In-Context Alignment: Chat with Vanilla Language Models Before Fine-Tuning

• Use decode_icl_llama.py to generate responses with Llama-2-vanilla models with in-context alignment.
• Use other decode_*.py files to generate responses with baseline models.
• Use eval_outputs.py for automatic evaluations.
• Example commands can be found in the headers of the files.
• For more details, please refer to our paper or email Han at xiaochuang.han@gmail.com :)