeBBoxWithVar()
    @     0x7fbeb51debd1 paddle::DetectionOutputLayer::forward()
    @     0x7fbeb523ffcd paddle::NeuralNetwork::forward()
    @     0x7fbeb5240ce3 paddle::GradientMachine::forwardBackward()
    @     0x7fbeb5516044 GradientMachine::forwardBackward()
    @     0x7fbeb50b1c79 _wrap_GradientMachine_forwardBackward
    @           0x4cb755 PyEval_EvalFrameEx
    @           0x4c2705 PyEval_EvalCodeEx
    @           0x4ca7df PyEval_EvalFrameEx
    @           0x4c2705 PyEval_EvalCodeEx
    @           0x4ca088 PyEval_EvalFrameEx
    @           0x4c2705 PyEval_EvalCodeEx
    @           0x4ca088 PyEval_EvalFrameEx
    @           0x4c9d7f PyEval_EvalFrameEx
    @           0x4c9d7f PyEval_EvalFrameEx
    @           0x4c9d7f PyEval_EvalFrameEx
    @           0x4c2705 PyEval_EvalCodeEx
    @           0x4c24a9 PyEval_EvalCode
    @           0x4f19ef (unknown)
    @           0x4ec372 PyRun_FileExFlags
    @           0x4eaaf1 PyRun_SimpleFileExFlags
    @           0x49e208 Py_Main
    @     0x7fbf044dc830 __libc_start_main
    @           0x49da59 _start
    @                0x0 (unknown)
Floating point exception