RuntimeError: shape '[80, -1, 8, 40]' is invalid for input of size 20971520

Because of this setting, transformer input's (hidden state) batch size turns into 16, but text embedding's (encoder_hidden_state) batch size remains 80, then causes that error.
How can I fix it?
Thanks!
RuntimeError: shape '[80, -1, 8, 40]' is invalid for input of size 20971520
Because of this setting, transformer input's (hidden state) batch size turns into 16, but text embedding's (encoder_hidden_state) batch size remains 80, then causes that error.
How can I fix it?
Thanks!