不论是训练还是推理，如果将model(x)放在dataset.create_dict_iterator()遍历的外面，程序正确运行。

​

但是只要将model(x)放入dataset.create_dict_iterator()的遍历后，就会报错，即使x并不是dataset中的数据。

每次运行的情况还不一样，有的时候直接卡在前向网络的Batchnorm计算中，程序不停止，也不报错，有的时候会报错RuntimeError: Illegal AnfNode for evaluating.node: @BatchNorm.20:фdout(type:Parameter), fg: BatchNorm.20 conf: Node: 0x18b20bbdb00/@BatchNorm.20:фdout-uid(2115), Context: 0x18b20b4fc70/{FuncGraph: BatchNorm.20 Args: [0]: AbstractTensor(shape: (1, 32, 40, 99), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), [1]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), [2]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), [3]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), [4]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), [5]: AbstractTuple{element[0]: AbstractTensor(shape: (1, 32, 40, 99), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[1]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[2]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[3]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[4]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue)}, [6]: AbstractTuple{element[0]: AbstractTensor(shape: (1, 32, 40, 99), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[1]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[2]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[3]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[4]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue)}, Parent: { Args: }}, FuncGraph: 0x18b20bb0cb0/BatchNorm.20 。