Llama adapter

[sywangyi]

What does this PR do?

Fixes # (issue)

Before submitting

• [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• [ ] Did you make sure to update the documentation with your changes?
• [ ] Did you write any new necessary tests?

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[emascarenhas]

Yi, I have 2 initial questions.

1. The default of 2 adapter_layers is that expected to give comparable results in terms of accuracy like LoRA? What do the authors of llama_adapter say about the default?
2. why the change from unittest to pytest in test_peft_inference.py?

[sywangyi]

Yi, I have 2 initial questions.

1. The default of 2 adapter_layers is that expected to give comparable results in terms of accuracy like LoRA? What do the authors of llama_adapter say about the default?
2. why the change from unittest to pytest in test_peft_inference.py?

1. please see https://github.com/OpenGVLab/LLaMA-Adapter, the simple fine-tuning code of LLaMA-Adapter on LLaMA-7B model, I just change default value of adapter_layer and adapter_len to adapt to the command.
2. for use of pytest.mark.parametrize. We can have only one method called test_text_generation_llama(with the @pytest.mark.parametrize(...) decorator to avoid defining one peft test per model