Confusion between Training and Trainable

[vishvapalsinh]

keras-bert/keras_bert/loader.py

Line 22 in f8bb7ab

def build_model_from_config(config_file,

In the above-defined function, I am not able to assimilate the use case of Training and Trainable.
Please correct me if my understandings are wrong

Case 1: training = True & trainable = True
My model will train all the weights and will write MLM and NSP layers output.
The model will also fine-tune all the weights from the beginning.

Case 2: training = False & trainable = False
Here my model will not train at all and also won't provide NSP layers output.

Case 3: training = True & trainable = False
In this case, the model will provide MLM and NSP but the weights of the model won't update. Therefore, the model will not finetune its wights according to tasks (i.e. classification or labeling).

Any help will be highly appreciated.
Thank you