在 Visual Studio 2017 中使用 LSTM 层创建 keras 模型时出现 NotImplementedError

我的代码是：

import tensorflow as tf
keras_model = tf.keras.Sequential([
    tf.keras.layers.Embedding(1000, 16),
    tf.keras.layers.LSTM(32),
])

我有一个错误：

NotImplementedError：无法转换符号张量 (lstm/strided_slice:0) 到 numpy 数组。此错误可能表明 您正在尝试将张量传递给 NumPy 调用，但该调用不受支持

调用堆栈：

Tensor.__array__ in ops line 848    Python  
_wrapreduction in fromnumeric line 90   Python  
prod in fromnumeric line 2962   Python  
prod in C:\Users\FA\Google Drive\Colab Notebooks\PoetryTransformer\Accentizer\<__array_function__ internals> line 6 Python  
_constant_if_small in array_ops line 2732   Python  
zeros in array_ops line 2794    Python  
wrapped in _tag_zeros_tensor in array_ops line 2747 Python  
wrapper in add_dispatch_support in dispatch line 201    Python  
create_zeros in _generate_zero_filled_state in recurrent line 2981  Python  
<listcomp> in nest line 635 Python  
map_structure in nest line 635  Python  
_generate_zero_filled_state in recurrent line 2984  Python  
_generate_zero_filled_state_for_cell in recurrent line 2968 Python  
LSTMCell.get_initial_state in recurrent line 2524   Python  
RNN.get_initial_state in recurrent line 646 Python  
RNN._process_inputs in recurrent line 862   Python  
LSTM.call in recurrent_v2 line 1108 Python  
Layer._functional_construction_call in base_layer line 1117 Python  
Layer.__call__ in base_layer line 926   Python  
RNN.__call__ in recurrent line 663  Python  
Sequential.add in sequential line 221   Python  
_method_wrapper in no_automatic_dependency_tracking in base line 457    Python  
Sequential.__init__ in sequential line 142  Python  
_method_wrapper in no_automatic_dependency_tracking in base line 457    Python  
ragged module line 5    Python  

我无法使用顺序 API 或功能 API 添加 LSTM 层。 定制模型效果很好。 这可能是什么原因？

张量流版本是2.3.0。 Python 版本 3.6。

该错误与 Visual Studio 有关（当我运行没有 VS 时没有错误）。