版本1.7.0 py3

推理模型在notebook里用cpu测试没有出现问题，但在在线推理上报错，conv_bn_layer来自官方resnet模型：

System error: Invoke operator conv2d error.
Python Call stacks: 
  File \"/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/framework.py\", line 2488, in append_op
    attrs=kwargs.get(\"attrs\", None))
  File \"/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layer_helper.py\", line 43, in append_op
    return self.main_program.current_block().append_op(*args, **kwargs)
  File \"/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/nn.py\", line 2803, in conv2d
    \"data_format\": data_format,
  File \"\", line 128, in conv_bn_layer
    data_format=data_format)
  File \"\", line 64, in net
    data_format=data_format)
...
C++ Call stacks: 
op conv2d does not have kernel for data_type[float]:data_layout[NCHW]:place[CPUPlace]:library_type[PLAIN] at [/paddle/paddle/fluid/framework/operator.cc:979]
PaddlePaddle Call Stacks: 
0       0x7ff1225fc788p void paddle::platform::EnforceNotMet::Init(std::string, char const*, int) + 360
1       0x7ff1225fcad7p paddle::platform::EnforceNotMet::EnforceNotMet(std::string const&, char const*, int) + 87
2       0x7ff123958b99p paddle::framework::OperatorWithKernel::ChooseKernel(paddle::framework::RuntimeContext const&, paddle::framework::Scope const&, boost::variant const&) const + 1289
3       0x7ff123959c38p paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, boost::variant const&, paddle::framework::RuntimeContext*) const + 728
4       0x7ff12395a241p paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, boost::variant const&) const + 529
5       0x7ff1239551e4p paddle::framework::OperatorBase::Run(paddle::framework::Scope const&, boost::variant const&) + 260
6       0x7ff12277c11ep paddle::framework::Executor::RunPreparedContext(paddle::framework::ExecutorPrepareContext*, paddle::framework::Scope*, bool, bool, bool) + 206
7       0x7ff12277f1afp paddle::framework::Executor::Run(paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool, std::vector > const&, bool) + 143
...

看起来是conv2d对某些参数没注册？不过既然cpu上能测得过没道理啊....

notebook上测试:

[inference_program, feed_target_names, fetch_targets] = fluid.io.load_inference_model('work/Online3', exe)
inputImg_data = np.zeros(25, 3, 640, 640).astype(np.uint8)
results = exe.run(inference_program, feed={feed_target_names[0]: inputImg_data}, fetch_list=fetch_targets)

输入转换器，输入数据是图像base64：

def reader_infer(data_args):
    """
    reader_inter
    示例为目标检测数据预处理
    :param data_args: 接口传入的数据，以k-v形式
    :return [[]], feeder
    """
    def reader():
        """
        reader
        :return:
        """
        inImg = fluid.layers.data('inImg', shape=[-1, 3, 640, 640], dtype='uint8')
        feeder = fluid.DataFeeder(place=fluid.CUDAPlace(0), feed_list=[inImg])
        img = base64_to_image(data_args["inImg"])
		# do something
        assert imgSet.shape[2] == imgSet.shape[3] and imgSet.shape[2] == 640
        return [[imgSet]], feeder  # imgSet.shape:(25, 3, 640, 640)
    return reader