Describe the bug

Dropout is always passed as a flat rate independent of training status.

https://github.com/speechbrain/speechbrain/blob/develop/speechbrain/nnet/attention.py#L1367-L1374

Expected behaviour

Dropout should be 0.0 when not in training mode.

To Reproduce

No response

Environment Details

No response

Relevant Log Output

Additional Context

No response