安装GPU版本后运行异常

[nuoline]

安装的GPU版本
运行：paddle train
--config=postag_train.py
--parallel_nn=1
--use_gpu=true
--dot_period=10
--log_period=1000
--test_period=0
--num_passes=10
错误提示如下：
python27-gcc482/bin/paddle: line 83: [: too many arguments
File "", line 3
if LooseVersion("0.10.0
^
SyntaxError: EOL while scanning string literal
I1115 15:42:03.529911 4739 Util.cpp:166] commandline: /home/ssd1/zhaizhouwei/python27-gcc482/bin/paddle_trainer --config=postag_train.py --parallel_nn=1 --use_gpu=true --dot_period=10 --log_period=1000 --test_period=0 --num_passes=10
W1115 15:42:11.678481 4739 CpuId.h:112] PaddlePaddle wasn't compiled to use avx instructions, but these are available on your machine and could speed up CPU computations via CMAKE .. -DWITH_AVX=ON
I1115 15:42:11.881070 4739 Trainer.cpp:165] trainer mode: Normal
F1115 15:42:11.883350 4739 hl_gpu_matrix_kernel.cuh:181] Check failed: cudaSuccess == err (0 vs. 8) [hl_gpu_apply_unary_op failed] CUDA error: invalid device function
*** Check failure stack trace: ***
@ 0x66fb8d google::LogMessage::Fail()
@ 0x67363c google::LogMessage::SendToLog()
@ 0x66f683 google::LogMessage::Flush()
@ 0x674b4e google::LogMessageFatal::~LogMessageFatal()
@ 0x940f04 hl_gpu_apply_unary_op<>()
@ 0x941245 paddle::BaseMatrixT<>::applyUnary<>()
@ 0x941473 paddle::BaseMatrixT<>::zero()
@ 0x7afd81 paddle::Parameter::enableType()
@ 0x7ab60c paddle::parameterInitNN()
@ 0x7add5a paddle::NeuralNetwork::init()
@ 0x7d402f paddle::ParallelNeuralNetwork::init()
@ 0x7d6a11 paddle::GradientMachine::create()
@ 0x8a7a08 paddle::TrainerInternal::init()
@ 0x8a40b2 paddle::Trainer::init()
@ 0x652a84 main
@ 0x7f07e3992bd5 __libc_start_main
@ 0x66d425 (unknown)
@ (nil) (unknown)
/home/ssd1/zhaizhouwei/python27-gcc482/bin/paddle: line 96: 4739 Aborted ${DEBUGGER} $PADDLE_BIN_PATH/paddle_trainer ${@:2}

[guoshengCS]

有类似的issue可以参考下 #2931 (comment)

[sweetsky0901]

可以查看一下你本地的cuda和cudnn的版本。
如果是通过pip install安装的，是需要要求cuda8和cudnn5的。
具体下载地址
cuda8的下载地址是：https://developer.nvidia.com/cuda-toolkit-archive
cudnn5的下载地址是：https://developer.nvidia.com/rdp/cudnn-archive

[Yancey1989]

这个issue由于长时间未更新先关闭了，如有依然有问题请重新打开它。