Varying input shape

[anasvaf]

Hello,

Thank you for translating the PyTorch script to TensorFlow :) I have a question about the input shape that we can feed to the ConformerBlock. The block, currently, can support an input shape of [Batch_Size, Time, Frequency] (in the case of audio). Assume that my Time dimension is not fixed, but is varying. Therefore, an example could be that i get [None(batch_size), None(Time_size), 128(Frequency)]. The block as it is does not support it. The error is occurring in the attention.py that the line seq = tf.range(n) does not support None values. Any ideas how to overcome this issue?