Run_squad script does not parse adapter args correctly and does not save adapters

[xplip]

🐛 Bug

Information

Model I am using (Bert, XLNet ...):
mBERT

Language I am using the model on (English, Chinese ...):
Arabic, but the issue is language/dataset independent

Adapter setup I am using (if any):
Arabic lang adapter from adapterhub, new squad task adapter

The problem arises when using:

• the official example scripts: (give details below)
• my own modified scripts: (give details below)

The tasks I am working on is:

• an official GLUE/SQUaD task:
• my own task or dataset: (give details below)
  Arabic Reading Comprehension Dataset (ARCD) but should be the same for any dataset

To reproduce

Steps to reproduce the behavior:

Issue of args not being parsed correctly:

1. Try to define language adapter by providing --load_lang_adapter [adapter name]
2. Script fails as argument parser expects it to be --load_language_adapter
3. Try to define language adapter by providing --load_language_adapter instead
4. Script fails as setup_task_adapter_training function of adapter_training.py expects it to be --load_lang_adapter

Issue of adapters not being saved:

1. Run script to finetune adapters
2. Script stores full model for each checkpoint and at the end of training

Expected behavior

I should be able to define the language adapter with the --load_lang_adapter flag and its config with the --lang_adapter_config flag. When using adapters to finetune my model, I would usually like to store the adapters, not the full model.

Environment info

• transformers version: 2.11.0
• Platform: macOS-10.15.6-x86_64-i386-64bit
• Python version: 3.8.5
• PyTorch version (GPU?): 1.5.1 (False)
• Tensorflow version (GPU?): 2.3.0 (False)
• Using GPU in script?: False
• Using distributed or parallel set-up in script?: False