Transformer inputs [batch_size, input_length, 1, hidden_dim]

[icecream0]

Description

Transformer inputs [batch_size, input_length, 1, hidden_dim], I want to know why inputs shape is not [batch_size, input_length, hidden_dim]， have some reasons？please tell me ,thank you ...

Environment information

OS: <your answer here>

$ pip freeze | grep tensor
# your output here

$ python -V
# your output here

For bugs: reproduction and error logs

# Steps to reproduce:
...

# Error logs:
...

[martinpopel]

I think all inputs are treated as two dimensional, so the framework can generalize to images easily.

[icecream0]

I think all inputs are treated as two dimensional, so the framework can generalize to images easily.

maybe，thanks a lot. transformer mainly used for nlp, so the design of framework make code reader more difficult