When toknization with prompt_prefix with \n in it, it is not tokenized correctly with llama tokenizer. Somehow they become \n\n in the tokenization.
I add this --prompt_prefix "\n<|user|>\n" --prompt_postfix "</s>\n<|assistant|>\n" in my script, but the decoded data looks like this
</s> \\n<|assistant|>\\n
When toknization with prompt_prefix with \n in it, it is not tokenized correctly with llama tokenizer. Somehow they become \n\n in the tokenization.
I add this
--prompt_prefix "\n<|user|>\n" --prompt_postfix "</s>\n<|assistant|>\n"in my script, but the decoded data looks like this</s> \\n<|assistant|>\\n