NousResearch

/

Llama-2-7b-chat-hf

like 146

If you are asking for the format of the prompts for fine-tuning, then from:
https://huggingface.co/blog/llama2

<s>[INST] <<SYS>>
{{ system_prompt }}
<</SYS>>

{{ user_message }} [/INST]

I don't see anything for the assistant response though, so maybe something like this(?)

<s>[INST] <<SYS>>
{{ system_prompt }}
<</SYS>>

{{ user_message }} [/INST]{{ assistant_response }}

That or with a \n inbetween

Can I get the dataset in CSV format to fine-tune the model?"

The dataset was never released by Meta. So, unfortunately, no

Getting a weird error: "The model did not return a loss from the inputs, only the following keys: logits. " ... Any ideas on what I can do to fix this?