ValueError: Tensor-typed variable initializers must either be wrapped in an init_scope or callable (e.g., `tf.Variable(lambda : tf.truncated_normal([10, 40]))`) when building functions. Please file a feature request if this restriction inconveniences you.

[ruthu1998]

I get this error while running the generate_alstmfcn function.
Please help.
x = AttentionLSTM(NUM_CELLS)(ip)

[kamran0310]

You might be using multi variate LSTM.
If above is the case you should import layer_utils_multivariate instead of layer_utils dependencies from utils directory.