Add adapter_summary() method

[calpt]

This PR introduces a new adapter_summary() method that provides a summary of the currently added adapters.

Example

model = AutoAdapterModel.from_pretrained("roberta-base")
for name, config in ADAPTER_CONFIG_MAP.items():
    model.add_adapter(name, config=config)
print(model.adapter_summary())

outputs...

Name                     Architecture         #Param      %Param  Active   Train
--------------------------------------------------------------------------------
pfeiffer                 bottleneck           894528       0.718       0       1
houlsby                  bottleneck          1789056       1.435       0       1
pfeiffer+inv             bottleneck          1190592       0.955       0       1
houlsby+inv              bottleneck          2085120       1.673       0       1
compacter++              bottleneck            28576       0.023       0       1
compacter                bottleneck            57088       0.046       0       1
prefix_tuning            prefix_tuning       9872384       7.920       0       1
prefix_tuning_flat       prefix_tuning        552960       0.444       0       1
parallel                 bottleneck          7091712       5.689       0       1
scaled_parallel          bottleneck          7091724       5.690       0       1
lora                     lora                 294912       0.237       0       1
mam                      union              22493984      18.046       0       1
--------------------------------------------------------------------------------
Full model                                 124645632     100.000               1

Other fixes

Fix get_adapter() for:

• prefix tuning
• adapters w. shared or invertible adapters

[JoPfeiff]

❤️