fix(llama-3-1): add trailing new line

[matoushavlena]

Description

It seems like the official llama 3.1 implementation ends the prompt with \n\n before passing it for inference. Currently only single \n is passed. This has an effect on the inference output.

Supporting evidence:

• https://github.com/meta-llama/llama-models/blob/659a13ae2099748d23088fbf9293521de569dd8c/models/llama3/api/chat_format.py#L29
• https://dataplatform.cloud.ibm.com/docs/content/wsj/analyze-data/fm-prompt-samples.html?context=wx#sample7a0
• https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct/blob/main/tokenizer_config.json