capi sequence方式调用出core

[fty8788]

仿照Paddle/paddle/capi/examples/model_inference/sequence/main.c的代码，改成DSSM模型的预测程序，input是两个ids：

CHECK(paddle_arguments_resize(in_args, 2));

  // Create input ids.
  int sentence_ids[] = {83, 48, 20, 84, 394, 853, 64, 53, 94};

  paddle_ivector sentence0 = paddle_ivector_create(
      sentence_ids, sizeof(sentence_ids) / sizeof(int), false, false);
  paddle_ivector sentence1 = paddle_ivector_create(
      sentence_ids, sizeof(sentence_ids) / sizeof(int), false, false);
  CHECK(paddle_arguments_set_ids(in_args, 0, sentence0));
  CHECK(paddle_arguments_set_ids(in_args, 1, sentence1));

  int seq_pos_array[] = {0, sizeof(sentence_ids) / sizeof(int)};

  paddle_ivector seq_pos0 = paddle_ivector_create(
      seq_pos_array, sizeof(seq_pos_array) / sizeof(int), false, false);
  paddle_ivector seq_pos1 = paddle_ivector_create(
      seq_pos_array, sizeof(seq_pos_array) / sizeof(int), false, false);

  CHECK(paddle_arguments_set_sequence_start_pos(in_args, 0, 0, seq_pos0));
  CHECK(paddle_arguments_set_sequence_start_pos(in_args, 1, 0, seq_pos1));

  paddle_arguments out_args = paddle_arguments_create_none();
  CHECK(paddle_gradient_machine_forward(machine,
                                        in_args,
                                        out_args,
                                        /* isTrain */ false));

运行出core：
F0821 11:17:06.300724 25829 Matrix.cpp:3273] Check failed: dynamic_cast<CpuIVector*>(&label)
*** Check failure stack trace: ***
    @     0x7ffff6cee87d  google::LogMessage::Fail()
    @     0x7ffff6cf232c  google::LogMessage::SendToLog()
    @     0x7ffff6cee3a3  google::LogMessage::Flush()
    @     0x7ffff6cf383e  google::LogMessageFatal::~LogMessageFatal()
    @     0x7ffff6c47899  paddle::CpuMatrix::oneHotCrossEntropy()
    @     0x7ffff6abc475  paddle::CostLayer::forward()
    @     0x7ffff6b65530  paddle::NeuralNetwork::forward()
    @     0x7ffff6a1c4e6  paddle_gradient_machine_forward
    @           0x401774  main
    @     0x7ffff6483bd5  __libc_start_main
    @           0x4011b9  (unknown)
    @              (nil)  (unknown)

Program received signal SIGABRT, Aborted.
0x00007ffff64973f7 in raise () from /opt/compiler/gcc-4.8.2/lib/libc.so.6
(gdb) bt
#0  0x00007ffff64973f7 in raise () from /opt/compiler/gcc-4.8.2/lib/libc.so.6
#1  0x00007ffff64987d8 in abort () from /opt/compiler/gcc-4.8.2/lib/libc.so.6
#2  0x00007ffff6cf7715 in google::DumpStackTraceAndExit() () from /home/yanchunwei/third_party/tengfei/Paddle/paddle/capi/examples/model_inference/usr/local/lib/libpaddle_capi_shared.so
#3  0x00007ffff6cee87d in google::LogMessage::Fail() () from /home/yanchunwei/third_party/tengfei/Paddle/paddle/capi/examples/model_inference/usr/local/lib/libpaddle_capi_shared.so
#4  0x00007ffff6cf232c in google::LogMessage::SendToLog() () from /home/yanchunwei/third_party/tengfei/Paddle/paddle/capi/examples/model_inference/usr/local/lib/libpaddle_capi_shared.so
#5  0x00007ffff6cee3a3 in google::LogMessage::Flush() () from /home/yanchunwei/third_party/tengfei/Paddle/paddle/capi/examples/model_inference/usr/local/lib/libpaddle_capi_shared.so
#6  0x00007ffff6cf383e in google::LogMessageFatal::~LogMessageFatal() () from /home/yanchunwei/third_party/tengfei/Paddle/paddle/capi/examples/model_inference/usr/local/lib/libpaddle_capi_shared.so
#7  0x00007ffff6c47899 in paddle::CpuMatrix::oneHotCrossEntropy(paddle::Matrix&, paddle::VectorT<int>&) () at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/math/Matrix.cpp:3273
#8  0x00007ffff6abc475 in paddle::CostLayer::forward(paddle::enumeration_wrapper::PassType) () at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/gserver/layers/CostLayer.cpp:50
#9  0x00007ffff6b65530 in paddle::NeuralNetwork::forward(std::vector<paddle::Argument, std::allocator<paddle::Argument> > const&, std::vector<paddle::Argument, std::allocator<paddle::Argument> >*, paddle::enumeration_wrapper::PassType) () at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/gserver/gradientmachines/NeuralNetwork.cpp:250
#10 0x00007ffff6a1c4e6 in paddle_gradient_machine_forward () at /home/yuyang/BuildAgent3/work/d55918cf60d51073/paddle/capi/gradient_machine.cpp:81
#11 0x0000000000401774 in main ()

[luotao1]

capi 模型文件要为infernce的而不是train的模型。
从core中看，CostLayer::forward()表示有cost层，这是train的模型。

[fty8788]

https://gist.github.com/fty8788/313a9e94e5738065f482436307da8a2a

[luotao1]

@Superjom 能请你看一下么， @fty8788 设置了is_infer=True也是一样的错误

[luotao1]

DSSM 模型已更新，修改见 PaddlePaddle/models#212 。可用最新模型试一下。

[Superjomn]

模型在非 capi 状态下，应该是 train/infer 没问题的吧？ @fty8788
如果是这样，主要看下 capi 是否有问题吧

[fty8788]

@luotao1 一样的错误

[fty8788]

@Superjom python版本没有问题，capi的问题

[fty8788]

删掉配置中的cost层就可以了，即使不返回，也不能存在